02.09.2015 Views

Thèse d'Habilitation à Diriger les Recherches Université Pierre et ...

Thèse d'Habilitation à Diriger les Recherches Université Pierre et ...

Thèse d'Habilitation à Diriger les Recherches Université Pierre et ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Thèse d’Habilitation à <strong>Diriger</strong> <strong>les</strong> <strong>Recherches</strong><br />

Université <strong>Pierre</strong> <strong>et</strong> Marie Curie — Paris VI<br />

Spécialité : Physique<br />

présentée par<br />

Silvère AKAMATSU<br />

pour obtenir l’Habilitation à <strong>Diriger</strong> <strong>les</strong> <strong>Recherches</strong><br />

de l’Université <strong>Pierre</strong>-<strong>et</strong>-Marie Curie (Paris VI)<br />

Etude expérimentale par observation en temps réel de la<br />

dynamique non-linéaire des fronts de solidification<br />

directionnelle d’alliages transparents<br />

Soutenue le 8 février 2008 devant le jury composé de<br />

Mme Martine BEN AMAR Présidente<br />

M. Yves BRECHET Rapporteur<br />

M. Yves COUDER Examinateur<br />

M. Heiner MUELLER-KRUMBHAAR Rapporteur<br />

M. Alain POCHEAU Examinateur<br />

M. Michel RAPPAZ Rapporteur


CURRICULUM VITAE<br />

M. Silvère AKAMATSU<br />

Institut des Nanosciences de Paris - CNRS UMR 7588<br />

Universités <strong>Pierre</strong> <strong>et</strong> Marie Curie (Paris VI) <strong>et</strong> Denis Diderot (Paris VII)<br />

Campus Boucicaut, 140 rue de Lourmel, 75015 Paris.<br />

Né le 26 juill<strong>et</strong> 1964 (43 ans)<br />

Marié, 3 enfants<br />

Diplômes Universitaires<br />

- Diplôme d’ingénieur de l’Ecole Centrale de Lyon (1984-1987)<br />

- DEA de Science des Matériaux (Université Lyon I ; 1987)<br />

- Doctorat de l’Université <strong>Pierre</strong> <strong>et</strong> Marie Curie (Paris VI) - Spécialité Chimie Physique<br />

- "Etude par microscopie de fluorescence des transitions de phases du premier<br />

ordre dans <strong>les</strong> monocouches de Langmuir d’acides gras <strong>et</strong> d’acyl amino acides" (janvier<br />

1992), sous la direction de F. Rondelez.<br />

Emplois <strong>et</strong> Fonctions<br />

- Septembre 1987 - janvier 1992 (thèse) : Bourse BDI cofinancée CNRS-ELF<br />

Interruption d’un an (décembre 1989-novembre 1990) pour le Service National<br />

- Mars 1992 - novembre 1992 : Stage post-doctoral au Laboratoire Léon Brillouin,<br />

CE Saclay, CEA.<br />

- Depuis décembre 1992 : Chargé de Recherche au CNRS au Groupe de Physique<br />

des Solides, puis à l’INSP<br />

CR2 du 1er décembre 1992 au 30 novembre 1996<br />

CR1 depuis le 1er décembre 1996<br />

Responsabilités<br />

Responsable de l’équipe "Surfaces, Interfaces, Croissance" du GPS (jusqu’en 2004).<br />

Co-organisateur du séminaire général de l’INSP (2004-2006)<br />

Encadrements de Recherche<br />

Stages de DEA <strong>et</strong> M2 :<br />

Physique des Solides : L. Pauchard (1993), C. Couvert (1994), S. Deudé (1995)<br />

Physique des Liquides : M. Ginibre (1993), G. Bout<strong>et</strong> (1996), S. Moulin<strong>et</strong> (1999),<br />

G. Guéna (2003), C. Defrenne (2007)<br />

Physique Théorique : N. Sator (1997), P. Tran Dinh (2006)<br />

Surfaces,... (Lyon I) : M. Perrut (2004)<br />

Thèses de Doctorat (co-direction avec G. Faivre) :<br />

M. Ginibre (Université Paris VI) "Diagramme des morphologies de croissance de l’alliage<br />

eutectique lamellaire CBr4-C2Cl6 en solidification directionnelle d’échantillons<br />

minces", soutenue le 11 décembre 1997.<br />

M. Perrut (Université Paris VI), "Dynamique des fronts de solidification eutectique<br />

2D", soutenue le 19 octobre 2007.


Stages post-doctoraux :<br />

- T. Ihle (Jülich), Simulations numériques de fronts de croissance dendritiques <strong>et</strong><br />

"en algue" (1996) ; co-resp. : C. Caroli, G. Faivre.<br />

- T. Börzsönyi (Budapest), Dynamique de fronts de solidification fac<strong>et</strong>tés (1999-<br />

2002) ; co-resp. : G.Faivre.<br />

- R. Gonzalez-Cinca (Barcelone), Simulation numérique de dynamique de joints de<br />

grains en solidification (2000) ; co-resp. : G.Faivre.<br />

- R. Folch (Barcelone), Simulation numérique de fronts eutectiques lamellaires (2000) ;<br />

co-resp. : G.Faivre, M. Plapp (LPMC, Ecole Polytechnique, Palaiseau).<br />

Eco<strong>les</strong><br />

- Ecole d’été de Physique Théorique "Solides loin de l’équilibre", Beg Rohu, France<br />

(cours de P. Nozières, C. Caroli <strong>et</strong> J. Langer), août 1989.<br />

- "International Summer School on Fundamental Problems in Statistical Mechanics<br />

VIII" Altenberg (Allemagne), 28 juin - 10 juill<strong>et</strong> 1993.<br />

- "Journées grenobloises de formation en physique non linéaire", Aix-<strong>les</strong>-Bains (France),<br />

31 janvier - 3 février 1994.<br />

Séjours à l’étranger, collaborations, contrats<br />

- Séjours (décembre 2002, juill<strong>et</strong>-août 2005, mars 2006, mars 2007) au Dept of Physics,<br />

IPST/IREAP, University of Maryland, College Park (USA), pour collaboration<br />

avec le Pr W. Losert.<br />

- Collaboration avec S. Rex <strong>et</strong> coll., ACCESS (Aix-la-Chapelle, Allemagne) ; proj<strong>et</strong><br />

DIRSOL/SEBA de l’ESA.<br />

- Contrat CNES/ESA, dans le cadre du GDR "Micropesanteur fondamentale <strong>et</strong><br />

appliquée"<br />

- Participation au GDR « Champ de phase »<br />

Principa<strong>les</strong> conférences<br />

- 5th ECIS Conference, Mayence, Allemagne, septembre 1991.<br />

- Congrès Général de Physique de la SFP, Marseille, sept. 1995 (conférencier invité).<br />

- Workshop on Instabilities, Chaos and Fractals in Crystal Growth, Zürich, mars<br />

1996.<br />

- Colloque annuel du Groupe Français de Croissance Cristalline, CRMC2, Marseille,<br />

2-4 avril 1997 (conférencier invité).<br />

- Patterns, non-linear dynamics and stochastic behaviour in spatially extended, complex<br />

systems (PNS’97), Budapest, 23-28 octobre 1997.<br />

- Co-organisateur d’un minicolloque aux Journées de la Matière Condensée de la<br />

SFP, Marseille (2002).<br />

- Matériaux 2002, Tours, octobre 2002. -Gordon Conference on Gravitational effects<br />

in physico-chemical systems, New London, Connecticut, USA, juill<strong>et</strong> 2003.<br />

- Journées de la Matière Condensée de la SFP, Nancy (2004).<br />

- Congrès annuel APS, Baltimore (mars 2006)<br />

- Congrès annuel TMS (invité), Orlando (février 2007)


Publications<br />

[1] B.G. Moore, C.M. Knobler, S. Akamatsu, F. Rondelez, Phase diagram of Langmuir<br />

monolayers of pentadecanoic acid : quantitative comparison of surface pressure<br />

and fluorescence microscopy results, J. Phys. Chem., 94, 4588 (1990).<br />

[2] J. Lucassen, S. Akamatsu, F. Rondelez, Formation, evolution and rheology of twodimensional<br />

foams in spread monolayers at the air-water interface, J. Coll. Interface<br />

Sci., 144, 434 (1991).<br />

[3] S. Akamatsu, F. Rondelez, Fluorescence microscopy evidence for two different LE-<br />

LC phase transitions in Langmuir monolayers of fatty acids, J. Physique II (Paris),<br />

1, 1309 (1991).<br />

[4] S. Akamatsu, F. Rondelez, Two-dimensional pattern formation in Langmuir monolayers,<br />

in Progress in Colloid and Polymer Science, 89, Steinkopff Verlag, Darmstadt,<br />

RFA (1992).<br />

[5] S. Akamatsu, O. Bouloussa, K. To, F. Rondelez, Two-dimensional dendritic<br />

growth in Langmuir monolayers of D-Myristoyl Alanine, Phys. Rev. A, 46, R4504<br />

(1992).<br />

[6] K. To, S. Akamatsu, F. Rondelez, Stripe phase in the Gas-Liquid coexistence<br />

region of Langmuir monolayers, Europhys. L<strong>et</strong>t., 21, 343 (1993).<br />

[7] S. Akamatsu, M. Ginibre, G. Faivre, Observation expérimentale de la morphologie<br />

"en algue" <strong>et</strong> d’états asymétriques appariés en solidification directionnelle de films<br />

minces d’alliages CBr4-C2Cl6, Ann. Phys. Fr., 19, 645 (1994).<br />

[8] S. Akamatsu, G. Faivre, T. Ihle, Symm<strong>et</strong>ry-broken fingers and seaweed patterns<br />

in thin-film directional solidification of a non-fac<strong>et</strong>ed cubic crystal, Phys. Rev. E,<br />

51, 4751 (1995).<br />

[9] S. Akamatsu <strong>et</strong> G. Faivre, Residual-impurity effects in directional solidification :<br />

long-lasting recoil of the front and nucleation-growth of gas bubb<strong>les</strong>, J. Phys. I France,<br />

6, 503 (1996).<br />

[10] M. Ginibre, S. Akamatsu, G. Faivre, Experimental d<strong>et</strong>ermination of the stability<br />

diagram of a lamellar eutectic growth front, Phys Rev E, 56, 780 (1997).<br />

[11] C. Counillon, L. Daud<strong>et</strong>, T. Podgorski, M.-C. Jullien, S. Akamatsu, L. Limat,<br />

Global drift of a circular array of liquid columns, Europhys. L<strong>et</strong>t., 40, 37-42 (1997).<br />

[12] S. Akamatsu, T. Ihle, Similarity law for the tilt angle of dendrites in directional<br />

solidification of non-axially-oriented crystals, Phys. Rev. E, 56, 4479 (1997).<br />

[13] S. Akamatsu, G. Faivre, Morphology diagrams in directional solidification of<br />

binary alloys, Anna<strong>les</strong> de Physique, Colloque C2, 22, 39 (1997).<br />

[14] S. Akamatsu, G. Faivre, Anisotropy-driven dynamics of cellular fronts in directional<br />

solidification in thin samp<strong>les</strong>, Phys. Rev. E, 58, 3302 (1998).<br />

[15] G. Faivre, S. Akamatsu, Thin-sample directional solidification of transparent<br />

alloys, Workshop on Solidification, Zermatt, Juill<strong>et</strong> 1998 (CD-rom)<br />

[16] S. Akamatsu, G. Faivre, Traveling waves, two-phase fingers, and eutectic colonies<br />

in thin-sample directional solidification of a ternary eutectic alloy, Phys. Rev. E, 61,<br />

3757 (2000).<br />

[17] S. Akamatsu, S. Moulin<strong>et</strong>, G. Faivre, The formation of Lamellar-eutectic grains<br />

in thin samp<strong>les</strong>, M<strong>et</strong>. Mater. Trans A, 32A, 2039 (2001).


[18] S. Akamatsu, S. Bottin-Rousseau, G. Faivre, La dynamique de solidification des<br />

eutectiques lamellaires : des échantillons minces aux systèmes massifs, J. Phys. IV<br />

France, 11 (2001).<br />

[19] T. Börzsönyi, S. Akamatsu, G. Faivre, Dynamics of a fac<strong>et</strong>ed nematic-smectic<br />

B front in thin-sample directional solidification, Phys. Rev. E, 65, 011702 (2002).<br />

[20] S. Bottin-Rousseau, S. Akamatsu, G. Faivre, Formation of grain subboundaries<br />

during directional solidification below the cellular-bifurcation threshold, Phys. Rev.<br />

B, 66, 054102 (2002).<br />

[21] T. Börzsönyi, S. Akamatsu, Surface effects in nucleation and growth of smectic-B<br />

crystals in thin samp<strong>les</strong>, Phys. Rev. E, 66, 051709 (2002).<br />

[22] S. Akamatsu, M. Plapp, G. Faivre, A. Karma, Pattern Stability and Trijunction<br />

Motion in Eutectic Solidification, Phys. Phys. Rev. E, 66, 30501 (2002).<br />

[23] T. Börzsönyi, S. Akamatsu, G. Faivre, Dynamics of a fac<strong>et</strong>ed nematic-smectic B<br />

front in thin-sample directional solidification, in Interface and Transport Dynamics,<br />

ed. H. Emmerich, B. Nestler, <strong>et</strong> M. Schreckenberg ; Lecture Notes in Computational<br />

Science and Engineering Vol. 32, Springer, 2003.<br />

[24] S. Akamatsu, M. Plapp, G. Faivre, A. Karma, Overstability of lamellar eutectic<br />

growth below the minimum-undercooling spacing, M<strong>et</strong>al. Mater. Trans. A, 35, 1815<br />

(2004).<br />

[25] S. Akamatsu, S. Bottin-Rousseau, G. Faivre, Experimental evidence for a zigzag<br />

bifurcation in bulk lamellar eutectic growth, Phys. Rev. L<strong>et</strong>ters, 93, 175701 (2004).<br />

[26] S. Akamatsu, S. Bottin-Rousseau, G. Faivre, Stability of lamellar eutectic growth<br />

in thick samp<strong>les</strong>, Phil. Mag., 86, 3703 (2006).<br />

[27] S. Akamatsu, K. Y. Lee, W. Losert, Control of eutectic solidification microstructures<br />

through laser spot perturbations, J. Cryst. Growth, 289, 331 (2006).<br />

[28] A. Parisi, M. Plapp, S. Akamatsu, S. Bottin-Rousseau, M. Perrut, G. Faivre,<br />

Three-dimensional phase-field simulations of eutectic solidification and comparison<br />

to in situ experimental observations, in "Modeling of Casting, Welding, and Advanced<br />

Solidification Processes - XI", pp. 417-424, edited by C.-A. Gandin and M.<br />

Bell<strong>et</strong>, The Minerals, M<strong>et</strong>al and Materials Soci<strong>et</strong>y, Warrendale, PA (2006).<br />

[29] A. J. Pons, A. Karma, S. Akamatsu, M. Newey, A. Pomerance, H. Singer <strong>et</strong> W.<br />

Losert, Feedback control of unstable cellular solidification fronts, Phys. Rev. E , 75,<br />

021602 (2007).<br />

[30] S. Akamatsu, S. Bottin-Rousseau, M. Perrut, G. Faivre, V.T. Witusiewicz, L.<br />

Sturz , Real-time study of thin and bulk eutectic growth in Succinonitrile-(D)Camphor<br />

alloys, J. Cryst. Growth, 299, 418 (2007).<br />

[31] S. Bottin-Rousseau, M. Perrut, C. Picard, S. Akamatsu, G. Faivre, An experimental<br />

m<strong>et</strong>hod for the in situ observation of eutectic growth patterns in bulk samp<strong>les</strong><br />

of transparent alloys, J. Cryst. Growth, 306, 465 (2007).


Résumé<br />

Nous présentons une étude expérimentale <strong>et</strong> fondamentale de la dynamique nonlinéaire<br />

des fronts de solidification directionnelle en échantillons minces (géométrie<br />

2D) d’alliages transparents non-fac<strong>et</strong>tés (en particulier CBr 4 -C 2 Cl 6 ), observées en<br />

temps réel. Des microstructures stationnaires périodiques à l’échelle du micron, typiques<br />

d’un système étendu maintenu hors d’équilibre, apparaissent au front de<br />

croissance cristalline à partir d’instabilités de l’interface solide-liquide gouvernées<br />

par la diffusion, restabilisées par la capillarité. Dans <strong>les</strong> alliages dilués, nos observations<br />

m<strong>et</strong>tent au clair l’influence de l’anisotropie interfaciale sur la stabilité des<br />

fronts cellulaires <strong>et</strong> dendritiques. En faisant varier l’anisotropie effective (2D) en<br />

fonction de l’orientation du cristal dans <strong>les</strong> échantillons minces, nous avons découvert<br />

de nouvel<strong>les</strong> transitions morphologiques vers des structures non-dendritiques,<br />

en particulier la structure "en algue" <strong>et</strong> le doublon à anisotropie interfaciale nulle.<br />

Nous étudions aussi un mécanisme de polygonisation dynamique (production de<br />

sous-joints de grains en cours de solidification). Dans <strong>les</strong> alliages eutectiques, nous<br />

dressons un diagramme des morphologies 2D étendues homogènes (bifurcations par<br />

brisure de symétrie) <strong>et</strong> localisées. Ces résultats sont appuyés par des simulations<br />

numériques quantitatives (collaborations). Nos études à long terme concernent 1-<br />

l’observation directe de la dynamique des structures de solidification eutectique lamellaires<br />

<strong>et</strong> fibreuses d’alliages transparents en échantillons massifs ; 2-la mesure<br />

de coefficients d’anisotropie <strong>et</strong> l’observation de transitions morphologiques dans des<br />

cristaux orientés d’alliages métalliques en échantillons minces ; 3-la solidification<br />

d’alliages fac<strong>et</strong>tés ; 4- la maîtrise des microstructures par micromanipulation.<br />

Summary<br />

We present an experimental and fundamental study of the nonlinear dynamics of<br />

directional solidification fronts in thin samp<strong>les</strong> (2D geom<strong>et</strong>ry) of nonfac<strong>et</strong>ed transparent<br />

alloys (in particular CBr 4 -C 2 Cl 6 ), by real-time observation. Steady, periodic<br />

structures (on a micron scale) of the crystal growth front, typical of a pattern formation<br />

in a spatially extended system maintained out of equilibrium, arise from<br />

diffusion controled morphological instabilities of the solid-liquid interface restabilized<br />

by the capillarity. In dilute alloys, we cast light on the influence of the interfacial<br />

anisotropy on the stability of cellular and dendritic structures. We find new morphologies,<br />

in particular the "seaweed" structure and the doublon for a vanishing<br />

anisotropy, by changing the orientation of the crystal, thus the effective (2D) anisotropy,<br />

in a thin sample. We also study a mechanism of formation of subboundaries<br />

in single crystals. In eutectic alloys, we could draw a morphology diagram of (2D)<br />

homogeneous, extended lamellar microstructures (symm<strong>et</strong>ry breaking bifurcations)<br />

and of various localized dynamical objects. These results are supported by quantitative<br />

numerical simulations (collaborations). Our long-term research studies will<br />

concern 1- the direct observation of the dynamics of lamellar and fibrous eutectic<br />

solidification structures of transparentalloys in semi-bulk samp<strong>les</strong> by in situ ; 2- the<br />

measurement of anisotropy coefficients and the direct observation of microstructure<br />

transitions in crystals of known orientation of m<strong>et</strong>allic alloys in thin samp<strong>les</strong> ; 3- the<br />

solidification of fac<strong>et</strong>ed alloys ; and 4- the feedback control of solidification microstructures<br />

with a new micromanipulation m<strong>et</strong>hod.


Remerciements<br />

Mes principaux remerciements vont à Gabriel Faivre, de qui j’ai appris la pratique<br />

d’une recherche scientifique rigoureuse, profonde <strong>et</strong> fertile.<br />

Je remercie chaleureusement Christiane Caroli, à qui je dois beaucoup.<br />

Je remercie Camille Cohen, ancien directeur du GPS, qui m’a accueilli comme<br />

jeune chercheur, ses successeurs Jean Klein <strong>et</strong> Claud<strong>et</strong>te Lapersonne, ainsi que Claudine<br />

Noguéra, directrice de l’INSP.<br />

Je remercie <strong>les</strong> membres de mon jury,en particulier mes rapporteurs.<br />

Que mes collaborateurs, parmi <strong>les</strong>quels, en premier lieu, Sabine Bottin-Rousseau,<br />

<strong>et</strong>, dans un grand désordre, M. Plapp, T. Börzsönyi, A. Karma, S. Rex, L. Sturz,<br />

V. Witusievicz, W. Losert, T. Ihle, L. Limat, M. Ginibre, M. Perrut, C. Picard, A.<br />

Fleury, F. Br<strong>et</strong>on <strong>et</strong> C. Bourgeois, soient assurés de ma reconnaissance.<br />

Je remercie B. Zappoli, pour le soutien que le CNES apporte depuis de nombreuses<br />

années à mes travaux.


Table des matières<br />

1 Introduction générale 10<br />

2 Eléments théoriques <strong>et</strong> méthodes expérimenta<strong>les</strong> 15<br />

2.1 Solidification directionnelle d’alliages non-fac<strong>et</strong>tés en géométrie 2D . . 15<br />

2.1.1 Alliages non-fac<strong>et</strong>tés . . . . . . . . . . . . . . . . . . . . . . . 15<br />

2.1.2 Géométrie 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . 17<br />

2.1.3 Equations de la solidification (alliages dilués) . . . . . . . . . . 17<br />

2.1.4 Front plan stationnaire <strong>et</strong> instabilité cellulaire . . . . . . . . . 18<br />

2.2 Méthodes expérimenta<strong>les</strong> . . . . . . . . . . . . . . . . . . . . . . . . . 20<br />

3 Dynamique des fronts de solidification d’alliages dilués en géométrie<br />

2D 22<br />

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22<br />

3.2 Fronts dendritiques, "structure en algue" <strong>et</strong> transitions morphologiques 24<br />

3.2.1 Monocristaux orientés . . . . . . . . . . . . . . . . . . . . . . 24<br />

3.2.2 Anisotropie interfaciale . . . . . . . . . . . . . . . . . . . . . . 26<br />

3.2.3 Principaux types de morphologies . . . . . . . . . . . . . . . . 29<br />

3.2.4 Morphologies dendritiques . . . . . . . . . . . . . . . . . . . . 29<br />

3.2.5 Estimation des coefficients d’anisotropie . . . . . . . . . . . . 34<br />

3.2.6 Doublon <strong>et</strong> structure en algue . . . . . . . . . . . . . . . . . . 35<br />

3.2.7 Diagrammes de morphologies – Questions ouvertes . . . . . . 40<br />

3.3 Influence de l’anisotropie interfaciale sur la stabilité des fronts cellulaires 42<br />

3.3.1 Mesure du diagramme de bifurcation cellulaire . . . . . . . . . 42<br />

3.3.2 Stabilité des fronts cellulaires . . . . . . . . . . . . . . . . . . 43<br />

3.3.3 Questions en suspens . . . . . . . . . . . . . . . . . . . . . . . 47<br />

3.4 Polygonisation dynamique en cours de croissance . . . . . . . . . . . 48<br />

3.4.1 Position du problème . . . . . . . . . . . . . . . . . . . . . . . 48<br />

3.4.2 Joints <strong>et</strong> sous-joints de grains . . . . . . . . . . . . . . . . . . 49<br />

3.4.3 Processus de création de sous-joints de grains . . . . . . . . . 50<br />

3.4.4 Dynamique spatio-temporelle de la polygonisation . . . . . . . 52<br />

3.4.5 Questions ouvertes . . . . . . . . . . . . . . . . . . . . . . . . 52<br />

3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54<br />

4 Fronts de solidification eutectique lamellaires en géométrie 2D 55<br />

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55<br />

4.2 Bases théoriques - Calcul de Jackson <strong>et</strong> Hunt . . . . . . . . . . . . . . 57<br />

8


4.3 Diffusion de la phase <strong>et</strong> élimination de lamel<strong>les</strong> . . . . . . . . . . . . . 59<br />

4.3.1 Mesure expérimentale de la courbe de surfusion . . . . . . . . 59<br />

4.3.2 Diffusion de la phase - Instabilité d’Eckhaus . . . . . . . . . . 61<br />

4.3.3 Elimination de lamel<strong>les</strong> . . . . . . . . . . . . . . . . . . . . . . 63<br />

4.4 Instabilités secondaires des fronts eutectiques lamellaires . . . . . . . 65<br />

4.4.1 Diagramme des morphologies . . . . . . . . . . . . . . . . . . 65<br />

4.4.2 Défauts dynamiques . . . . . . . . . . . . . . . . . . . . . . . 67<br />

4.5 Conclusion <strong>et</strong> questions ouvertes . . . . . . . . . . . . . . . . . . . . . 68<br />

5 Etudes en cours - Proj<strong>et</strong>s de recherche 71<br />

5.1 Observation directe des fronts de solidification eutectique en échantillons<br />

massifs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71<br />

5.2 Solidification d’alliages métalliques en échantillons minces . . . . . . . 76<br />

5.3 Alliages fac<strong>et</strong>tés . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78<br />

5.4 Maîtrise des microstructures de solidification par micromanipulation . 79<br />

6 Conclusion 82<br />

9


Chapitre 1<br />

Introduction générale<br />

La formation des microstructures de solidification est le résultat d’un phénomène<br />

de structuration non-linéaire d’un système étendu maintenu hors d’équilibre<br />

(out-of-equilibrium pattern formation) [1, 2, 3, 4, 5]. La solidification –croissance<br />

d’un solide cristallin à partir d’un liquide– se produit par propagation de l’interface<br />

solide-liquide. Elle s’accompagne, dans <strong>les</strong> mélanges, du rej<strong>et</strong> de chaleur <strong>et</strong> d’espèces<br />

chimiques, dont le transport est assuré par la diffusion. L’alliage ainsi produit<br />

n’est généralement pas homogène : on y observe, sur une grande gamme d’échel<strong>les</strong><br />

souvent proche du micron, des modulations de composition, une dispersion de différentes<br />

phases cristallines, une structure de grains ou des arrangements de défauts<br />

de réseau [6, 7, 8]. Ces microstructures sont la trace, figée dans le solide, d’instabilités<br />

morphologiques du front de solidification, c’est-à-dire de modulations, à la<br />

même échelle, de la forme de l’interface solide-liquide <strong>et</strong> du champ de concentration<br />

associé. Dans <strong>les</strong> alliages non-fac<strong>et</strong>tés (<strong>les</strong> métaux par exemple), quand l’interface<br />

reste proche de l’équilibre local, ces instabilités sont gouvernées par le couplage entre<br />

<strong>les</strong> gradients de diffusion <strong>et</strong> la vitesse de propagation de l’interface [9, 10, 11, 12].<br />

Les microstructures se restabilisent sous l’eff<strong>et</strong> des forces capillaires. De c<strong>et</strong>te compétition<br />

dynamique entre diffusion <strong>et</strong> capillarité résultent des formes de croissance<br />

variées dont la taille n’est fixée qu’en ordre de grandeur. Expérimentalement, la distribution<br />

spatiale des microstructures varie en fonction des conditions aux limites<br />

<strong>et</strong> de l’histoire de l’expérience. C<strong>et</strong>te "multistabilité" est une caractéristique fondamentale<br />

d’un problème non-linéaire à frontière libre, dont <strong>les</strong> solutions n’obéissent<br />

à aucun principe de minimisation.<br />

Nous présentons ici une revue expérimentale de la formation des microstructures<br />

de solidification par une méthode de laboratoire, la solidification directionnelle en<br />

échantillons minces d’alliages transparents, ou SDEM (Fig. 1.1). C<strong>et</strong>te méthode, inventée<br />

par Jackson <strong>et</strong> Hunt dans <strong>les</strong> années 1960 [13, 14], a été reprise par différentes<br />

équipes (notamment cel<strong>les</strong> de W. Kurz [15], R. Trivedi [16], H. Cummins [17], A. Libchaber<br />

[18], C. Guthmann [19], G. Faivre [20], P. Oswald [21] <strong>et</strong> A. Pocheau [22]) à<br />

partir des années 1980. Sous un éclairage nouveau, celui de la physique non-linéaire,<br />

<strong>les</strong> observations obtenues en SDEM apportent des réponses souvent inattendues à<br />

des questions d’intérêt fondamental, parfois anciennes, mais non résolues jusque là.<br />

Notre contribution dans ce domaine concerne deux types de microstructures de solidification,<br />

cellulaires <strong>et</strong> dendritiques d’une part, <strong>et</strong> eutectiques lamellaires d’autre<br />

10


part. Deux grands thèmes seront considérés : la dynamique non-linéaire des fronts<br />

périodiques de grande extension, <strong>et</strong> l’influence de l’anisotropie interfaciale sur la<br />

stabilité des microstructures. Des questions plus spécifiques, mais importantes en<br />

métallurgie, seront aussi abordées ici avec plus de précision. L’accent sera mis sur la<br />

découverte de nouvel<strong>les</strong> morphologies de croissance cristalline, <strong>et</strong> sur la mesure de<br />

quantités physiques diffici<strong>les</strong> d’accès, tel<strong>les</strong> que <strong>les</strong> propriétés des interfaces solideliquide<br />

<strong>et</strong> leurs anisotropies. L’intérêt propre des échantillons minces sera souligné.<br />

Le succès d’études expérimenta<strong>les</strong> <strong>et</strong> numériques menées en commun, <strong>et</strong> aboutissant<br />

à des conclusions en accord (semi) quantitatif, sera aussi mis en valeur. Le ssimulations<br />

numériques ont été réalisées par des méthodes de "suivi de front" (front<br />

tracking) <strong>et</strong> de "champ de phase" (phase field model) [23, 24, 25, 26]– réalisées par<br />

des collaborateurs théoriciens-numériciens (entre autres, H. Müller-Krumbhaar, T.<br />

Ihle, C. Misbah, K. Kassner, A. Karma, <strong>et</strong> M. Plapp) Nous ferons enfin une part<br />

importante à nos proj<strong>et</strong>s de recherche, parmi <strong>les</strong>quels l’observation directe de la formation<br />

des microstructures de solidification en échantillons massifs témoigne d’une<br />

complexité théorique <strong>et</strong> expérimentale croissante.<br />

Fig.1.1 – Solidification directionnelle en échantillons minces (SDEM). L’axe x est parallèle aux<br />

isothermes <strong>et</strong> perpendiculaire à z (axe du gradient thermique) <strong>et</strong> à y (axe transverse). V : vitesse de<br />

translation de l’échantillon selon z. Le gradient de température s’établit dans une fenêtre (≈ 1 cm)<br />

à bords plans entre <strong>les</strong> blocs froid <strong>et</strong> chaud. Observation : vidéomicroscopie en lumière transmise.<br />

En solidification directionnelle, on fait croître le cristal à vitesse constante V<br />

dans la direction z d’un gradient de température uniaxe G fixe. Dans <strong>les</strong> régimes<br />

stationnaires, qui s’établisent à l’issue de courts transitoires, le front de solidification<br />

se propage à la vitesse V par rapport au liquide –il est immobile dans le repère du laboratoire.<br />

La SDEM perm<strong>et</strong> de visualiser la forme de l’interface solide-liquide in situ<br />

<strong>et</strong> de suivre en temps réel l’évolution du front de solidification, sur une large gamme<br />

d’échel<strong>les</strong> caractéristiques temporel<strong>les</strong> <strong>et</strong> spatia<strong>les</strong>. En échantillon mince (d’épaisseur<br />

typique 10 µm), la dynamique de solidification se déroule en géométrie (quasi) bidimensionnelle<br />

(2D). La convection dans le liquide est bloquée. Le gradient thermique<br />

est constant : la chaleur latente s’évacue rapidement dans <strong>les</strong> parois <strong>et</strong> la diffusion<br />

du soluté domine la dynamique. Le front de solidification est plan à grande échelle,<br />

mais se structure dans une direction, à une (p<strong>et</strong>ite) échelle (entre 1 <strong>et</strong> 100 µm) qui<br />

dépend de l’alliage, de V (


Fig.1.2 – Structures de solidification périodiques stationnaires (SDEM ; alliages CBr 4 -C 2 Cl 6 ).<br />

De gauche à droite : structures cellulaire, dendritique <strong>et</strong> eutectique lamellaire. Le liquide est en<br />

haut. Le front de solidification progresse vers le haut. Le solide <strong>et</strong> le liquide sont transparents, mais<br />

d’indices optiques différents. Les interfaces apparaissent comme des lignes noires. Ces micrographies<br />

montrent une partie de l’échantillon, de largeur totale proche de 1 cm. Ces caractéristiques valent,<br />

sauf mention particulière, pour <strong>les</strong> micrographies suivantes.<br />

Les questions que pose la formation des structures de solidification sont extrêmement<br />

diverses. D’une manière générale, la résolution des équations de la solidification<br />

montre l’existence de microstructures périodiques stationnaires (cellulaires,<br />

dendrites ou eutectiques lamellaires, par exemple), correspondant à des branches de<br />

solutions s’étendant sur un continuum de la période λ. En revanche, l’expérience<br />

montre que λ reste proche, en moyenne, <strong>et</strong> très approximativement, de valeurs remarquab<strong>les</strong>.<br />

Ceci pose un problème général de "sélection", formulé dans le passé<br />

sous la forme d’une conjecture empirique, mais sans support théorique rigoureux,<br />

selon laquelle il existe des mécanismes généraux menant à une microstructure de<br />

forme <strong>et</strong> de taille uniques, à paramètres de contrôle fixés. Les développements récents<br />

perm<strong>et</strong>tent en fait d’invalider c<strong>et</strong>te hypothèse.<br />

En géométrie 2D, <strong>les</strong> structures de solidification stationnaires adm<strong>et</strong>tent un large<br />

domaine de stabilité, borné par des instabilités morphologiques provoquées par de<br />

faib<strong>les</strong> variations des paramètres de contrôle. De nombreux aspects de la dynamique<br />

des fronts de solidification en découlent. En solidification d’un alliage binaire isotrope,<br />

à cinétique interfaciale rapide, sans convection, sans eff<strong>et</strong>s de bords (système<br />

infini) –ce qui peut définir un système "idéal"– on observe une phénoménologie d’instabilités<br />

dont l’analyse générale repose sur des fondements bien établis sur un plan<br />

théorique (voir <strong>les</strong> Réfs. [2, 27, 28, 29]). Les références expérimenta<strong>les</strong> sont nombreuses,<br />

en particulier en convection hydrodynamique entr<strong>et</strong>enue dans le champ de<br />

pesanteur (de symétrie axiale, comme la solidification directionnelle) –par exemple<br />

<strong>les</strong> instabilités de Rayleigh-Bénard, de Faraday, de digitation dirigée ou des colonnes<br />

liquides [30, 31, 32, 33]. C<strong>et</strong>te analyse conduit à classer <strong>les</strong> différents types d’instabilités<br />

des structures périodiques en deux grandes catégories : d’une part des instabilités<br />

"secondaires" par brisure de symétrie, <strong>et</strong>, d’autre part, des instabilités "de<br />

phase". Les premières peuvent donner lieu à bifurcation, le système se restabilisant<br />

alors en une nouvelle structure stationnaire (ou permanente), dérivante ou oscillante.<br />

El<strong>les</strong> sont décrites par des "équations d’amplitude", dont <strong>les</strong> termes peuvent<br />

être déduits par des arguments de symétrie. Ces équations prédisent aussi des phénomènes<br />

spatio-temporels complexes, loin d’un seuil de bifurcation (parois dynamiques<br />

[18, 20, 34] <strong>et</strong> ondes solitaires [29]). En revanche, l’instabilité associée à la "diffusion<br />

de la phase", dite instabilité d’Eckhaus, ne conduit pas, elle, à la stabilisation d’une<br />

nouvelle structure (elle est, par exemple, responsable de l’élimination de lamel<strong>les</strong> à<br />

la limite de stabilité inférieure des fronts eutectiques lamellaires).<br />

12


En pratique, <strong>les</strong> conditions expérimenta<strong>les</strong> peuvent s’écarter sensiblement de la<br />

situation idéale. Dans ce cas, on ne peut plus négliger l’eff<strong>et</strong> de certaines "perturbations"<br />

sur la dynamique de solidification, comme nous l’avons tacitement fait jusqu’ici.<br />

L’anisotropie interfaciale, c’est-à-dire celle des propriétés de l’interface solideliquide<br />

(tension de surface, cinétique interfaciale), d’origine cristallographique, peut<br />

être vue comme une de ces perturbation –sans doute la principale. Dans certains<br />

cas (eutectiques lamellaires), c<strong>et</strong>te anisotropie, qui est de très faible amplitude pour<br />

<strong>les</strong> cristaux non-fac<strong>et</strong>tés, ne fait que biaiser légèrement la dynamique des microstructures<br />

de solidification, qui reste proche d’une dynamique idéale. Elle tient, en<br />

revanche, un rôle décisif en croissance dendritique, ce que nous développerons plus<br />

loin. Par ailleurs, l’étude des imperfections d’origine instrumentale nous a également<br />

amené à des découvertes surprenantes –par exemple celle des "eutectiques gazeux"<br />

lors de l’étude de l’eff<strong>et</strong> des impur<strong>et</strong>és (gaz) résiduel<strong>les</strong> (voir [36] <strong>et</strong> <strong>les</strong> références<br />

incluses) –mais elle sort du contexte de ce mémoire.<br />

Le plan du mémoire est le suivant. Dans le Chapitre 2, nous préciserons succintement<br />

le cadre théorique de nos études. Nous présenterons <strong>les</strong> méthodes expérimenta<strong>les</strong>,<br />

sans entrer dans le détail des progrès réalisés (purification des matériaux, automatisation<br />

des microdéplacements, acquisition <strong>et</strong> traitement d’image numérique).<br />

Le suj<strong>et</strong> principal du Chapitre 3 (Réfs. [35] à [38]) est l’étude des transitions morphologiques<br />

des fronts de solidification directionnelle d’alliages CBr 4 -C 2 Cl 6 dilués<br />

au-dessus du seuil cellulaire V c en fonction de l’anisotropie interfaciale. Ce travail<br />

m<strong>et</strong> à profit la possibilité qu’offrent <strong>les</strong> échantillons minces de faire varier l’anisotropie<br />

interfaciale effective dans le plan de l’expérience, indépendamment des autres<br />

paramètres, en fonction de l’orientation du monocristal. Nous nous intéresserons<br />

aux morphologies de solidification à forte vitesse (§ 3.2), puis à la stabilité des fronts<br />

cellulaires (§ 3.3). Le lien de ces morphologies avec l’anisotropie interfaciale est<br />

bien établi théoriquement, mais mal étayé expérimentalement parce que l’anisotropie<br />

interfaciale est une quantité très difficile à mesurer. Nous montrerons <strong>les</strong> progrès<br />

réalisés, notamment la découverte expérimentale de la "structure en algue" <strong>et</strong> du<br />

"doublon", qui remplace la dendrite à anisotropie interfaciale nulle. La comparaison<br />

semi-quantitative entre observations <strong>et</strong> simulations numériques perm<strong>et</strong> d’estimer<br />

la valeur des coefficients d’anisotropie. Nous mentionnerons enfin une étude, indépendante<br />

des précédentes, d’un phénomène de formation de sous-joints de grains<br />

(polygonisation dynamique) dans des monocristaux (§ 3.4) [39].<br />

Le Chapitre 4 sera consacré à la dynamique des fronts eutectiques lamellaires<br />

(Réf. [40] à [43]). Après une brève présentation, nous m<strong>et</strong>trons en valeur l’étude de la<br />

diffusion de la phase <strong>et</strong> de l’instabilité d’Eckhaus ou d’élimination de lamel<strong>les</strong> (§ 4.3),<br />

puis l’étude complète de la dynamique des instabilités secondaires (§4.4). L’étude<br />

des stades initiaux de la solidification eutectique (formation de grains eutectiques<br />

par invasion-branchement) [45], <strong>et</strong> celle de l’influence d’impur<strong>et</strong>és résiduel<strong>les</strong> en solidification<br />

d’eutectiques binaires (formation de "colonies" ou cellu<strong>les</strong> eutectiques)<br />

[44] sont l’obj<strong>et</strong> de deux artic<strong>les</strong> placés en Annexe.<br />

Les proj<strong>et</strong>s de recherche à court terme, ou en prolongation directe de nos travaux<br />

précédents, seront mentionnés dans le cours des chapitres §3 <strong>et</strong> §4. Les proj<strong>et</strong>s à plus<br />

long terme, <strong>et</strong> <strong>les</strong> études en cours sur <strong>les</strong>quels ils s’appuient –observation in situ des<br />

13


fronts de solidification eutectique en échantillons massifs (Réfs. [46] à [49]), étude<br />

quantitative des eff<strong>et</strong>s d’anisotropie en solidification d’alliages dilués <strong>et</strong> eutectiques<br />

en solidification directionnelle d’alliages métalliques en échantillons minces, étude<br />

de l’interaction entre la croissance fac<strong>et</strong>tée <strong>et</strong> <strong>les</strong> instabilités diffusives [50, 51], <strong>et</strong><br />

maîtrise des microstructures par micromanipulation (en prolongement d’une collaboration<br />

avec W. Losert) [52, 53]– feront l’obj<strong>et</strong> du Chapitre 5, complété par trois<br />

artic<strong>les</strong> en Annexe.<br />

14


Chapitre 2<br />

Eléments théoriques <strong>et</strong> méthodes<br />

expérimenta<strong>les</strong><br />

2.1 Solidification directionnelle d’alliages non-fac<strong>et</strong>tés<br />

en géométrie 2D<br />

2.1.1 Alliages non-fac<strong>et</strong>tés<br />

On peut schématiquement classer <strong>les</strong> cristaux en deux catégories [12, 54, 55, 56] :<br />

<strong>les</strong> cristaux à interface solide-liquide respectivement non-fac<strong>et</strong>tée <strong>et</strong> fac<strong>et</strong>tée. Les<br />

cristaux non-fac<strong>et</strong>tés, auxquels ce mémoire s’intéresse, vérifient deux propriétés :<br />

1- Les cinétiques interfacia<strong>les</strong> sont rapides. La forme communément utilisée de la loi<br />

exprimant la dépendance de la vitesse de propagation v de l’interface solide-liquide<br />

en <strong>les</strong> différences de potentiel chimique des constituants entre <strong>les</strong> deux phases est<br />

l’équation de Gibbs-Thomson, qui s’écrit :<br />

∆T = ∆T sol + ∆T cap + ∆T k (2.1)<br />

où ∆T est la différence entre la température de référence <strong>et</strong> celle de l’interface. Le<br />

terme ∆T sol = m(C i −C 0 ) (m : pente algébrique du liquidus) décrit le diagramme de<br />

phase. Le terme capillaire ∆T cap = a GT κ est proportionnel à la courbure de l’interface<br />

κ <strong>et</strong> le coefficient capillaire (ou de Gibbs-Thomson) a GT est proportionnel à la<br />

tension de surface γ. Le terme cinétique ∆T k mesure, lui, l’écart à l’équilibre (local)<br />

nécessaire à la propagation de l’interface. Au premier ordre (cinétique linéaire), on a<br />

∆T k = βv [57]. Le coefficient de cinétique linéaire β ne dépasse pas, disons, quelques<br />

10 −3 Ksµm −1 , <strong>et</strong> ∆T k est p<strong>et</strong>it devant <strong>les</strong> autres termes. On peut parfois le négliger<br />

(équilibre local à l’interface).<br />

2- Les deux quantités γ <strong>et</strong> β ont une faible anisotropie. El<strong>les</strong> dépendent de l’orientation<br />

de l’interface par rapport au réseau cristallin, en respectant ses symétries,<br />

mais leurs variations ne dépassent pas un seuil déterminé, de l’ordre du pourcent.<br />

La forme d’équilibre d’un cristal non-fac<strong>et</strong>té, déduit de la fonction γ(n), où n est<br />

le vecteur normal à l’interface ("construction de Wulff") [12], s’écarte peu d’une<br />

sphère.<br />

La croissance des cristaux non-fac<strong>et</strong>tés est pratiquement synonyme de croissance<br />

limitée par la diffusion. Le mécanisme élémentaire de l’instabilité de type diffusif<br />

15


est le suivant. A l’équilibre, un solide <strong>et</strong> un liquide en coexistence n’ont pas la<br />

même concentration. En cours de solidification, la chaleur latente <strong>et</strong> <strong>les</strong> espèces<br />

chimiques rej<strong>et</strong>ées par le solide sont évacuées par diffusion dans le liquide (<strong>et</strong> dans<br />

<strong>les</strong> parois, en ce qui concerne la chaleur). Les gradients de soluté permanents qui<br />

se construisent près de l’interface dominent la dynamique. Leur valeur augmente<br />

(en valeur absolue) quand la vitesse de propagation V de l’interface augmente (loi<br />

de conservation). Au-delà d’une vitesse-seuil V sc , dite de surfusion constitutionnelle,<br />

ces gradients sont suffisamment forts pour qu’une partie du liquide, très enrichi en<br />

soluté, soit métastable (en surfusion) par rapport au solide. C’est une condition<br />

nécessaire (mais non suffisante) pour que l’interface se déstabilise par un "eff<strong>et</strong> de<br />

pointe" dynamique (voir §2.1.4).<br />

La croissance gouvernée par la diffusion est celle que l’on observe dans <strong>les</strong> alliages<br />

métalliques (Fig. 2.1) [6]. Beaucoup d’entre eux ont une grande importance<br />

industrielle. Certains alliages binaires (Sn-Pb ; In-Bi) ou multicomposants servent<br />

aussi de systèmes modè<strong>les</strong> en SDEM [59, 60, 61] (voir aussi §6) ou en solidification<br />

en échantillons massifs [58]. Les méthodes récentes, utilisant le rayonnement X<br />

synchrotron, qui perm<strong>et</strong>tent maintenant de visualiser in situ la croissance d’alliages<br />

métalliques [62, 63, 64] restent, bien sûr, assez lourdes à m<strong>et</strong>tre en oeuvre. En dehors<br />

des métaux, on peut citer <strong>les</strong> études (géométrie 3D) de la solidification de gaz rares<br />

[65, 66]. Enfin, <strong>les</strong> observations de Van Suchteleen de SDEM d’alliages eutectiques<br />

de sels en échantillons minces (non publiées), ont servi de précurseur aux études<br />

poursuivies dans notre laboratoire [67].<br />

Fig.2.1 – Microstructures de solidification d’alliages métalliques. De gauche à droite : structures<br />

cellulaire (alliage à base Al), dendritique (alliage à base Co), <strong>et</strong> eutectique lamellaire (Sn-Pb).<br />

Nous avons utilisé principalement l’alliage binaire CBr 4 -C 2 Cl 6 [13, 14, 16, 68, 69],<br />

un des alliages organiques transparents non-fac<strong>et</strong>tés à bas point de fusion (m<strong>et</strong>allic<br />

analogs) <strong>les</strong> plus utilisés en recherche fondamentale –parmi <strong>les</strong>quels <strong>les</strong> alliages à base<br />

de succinonitrile tiennent une place importante (voir, par exemple, <strong>les</strong> Réfs [15, 17,<br />

22, 48, 70, 71, 72]). Ils forment des cristaux moléculaires (phases dites "plastiques")<br />

de haute symétrie (cubique le plus souvent) [73]. Beaucoup d’entre eux ont été<br />

découverts par Jackson [13], mais une récente étude a permis d’en caractériser de<br />

nouveaux [74]. Ils perm<strong>et</strong>tent d’observer en temps réel (<strong>et</strong> non pas post mortem ; Fig.<br />

2.1) une dynamique de solidification représentative, à des rapports d’échelle près, de<br />

la croissance non-fac<strong>et</strong>tée.<br />

16


2.1.2 Géométrie 2D<br />

L’utilisation des échantillons minces (épaisseur e < 50 µm) introduit plusieurs<br />

éléments de simplification, par rapport à la solidification en échantillons massifs :<br />

la convection dans le liquide est bloquée par eff<strong>et</strong> de couche visqueuse aux parois<br />

(sauf en présence d’interfaces fluide-fluide [36]) ; la géométrie est bidimensionnelle<br />

(2D) ; <strong>les</strong> isothermes sont planes <strong>et</strong> alignées selon x ; on peut obtenir, relativement<br />

simplement, de grands monocristaux (§3.2). La géométrie 2D est assurée par un<br />

alignement de l’interface solide-liquide perpendiculairement au plan de l’échantillon<br />

(aux eff<strong>et</strong>s de ménisque près) dû la condition de flux diffusif de soluté nul aux<br />

parois. Des "eff<strong>et</strong>s 3D" sensib<strong>les</strong> ne surviennent que lorsque la taille caractéristique<br />

des microstructures devient plus p<strong>et</strong>ite que e.<br />

Fig.2.2 – Schéma de principe de la solidification directionnelle (échantillons épais <strong>et</strong> minces).<br />

ζ : forme de l’interface. C : champ de concentration. Le champ de température est imposé de<br />

l’extérieur. ˆn : vecteur unitaire normal à l’interface. z : direction de tirage <strong>et</strong> du gradient thermique ;<br />

l’axe x <strong>et</strong> l’axe y (direction transverse) sont dans le plan des isothermes.<br />

2.1.3 Equations de la solidification (alliages dilués)<br />

Nous considérons la solidification directionnelle d’un alliage binaire dilué isotrope,<br />

de concentration C 0 , en géométrie 2D. Les échanges sont entièrement diffusifs.<br />

La chaleur latente est négligée, ainsi que <strong>les</strong> différences de conductivité thermique<br />

dans le liquide <strong>et</strong> le solide. Le champ de température T − T 0 = Gz (T 0 est la température<br />

choisie comme origine) est fixé dans le repère du laboratoire. Les deux<br />

paramètres de contrôle sont, outre C 0 , la vitesse de tirage de l’échantillon V <strong>et</strong> le<br />

gradient thermique G. On note ζ ≡ ζ(x, t) (forme du front) <strong>et</strong> C(x, z, t) (champ de<br />

concentration dans le liquide) <strong>les</strong> deux inconnues du problème (Fig. 2.2).<br />

Diffusion du soluté<br />

L’équation de la diffusion du soluté dans le liquide s’écrit<br />

17


(approximation des solutions diluées), dans le repère du laboratoire :<br />

D∆C + V ∂C<br />

∂z = ∂C<br />

∂t , (2.2)<br />

où D est le coefficient de diffusion du soluté dans le liquide. On néglige la diffusion<br />

dans le solide (modèle unilatéral), ce qui est réaliste pour <strong>les</strong> matériaux ordinaires.<br />

Conservation de la matière L’équation de conservation du soluté à l’interface<br />

s’écrit :<br />

D∇C i • ˆn = (Ci S − C i )( ˙ζ + V ) • ˆn . (2.3)<br />

où C i <strong>et</strong> Ci<br />

S sont la concentration du liquide <strong>et</strong> du solide à l’interface, ∇C i le gradient<br />

dans le liquide à l’interface, <strong>et</strong> ˆn le vecteur unitaire normal à l’interface.<br />

Equation de Gibbs-Thomson L’équation de Gibbs-Thomson (éq. 2.1) s’écrit :<br />

Gζ = T i − T 0 = m(C i − C 0 ) + a GT κ − β( ˙ζ + V ) , (2.4)<br />

où T i est la température de l’interface. On prend C 0 (concentration moyenne) comme<br />

référence, donc T 0 = TC l 0<br />

(température du liquidus). Pour un système isotrope,<br />

a GT ≡ a 0 = γ 0 T f /L v = γ 0 /∆S V (L v <strong>et</strong> ∆S v : chaleur latente <strong>et</strong> entropie de fusion<br />

par unité de volume, respectivement ; γ 0 : tension de surface) <strong>et</strong> β ≡ β 0 sont des<br />

constantes.<br />

Longueurs caractéristiques La longueur de diffusion l d = D/V provient du<br />

terme d’advection de l’éq. 2.2. Trois autres longueurs sont tirées de l’éq. 2.4 : la<br />

longueur thermique l t = ∆T 0 /G, où ∆T 0 = T l (C 0 )−T s (C 0 ) [T s (C 0 ) : température du<br />

solidus pour C 0 ] s’écrit (version linéarisée) ∆T 0 = m∆C 0 = m 1−K C K 0 (K = m/m s :<br />

coefficient de partage de l’alliage ; m s : pente du solidus) ; la longueur capillaire<br />

d 0 = a 0 /∆T 0 ; la longueur cinétique l cin = Dβ 0 /∆T 0 .<br />

2.1.4 Front plan stationnaire <strong>et</strong> instabilité cellulaire<br />

Dans une expérience-type, on laisse d’abord le système s’équilibrer à l’arrêt (V =<br />

0), après une fusion partielle –on conserve un "talon" solide qui servira de germe pour<br />

la solidification. A l’équilibre, la concentration du liquide est C L = C 0 , celle du solide<br />

C s = KC 0 , <strong>et</strong> la température de l’interface T i = T l (C 0 ). En régime stationnaire du<br />

front plan, à l’interface, on a nécessairement C s = C 0 , C L = ( C 0 /K <strong>et</strong> T i = T s (C 0 )<br />

(Fig. 2.3). Le liquide n’est plus homogène selon z : C(z) = C 0 1 +<br />

1−K d)<br />

K e−z/l [z = 0<br />

correspond à T s (C 0 )]. Ce profil concentration se constitue durant un transitoire<br />

("redistribution de soluté") d’amplitude totale l t , durant lequel la température de<br />

l’interface passe de T l (C 0 ) à T s (C 0 ), donc recule dans le repère du laboratoire, <strong>et</strong> dont<br />

la forme est approximativement exponentielle, de temps caractéristique τ = D/V 2<br />

si K est proche de 1 (cas de CBr 4 -C 2 Cl 6 ) [76, 77, 78, 79].<br />

Le front plan stationnaire n’est stable que pour de faib<strong>les</strong> valeur de V . L’analyse<br />

de stabilité linéaire du front plan [3, 10, 11, 80] montre qu’à G donné, il existe une<br />

vitesse-seuil V c en dessous de laquelle le front plan est stable, mais au-dessus de<br />

laquelle il se déstabilise. Au premier ordre, on trouve :<br />

V c ≈ D ( ) ]<br />

d0 K<br />

[1 2 1/3<br />

+ 3<br />

. (2.5)<br />

l t 4l t<br />

18


Fig.2.3 – Régimes stationnaires ; solidification directionnelle d’alliages dilués. A l’arrêt (V = 0) :<br />

solide en équilibre avec le liquide homogène. Régime de front plan monophasé (0 < V < V c ).<br />

Instabilité cellulaire (V > V c ). Front dendritique (V >> V c ).<br />

Au seuil, il existe une seule longueur d’onde instable ("critique") λ c proportionnelle<br />

à V −2/3<br />

c . Le vecteur d’onde critique k c = 2π/λ c s’exprime comme suit :<br />

[ ] KD<br />

2 1/3 ( ) [ 1/3 ( ) ]<br />

K<br />

d0 K 2 1/3 2/3<br />

k c ≈<br />

≈<br />

1 + 3<br />

. (2.6)<br />

2d 0 Vc<br />

2 2d 0 lt<br />

2 4l t<br />

Les variab<strong>les</strong> V <strong>et</strong> G n’interviennent pas séparément. Le paramètre de contrôle<br />

pertinent (à d 0 fixée) est la variable réduite µ = l t /l d = V/V sc , proportionnelle à V/G.<br />

La vitesse-seuil de surfusion constitutionnelle V sc = D/l t [9] est celle pour laquelle<br />

le gradient de concentration à l’interface est égal (en valeur absolue) à |m|G, soit<br />

l t = l d . L’approximation V c ≈ V sc (µ c ≈ 1), qui revient à négliger le terme capillaire<br />

dans l’éq. 2.5, sous-estime la valeur du seuil de 10 à 20% typiquement (Table 2.1).<br />

Elle ne se justifie pas en général, mais est suffisante (<strong>et</strong> facilement utilisable), vu la<br />

marge d’erreur expérimentale habituelle.<br />

) (<br />

∆T [K] l t [µm] d 0 [µm] V sc [µms −1 ] λ c [µm] V c [µms −1 ] 3<br />

1/3<br />

d 0 K 2<br />

4l t<br />

1.87 170 0.046 2.95 96 3.24 0.10<br />

Tab. 2.1 – Grandeurs caractéristiques (voir texte) d’un alliage CBr 4 -C 2 Cl 6 . C 0 = 0.08 (fraction<br />

molaire) ; G = 110 K/cm. Le calcul à partir de l’équation de dispersion sans approximation donne<br />

V c = 3.29 µms −1 <strong>et</strong> λ c = 90.6µm.<br />

Pour V > V c , l’intervalle de longueurs d’onde instab<strong>les</strong> s’élargit rapidement –<br />

ceci est une conséquence de la grande disparité des échel<strong>les</strong> de la diffusion <strong>et</strong> de la<br />

capillarité (cf Fig. 3.28 plus loin). Il ne contient pas k = 0. Le gradient thermique<br />

(la capillarité) stabilise le front par rapport aux p<strong>et</strong>its (grands) vecteurs d’onde –<br />

mais la "restabilisation absolue" (ferm<strong>et</strong>ure du domaine d’instabilité vers <strong>les</strong> fortes<br />

valeurs de V à G donné) n’est pas observable dans <strong>les</strong> matériaux <strong>et</strong> <strong>les</strong> conditions<br />

19


considérés ici. Le mode "dangereux" (de plus fort taux d’amplification) varie en<br />

(d 0 l d ) 1/2 . Expérimentalement, dans <strong>les</strong> premiers stades de l’instabilité, c’est souvent<br />

ce mode qui émerge. En revanche, la structure cellulaire restabilisée peut adopter<br />

une valeur de λ assez sensiblement différente. L’instabilité cellulaire se présente,<br />

peu au-dessus du seuil, sous la forme d’une bifurcation, dont le caractère direct ou<br />

indirect dépend de la valeur du coefficient de partage K. Pour K proche de 1 (cas<br />

de CBr 4 -C 2 Cl 6 ), la bifurcation est directe, ce que confirment <strong>les</strong> expériences (elle est<br />

indirecte pour K proche de 0) [80]. Le régime faiblement non-linéaire qui, dans le<br />

cas de CBr 4 -C 2 Cl 6 , se limite aux valeurs de V tel<strong>les</strong> que V/V c − 1 est de l’ordre de<br />

10 −2 [81], n’est pas observable.<br />

2.2 Méthodes expérimenta<strong>les</strong><br />

Les caractéristiques de l’alliage binaire CBr 4 -C 2 Cl 6 (Fig. 2.4) ont été (re)mesurées<br />

par Mergy <strong>et</strong> al. [68, 69]. Le diagramme de phase présente un plateau eutectique<br />

étroit, entre deux phases α (riche en CBr 4 ) <strong>et</strong> β (riche en C 2 Cl 6 ) qui cristallisent dans<br />

des systèmes cubiques –α est cubique face centrée [82], β cubique centrée. L’étude des<br />

fronts cellulaires <strong>et</strong> dendritiques (§3) a été faite dans la phase α (0 < C 0 < 8 mol%).<br />

120<br />

110<br />

T (°C)<br />

100<br />

90<br />

80<br />

α<br />

Liquide<br />

β<br />

70<br />

0 5 10 15 20 25 30 35<br />

C (mol%)<br />

Fig.2.4 – Diagramme de phase de l’alliage CBr 4 -C 2 Cl 6 [68]. On note C α <strong>et</strong> C β <strong>les</strong> bornes du<br />

plateau eutectique (température T E ), <strong>et</strong> C E la concentration au point eutectique.<br />

Les échantillons minces sont faits de deux lamel<strong>les</strong> de verre séparées par deux<br />

espaceurs qui délimitent un espace utile d’environ 60 mm × 5 mm × 12 µm. On<br />

peut découper <strong>les</strong> espaceurs en forme d’entonnoir de sorte à former un sélecteur de<br />

cristaux (cf Fig. 3.4 plus loin). Le banc de solidification directionnelle en échantillons<br />

minces (Fig. 2.6) est formé de deux blocs de cuivre régulés en température –un bloc<br />

chaud (chauffage résistif commandé par PID) <strong>et</strong> un bloc froid (circulation d’eau)–<br />

fixés sur un socle isolant, séparés par une fenêtre réglable (5–10 mm) où s’établit le<br />

gradient. Chaque bloc est composé de deux pièces entre <strong>les</strong>quel<strong>les</strong> glisse l’échantillon.<br />

Les deux pièces du bloc chaud sont régulées séparément pour compenser d’éventuels<br />

biais thermiques transverses (ceci est naturellement encore plus important pour <strong>les</strong><br />

expériences faites avec des échantillons semi-épais ; voir §5.1). L’échantillon est guidé<br />

en translation <strong>et</strong> déplacé par un moteur à courant continu (V = 0.01 − 200 µms −1 ).<br />

Le tout est fixé sur une platine XY motorisée pour pouvoir balayer l’ensemble de<br />

20


Fig.2.5 – Echantillon mince (<strong>les</strong> joints de colle<br />

ne sont pas représentés).<br />

Fig.2.6 – Banc de solidification directionnelle.<br />

l’échantillon (la fenêtre d’observation de l’ensemble microscope+caméra peut varier<br />

de 0.250 à 1 mm, en fonction de l’objectif). Le profil de température est linéaire à<br />

l’arrêt au voisinage de la position de l’interface, <strong>et</strong> se déforme peu en cours de tirage.<br />

Une étape-clé de la préparation des expériences de SDEM est la purification des<br />

produits de départ (par fusion de zone, sublimation ou simple distillation). Le remplissage<br />

des échantillons se fait sous atmosphère contrôlée <strong>et</strong> l’échantillon est ensuite<br />

scellé. Malgré ces précautions, subsistent des impur<strong>et</strong>és résiduel<strong>les</strong> "gazeuses" (argon,<br />

gaz de l’air), en quantité très faible , mais suffisante pour perturber la solidification.<br />

Par une caractérisation in situ des échantillons [36] (voir aussi [68, 83, 84, 85]),<br />

en considérant qu’une impur<strong>et</strong>é, notée X, domine <strong>les</strong> eff<strong>et</strong>s mesurés, on trouve, dans<br />

<strong>les</strong> meilleurs cas, que sa concentration molaire C X0 ne dépasse pas ≈ 5 × 10 −4 . Nous<br />

avons estimé <strong>les</strong> principa<strong>les</strong> caractéristiques d’un diagramme de phase CBr 4 −X<br />

postulé. Il présente un plateau de coexistence liquide-solide-vapeur, <strong>et</strong> rend compte<br />

des phénomènes observés en présence d’une quantité importante d’impur<strong>et</strong>é : la<br />

germination de bul<strong>les</strong> dans le liquide <strong>et</strong> la croissance couplée solide-vapeur (Figure<br />

2.7) [36, 86]. Il serait intéressant de poursuivre l’étude de ces "eutectiques gazeux"<br />

[86, 87, 88, 89, 90] <strong>et</strong> de la formation de pores dans le solide en cours de solidification<br />

(voir, par exemple, [91, 92]).<br />

Fig.2.7 – Bul<strong>les</strong> tubulaires (eutectique gazeux) stationnaires dans des échantillons capillaires de<br />

CBr 4 chargé en gaz résiduel.<br />

21


Chapitre 3<br />

Dynamique des fronts de<br />

solidification d’alliages dilués en<br />

géométrie 2D<br />

3.1 Introduction<br />

Les fronts cellulaires <strong>et</strong> dendritiques (Fig. 3.1) sont des structures stationnaires<br />

périodiques (de période λ : espacement inter-cellulaire ou dendritique) caractéristiques<br />

de la solidification directionnelle d’alliages dilués non-fac<strong>et</strong>tés au-dessus du<br />

seuil cellulaire V c [9, 10, 11, 3, 80, 81]. La forme de la microstructure restabilisée<br />

est très non-linéaire, donc difficile à prévoir, même très peu au-dessus du seuil. En<br />

pratique, λ reste assez proche d’une quantité (mode dangereux) déterminée par une<br />

compétition entre la diffusion <strong>et</strong> <strong>les</strong> forces capillaires, mais peut varier en fonction<br />

des conditions aux limites <strong>et</strong> de l’histoire de l’expérience. Ces structures subissent<br />

différents types de transformation morphologique en fonction des caractéristiques<br />

de l’alliage <strong>et</strong> des paramètres de contrôle, dont la transition, lorsqu’on augmente V ,<br />

des cellu<strong>les</strong> "peu profondes" vers <strong>les</strong> dendrites. Les couplages diffusifs, forts dans<br />

<strong>les</strong> fronts cellulaires (λ < l d ) [93, 94], s’affaiblissent dans <strong>les</strong> fronts dendritiques<br />

(V >> V c ; λ > l d ). La forme des "doigts" n’est alors plus déterminée par <strong>les</strong> interactions<br />

entre voisines, mais par des eff<strong>et</strong>s non-linéaires à la pointe (échelle locale).<br />

Le fait remarquable, établi dans <strong>les</strong> années 1980, est que la forme dendritique doit<br />

son existence <strong>et</strong> sa stabilité à un p<strong>et</strong>it paramètre, l’anisotropie interfaciale.<br />

La croissance dendritique a fait l’obj<strong>et</strong> d’un nombre important d’études expérimenta<strong>les</strong><br />

(voir, par exemple, [66, 95, 96, 97, 98]), théoriques [99, 100, 101, 102,<br />

103, 104, 105, 106] <strong>et</strong> numériques [107, 24, 108, 109] en croissance libre (croissance<br />

d’un monocristal isolé dans le liquide uniformément sous-refroidi). La dendrite libre<br />

est une forme de croissance stationnaire en pointe (ou "aiguille") de profil approximativement<br />

parabolique affectée d’un branchement latéral. Sauf cas particulier, ce<br />

branchement s’amplifie "sur place", à l’arrière de la pointe, sans perturber son avancée<br />

(instabilité convective). Dans la plupart des matériaux formant des cristaux de<br />

symétrie élevée, <strong>les</strong> dendrites libres croissent le long d’un axe cristallographique de<br />

haute symétrie. La théorie montre qu’on ne peut trouver de solution dendritique stationnaire<br />

–très proche du cristal aiguille trouvé dans l’approximation de croissance<br />

22


Fig.3.1 – Structures de solidification (SDEM) d’un cristal d’orientation axiale(alliage CBr 4 -<br />

8 mol% C 2 Cl 6 ). G = 110 Kcm −1 ; V c ≈ 2.2 µms −1 . a) cellu<strong>les</strong> peu profondes symétriques<br />

(V = 4.4 µms −1 ) ; b) dendrites symétriques (V = 31 µms −1 ).<br />

diffusive sans tension de surface [99]– qu’en présence d’une certaine amplitude de<br />

l’anisotropie interfaciale.<br />

En l’absence d’anisotropie, la solution dendritique disparaît. Ceci soulève naturellement<br />

la question de savoir quelle forme de solidification la remplace dans un système<br />

isotrope. La réponse théorique à c<strong>et</strong>te question est venue de l’étude des formes<br />

stationnaires de croissance (libre) dans un canal (2D) [110, 111, 112, 113, 114]. Dans<br />

ces calculs on fait varier la surfusion ∆T , la largeur du canal w <strong>et</strong> l’anisotropie interfaciale.<br />

Ceci a conduit à la découverte du "doublon" (Fig. 3.2), sorte de dendrite<br />

fendue, favorisée par une faible anisotropie <strong>et</strong> un fort sous-refroidissement, résultant<br />

de l’accolement en image miroir de deux doigts à symétrie brisée (SB) –la terminaison<br />

en "on" marque le caractère local de l’obj<strong>et</strong>, au même titre que la dendrite<br />

(voir aussi [115, 116]). Les caractéristiques morphologiques (rayon en tête, largeur<br />

du canal central) du doublon sont fixées, sauf sa direction de croissance.<br />

300<br />

200<br />

z<br />

100<br />

0<br />

100 200 300 400 500 600 700<br />

x<br />

Fig.3.2 – Doublon en croissance libre dans un canal (simulation numérique [111]).<br />

Nous proposons ici une étude expérimentale des morphologies dendritiques <strong>et</strong><br />

non-dendritiques en SDEM d’alliages CBr 4 -C 2 Cl 6 dilués. En solidification directionnelle,<br />

pour un alliage donné, <strong>les</strong> paramètres de contrôle sont V , G, <strong>et</strong> la période<br />

spatiale λ (analogue au paramètre w en croissance dans un canal avec conditions périodiques<br />

aux parois), auxquels s’ajoute toujours l’anisotropie interfaciale. La valeur<br />

du sous-refroidissement caractéristique des structures (pointe des dendrites ou des<br />

cellu<strong>les</strong>) s’ajuste, une fois ces paramètres fixés. Dans la limite où V >> V c , <strong>les</strong> eff<strong>et</strong>s<br />

du gradient thermique s’affaiblissent : il est alors possible de transposer approximativement<br />

<strong>les</strong> résultats obtenus en croissance libre dans un canal en échangeant le rôle<br />

du sous-refroidissement <strong>et</strong> de la vitesse de croissance [117, 118].<br />

23


Le but principal de ce chapitre est d’apporter de nouvel<strong>les</strong> preuves expérimenta<strong>les</strong><br />

de la sensibilité des formes de solidification à l’anisotropie capillaire <strong>et</strong> cinétique. Il<br />

existe peu, dans la littérature, d’études comparab<strong>les</strong> (voir l’étude-précurseur de la<br />

réf. [119] <strong>et</strong> l’étude plus récente de la réf. [136]). Les études des réfs. [22, 71, 120, 121,<br />

122] ont permis d’explorer différentes branches de solutions stationnaires ou oscillantes,<br />

principalement à anisotropie constante. En revanche, nos résultats bénéficient<br />

du support du calcul numérique. La partie 3.2 concerne <strong>les</strong> structures observées pour<br />

V >> V c . Nous expliquerons d’abord comment, dans <strong>les</strong> échantillons minces, on peut<br />

faire varier l’anisotropie interfaciale 2D en changeant l’orientation du cristal. Les résultats<br />

centraux sont la découverte expérimentale de la structure "en algue" <strong>et</strong> du<br />

doublon à anisotropie faible, <strong>et</strong> de la structure "dégénérée". Nous amorçons une discussion<br />

sur la mesure des coefficients d’anisotropie capillaire <strong>et</strong> cinétique, <strong>et</strong> de leur<br />

influence relative –dans CBr 4 -C 2 Cl 6 , l’anisotropie cinétique semble être dominante.<br />

Dans la partie 3.3, nous étudierons la dynamique des fronts cellulaires en SDEM<br />

de l’alliage CBr 4 -C 2 Cl 6 en fonction de l’anisotropie interfaciale, à la lumière des<br />

résultats précédents.<br />

Nous proposons enfin (§3.4) une étude, indépendante de ce qui précède, d’un phénomène<br />

de "polygonisation dynamique" de monocristaux en cours de solidification.<br />

Ce phénomène, la création de joints <strong>et</strong> sous-joints de grains dans des monocristaux,<br />

fait intervenir un couplage de la dynamique diffusive avec <strong>les</strong> défauts de réseau –<br />

dislocations <strong>et</strong> sous-joints de grains. On donne ainsi à ce problème complexe, dont<br />

la formulation initiale remonte aux travaux historiques de Chalmers [6], Franck [54]<br />

<strong>et</strong> Jackson [123], une explication principalement fondée sur la dynamique diffusive<br />

du front.<br />

3.2 Fronts dendritiques, "structure en algue" <strong>et</strong> transitions<br />

morphologiques<br />

3.2.1 Monocristaux orientés<br />

La Figure 3.3(haut) montre un front de solidification dans un échantillon mince<br />

polycristallin. Aux différents grains correspondent différentes morphologies de croissance<br />

: ceci montre la forte dépendance des structures de solidification 1D en l’orientation<br />

du cristal –<strong>et</strong> <strong>les</strong> perturbations engendrées par <strong>les</strong> joints de grains. Pour<br />

obtenir des monocristaux de grande taille (Fig. 3.3, bas), on utilise un sélecteur en<br />

forme d’entonnoir (Fig. 3.4).<br />

Pour spécifier l’orientation relative du cristal <strong>et</strong> du dispositif de croissance, nous<br />

utilisons la notation suivante. Pour simplifier, nous nous limitons au cas où un plan<br />

réticulaire (hkl) est parallèle au plan de l’échantillon (plan xz) <strong>et</strong> une direction<br />

réticulaire [uvw] est parallèle à la direction de croissance z. Bien entendu, [uvw]<br />

doit appartenir à (hkl). On a donc hu + kv + lw = 0. Nous notons alors (hkl)[uvw]<br />

l’orientation du cristal. L’ensemble des orientations que l’on peut obtenir à partir de<br />

celle-ci par <strong>les</strong> symétries du cristal est noté {hkl}〈uvw〉. Dans le plan de l’échantillon,<br />

nous repérons <strong>les</strong> directions par l’angle θ qu’el<strong>les</strong> font avec l’axe z. Par exemple,<br />

24


Fig.3.3 – Fronts de solidification (SDEM ; CBr 4 -C 2 Cl 6 ; V ≈ 20µms −1 ). Haut : polycristal (largeur<br />

: 6.8 mm). Bas : monocristal (largeur : 5.5 mm).<br />

Fig.3.4 – A gauche : Expansion d’un monocristal dans l’"entonnoir" à la sortie du canal sélecteur.<br />

A droite : Indexation de l’orientation (hkl)[uvw] d’un cristal dans un échantillon mince.<br />

l’axe x, parallèle au front <strong>et</strong> orienté de gauche à droite correspond ainsi à θ = −π/2<br />

(Fig. 3.4).<br />

Il faut, pour analyser nos observations, trouver un moyen de déterminer l’orientation<br />

des cristaux. Dans <strong>les</strong> échantillons d’alliages CBr 4 -C 2 Cl 6 , nous n’avons pas<br />

tenté de le faire par diffraction X (la mesure devrait avoir lieu in situ car l’alliage<br />

subit une transition de phase solide-solide vers 42˚C). Nous avons utilisé un moyen<br />

indirect : la mesure de l’orientations des fac<strong>et</strong>tes d’inclusions gazeuses dans le solide.<br />

Pavlovska <strong>et</strong> Nemov [124] ont trouvé que des inclusions gazeuses au repos dans le<br />

CBr 4 ont de grandes fac<strong>et</strong>tes {111} <strong>et</strong> de p<strong>et</strong>ites fac<strong>et</strong>tes {100}. Dans nos expériences,<br />

nous trouvons que <strong>les</strong> fac<strong>et</strong>tes principa<strong>les</strong> sont des fac<strong>et</strong>tes {100} (cela peut<br />

provenir de ce que <strong>les</strong> inclusions migrent dans le gradient de température ; ce ne sont<br />

pas des formes d’équilibre).<br />

Fig.3.5 – Inclusions gazeuses fac<strong>et</strong>tées dans des cristaux de phase α (CBr 4 -C 2 Cl 6 ) migrant dans<br />

le gradient thermique : a) cristal (111) ; b) cristal d’orientation "dégénérée".<br />

Ce moyen nous a permis d’orienter avec certitude certains cristaux. Dans le cas<br />

où <strong>les</strong> fac<strong>et</strong>tes des inclusions gazeuses forment des ang<strong>les</strong> d’environ 120 o entre el<strong>les</strong><br />

25


(Fig. 3.5), on sait que l’orientation du cristal est très proche de (111) (l’axe [111] est<br />

d’ordre trois). Quand deux fac<strong>et</strong>tes forment un angle d’environ 45 o avec l’axe z, <strong>et</strong><br />

que l’intersection d’une troisième avec <strong>les</strong> plaques de verre est (presque) parallèle à<br />

x, alors on sait que l’orientation est proche de l’orientation dite "dégénérée" (plan<br />

[001], z parallèle à [110], <strong>les</strong> axes [100] <strong>et</strong> [010] placés symétriquement de part <strong>et</strong><br />

d’autre de z). Nous avons effectué nos observations dans des échantillons orientés de<br />

c<strong>et</strong>te manière aussi près que possible de ces orientations particulières.<br />

3.2.2 Anisotropie interfaciale<br />

Pour un système anisotrope, la tension de surface est une fonction de la normale<br />

unitaire à l’interface ˆn : γ ≡ γ(ˆn). L’équation 2.4 (sans cinétique) doit être écrite en<br />

incluant non pas la tension, mais la raideur de surface, ce qui donne (en 3D) :<br />

T i − T 0 = m(C i − C 0 ) +<br />

2∑<br />

a ɛj κ j , (3.1)<br />

où l’indice j = 1, 2 se rapporte aux deux directions principa<strong>les</strong> ι j de courbure de<br />

l’interface, κ j désigne <strong>les</strong> deux courbures principa<strong>les</strong>, <strong>et</strong> :<br />

[<br />

]<br />

a ɛj = T f /L γ(ˆn) + ∂θ 2 j<br />

γ(ˆn) , (3.2)<br />

où ∂θ 2 j<br />

désigne la dérivée partielle seconde de γ(ˆn) par rapport aux ang<strong>les</strong> mesurés<br />

dans <strong>les</strong> plans (ˆnι j ).<br />

La tension de surface respectant la symétrie du cristal peut être développée sur<br />

une base de fonctions angulaires de même symétrie. Dans des cristaux de symétrie<br />

cubique complète, <strong>les</strong> opérations de symétrie ont pour eff<strong>et</strong> de permuter <strong>et</strong>(ou) de<br />

changer le signe des composantes (n 1 , n 2 , n 3 ) de ˆn sur <strong>les</strong> axes du cube. Toute combinaison<br />

des n 2 i invariante par rapport aux permutations est "cubique" (respecte<br />

la symétrie du cristal cubique). Une base de tel<strong>les</strong> combinaisons rangées par ordre<br />

décroissant habituellement utilisée est [125] :<br />

j=1<br />

Q = n 4 1 + n 4 2 + n 4 3 ; S = n 2 1n 2 2n 2 3 ; T = n 8 1 + n 8 2 + n 8 3 ; ... (3.3)<br />

On vérifie que, puisque n 2 1 + n 2 2 + n 2 3 = 1, il existe, pour chaque degré, une seule<br />

fonction angulaire cubique indépendante (par exemple, n 2 1n 2 2 + n 2 1n 2 3 + n 2 2n 2 3 = 1/2 −<br />

1/2Q). Aucun principe n’impose que <strong>les</strong> coefficients du développement de γ(ˆn) sur<br />

une telle base décroisse de façon monotone avec le degré de la fonction. Cependant,<br />

dans le domaine des morphologies de solidification non-fac<strong>et</strong>tée, toutes <strong>les</strong> études<br />

menées jusqu’ici indiquent que seuls <strong>les</strong> coefficients des deux fonctions de plus bas<br />

degré (Q <strong>et</strong> S) ont besoin d’être pris en compte.<br />

Diverses combinaisons linéaires de Q <strong>et</strong> S ont été utilisées par différents auteurs.<br />

Lorsqu’on travaille en géométrie 2D, <strong>les</strong> fonctions :<br />

⎧<br />

⎨ f 4 = 4Q − 3<br />

(3.4)<br />

⎩<br />

f 6 = −108S + 1<br />

26


sont particulièrement commodes car leurs formes dans <strong>les</strong> plans de haute symétrie<br />

sont très simp<strong>les</strong>. Par exemple, on a f 4 = cos4θ <strong>et</strong> f 6 = 1 pour une orientation<br />

{001}〈100〉, <strong>et</strong> f 4 = −1 <strong>et</strong> f 6 = cos6θ pour une orientation {111}〈1¯10〉. Ces fonctions<br />

ont cependant le désavantage de ne pas être de moyenne nulle. Dans <strong>les</strong> réfs. [126,<br />

127], Hoyt <strong>et</strong> coll. utilisent <strong>les</strong> fonctions suivantes :<br />

⎧<br />

⎨ f4 0 = Q − ¯Q<br />

⎩<br />

f6 0 = − 17 + 3Q + 66S = 3(Q − ¯Q) + 66(S − ¯S)<br />

(3.5)<br />

7<br />

où ¯Q = 3/5 <strong>et</strong> ¯S = 1/105 sont <strong>les</strong> moyennes angulaires de Q <strong>et</strong> S. Ces fonctions<br />

sont de moyennes nulle <strong>et</strong> orthogona<strong>les</strong> entre el<strong>les</strong>. On pose :<br />

γ(ˆn) = ¯γ ( 1 + ɛ 0 4f 0 4 + ɛ 0 6f 0 6<br />

)<br />

= ¯γ (1 + ɛ0 + ɛ 4 f 4 + ɛ 6 f 6 ) . (3.6)<br />

On obtient alors <strong>les</strong> relations suivantes :<br />

⎧<br />

ɛ 0 = 3<br />

20<br />

⎪⎨<br />

ɛ0 4 + 109<br />

252 ɛ0 6<br />

ɛ 4 = 1 4 ɛ0 4 + 3 4 ɛ0 6<br />

⎪⎩<br />

ɛ 6 = − 11<br />

18 ɛ0 6 .<br />

(3.7)<br />

A titre d’exemple, nous avons reporté dans le Tableau 3.1 <strong>les</strong> valeurs des deux<br />

ensemb<strong>les</strong> de coefficients correspondant aux études des réfs. [127] (calculs de dynamique<br />

moléculaire) <strong>et</strong> [128] (expériences). On voit que dans <strong>les</strong> systèmes considérés<br />

(Cu, Cu-Ni <strong>et</strong> Al-Sn), le coefficient d’ordre 6 n’est pas négligeable. Cependant, <strong>les</strong><br />

expériences dans CBr 4 -C 2 Cl 6 présentées dans la suite ne m<strong>et</strong>tent pas en évidence<br />

d’eff<strong>et</strong> attribuable à ɛ 6 . Nous nous en tiendrons donc à l’hypothèse simplificatrice<br />

selon laquelle, dans CBr 4 -C 2 Cl 6 , seul ɛ 4 est non nul (> 0). Le diagramme de Wulff<br />

correspondant est schématisé dans la Fig. 3.6. Dans ce cas, γ(ˆn) est maximum selon<br />

<strong>les</strong> axes 〈100〉, minimum selon 〈111〉, <strong>et</strong> présente un point selle le long de chaque axe<br />

〈110〉 (Fig. 3.6).<br />

matériau ɛ 0 4 × 10 2 ɛ 0 6 × 10 2 ɛ 0 × 10 2 ɛ 4 × 10 2 ɛ 6 × 10 2<br />

Ni a 9.0 -1.1 0.87 1.43 0.67<br />

Ni − Cu a 7.2 -0.7 0.78 1.28 0.43<br />

Al − 5wt%Sn b 1.8 -1.1 -0.21 -0.39 0.68<br />

Tab. 3.1 – Valeur de coefficients d’anisotropie capillaire, d’après <strong>les</strong> réfs. [127] (a) <strong>et</strong> [128] (b).<br />

En géométrie 2D, dans le plan de l’échantillon, l’équation de Gibbs-Thomson<br />

s’écrit :<br />

T i − T 0 = m(C i − C 0 ) + a(θ)κ , (3.8)<br />

où le coefficient de Gibbs-Thomson a(θ) est proportionnel à la raideur de surface<br />

τ(θ) qui est maintenant définie simplement par :<br />

τ(θ) ≡ γ + γ ′′ , (3.9)<br />

27


Fig.3.6 – Diagramme de Wulff schématique d’un cristal cubique faiblement anisotrope, d’harmonique<br />

sphérique d’ordre 4 dominante.<br />

où γ ≡ γ(θ) <strong>et</strong> γ ′′ ≡ ∂2 γ<br />

∂θ 2 . La longueur capillaire d(θ) est proportionnelle à a(θ), donc<br />

possède la même anisotropie. On veut connaître la variation de a(θ) pour une orientation<br />

quelconque du cristal par rapport à l’échantillon. Considérons d’abord certaines<br />

orientations particulières. Pour des orientations "axia<strong>les</strong>" {001}〈100〉, prises comme<br />

référence, on a :<br />

γ(θ) = γ 0 [1 + ɛ 4 cos4θ] , (3.10)<br />

donc<br />

τ(θ) = γ 0 [1 − 15ɛ 4 cos4θ]. (3.11)<br />

L’anisotropie effective dans le plan ɛ 4 est alors à son maximum. Un autre cas intéressant<br />

est celui où l’axe [100] reste dans le plan de l’échantillon, mais est incliné<br />

d’un angle quelconque α 0 par rapport à z :<br />

τ(θ) = γ 0 [1 − 15ɛ 4 cos4(θ − α 0 )] . (3.12)<br />

Parmi ces orientations, celle où l’axe [110] est parallèle à z, appelée "orientation<br />

dégénérée" (001)[110] est particulière : <strong>les</strong> deux minima de τ sont disposés symétriquement<br />

par rapport à z –l’axe [110] est parallèle à z <strong>et</strong> <strong>les</strong> axes [100] <strong>et</strong> [010] sont<br />

disposés à ±45 o de l’axe z.<br />

Pour un cristal d’orientation {111}〈uvw〉 [qu’on notera simplement {111} ou<br />

(111)], la composante d’ordre 4 de τ(θ) est nulle. Dans l’approximation proposée ici,<br />

le système est alors isotrope, ce qui peut s’expliquer par un argument de symétrie<br />

simple, puisque l’axe [111] est un axe de symétrie 3, alors que le réseau est cubique.<br />

Il convient de garder à l’esprit qu’une p<strong>et</strong>ite composante d’anisotropie ("résiduelle")<br />

d’ordre 6 est en réalité présente dans ces orientations (111) de CBr 4 -C 2 Cl 6 .<br />

Pour une orientation quelconque, on peut représenter l’anisotropie interfaciale<br />

par un développement de Fourier en θ de la tension <strong>et</strong> de la raideur de surface, qui<br />

s’écrivent, à l’ordre le plus bas :<br />

γ(θ) ≈ γ 0 {1 + ɛ[5β 2 cos2(θ − α 0 − α 2 ) + cos4(θ − α 0 )]} , (3.13)<br />

τ(θ) ≈ τ 0 {1 − 15ɛ[β 2 cos2(θ − α 0 − α 2 ) + cos4(θ − α 0 )]} . (3.14)<br />

Dans ces expressions, τ 0 , α 2 , β 2 , <strong>et</strong> surtout ɛ(< ɛ 4 ) dépendent de la "désorientation"<br />

du cristal "hors du plan" de l’échantillon (c’est-à-dire l’angle entre la direction [100]<br />

28


<strong>et</strong> l’axe transverse y), <strong>et</strong> α 0 de sa désorientation "dans le plan" (voir §3.2.4). Puisque<br />

ɛ 4 est très inférieur à 1, alors γ 0 <strong>et</strong> τ 0 sont à peu près égaux aux valeurs moyennes<br />

nomina<strong>les</strong>. On trouve aussi que β 2 est p<strong>et</strong>it près de l’orientation "dégénérée" (α 0 =<br />

45 o ) (001)[110], ce qui justifie l’utilisation de l’expression 3.12 (Figure 3.7). Enfin,<br />

le rapport ɛ/ɛ 4 est presque toujours proche de 1, sauf dans le voisinage proche de<br />

l’orientation (111), où il tombe à zéro. La variation de ɛ/ɛ 4 est d’ailleurs très rapide<br />

dans c<strong>et</strong>te région, alors qu’elle est lente autour de l’orientation axiale (Fig. 3.7).<br />

Fig.3.7 – Paramètres ɛ/ɛ 4 <strong>et</strong> β 2 (voir texte), décrivant l’anisotropie 2D en fonction de l’orientation<br />

de la normale y au plan de l’échantillon par rapport au cristal. Insert : trajectoire de y représentée<br />

dans une projection stéréographique.<br />

Pour fixer <strong>les</strong> idées, nous nous sommes limités à l’anisotropie de tension de surface.<br />

Nous verrons que, dans le cas de l’alliage CBr 4 -C 2 Cl 6 , <strong>les</strong> phénomènes observés<br />

sont principalement dus à l’anisotropie du coefficient de cinétique linéaire β(ˆn). Les<br />

règ<strong>les</strong> de symétrie, détaillées ci-dessus pour τ(ˆn) doivent être appliquées à β(ˆn). En<br />

particulier, on pourra écrire, en 2D dans une orientation {001}〈100〉 :<br />

β(θ) ≈ β 0 [1 + ɛ k cos4θ] , (3.15)<br />

où ɛ k est le coefficient d’anisotropie cinétique (d’ordre 4) maximum.<br />

3.2.3 Principaux types de morphologies<br />

La Figure 3.8 montre cinq morphologies-types de fronts de solidification monophasée<br />

de l’alliage CBr 4 -C 2 Cl 6 correspondant à cinq orientations-types du cristal.<br />

El<strong>les</strong> ont été observées loin au-dessus du seuil de l’instabilité cellulaire (V > 6V c )<br />

dans de grands monocristaux. Nous allons <strong>les</strong> considérer à tour de rôle dans la suite.<br />

Par anticipation, nous faisons aussi figurer dans la même page <strong>les</strong> cinq structures<br />

de fronts cellulaires observées à des vitesses proches de V c dans <strong>les</strong> mêmes cristaux<br />

(Fig. 3.9).<br />

3.2.4 Morphologies dendritiques<br />

Support théorique<br />

Pour la plupart des orientations des cristaux, nous observons des morphologies<br />

de croissance de type dendritique. Leur caractérisation fait appel aux résultats théoriques<br />

suivants. Nous considérons des cristaux tels qu’en croissance libre, <strong>les</strong> dendrites<br />

suivent des axes cristallographiques de haute symétrie, correspondant à des<br />

29


Fig.3.8 – Structures de fronts de solidification à haute vitesse observées dans cinq monocristaux<br />

de différentes orientations : a) dendrites symétriques dans un cristal d’orientation quasi axiale<br />

(V = 31.1 µms −1 ) ; b) dendrites inclinées dans un cristal d’orientation non axiale à anisotropie<br />

forte (V = 23.9 µms −1 ) ; c) structure en algue dans un cristal d’orientation proche de (111) à<br />

anisotropie quasi nulle (V = 31.1 µms −1 ) ; d) coexistence de dendrites inclinées à gauche <strong>et</strong> à<br />

droite dans un cristal d’orientation proche d’une orientation "dégénérée" (V = 31 µms −1 ; <strong>les</strong><br />

dendrites inclinées à gauche ont dominé après un certain temps) ; e) dendrites inclinées dans une<br />

orientation correspondant à une anisotropie 2D faible (V = 31.1 µms −1 ). En a) <strong>et</strong> c), l’absence<br />

de branches secondaires détectab<strong>les</strong> est due à la prédominance des eff<strong>et</strong>s cinétiques.<br />

Fig.3.9 – Fronts cellulaires observés dans <strong>les</strong> même cristaux que ceux de la Fig. 3.8 : a) cellu<strong>les</strong><br />

symétriques (V = 3.4 µms −1 ) ; b) cellu<strong>les</strong> inclinées (V = 4.4 µms −1 ) ; c) structure de front<br />

cellulaire instationnaire (V = 3.77 µms −1 ) ; d) structure de front cellulaire instationnaire avec<br />

domaines inclinés (V = 3.77 µms −1 ) ; e) front cellulaire avec domaines d’inclinaison stationnaires<br />

–dérivant à vitesse latérale constante– (V = 3.44 µms −1 ). Micrographies observées à la fin des<br />

expériences (loin d’un transitoire initial).<br />

30


maxima de tension de surface (minima de la raideur de surface) –ou des minima<br />

du coefficient cinétique. A sous-refroidissement ∆ donné, la vitesse d’avancée v <strong>et</strong><br />

le rayon de courbure ρ en tête sont fixés. On peut calculer la forme dendritique<br />

stationnaire indépendamment du branchement secondaire –une instabilité latérale<br />

convective, qui s’amplifie sur place, à l’arrière de la pointe. Ivantsov a montré qu’un<br />

"cristal aiguille" de profil parabolique en mouvement stationnaire est solution du<br />

problème diffusif 2D, sans tension de surface [99]. En découle un première relation<br />

entre ∆, v <strong>et</strong> ρ :<br />

∆ = (πp) 1/2 e p erfc( √ p) , (3.16)<br />

où p = ρ Iv /l d = ρ Iv v/D, ρ Iv étant le rayon d’Ivantsov. Dans la limite p → 0,<br />

c<strong>et</strong>te expression donne ρ Iv V ∼ p ∼ ∆ 2 . La forme parabolique n’est cependant pas<br />

solution de l’équation de Gibbs-Thomson. On trouve une solution, proche de la<br />

parabole d’Ivantsov, si l’on fait intervenir, dans c<strong>et</strong>te équation, la tension de surface<br />

<strong>et</strong>/ou la cinétique interfaciale avec leurs termes d’anisotropie (on ne trouve pas de<br />

solution avec une tension de surface isotrope) [100, 101, 102]. La deuxième relation<br />

entre ∆, v <strong>et</strong> ρ découle d’un critère d’existence de la solution du problème (solvability<br />

criterion), qui s’exprime de la manière suivante :<br />

σ ∗ = d 0 D/ρ 2 IvV ≈ constante , (3.17)<br />

pour une dendrite d’origine capillaire. La constante σ* (solvability constant) est une<br />

caractéristique de l’alliage, qui dépend, entre autres, du coefficient d’anisotropie.<br />

Pour un cristal cubique, <strong>et</strong> si on ne r<strong>et</strong>ient que le seul coefficient d’anisotropie d’ordre<br />

4, ɛ 4 (voir éq. 3.10), on trouve :<br />

σ ∗ ∼ ɛ −7/4<br />

4 , (3.18)<br />

dans la limite où ɛ 4 est très p<strong>et</strong>it. Ce calcul montre qu’il n’existe pas de solution<br />

dendritique en l’absence d’anisotropie. Il souligne aussi le caractère local de la dynamique<br />

de la dendrite. L’analyse de la croissance dendritique en présence d’une<br />

anisotropie du coefficient cinétique conduit à un résultat similaire [103]. L’extension<br />

de c<strong>et</strong>te analyse aux fortes valeurs de la surfusion, à des formes quelconques de l’anisotropie<br />

interfaciale, capillaire <strong>et</strong>/ou cinétique, <strong>et</strong> ce en 2D ou 3D, ne peut se passer<br />

du calcul numérique [24, 109, 110, 118]. En pratique, le calcul numérique montre que<br />

le rayon de courbure de la parabole ajustée à la forme observée expérimentalement<br />

(ou numériquement) peut être utilisée faite sans introduire d’erreur (en particulier<br />

sur la valeur de σ*) notablement supérieure à l’incertitude expérimentale courante.<br />

Dendrites proches de l’orientation axiale<br />

Nous prenons comme orientation de référence l’orientation {001}〈100〉, dite axiale,<br />

pour laquelle <strong>les</strong> dendrites sont alignées avec l’axe de solidification z. En pratique, on<br />

observe presque toujours des dendrites inclinées d’un angle α avec z, <strong>et</strong> dérivant avec<br />

une vitesse latérale v d le long de l’axe x. Ceci indique que le cristal a une certaine<br />

désorientation dans le plan de l’échantillon par rapport à l’orientation axiale standard.<br />

Nous nous intéressons d’abord aux cas où c<strong>et</strong>te désorientation, mesurée par<br />

l’angle α 0 de l’éq. 3.12, est assez faible (orientations "quasi axia<strong>les</strong>"), pour pouvoir<br />

analyser la forme de la pointe des dendrites sans tenir compte de leur asymétrie.<br />

31


Fig.3.10 – Dendrites (SDEM ; CBr 4 -C 2 Cl 6 8 mol%).<br />

A gauche : V = 30 µms −1 ; dimension horizontale :<br />

154 µm ; orientation proche de {001}〈100〉. A droite :<br />

cristal fortement désorienté hors du plan par rapport à<br />

{001}〈100〉 ; V = 30 µms −1 .<br />

Fig.3.11 – Dendrite de succinonitrile<br />

non purifié (largeur : 110 µm) ;<br />

insert : ajustement d’une parabole sur<br />

la pointe (graduation : 5 µm).<br />

-Cristaux de faible désorientation hors du plan par rapport à {001}〈100〉 (anisotropie<br />

maximum). Nous mesurons (par un ajustement parabolique) ρ 2 M V = 700 ±<br />

100 µm 3 s −1 , d’où σ∗ = d 0 l d /ρ 2 ≈ 0.06 ± 0.01, ce qui est assez proche de valeurs<br />

trouvées par d’autres auteurs dans du CBr 4 non purifié [129, 130]. Cependant, la<br />

forme de la pointe des dendrites, d’apparence "triangulaire" (ceci s’accentue quand<br />

V augmente) <strong>et</strong> dépourvue ou presque de branches secondaires (Fig. 3.8a <strong>et</strong> 3.10a),<br />

est caractéristique, d’après <strong>les</strong> résultats numériques de Classen <strong>et</strong> al [118], de dendrites<br />

déterminées par l’anisotropie du coefficient cinétique. D’après ces auteurs,<br />

ceci n’entre cependant pas en contradiction avec des eff<strong>et</strong>s cinétiques moyens faib<strong>les</strong><br />

(p<strong>et</strong>ite valeur de β 0 ), non mesurab<strong>les</strong> directement.<br />

-Cristaux de désorientation sensible hors du plan. Les dendrites (quasi axia<strong>les</strong>) ont<br />

une pointe plus arrondie, <strong>et</strong> leur branchement secondaire est ample (Fig. 3.10) –noter<br />

le mécanisme de croissance d’une branche tertiaire (tail instability) qui perm<strong>et</strong> de<br />

réduire l’espacement dendritique λ, moins active pour <strong>les</strong> dendrites de faible désorientation.<br />

Ces caractéristiques confirment que l’anisotropie est d’amplitude plus<br />

faible que pour des orientations proches de {001}〈100〉. Nous avons aussi émis l’hypothèse<br />

qu’il s’agit de dendrites d’origine capillaire –on note aussi la ressemblance<br />

entre ces dendrites <strong>et</strong> cel<strong>les</strong> observées dans le succinonitrile (Fig. 3.11) [15, 22, 121,<br />

122, 131, 132]). Nous n’en avons pas fait d’étude systématique.<br />

Soulignons à nouveau que notre analyse repose sur l’hypothèse que <strong>les</strong> minima<br />

du coefficient cinétique suivent, comme τ, <strong>les</strong> directions [100]. Ceci n’a pas de caractère<br />

général. Certains systèmes (NH 4 Cl) sont connus pour présenter des transitions<br />

morphologiques m<strong>et</strong>tant en jeu différents axes de croissance dendritique, correspondant,<br />

très vraisemblablement, à des extrema capillaire <strong>et</strong> cinétique, respectivement<br />

[133, 134]. Notre hypothèse reste la plus vraisemblable pour le système CBr 4 -C 2 Cl 6 ;<br />

aucune observation n’est venue la contredire.<br />

Dendrites inclinées<br />

L’angle α de dérive des dendrites inclinées (Fig. 3.8b), défini par tanα = v d /V ,<br />

est d’autant plus grand que la désorientation du cristal dans le plan est forte [37].<br />

Autrement dit, il dépend de l’angle α 0 de la direction c du minimum de la raideur<br />

32


tanα<br />

0.7<br />

0.6<br />

0.5<br />

0.4<br />

0.3<br />

0.2<br />

0.1<br />

0<br />

0 1 2 3 4 5 6 7 8<br />

P<br />

Fig.3.12 – Réseau régulier de dendrites inclinées<br />

(SDEM de CBr 4 –8mol%C 2 Cl 6 ) pour<br />

trois valeurs de la vitesse de tirage : a) 9.94<br />

µms −1 ; b) 13.5 µms −1 ; c) 23.9 µms −1 . α :<br />

angle d’inclinaison des dendrites ; α 0 : angle<br />

entre l’axe z <strong>et</strong> l’axe c (voir texte).<br />

Fig.3.13 – Angle d’inclinaison des dendrites α<br />

vs P (disques). Ligne continue : ligne joignant<br />

<strong>les</strong> points obtenus par simulation numérique (T.<br />

Ihle) pour α 0 = 30 o . Les données de la réf. [137]<br />

(dendrites capillaires) sont aussi reportées (carrés).<br />

de surface τ (ou de β) par rapport à l’axe de tirage z –c est proche de la projection<br />

de l’axe [100] dans le plan de l’échantillon, mais pas exactement confondue avec<br />

elle. On observe que α reste toujours plus p<strong>et</strong>it que α 0 (Fig. 3.12) <strong>et</strong> augmente non<br />

seulement quand V augmente, mais aussi quand λ augmente. Le résultat le plus<br />

surprenant est que <strong>les</strong> données tombent sur une même courbe "maîtresse" lorsqu’on<br />

<strong>les</strong> représente en fonction du Pécl<strong>et</strong> du réseau P = λ/l d (Fig. 3.13). L’inclinaison<br />

des dendrites dépend donc principalement, <strong>et</strong> fortement, du couplage diffusif entre<br />

voisines. Le couplage diffusif diminue, <strong>et</strong> α tend vers α 0 , quand P augmente (α ≈ α 0<br />

pour P > 7).<br />

L’inclinaison des dendrites ne dépend pas de manière mesurable de la valeur du<br />

gradient G (pour V > 3V c ). Ceci vient sans doute du fait que la forme des dendrites<br />

est déterminée localement à la pointe (ρ


α (deg)<br />

40<br />

35<br />

30 Pé = 2.93<br />

25<br />

20<br />

15<br />

10<br />

5<br />

0<br />

0 10 20 30 40 50<br />

α o<br />

(deg)<br />

α (deg)<br />

25<br />

20<br />

15<br />

α 0 = 30°<br />

ε<br />

10<br />

k /ε c = 2<br />

Pé = 2.925<br />

5<br />

0<br />

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14<br />

ε k<br />

Fig.3.14 – Résultats numériques (T. Ihle). A gauche : angle α d’inclinaison des dendrites en<br />

fonction de l’angle cristallographique α 0 (P = 2.925). Ligne pointillée : droite passant par <strong>les</strong> deux<br />

premiers points. A droite : α en fonction du coefficient d’anisotropie cinétique ɛ k (P = 2.925 ;<br />

α 0 = 30 o ). Le rapport ɛ k /ɛ 4 est maintenu constant.<br />

[137]. Il est important de remarquer que, dans le travail de la réf. [137], <strong>les</strong> calculs ont<br />

été réalisés en incluant une anisotropie purement capillaire (15ɛ 4 = 0.1). Un élément<br />

de réponse est donné par la Fig. 3.14, qui montre, d’après <strong>les</strong> calculs numériques<br />

de T. Ihle, que l’angle de dérive des dendrites (à α 0 <strong>et</strong> P fixés) dépend, certes, du<br />

coefficient d’anisotropie cinétique, mais varie peu dans la gamme, autour de ɛ k = 0.1,<br />

qui correspond aux valeurs réalistes pour CBr 4 -C 2 Cl 6 . Il faudrait vérifier s’il en est<br />

de même<br />

Il existe cependant une différence sensible entre ces résultats <strong>et</strong> nos observations :<br />

la valeur de l’angle d’inclinaison limite de dendrites peut atteindre 70 o dans CBr 4 -<br />

C 2 Cl 6 (Fig. 3.15), mais ne dépasse pas 40˚ dans <strong>les</strong> autres cas. Il est possible que<br />

c<strong>et</strong>te limite soit déterminée principalement par l’ampleur des branches secondaires<br />

<strong>et</strong> à une tendance plus ou moins forte à la formation de doublons.L’énoncé d’une<br />

conclusion claire reste suspendu à des mesures précises de l’anisotropie interfaciale.<br />

Structure "dégénérée"<br />

Les dendrites fortement inclinées de la Figure 3.15 ont été observées, pour une<br />

forte vitesse de tirage (7V c < V ), dans un cristal d’orientation proche de l’orientation<br />

orientation "dégénérée" (001)[110]. Il présente donc deux minima de τ pour des<br />

directions c 1 <strong>et</strong> c 2 à peu près symétriques par rapport à z. Pour ce type d’orientation,<br />

lorsqu’on baisse la vitesse (2V c < V < 7V c ), une transition se produit vers une<br />

structure instationnaire, appelée, par extension, "structure dégénérée" (Fig. 3.16).<br />

C<strong>et</strong>te structure porte la marque des orientations des deux extrema de tension de surface<br />

(<strong>et</strong>/ou du coefficient cinétique) sous la forme de doigts inclinés dérivants (voir<br />

aussi <strong>les</strong> réfs. [138] <strong>et</strong> [139]). La structure dégénérée <strong>et</strong> la structure en algue se ressemblent,<br />

mais ne se confondent pas, la seconde étant fondamentalement libre d’eff<strong>et</strong><br />

d’anisotropie. La transition structure dégénérée-dendrites inclinées n’est reproduite<br />

numériquement qu’à condition d’introduire une anisotropie cinétique.<br />

3.2.5 Estimation des coefficients d’anisotropie<br />

Les observations révélatrices du caractère cinétique des dendrites de l’alliage<br />

CBr 4 -C 2 Cl 6 sont 1-la forme "triangulaire" des pointes dendrite <strong>et</strong> la faible ampleur<br />

34


Fig.3.15 – Dendrites très inclinées. Cristal d’orientation<br />

proche de (001)[110] (V = 32 µms −1 ≈ 17V c ).<br />

Une "source" de dendrites inclinées dans <strong>les</strong> deux directions<br />

est visible au centre de l’image (il n’y a pas de<br />

Fig.3.16 – Structure dégénérée ; cristal<br />

d’orientation proche de (001)[110]<br />

joint de grains). Dimension horizontale : 2,15 mm.<br />

(V = 7 µms −1 ≈ 3, 7V c ).<br />

des branches secondaires, 2-la forte valeur de l’angle limite de dérive des dendrites<br />

inclinées, <strong>et</strong> 3-le fait que l’angle α ne tend pas vers zéro quand V se rapproche de<br />

V c (cf §3.3.2).<br />

En recherchant la concordance entre résultats expérimentaux <strong>et</strong> numériques, nous<br />

avons pu évaluer la valeur moyenne du coefficient cinétique β 0 ≈ 4.5 × 10 −3 Ksµm −1<br />

<strong>et</strong> un encadrement de son anisotropie : 0.06 < ɛ k < 0.12. La valeur de la tension<br />

de surface moyenne (ou la longueur capillaire d 0 ≈ 10.5 ± 0.8 nm) de CBr 4 a été<br />

mesurée indépendamment [39, 68, 140]. La prédominance du coefficient ɛ k empêche<br />

une estimation fiable du coefficient d’anisotropie capillaire ɛ 4 . Des résultats numériques<br />

satisfaisants ont été obtenus en posant ɛ 4 = ɛ k /30 ≈ 0.003, en accord avec<br />

notre estimation précédente.<br />

La valeur –assez élevée– de β 0 correspond à un sous-refroidissements d’origine<br />

cinétique de plusieurs degrés pour V = 1 mms −1 . C<strong>et</strong> eff<strong>et</strong> paraît grand – le rapport<br />

entre la longueur cinétique <strong>et</strong> la longueur capillaire, l k /d 0 = Dβ 0 /a 0 , est d’à peu<br />

près 20– mais reste, en pratique, trop p<strong>et</strong>it pour être mesurable directement (on<br />

reste loin de la croissance fac<strong>et</strong>tée). Les éléments de comparaison sont assez rares.<br />

La comparaison semi-quantitative entre des calculs de champ de phase <strong>et</strong> des observations<br />

dans le nickel pur ont permis d’estimer β 0 ≈ 2.2 × 10 −6 Ksµm −1 <strong>et</strong> ɛ k ≈ 0.13<br />

(voir [109] <strong>et</strong> références incluses). D’après ce résultat, l’eff<strong>et</strong> cinétique moyen est<br />

donc considérablement moindre à faible vitesse (ou sous-refroidissement) dans Ni<br />

que dans CBr 4 , mais l’anisotropie cinétique est très proche. Dans le succinonitrile,<br />

l’anisotropie capillaire ɛ 4 ≈ 0.5±0.2% mesurée expérimentalement est confirmée par<br />

des calculs de dynamique moléculaire [141], mais il n’existe à notre connaissance pas<br />

d’estimation du coefficient cinétique (ni de son anisotropie). Dans le cas des molécu<strong>les</strong><br />

discotiques des Réfs. [142, 143], le coefficient cinétique β 0 est estimé à environ<br />

7.7 × 10 −3 Ksµm −1 (dans le plan de base hexagonal).<br />

3.2.6 Doublon <strong>et</strong> structure en algue<br />

Principa<strong>les</strong> observations<br />

Le résultat central de c<strong>et</strong>te étude est l’identification expérimentale de la transition<br />

dendrite/doublon (ou structure en algue). C<strong>et</strong>te transition est essentiellement<br />

indépendante de la discussion précédente sur la compétition entre anisotropies capillaire<br />

<strong>et</strong> cinétique, puisqu’elle est observée lorsque <strong>les</strong> deux coefficients d’anisotropie<br />

35


(en tout cas d’ordre 4) deviennent très p<strong>et</strong>its. Cela répond à la question de la forme<br />

du front de solidification quand la solution dendritique disparaît ou devient instable.<br />

Fig.3.17 – A gauche : structure en algue dans un monocristal d’orientation (proche de) (111),<br />

d’anisotropie 2D quasi nulle. A droite : structures en algue <strong>et</strong> dendritique dans un bicristal ; noter<br />

la différence de sous-refroidissement.<br />

Expérimentalement, on peut faire varier continûment l’anisotropie interfaciale,<br />

en 2D, en passant d’une orientation axiale (001)[100], où le coefficient d’anisotropie<br />

ɛ = ɛ 4 est à son maximum, à une orientation (111), où le coefficient d’anisotropie<br />

d’ordre 4 est nul (rotation du cristal autour d’un axe [110]). La Figure 3.17 montre<br />

un exemple de la structure en algue (seaweed pattern) observée dans un cristal proche<br />

de l’orientation (111). C<strong>et</strong>te structure est semblable à une morphologie appelée par<br />

le passé dense-branching morphology, qui n’avait été caractérisée, à l’époque, que<br />

par l’existence de nombreux branchements <strong>et</strong> divisions en pointe de doigts.<br />

Fig.3.18 – Diagrammes spatio-temporels de structures en algue (CBr 4 -8mol%C 2 Cl 6 ). Trajectoires<br />

des centres des canaux liquides à grande (a) <strong>et</strong> p<strong>et</strong>ite (b) échelle, dans une même expérience (V =<br />

64 µms −1 ≈ 34V c ; l d = 7.8 µm ; τ d = 0.12 s). Lignes épaisses (fines) : canaux larges (fins).<br />

Diagramme c) : enregistrement à interval<strong>les</strong> de temps réguliers de l’intersection d’une ligne parallèle<br />

à x un peu en arrière de la tête de la structure avec le profil du front, dans une structure en algue<br />

présentant une légère dérive due à une anisotropie résiduelle (V = 32 µms −1 ; G = 90 Kcm −1 ).<br />

On distingue deux échel<strong>les</strong> dans c<strong>et</strong>te structure. La plus grande (≈ 100 µm pour<br />

V = 30 µms −1 ) correspond à une large cellulation ("cellu<strong>les</strong> en algue"), de taille <strong>et</strong><br />

de dynamique comparab<strong>les</strong> à cel<strong>les</strong> de fronts dendritiques ou de cellu<strong>les</strong> profondes.<br />

La plus p<strong>et</strong>ite échelle est celle de fins canaux liquides se formant à l’intérieur des<br />

36


grandes cellu<strong>les</strong>. Deux éléments attestent la permanence de la structure : (i) la large<br />

cellulation est stationnaire sur de longs temps (relativement au temps de diffusion<br />

τ = D/V 2 ), <strong>et</strong> (ii) la largeur des canaux de liquide intracellulaires reste constante.<br />

Le diagramme spatio-temporel de la Fig.3.18 montre que la dynamique désordonnée<br />

de la structure en algue à p<strong>et</strong>ite échelle provient de la continuelle alternance de formation<br />

(par branchement en pointe ou tip splitting) <strong>et</strong> d’élimination des fins canaux<br />

liquides [121, 122]. Certains des fins canaux liquides identifiés dans la Fig.3.18, de<br />

largeur bien déterminée, ont, eux aussi, un temps de persistance largement supérieur<br />

à τ. Notre but sera de montrer que l’on peut identifier ces fins canaux aux canaux<br />

centraux de doublons.<br />

Ces caractéristiques s’observent sur une large gamme de vitesses (V > 5V c ; Fig.<br />

3.19). A plus basse vitesse (Fig. 3.19a), on trouve une dynamique (instationnaire)<br />

de front cellulaire (§3.3.2). A forte vitesse (Fig. 3.19d), <strong>les</strong> caractéristiques de la<br />

structure deviennent plus p<strong>et</strong>ites que l’épaisseur de l’échantillon <strong>et</strong> la dynamique<br />

devient 3D.<br />

Fig.3.19 – Structure en algue (cristal près d’une orientation 111. (a) V = 8.6 µms −1 (≈ 4.5V c ) ;<br />

(b) V = V = 29 µms −1 (≈ 15V c ) ; (c) V = 64 µms −1 (≈ 34V c ) ; (d) V = V = 100 µms −1 (≈ 53V c ).<br />

Caractérisation <strong>et</strong> stabilité du doublon<br />

D’après la théorie, le doublon (Fig. 3.2) résulte de l’accolement en images miroir<br />

de deux doigts à symétrie brisée (SB). La forme du doigt SB (ɛ = 0) a été calculée,<br />

dans une certaine approximation, par Ben Amar <strong>et</strong> Brener [112]. A ∆ donnée, la<br />

vitesse de croissance v, le rayon de courbure en tête ρ <strong>et</strong> la largeur du canal central<br />

du doublon w c sont sélectionnés [le calcul approché montre que V ∼ ∆ 9 , ce qui est<br />

une variation bien plus rapide que celle de la dendrite (V ∼ ∆ 4 )].<br />

En croissance libre dans un canal, on trouve numériquement que <strong>les</strong> doigts SB<br />

n’existent qu’au-dessus d’une valeur critique ∆ c du sous-refroidissement, relativement<br />

forte (≈ 0.7), <strong>et</strong> qui augmente avec l’intensité de l’anisotropie. D’après Kupferman<br />

<strong>et</strong> al [114], <strong>les</strong> doigts SB apparaissent par une bifurcation indirecte à partir<br />

de la branche dendritique pour ɛ > 0.1. Le doigt SB ne cesse donc pas d’exister<br />

à anisotropie non nulle. Il existe un domaine de "métastabilité" entre dendrites <strong>et</strong><br />

doigts SB. En revanche, la direction de croissance du doublon est libre (ce suggèrent<br />

<strong>les</strong> observations préliminaires de la Fig. 3.20). Elle dépend des conditions aux limites,<br />

<strong>et</strong> est très sensible à des hétérogénéités accidentel<strong>les</strong>. C<strong>et</strong>te indétermination<br />

37


en orientation du doublon explique l’existence des régimes non stationnaires, fortement<br />

ramifiés, rappelant des formes d’algues marines –d’où leur nom de structure<br />

en algue donné par H. Müller-Krumbhaar <strong>et</strong> coll. C<strong>et</strong>te dégénérescence est levée<br />

en présence d’une anisotropie non nulle [144]. La direction de croissance du doublon<br />

est alors stabilisée dans l’axe de l’extremum de tension de surface. Notons que<br />

des formes analogues au doublon en géométrie 3D ("triplons" <strong>et</strong> "quadruplons")<br />

ont été observées numériquement [145], <strong>et</strong> observées expérimentalement, au moins<br />

transitoirement [146].<br />

Fig.3.20 – Doublon quasi stationnaire (profils binarisés à différents instants ; à gauche) <strong>et</strong> changeant<br />

de direction de croissance (à droite). Croissance libre d’un alliage dilué CBr 4 -C 2 Cl 6 en<br />

échantillon mince.<br />

La Figure 3.21a montre un détail d’une structure en algue en SDEM de l’alliage<br />

CBr 4 -C 2 Cl 6 : une forme proche du doublon y est n<strong>et</strong>tement reconnaissable. Les deux<br />

doigts SB qui le compose subissent, cependant, un fort branchement secondaire qui<br />

remonte presque jusqu’en tête. On r<strong>et</strong>rouve ce même phénomène dans <strong>les</strong> simulations<br />

numériques (Figure 3.21b), lors du transitoire précédant la stabilisation ultérieure<br />

d’un doublon. La similitude entre le calcul <strong>et</strong> l’expérience est frappante. En solidification<br />

directionnelle, la dynamique de la structure en algue est donc essentiellement<br />

celle de doublons transitoires.<br />

Nous pouvons caractériser <strong>les</strong> doublons en référence aux résultats en croissance<br />

libre dans un canal (V >> V c ) :<br />

-la solidification de bicristaux (Figure 3.17) montre que la surfusion des pointes de<br />

dendrites est plus faible que celle de la structure en algue –ce que confirment des<br />

observations similaires dans un échantillon de succinonitrile (Fig. 3.22). La différence<br />

de sous-refroidissement δ∆ varie de 0.12 à 0.16 quand le rapport V/V c passe de 5<br />

a) b)<br />

Fig.3.21 – Doublon transitoire : images expérimentale (SDEM ; à gauche) <strong>et</strong> numérique (croissance<br />

libre dans un canal ; à droite).<br />

38


à 50. Elle est toujours plus forte que celle mesurée entre deux grains dendritiques<br />

quelconques, en accord avec la théorie.<br />

Fig.3.22 – Structure en algue (cristal de gauche) <strong>et</strong> dendrites (cristal de droite) dans un bicristal<br />

de succinonitrile.<br />

-le temps de vie des fins canaux liquides centraux des doublons est supérieur au<br />

temps de diffusion. Leur largeur W diminue en fonction de V , en accord avec une<br />

sélection de la valeur de W . La différence entre <strong>les</strong> mesures expérimenta<strong>les</strong> <strong>et</strong> <strong>les</strong><br />

données numériques peut sans doute être partiellement attribué à des eff<strong>et</strong>s 3D.<br />

-<strong>les</strong> doublons axiaux sont instab<strong>les</strong> vis-à-vis d’une inclinaison de leur direction de<br />

croissance, c<strong>et</strong>te instabilité étant responsable de l’instationnarité de la structure en<br />

algue à grande échelle.<br />

En résumé, la structure en algue, observée pour des cristaux d’orientation proche<br />

de (111), résulte d’une dynamique instationnaire de doublons transitoires, couplée<br />

à une cellulation à grande échelle, quasi stationnaire, mais dont on ne connaît pour<br />

l’instant pas l’origine précise. C<strong>et</strong>te cellulation peut éventuellement être comprise<br />

comme une instabilité de type cellulaire d’un front effectif de doublons (comparable,<br />

en cela, à la formation des "colonies" dans <strong>les</strong> eutectiques lamellaires ternaires). Les<br />

observations récentes de Bodenschatz <strong>et</strong> coll. [121, 122] (voir Fig. 3.22) dans des<br />

alliages de succinonitrile sont très similaires aux nôtres. Ceci montre le grand degré<br />

de généralité de nos résultats.<br />

Doublon <strong>et</strong> anisotropie<br />

Comme cela a été souligné plus haut (voir Fig. 3.7), la variation du coefficient<br />

d’anisotropie (d’ordre 4) autour de son minimum (nul) en l’orientation (111) est<br />

rapide –<strong>et</strong> même singulière. L’écart admissible à l’orientation (111) pour que <strong>les</strong> eff<strong>et</strong>s<br />

d’une anisotropie résiduelle restent faible sur la dynamique du front de solidification<br />

est donc très p<strong>et</strong>it. Ceci explique qu’en pratique, même en échantillon mince, le<br />

nombre des grains "bien orientés" parmi tous ceux du polycristal initial est faible.<br />

De plus, à défaut de pouvoir imposer strictement l’orientation du cristal, il reste<br />

donc presque toujours une anisotropie rémanente (d’ordre 4), qui est sans doute<br />

à l’origine de l’existence de structures en algue dérivantes (Figs. 3.18 <strong>et</strong> 3.23). Se<br />

pose également l’influence d’une possible composante non nulle de termes d’ordre<br />

supérieur (par exemple 6) dans le plan (111) [121, 122].<br />

Dans des cristaux d’orientation un peu plus éloignée de (111), mais présentant<br />

une anisotropie trop faible pour stabiliser des dendrites, des réseaux de doublons stationnaires,<br />

éventuellement inclinés (Fig. 3.23), sont observés. Les doigts symétriques,<br />

ne revanche, sont instab<strong>les</strong> (Fig. 3.24) <strong>et</strong> se divisent spontanément (sous l’eff<strong>et</strong> du<br />

39


Fig.3.23 – Structure en algue inclinée ; doublons stationnaires ; doublons dendritiques.<br />

bruit thermique). Ceci tend à prouver 1-que l’anisotropie ne détruit pas la solution<br />

doublon, mais stabilise sa direction de croissance, <strong>et</strong> 2-qu’il existe un intervalle de<br />

valeurs de ɛ à l’intérieur duquel seuls <strong>les</strong> doublons peuvent être stab<strong>les</strong>.<br />

Fig.3.24 – A gauche : formation d’un doublon quasi stationnaire par instabilité d’un doigt symétrique<br />

au sein d’une structure de doublons inclinés [cristal désorienté par rapport à (111)]. A droite :<br />

formation d’un doublon par une perturbation locale (accidentelle) de la pointe d’une dendrite par<br />

une poussière.<br />

Lorsque l’anisotropie augmente, dans des orientations proches de {001}〈100〉, on<br />

observe la formation d’un doublon à la suite d’une forte perturbation (Fig. 3.24) ou,<br />

en augmentant V , par appariement de deux dendrites préexistantes (Fig. 3.23).<br />

3.2.7 Diagrammes de morphologies – Questions ouvertes<br />

L’observation de fronts de solidification de monocristaux en SDEM d’alliages<br />

dilués de CBr 4 -C 2 Cl 6 bien au-delà du seuil cellulaire a permis l’identification de<br />

deux nouveaux types de morphologies 2D, en plus des fronts dendritiques (axiaux<br />

ou inclinés) : la "structure dégénérée" <strong>et</strong> la "structure en algue". Ces morphologies<br />

sont représentatives des morphologies observées couramment dans <strong>les</strong> cristaux nonfac<strong>et</strong>tés<br />

de symétrie cubique. Grâce aux échantillons minces, il est possible de passer<br />

de l’une à l’autre (en géométrie 2D) en fonction de l’orientation du cristal. On peut<br />

synthétiser c<strong>et</strong> ensemble de résultats dans des diagrammes de morphologies (Fig.<br />

3.25), qui présentent <strong>les</strong> principa<strong>les</strong> transitions morphologiques dans des orientations<br />

proches des orientations de haute symétrie.<br />

Les simulations numériques réalisées par T. Ihle (code de suivi de front) avec <strong>les</strong><br />

paramètres de l’alliage CBr 4 -C 2 Cl 6 ont reproduit <strong>les</strong> principa<strong>les</strong> morphologies trouvées<br />

expérimentalement. Leurs caractéristiques sont essentiellement indépendantes<br />

des spécificités de l’alliage CBr 4 -C 2 Cl 6 . La structure "dégénérée" (Fig. 3.26b) <strong>et</strong> la<br />

structure en algue à faible vitesse (Fig. 3.26c) sont semblab<strong>les</strong> aux observations expérimenta<strong>les</strong><br />

(Figs. 3.16 <strong>et</strong> 3.19a). Seule la structure de la Fig. 3.26d fait apparaître une<br />

40


Fig.3.25 – Diagrammes de stabilité des morphologies 1D des fronts de solidification d’alliages<br />

CBr 4 -C 2 Cl 6 pour V >> V c en fonction de l’orientation.<br />

différence sensible avec <strong>les</strong> structures en algue (pour V >> V c ) de nos expériences,<br />

ce qui peut provenir d’eff<strong>et</strong>s 3D ou d’un eff<strong>et</strong> d’une anisotropie résiduelle, capillaire<br />

ou cinétique, d’ordre supérieur à 4 (6 ou même 8). Nous n’avons pas suffisamment<br />

d’éléments pour conclure sur ce point. Enfin, certaines caractéristiques des structures<br />

dendritiques (forme "triangulaire", forte inclinaison limite ; Fig. 3.26a) ne peuvent<br />

être obtenues qu’en incluant une anisotropie cinétique. Par manque d’éléments de<br />

comparaison, on ne sait pas dire, à l’heure actuelle, à quel point c<strong>et</strong>te propriété est<br />

spécifique à l’alliage CBr 4 -C 2 Cl 6 .<br />

Fig.3.26 – Simulations numériques (T. Ihle). a) Dendrites inclinées (anisotropie cinétique). b)<br />

Structure "dégénérée". c) Structure en algue à faible vitesse. d) Structure en algue à forte vitesse.<br />

Il reste des zones d’incertitude, pour certaines orientations ou pour des vitesses<br />

relativement basses (typiquement 5V c –7V c ) dans le régime des cellu<strong>les</strong> profondes [22]<br />

–suj<strong>et</strong> de la "transition cellu<strong>les</strong>/dendrites". Nous n’avons pas abordé le problème spécifique<br />

du branchement secondaire dans <strong>les</strong> structures dendritiques [147]. Il manque,<br />

41


pour une compréhension plus générale, des données expérimenta<strong>les</strong> du même type<br />

que cel<strong>les</strong> que nous avons obtenues pour CBr 4 -C 2 Cl 6 , mais dans d’autres alliages ;<br />

pour lever certaines ambiguïtés, il faudra aussi, dans <strong>les</strong> études à venir, obtenir des<br />

renseignements précis sur l’orientation des cristaux. C’est ce que nous proj<strong>et</strong>ons de<br />

faire en utilisant des échantillons minces d’alliages métalliques (voir §5.2).<br />

3.3 Influence de l’anisotropie interfaciale sur la stabilité<br />

des fronts cellulaires<br />

3.3.1 Mesure du diagramme de bifurcation cellulaire<br />

En SDEM d’alliages dilués, peu au-dessus de V c , on observe des fronts périodiques<br />

de cellu<strong>les</strong> peu profondes. Ces structures sont caractéristiques de la solidification directionnelle<br />

<strong>et</strong> sont sensib<strong>les</strong> au gradient de température. L’interaction par diffusion<br />

dans le liquide entre la tête <strong>et</strong> le fond de la cellulation est forte, ainsi qu’entre<br />

cellu<strong>les</strong> voisines, contrairement aux fronts dendritiques [94]. Les structures cellulaires<br />

peuvent être stationnaires (Fig. 3.1), mais pas dans toutes <strong>les</strong> conditions de<br />

croissance. Leur dynamique présente certaines caractéristiques prévues par la phénoménologie<br />

générale des fronts modulés en géométrie 2D, mais le résultat le plus<br />

frappant, encore une fois, est leur forte sensibilité à l’anisotropie interfaciale. Nous<br />

caractériserons d’abord la bifurcation cellulaire dans l’alliage CBr 4 -C 2 Cl 6 , puis nous<br />

étudierons la stabilité des fronts cellulaires en régime non-linéaire (V ≈ 2V c ).<br />

La mesure de la bifurcation cellulaire –variation de l’amplitude A des cellu<strong>les</strong> en<br />

fonction de V – (Fig. 3.27) donne la valeur du seuil V c . L’incertitude sur V c a deux<br />

origines, d’une part la durée des transitoires, <strong>et</strong> d’autre part, la pré-cellulation due à<br />

la formation de sous-joints de grains, qui rend la bifurcation imparfaite (§3.4). Une<br />

méthode, meilleure, de mesure de V c , mise en oeuvre par Losert <strong>et</strong> al. [78] consiste à<br />

mesurer le taux de relaxation (pour V < V c ) ou d’amplification (pour V > V c ) d’une<br />

p<strong>et</strong>ite perturbation imposée volontairement au front plan.<br />

100<br />

80<br />

A (µm)<br />

60<br />

40<br />

20<br />

0<br />

1 2 3 4 5 6 7<br />

V (µm s -1 )<br />

Fig.3.27 – A gauche : Structures cellulaires axia<strong>les</strong> (CBr 4 -C 2 Cl 6 8 mol%). V = 2.15 µms −1 (a) ;<br />

2.8 µms −1 (b) ; 3.1 µms −1 (c) <strong>et</strong> 4.44 µms −1 (d). A droite : Amplitude A des cellu<strong>les</strong> en fonction de<br />

V (sauts de vitesse décroissants). Cerc<strong>les</strong> <strong>et</strong> triang<strong>les</strong> : échantillons différents. Barres vertica<strong>les</strong> :<br />

dispersion des données. Barres horizonta<strong>les</strong> : incertitude sur V . La mesure du seuil (≈ 2.1 ±<br />

0.2 µms −1 ) est inférieure à la valeur théorique (≈ 3.2µms −1 ) à cause des eff<strong>et</strong>s d’impur<strong>et</strong>és.<br />

Même très près de V c , la forme du front est loin d’une sinusoïde. Les régimes<br />

linéaire <strong>et</strong> faiblement non-linéaire sont clairement inaccessib<strong>les</strong> : l’élargissement ra-<br />

42


pide de l’intervalle de longueur d’ondes (linéairement) instab<strong>les</strong> favorise l’apparition<br />

précoce de non-linéarités. Des mesures répétées montrent que, en régime quasi stationnaire,<br />

l’espacement λ peut varier (à V/G donné) à l’intérieur d’un intervalle dont<br />

la moyenne est proche du mode dangereux (sensiblement inférieur à λ c ). C<strong>et</strong> intervalle<br />

est assez étroit, <strong>et</strong> varie peu avec V (Figure 3.28) [38, 148, 149]. En revanche,<br />

on voit qu’il dépend de l’orientation du cristal.<br />

12<br />

12<br />

10<br />

10<br />

V (µm s-1)<br />

8<br />

6<br />

V (µm s-1)<br />

8<br />

6<br />

4<br />

λ c<br />

nom<br />

4<br />

2<br />

0 100 200 300 400 500 600<br />

λ (µm)<br />

2<br />

λ c<br />

0<br />

0 20 40 60 80 100 120<br />

λ (µm)<br />

Fig.3.28 – Bifurcation cellulaire (CBr 4 -C 2 Cl 6 8 mol%). A gauche : limites de stabilité linéaire<br />

(ligne continue) <strong>et</strong> mode dangereux (ligne pointillée). Mesures expérimenta<strong>les</strong> dans un cristal axial<br />

(carrés) <strong>et</strong> non-axial (triang<strong>les</strong>). Symbo<strong>les</strong> ouverts (fermés) : valeurs maxima<strong>les</strong> (minima<strong>les</strong>) à<br />

V donnée. A droite : détail du même diagramme. Lignes épaisses (fines) : calculs sans (avec)<br />

impur<strong>et</strong>és résiduel<strong>les</strong> (3 × 10 −4 mol −1 ).<br />

3.3.2 Stabilité des fronts cellulaires<br />

Position du problème<br />

Les premières études de la dynamique spatio-temporelle des fronts cellulaires<br />

ayant mis en évidence des phénomènes de diffusion de la phase (instabilité d’Eckhaus<br />

<strong>et</strong> élimination de cellu<strong>les</strong>) <strong>et</strong> des instabilités secondaires (dérive <strong>et</strong> oscillations),<br />

ainsi que la formation de domaines dérivants, ont été faites en SDEM d’une molécule<br />

mésomorphe présentant une transition liquide isotrope-liquide nématique [18]. Les<br />

propriétés de ce système (diffusion symétrique) limitent <strong>les</strong> eff<strong>et</strong>s non-linéaires <strong>et</strong> facilitent<br />

l’analyse phénoménologique. Cependant, leur dynamique est compliquée par<br />

des eff<strong>et</strong>s spécifiques d’élasticité d’interface <strong>et</strong> à l’existence de courants de convection<br />

(interface fluide-fluide). Plus récemment, la sensibilité des fronts cellulaires à<br />

l’anisotropie interfaciale a été étudiée théoriquement, par simulations numériques<br />

[148], en fonction de V/G <strong>et</strong> du coefficient d’anisotropie (axiale) ɛ 4 , pour un système<br />

à diffusion symétrique. Les deux résultats <strong>les</strong> plus importants sont <strong>les</strong> suivants :<br />

1-à anisotropie nulle ou très faible, le "ballon" de stabilité des fronts cellulaires se<br />

réduit à une très p<strong>et</strong>ite région du diagramme V/G-k, tout près du seuil, qui se ferme<br />

pour une valeur de V , typiquement 0.1V c . Très probablement, on trouverait une<br />

extension encore plus p<strong>et</strong>ite pour un système usuel, sans diffusion notoire dans le<br />

solide (qui a un eff<strong>et</strong> stabilisant). Le domaine de stabilité ne s’étend <strong>et</strong> s’ouvre vers<br />

<strong>les</strong> fortes valeurs de V/G qu’au-delà d’une certaine anisotropie.<br />

2-<strong>les</strong> limites de stabilité correspondent à des modes de doublement de période. La<br />

limite inférieure coïncide avec le seuil λ c d’une instabilité non-oscillante ("station-<br />

43


naire"), notée 2λS (voir §5.4), qui conduit à l’élimination d’une cellule sur deux<br />

(dans un front périodique). La limite supérieure correspond à une instabilité oscillante,<br />

2λO, qui se restabilise sur un cycle limite. Les calculs <strong>et</strong> des observations<br />

expérimenta<strong>les</strong> montrent aussi l’existence d’une branche de "doubl<strong>et</strong>s" (cellu<strong>les</strong> à<br />

symétrie brisée accolées deux à deux) [120, 149].<br />

Principaux types de dynamique de structures cellulaires<br />

La Figure 3.9 montre <strong>les</strong> cinq types de fronts cellulaires observés pour une vitesse<br />

de tirage légèrement supérieure au seuil cellulaire (1.1V c < V < 4V c ) dans <strong>les</strong> mêmes<br />

cristaux que la Figure 3.8. Les Figures 3.29 <strong>et</strong> 3.30 rassemblent <strong>les</strong> diagrammes<br />

spatio-temporels correspondants. Nous allons considérer tour à tour <strong>les</strong> structures<br />

cellulaires stationnaires axia<strong>les</strong> (cristal a), puis instationnaires (dans un cristal à<br />

faible anisotropie ; cristal c), <strong>et</strong> enfin dérivantes (cristaux b, d <strong>et</strong> e).<br />

Cellu<strong>les</strong> axia<strong>les</strong><br />

Le diagramme spatio-temporel d’un front cellulaire dans un cristal axial de forte<br />

anisotropie (Fig. 3.29) perm<strong>et</strong> montre la stationnarité de la structure. La forme des<br />

cellu<strong>les</strong> est symétrique (Fig. 3.1). Une telle preuve expérimentale, en accord avec la<br />

théorie, n’avait jamais été apportée auparavant.<br />

La mesure de la distribution spatiale de l’espacement λ(x) montre, qu’en fonction<br />

des conditions initia<strong>les</strong>, des modulations de l’espacement à grande longueur d’onde<br />

subsistent sur d’assez longs temps –ces modulations sont le résultat de l’élimination<br />

d’un certain nombre de cellu<strong>les</strong> (l’événement isolé visible sur la Fig. 3.29 est le<br />

dernier de ce type à se produire dans c<strong>et</strong>te expérience) suivant le transitoire initial.<br />

Ces modulations relaxent par un processus général de diffusion de la phase (que<br />

nous n’avons pas étudié en détail). Ceci nous assure de sa stabilité. Pour explorer<br />

<strong>les</strong> limites de stabilité des fronts cellulaires axiaux, nous avons testé leur réaction<br />

à des sauts de vitesse. La limite de stabilité inférieure (élimination) se déplace vers<br />

<strong>les</strong> grandes valeurs de λ quand V augmente. Nous n’avons pas observé de mode<br />

d’instabilité secondaire se développer de manière permanente, mais seulement un<br />

certain type d’onde solitaire (Fig. 3.31). Ceci est sans doute à m<strong>et</strong>tre en relation<br />

avec la grande stabilité des dendrites axia<strong>les</strong>.<br />

Fronts cellulaires non stationnaires (faible anisotropie)<br />

Le contraste entre <strong>les</strong> deux diagrammes spatio-temporels de la Fig. 3.29 est frappant.<br />

A très faible anisotropie, on n’obtient pas de front stationnaire (Fig. 3.32b),<br />

en bon accord, à nouveau, avec la théorie. On identifie c<strong>et</strong>te fois des oscillations à<br />

doublement de période spatiale (2λO), couplées à des événements de disparition <strong>et</strong><br />

de création (tip splitting) de cellule, ainsi que des ondes solitaires (Fig. 3.31a). Ces<br />

éléments de dynamique spatio-temporelle (y compris <strong>les</strong> ondes solitaires) sont caractéristiques<br />

du voisinage d’une bifurcation oscillante [29, 150], prévue par l’étude<br />

numérique [149]. C<strong>et</strong>te dynamique désordonnée est, par ailleurs, n<strong>et</strong>tement différente<br />

de celle de la structure en algue à plus forte vitesse dans un même type de cristal<br />

44


Fig.3.29 – Diagrammes spatio-temporels (temps vers le haut ; voir Fig. 3.9). A gauche : cellu<strong>les</strong><br />

axia<strong>les</strong> (cristal a). A droite : cellu<strong>les</strong> instationnaires [cristal c (111)]. L’échelle verticale est très<br />

comprimée <strong>et</strong> exagère l’angle d’inclinaison des cellu<strong>les</strong>.<br />

Fig.3.30 – Diagrammes spatio-temporels (à gauche). A droite : graphes rassemblant <strong>les</strong> mesures<br />

de l’angle d’inclinaison α des cellu<strong>les</strong> en fonction de λ. Cristaux b, d <strong>et</strong> e de la Fig. 3.9. Cristal<br />

b : la courbe passant à travers <strong>les</strong> données (carrés) est une moyenne dans le temps des mesures ;<br />

<strong>les</strong> autres symbo<strong>les</strong> sont issus d’une autre expérience à V = 3.4 µms −1 (cerc<strong>les</strong>) <strong>et</strong> V = 3.7 µms −1<br />

(disques). Cristal d : α positif correspond à une dérive vers la gauche. Cristal e : mesures (moyennes<br />

dans le temps) faites avant (disques) <strong>et</strong> après (cerc<strong>les</strong>) un saut de vitesse de 3.4 à 3.75 µms −1 .<br />

45


Fig.3.31 – Diagrammes spatio-temporels. a) Cristal non axial, faiblement anisotrope. b) Dérive<br />

<strong>et</strong> oscillation de type 1λ0 (cristal de la Fig. 3.9d). c) Onde solitaire dans un cristal axial de forte<br />

anisotropie.<br />

(Fig. 3.18). Il n’y a pas, par exemple, de structure analogue à celle du doublon (ou<br />

du doubl<strong>et</strong>).<br />

Fig.3.32 – a) Cellu<strong>les</strong> inclinées ; cristal non axial. b) Cellu<strong>les</strong> instationnaires ; cristal (111). V =<br />

3.1 µms −1<br />

Fronts cellulaires dérivants<br />

L’anisotropie interfaciale joue également un grand rôle dans la dynamique des<br />

fronts cellulaires dérivants (cristaux d’orientation non axiale). La dérive des cellu<strong>les</strong><br />

est couplée à une asymétrie (inclinaison) marquée des cellu<strong>les</strong>, suivant la même<br />

direction que <strong>les</strong> dendrites du même cristal (Fig. 3.32a). Dans le détail, cependant,<br />

la dynamique spatio-temporelle de ces fronts dépend sensiblement de la force de<br />

l’anisotropie <strong>et</strong> du degré de symétrie de l’orientation.<br />

Pour une forte anisotropie (Figs. 3.9b, 3.32a <strong>et</strong> 3.30"cristal b"), la structure<br />

dérive en bloc. L’angle α de dérive des cellu<strong>les</strong> augmente quand V ou λ augmentent<br />

(Fig. 3.30"cristal b"), mais, à la différence des dendrites (§3.2.4), il augmente moins<br />

vite avec V qu’avec λ. L’angle α augmente quasi linéairement avec λ à V fixée,<br />

dans l’intervalle accessible, mais <strong>les</strong> mesures ne s’extrapolent pas à zéro. Des calculs<br />

numériques préliminaires ont montré que α ne s’annule également pas quand V<br />

tend vers V c . Ces observations sont compatib<strong>les</strong> avec l’hypothèse d’une anisotropie<br />

cinétique dominante dans CBr 4 -C 2 Cl 6 : seule l’anisotropie cinétique intervient au<br />

premier ordre dans l’équation de Gibbs-Thomson –l’anisotropie capillaire, présente<br />

dans le terme, non-linéaire, de courbure, est inactive au seuil [151].<br />

46


Pour des cristaux proches d’une orientation "dégénérée" (Figs. 3.9d <strong>et</strong> 3.30),<br />

on observe une dynamique désordonnée à grande échelle, mais contenant de larges<br />

domaines de cellu<strong>les</strong> dérivantes (dans <strong>les</strong> deux directions symétriques). Les deux<br />

directions d’inclinaison apparaissent comme deux branches séparées. El<strong>les</strong> ne sont<br />

pas en image miroir l’une de l’autre à cause de la désorientation du cristal hors<br />

du plan. Des oscillations préservant la période spatiale de base (notées "1λO") se<br />

superposent, de plus, à la dérive des cellu<strong>les</strong> (Fig. 3.31b).<br />

Dans certains cristaux, dont <strong>les</strong> ang<strong>les</strong> de désorientation dans <strong>et</strong> hors du plan<br />

sont tous deux forts, on observe une décomposition spatio-temporelle en domaines<br />

inclinés <strong>et</strong> domaines axiaux (Fig. 3.9e <strong>et</strong> 3.30"cristal e"). C<strong>et</strong>te dynamique rappelle<br />

celle observée dans <strong>les</strong> eutectiques lamellaires (§4.4). La variation α en fonction<br />

de λ (à l’intérieur des domaines) montre un comportement de bifurcation directe :<br />

l’angle de dérive s’annule, c<strong>et</strong>te fois, pour une valeur seuil, assez bien définie, de λ.<br />

Ceci suggère que, pour de fortes désorientations hors du plan, l’anisotropie capillaire<br />

est dominante. Ce schéma de bifurcation est compatible, du point de vue de la<br />

phénoménologie générale, avec une dynamique de domaines stationnaires. Cependant<br />

le rôle de l’anisotropie est incontestable dans la situation présente.<br />

3.3.3 Questions en suspens<br />

Nous avons montré deux points essentiels, en accord compl<strong>et</strong> avec l’étude numérique<br />

de Kopczynski <strong>et</strong> al. [148] : (i) en SDEM (géométrie 2D), la stabilité des fronts<br />

cellulaires près du seuil dépend très sensiblement de l’anisotropie <strong>et</strong> de l’orientation<br />

du cristal ; (ii) la dynamique spatio-temporelle de ces fronts obéit à une phénoménologie<br />

générale associée aux bifurcations par brisure de symétrie (oscillation à<br />

doublement de période <strong>et</strong> inclinaison).<br />

Il reste certains points obscurs. Nous n’avons pas observé de réseau stationnaire<br />

de doubl<strong>et</strong>s, trouvés numériquement comme une branche possible de solutions stationnaires<br />

[149], <strong>et</strong> observés dans des alliages à base de succinonitrile [120]. Dans nos<br />

expériences, seuls des doubl<strong>et</strong>s isolés ont été observés transitoirement. Deux explications<br />

sont possib<strong>les</strong>. La première serait qu’une anisotropie cinétique ne perm<strong>et</strong>trait<br />

pas la stabilité de tels doubl<strong>et</strong>s. L’autre explication, plus vraisemblable, vient de<br />

l’étroitesse, prédite théoriquement, de l’intervalle de stabilité des doubl<strong>et</strong>s. La question<br />

de la transition cellu<strong>les</strong>-dendrites, c’est-à-dire celle du raccord entre <strong>les</strong> branches<br />

de solutions cellulaires proches du seuil <strong>et</strong> <strong>les</strong> solutions (dendrites ou doublons) à<br />

plus fortes vitesses reste, également, en suspens.<br />

Enfin, en général, des variations de l’espacement intercellulaire subsistent à grande<br />

échelle, alors même que la diffusion de phase est visiblement active pour des modulations<br />

de plus courte longueur d’onde. Cela peut provenir de la formation de<br />

sous-joints de grains, durant le transitoire initial, auxquels s’ancrent <strong>les</strong> fonds de<br />

sillons intercellulaires. Ces sous-joints ont une dynamique propre <strong>et</strong> biaisent (ou<br />

bloquent) le phénomène général de diffusion de la phase. Le processus de formation<br />

des sous-joints de grains est décrit <strong>et</strong> étudié dans la partie suivante.<br />

47


3.4 Polygonisation dynamique en cours de croissance<br />

3.4.1 Position du problème<br />

La formation de défauts de réseau plans –joints <strong>et</strong> sous-joints de grains– est<br />

une des préoccupations majeures en métallurgie d’élaboration. Les propriétés de<br />

plasticité <strong>et</strong> de résistance à la rupture des pièces métalliques en dépendent largement.<br />

Nous considérons ici la polygonisation de monocristaux en cours de solidification,<br />

un phénomène qui a révélé par des métallographies de monocristaux métalliques<br />

solidifiés directionnellement (à partir d’un cristal de référence orienté), mais dont<br />

l’origine est incertaine (voir par ex. [153]) : partant d’un monocristal, on r<strong>et</strong>rouve<br />

un solide constitué d’une mosaïque de p<strong>et</strong>its grains faiblement désorientés.<br />

La première étude expérimentale in situ d’eff<strong>et</strong>s de polygonisation en cours de<br />

croissance (alliages à base d’aluminium) a été faite par Grange <strong>et</strong> al [62], grâce à<br />

l’emploi d’une imagerie en topographie par rayonnement X synchrotron (méthode<br />

de Lang). Dans des échantillons semi-minces (≈ 200µm), en régime de cellu<strong>les</strong> profondes,<br />

on voit des faisceaux de dislocations s’accumuler en fond de sillons intercellulaires,<br />

<strong>et</strong> la formation d’un sous-joint de grains, ou joint de grains de faible<br />

désorientation, au pied d’une cellule. L’origine du mécanisme reste encore mal connu<br />

–certaines hypothèses invoquent l’action d’écoulements dans le liquide [64].<br />

Nous nous intéressons ici à ce qui se produit en échantillons minces en l’absence<br />

de cellu<strong>les</strong>, en régime de front plan (V < V c ). Un mécanisme plausible de polygonisation<br />

serait une instabilité élastico-diffusive (de type Asaro-Tiller-Grinfeld) induite<br />

par un couplage entre des contraintes (d’origine inconnue) <strong>et</strong> la dynamique diffusive<br />

du front [154]. Il semble, d’après nos observations, que ce type d’instabilité,<br />

envisageable dans des matériaux beaucoup moins plastiques que <strong>les</strong> métaux ou nos<br />

matériaux modè<strong>les</strong>, soit précédé d’un autre, appelé ici "polygonisation dynamique".<br />

L’observation centrale est la suivante : après une solidification à V < V c , on observe,<br />

assez régulièrement disposés le long de l’interface, une collection de très p<strong>et</strong>ites "encoches"<br />

(cusps) ou sillons, qu’on interprète comme résultant de l’affleurement de<br />

sous-joints de grains à l’interface (Fig. 3.33). Notre but est d’étayer expérimentalement<br />

<strong>et</strong> théoriquement une interprétation de ce phénomène sur la base d’un couplage<br />

entre dynamique diffusive du front <strong>et</strong> défauts de réseau.<br />

Fig.3.33 – Trois instantanés d’un front de solidification (CBr 4 non dopé ; G = 110 Kcm −1 ). a)<br />

Au repos (V = 0) avant solidification. Un joint de grain est en train de migrer vers la droite. b)<br />

Après 30 min de solidification (V = 6.5 µms −1 ). Cinq sillons peu profonds sont apparus. c) Après<br />

un saut de vitesse vers 20 µms −1 , au-dessus du seuil cellulaire (impur<strong>et</strong>és résiduel<strong>les</strong>). Les sillons<br />

intercellulaires sont associés à un sous-joint (sauf le premier <strong>et</strong> le troisième à partir de la droite).<br />

48


3.4.2 Joints <strong>et</strong> sous-joints de grains<br />

L’existence d’un sillon à l’affleurement d’un joint ou d’un sous-joint de grains<br />

(Fig. 3.34) à l’interface solide-liquide est imposée par la condition d’équilibre le long<br />

de la ligne de jonction (Fig. 3.35) : γ GB /γ = 2sinα GB , où γ GB est la tension de<br />

surface du joint de grains <strong>et</strong> α GB = tan −1 (dζ/dx)| x=0 l’angle d’accrochage (x =<br />

0 est la position du joint de grains) [12]. L’angle α GB est de 90 o pour un joint<br />

de grains de grande désorientation (γ GB = 2γ près de la température de fusion),<br />

qui est dit "fondu", ou "mouillé", car il possède une structure proche du liquide à<br />

l’échelle microscopique [155]. L’analyse de la forme du ménisque associé à un joint<br />

de grains est un moyen classique pour mesurer la tension de surface solide-liquide ;<br />

nous trouvons γ ≈ 8.7 ± 1 mNm −1 , en bon accord avec <strong>les</strong> estimations de la Réf.<br />

[68] (voir aussi [156, 157]).<br />

z (µm)<br />

• GB : joint de grain<br />

• SB : sous-joint<br />

• T : faisceau de dislocations<br />

x (µm)<br />

Fig.3.34 – Profil binarisé de l’interface solide-liquide (micrographie du haut) à V = 0 (C ∞ = 1% ;<br />

G = 44 Kcm −1 ). GB : joint de grains de grande désorientation ; SB : sous-joints de grains ; T :<br />

cuv<strong>et</strong>te due à une modulation continue de la densité de dislocations. La présence des joints de grains<br />

dans le solide près de l’interface est attestée par des lignes faiblement contrastées courant le long<br />

de l’intersection du joint de grains avec <strong>les</strong> plaques de verre, décorée par de fins canaux liquides.<br />

Fig.3.35 – Ménisque d’un (sous-)joint de grains (interfaces isotropes). γ : tension de surface<br />

solide-liquide ; γ GB : tension de surface du joint ; α : angle de raccord ; h : profondeur du ménisque.<br />

En pratique, on distingue <strong>les</strong> sous-joints de grains des joints de grande désorientation<br />

par la faible profondeur h GB du ménisque associé (Fig. 3.34). C<strong>et</strong>te hauteur<br />

est donnée par h GB = √ 2d m sin α GB<br />

2<br />

, avec d m = √ 2a o /G. On a aussi, pour un sous-<br />

49


joint, α SB ≈ γ SB /2γ ≈ √ 2h GB /d m <strong>et</strong> γ SB < 2γ. L’angle en fond de sillon est très<br />

p<strong>et</strong>it quand γ SB est très faible.<br />

En plus de joints <strong>et</strong> sous-joints de grains, on remarque aussi, le long de l’interface<br />

solide-liquide, la présence de légères dépressions, de profondeur submicronique, mais<br />

d’extension latérale comparable à celle du ménisque des joints de grains. El<strong>les</strong> persistent<br />

à l’équilibre, <strong>et</strong> on <strong>les</strong> r<strong>et</strong>rouve après fusion. El<strong>les</strong> sont donc liées à des défauts<br />

du solide, vraisemblablement des faisceaux de dislocations non encore arrangées en<br />

sous-joints. On suppose que la profondeur δζ de la dépression est déterminée par<br />

un accroissement de la densité de dislocations n dislo , <strong>et</strong> on estime la différence de<br />

température δT = Gδζ par δT = δT dislo = ∆E dislo /∆S f , où ∆E dislo est l’accroissement<br />

d’énergie libre du solide dû à la présence des dislocations <strong>et</strong> ∆S f l’entropie de<br />

fusion. Pour un cristal cfc :<br />

∆E dislo /∆S f ≈ G e b 2 n dislo T f /L v , (3.19)<br />

où G e est le module élastique de cisaillement du solide, b le vecteur de Bürgers des<br />

dislocations, T f la température de fusion, <strong>et</strong> L v la chaleur latente de la transition<br />

(par unité de volume). D’après <strong>les</strong> valeurs disponib<strong>les</strong> dans la littérature (métaux<br />

cfc), on peut prendre G e /L v ≈ 25 (l’erreur commise ne peut excéder un facteur 2)<br />

[158]. On sait, pour CBr 4 , que b = 1 < 110 > ≈ 0.634 nm <strong>et</strong> T 2 f ≈ 360 K. Pour<br />

δζ ≈ 0.3 µm <strong>et</strong> G = 40 K, soit δT ≈ 1 mK, on en déduit n dislo ≈ 10 7 cm −2 . C<strong>et</strong>te<br />

valeur excède largement celle, disons ≈ 10 5 cm −2 , que l’on trouve communément dans<br />

<strong>les</strong> cristaux métalliques après solidification (sans toujours en connaître l’origine).<br />

Ici, le mécanisme de multiplication des dislocations nécessaire à ce phénomène reste<br />

inconnu ; pour amorcer une discussion à ce suj<strong>et</strong>, nous donnerons, à la fin de ce<br />

chapitre, un exemple où le cristal subit une déformation plastique imposée.<br />

La structure interne d’un sous-joints de grains est un arrangement régulier de<br />

dislocations, qui perm<strong>et</strong> l’écrantage du champ élastique à grande distance associé aux<br />

dislocations, donc un gain énergétique très sensible. A l’aide de la formule de Read-<br />

Shockley [159], on estime l’ordre de grandeur de la distance inter-dislocation dans<br />

des sous-joints correspondant à une profondeur de ménisque tout juste détectable<br />

(0.3 µm) à environ 0.2 µm (moins d’une centaine de dislocations empilées dans une<br />

l’épaisseur des échantillons minces). On estime aussi que pour créer un arrangement<br />

de sous-joints distants de 50 µm tels que h m ≈ 1 µm, on doit disposer d’une densité<br />

de dislocations (de même signe) d’environ 5 × 10 3 cm −2 , ce qui n’excède pas une<br />

valeur "habituelle". Reste à expliquer le processus de polygonisation lui-même.<br />

3.4.3 Processus de création de sous-joints de grains<br />

Après fusion directionnelle (partielle) rapide d’un échantillon polycristallin, <strong>les</strong><br />

joints de grains préexistants se réarrangent rapidement (leur grande mobilité serait<br />

un suj<strong>et</strong> d’étude en soi). C<strong>et</strong>te recristallisation "efface" <strong>les</strong> sous-joints pré-existants<br />

qui, eux, ont une mobilité réduite. Il n’en subsiste pratiquement pas au moment où<br />

l’on démarre la solidification (Figure 3.33a). On se r<strong>et</strong>rouve, au moment de démarrer<br />

la solidification, avec quelques cristaux séparés de joints de grains distants de<br />

plusieurs 100 µm.<br />

50


8<br />

6<br />

z (µm)<br />

4<br />

2<br />

t=0<br />

t=80s<br />

0<br />

t=160s<br />

t=200<br />

t=280s<br />

!2<br />

t=330s<br />

t=370s<br />

0 50 100 150<br />

x (µm)<br />

Fig.3.36 – Profils d’un front de solidification près<br />

d’un joint de grains, 10 s après le départ (V =<br />

2.15 µms −1 . Les profils successifs ont été décalés<br />

vers le bas de quantités arbitraires. C ∞ = 1%.<br />

G = 44 Kcm −1 .<br />

Fig.3.37 – Déformation stationnaire de<br />

l’interface à proximité d’un joint de grains<br />

pour V < V c . Les défauts de réseau s’accumulent<br />

dans <strong>les</strong> creux en restant perpendiculaires<br />

au front.<br />

En cours de solidification (V < V c ), on observe, à proximité, <strong>et</strong> de part <strong>et</strong> d’autre<br />

des joints de grains qui subsistent, l’apparition de sillons caractéristiques de sousjoints.<br />

Les deux éléments de base du mécanisme de formation de sous-joints sont :<br />

1-une déformation d’origine diffusive du front de solidification au voisinage d’un<br />

joint de grains, <strong>et</strong> 2-une accumulation des défauts de réseau dans <strong>les</strong> creux de c<strong>et</strong>te<br />

déformation. L’interface se déforme d’abord légèrement au voisinage du joint de<br />

grains, prenant l’allure d’une sinusoïde amortie (stationnaire) de pseudo-période<br />

proche de la longueur d’onde critique λ c de l’instabilité cellulaire (Figure 3.36).<br />

Ceci est un eff<strong>et</strong> précurseur connu de l’instabilité cellulaire en dessous de V c induit<br />

par le ménisque [160, 161]. Si l’on arrête la solidification à ce moment, l’interface<br />

r<strong>et</strong>rouve sa forme initiale. Si l’on poursuit la solidification, le premier creux s’amplifie<br />

sensiblement. Si l’on s’arrête, une légère dépression d’amplitude submicrométrique<br />

telle que celle analysée plus haut (Fig. 3.34) subsiste à l’arrêt, ce qui signifie une<br />

accumulation de dislocations (Fig. 3.37). Si l’on poursuit encore, un sillon (donc<br />

un sous-joint) se forme dans le creux-précurseur. Le processus peut se répéter <strong>et</strong><br />

se propager le long de l’interface. On r<strong>et</strong>rouve <strong>les</strong> sillons au même emplacement au<br />

cours d’une fusion.<br />

Pour analyser ce phénomène à partir du calcul de Corriel <strong>et</strong> Sekerka [160] <strong>et</strong><br />

en tenant compte de la présence de défauts de réseau, on peut supposer (c’est vrai<br />

à l’équilibre pour des déformations de grande échelle) que <strong>les</strong> dislocations (ou des<br />

sous-joints "élémentaires", de très faible désorientation) restent perpendiculaires à<br />

l’interface en cours de croissance. Par un argument déjà utilisé par le passé [14, 167]<br />

(cf. §4.3.2), on s’aperçoit que <strong>les</strong> défauts doivent s’accumuler dans <strong>les</strong> creux de la<br />

déformation imposée. La densité de défauts augmente dans le creux, donc le potentiel<br />

chimique du solide aussi, ce qui abaisse sa température d’équilibre avec le liquide.<br />

L’interface recule donc dans <strong>les</strong> creux, qui s’approfondissent encore. On obtient là<br />

clairement un mécanisme d’instabilité. On peut formaliser ce schéma par un calcul<br />

de stabilité linéaire [39]. Pour V proche de V sc < V c , la distance caractéristique entre<br />

<strong>les</strong> sous-joints formés par c<strong>et</strong>te dynamique est proche de λ c .<br />

51


3.4.4 Dynamique spatio-temporelle de la polygonisation<br />

Un diagramme spatio-temporel simplifié comme celui de la Figure 3.38 montre la<br />

structure de "rivières" que laissent <strong>les</strong> sous-joints formés dans le solide. Au moment<br />

de leur création, <strong>les</strong> sous-joints sont bien distants d’une valeur, proche de λ c , qui<br />

dépend peu de V . Cependant, <strong>les</strong> sous-joints ne sont pas rectilignes : <strong>les</strong> sillons<br />

dérivent le long de l’interface <strong>et</strong> peuvent changer de direction. Nous l’expliquons<br />

comme suit.<br />

Fig.3.38 – Trajectoires des sillons de joints <strong>et</strong> sous-joints de grains. Les sauts de vitesse (V =<br />

1.8, 2.15, 2.8 <strong>et</strong> 3.8 µms −1 ) sont marqués. La dernière valeur de V est au-dessus de V c . C ∞ = 8%.<br />

G = 110 Kcm −1 . Flèches : joints de grains préexistants. En haut : front cellulaire à la fin de<br />

l’expérience.<br />

Contrairement aux joints de grains de grande désorientation, <strong>les</strong> sous-joints qui<br />

débouchent à l’interface solide-liquide sont souvent inclinés par rapport à z. Ceci<br />

est dû à l’anisotropie de tension de surface du sous-joint lui-même, qui peut être<br />

très forte, voire singulière (contrairement à celle des joints mouillés). En toute rigueur,<br />

l’équilibre (équation de Young-Herring) en fond de sillon s’écrit avec le vecteur<br />

ˆσ GB = ˆγ GB + ˆγ GB ′, qui est très différent de ˆγ GB en norme <strong>et</strong> direction (c<strong>et</strong> eff<strong>et</strong> de<br />

l’anisotropie est bien plus marqué que pour l’interface solide-liquide, ou pour <strong>les</strong><br />

joints de grains). Comme <strong>les</strong> sous-joints ne sont pas mobi<strong>les</strong>, la dérive d’un sillon<br />

est déterminée par l’inclinaison du sous-joint (donc du vecteur ˆσ GB ) associé. En<br />

cours de solidification, l’inclinaison d’un sous-joint dépend cependant de la vitesse<br />

de solidification (on peut le vérifier par un calcul approché). Enfin, la dérive des<br />

sous-joints <strong>les</strong> amène à se rencontrer entre eux <strong>et</strong> à fusionner, donc à changer de<br />

nature <strong>et</strong> d’anisotropie ; la vitesse de dérive change alors (éventuellement de signe).<br />

3.4.5 Questions ouvertes<br />

Plusieurs questions restent en suspens, notamment celle de l’origine des dislocations<br />

alimentant le processus de polygonisation en solidification directionnelle<br />

[162, 123]. Nous proposons, pour terminer, un exemple, assez spectaculaire, où nous<br />

avons pu provoquer la polygonisation d’un cristal en cours de croissance après une<br />

52


déformation plastique dont on connaît l’origine (sinon l’amplitude). Il s’agit d’une<br />

expérience de croissance en échantillon mince d’un cristal de phase α d’un alliage<br />

dilué CBr 4 -C 2 Cl 6 (1 mol%) par eff<strong>et</strong> Clapeyron. Ordinairement, en croissance libre,<br />

on fait croître un monocristal dans le liquide uniformément sous-refroidi en abaissant<br />

la température de l’ensemble de l’échantillon. Notre dispositif a été modifié<br />

selon un modèle développé dans l’équipe d’A. Buka à Budapest (<strong>et</strong> mis au point à<br />

Paris par T. Börzsönyi) pour faire varier la pression (à température constante) dans<br />

une boîte étanche reliée à une arrivée d’air comprimé de pression variable contenant<br />

l’échantillon. La pression se transm<strong>et</strong> instantanément au liquide (<strong>et</strong> à un p<strong>et</strong>it<br />

cristal préalablement équilibré) si l’échantillon est ouvert, <strong>et</strong> la croissance du solide<br />

démarre sans délai. Ce n’est pas le cas dans <strong>les</strong> échantillons de CBr 4 -C 2 Cl 6 , qui sont<br />

entièrement scellés pour éviter l’évaporation des composants.<br />

Fig.3.39 – Fusion partielle, croissance <strong>et</strong> polygonisation d’un cristal de phase α (CBr 4 -C 2 Cl 6 ) en<br />

croissance libre par eff<strong>et</strong> Clapeyron. a) t = 0 : cristal à l’équilibre ; b) t = 5 s : après application<br />

de la pression (∆P ≈ 0.8 atm), le cristal a diminué de taille (profil binarisé : cristal initial). c)<br />

t = 100 s ; d) t = 780 s (taille finale). Barre : 100 µm.<br />

La Figure 3.39 montre la croissance d’un cristal initialement équilibré avec le<br />

liquide à T ≈ T liq <strong>et</strong> P = P atm après application de la pression (∆P ≈ 0.8 atm).<br />

L’étape finale de croissance est, globalement, conforme à ce qu’on attend. Le cristal<br />

cesse de croître lorsque son rayon a augmenté d’environ 75 µm. On estime, avec<br />

une valeur raisonnable du coefficient de Clapeyron ∆T/δP ≈ 0.02 K/atm (mesurée<br />

pour le succinonitrile dans la ref. [182]), que le sous-refroidissement équivalent est<br />

de 0.02 K. On estime aussi la valeur du gradient thermique résiduel G dans notre<br />

dispositif de croissance libre à G ≈ 0.1 Kmm −1 . L’accroissement de rayon du cristal<br />

devrait alors être d’environ 150 µm, ce qui s’accorde raisonnablement avec la mesure.<br />

En revanche, la fusion partielle observée dans <strong>les</strong> premières secondes (réduction du<br />

rayon δr ≈ 5 µm) <strong>et</strong> la forme finale du cristal sont surprenants. Nous en proposons<br />

l’interprétation suivante. L’augmentation de pression dans la boîte induit, dans <strong>les</strong><br />

premiers instants, un léger déplacement des deux plaques de verre l’une par rapport<br />

à l’autre, <strong>et</strong> fait subir une contrainte de cisaillement au cristal (<strong>et</strong> pas seulement<br />

un chargement hydrostatique). La déformation plastique qui s’ensuit s’accompagne<br />

d’une augmentation de la densité de dislocations qui, d’après l’Eq. 3.19, est de l’ordre<br />

de 10 7 cm −2 –elle est comparable à l’estimation de n dislo pour <strong>les</strong> p<strong>et</strong>ites dépressions<br />

observées en solidification directionnelle. La croissance du cristal après c<strong>et</strong>te étape<br />

indique une réorganisation des défauts dans le cristal, qui conduit, en particulier,<br />

à la formation de sous-joints qui déterminent la forme du cristal à grande échelle<br />

(Figs. 3.39c <strong>et</strong> 3.39d). On ne trouve aucune trace de ces sous-joints, à la fusion (non<br />

montrée), dans la partie centrale du cristal. La formation des sous-joint est donc<br />

53


ien liée à la dynamique de solidification, comme nous pensons l’avoir aussi montré<br />

en solidification directionnelle.<br />

Dans ce processus, <strong>les</strong> parois ont un rôle important qui reste à être précisé dans<br />

<strong>les</strong> expériences de solidification directionnelle. Ce rôle n’est peut-être pas totalement<br />

indépendant de l’épaisseur e des échantillons. L’observation d’une dynamique de<br />

polygonisation en échantillons massifs reste à faire.<br />

3.5 Conclusion<br />

Nous avons montré dans ce chapitre de quelle manière <strong>les</strong> morphologies de solidification<br />

directionnelle des alliages binaires dilués non-fac<strong>et</strong>tés en échantillons minces,<br />

c’est-à-dire en géométrie 2D, dépendent de l’anisotropie des propriétés interfacia<strong>les</strong><br />

–la tension de surface <strong>et</strong> la cinétique interfaciale (linéaire), y compris près du seuil<br />

cellulaire V c . La dendrite, forme de croissance liée à une assez forte anisotropie, n’est<br />

pas la seule forme que l’on puisse observer loin au-dessus de V c . En particulier, quand<br />

l’anisotropie devient faible, on observe une autre forme bien caractérisée, le doublon,<br />

dont l’instabilité vis-à-vis d’un changement de direction de croissance donne lieu à<br />

une dynamique instationnaire à grande échelle (structure en algue). Doublons <strong>et</strong><br />

dendrites peuvent coexister dans un certain intervalle de valeurs des coefficients<br />

d’anisotropie <strong>et</strong> de vitesse de solidification. Ces observations, qui n’avaient jamais<br />

été faites, sont conformes à des résultats théoriques récents, <strong>et</strong> sont corroborées par<br />

le calcul numérique.<br />

La mesure des coefficients d’anisotropie est un élément crucial pour la prédiction<br />

de la forme des microstructures sur le plan quantitatif. Nous en avons donné un<br />

exemple pour l’alliage CBr 4 -C 2 Cl 6 . Nous m<strong>et</strong>tons en valeur l’utilisation des échantillons<br />

minces, qui perm<strong>et</strong>tent d’utiliser la variation de caractéristiques morphologiques<br />

<strong>et</strong> de transitions morphologiques franches pour différencier l’influence de la<br />

tension de surface <strong>et</strong> celle de la cinétique interfaciale en fonction de l’orientation du<br />

cristal. En revanche, la méthode, telle qu’elle est employée actuellement, ne perm<strong>et</strong><br />

pas de connaître avec exactitude l’orientation des cristaux. Notre idée, à terme, est<br />

de réussir à coupler l’observation des morphologies de solidification <strong>et</strong> des mesures,<br />

par exemple par diffraction de rayons X, donnant directement l’orientation des cristaux<br />

(<strong>et</strong> révélant la présence de défauts de réseau). Nous en reparlerons dans le<br />

Chapitre 5. Depuis nos travaux, des efforts expérimentaux <strong>et</strong> numériques (en croissance<br />

libre) ont été faits dans d’autres équipes de recherche pour étudier <strong>les</strong> eff<strong>et</strong>s<br />

de l’anisotropie interfaciale sur <strong>les</strong> morphologies de solidification en géométrie 3D<br />

[108, 109, 145, 146]. Nous ne développerons pas ce suj<strong>et</strong> plus avant.<br />

54


Chapitre 4<br />

Fronts de solidification eutectique<br />

lamellaires en géométrie 2D<br />

4.1 Introduction<br />

La solidification d’un alliage binaire non-fac<strong>et</strong>té dont la concentration se situe à<br />

l’intérieur d’un plateau eutectique du diagramme de phases donne généralement lieu<br />

à la formation de microstructures biphasées dans le solide. La proportion moyenne<br />

des deux solides est donnée par la conservation de la masse, mais leur répartition<br />

spatiale est le résultat d’une dynamique de solidification couplée, où <strong>les</strong> deux phases<br />

solides, notées α <strong>et</strong> β, de composition <strong>et</strong> de structure cristalline différentes, coexistent<br />

en permanence au front <strong>et</strong> croissent simultanément en échangeant du soluté<br />

par diffusion dans le liquide. En solidification directionnelle en container massif, il<br />

existe deux types de structures eutectiques stationnaires de haute symétrie : <strong>les</strong><br />

structures lamellaires –alternance de lamel<strong>les</strong> α <strong>et</strong> β dans une direction– <strong>et</strong> fibreuses<br />

–arrangement de symétrie hexagonale de fibres d’une des deux phases solides dans<br />

une matrice continue de l’autre (Fig. 4.1). On r<strong>et</strong>rouve des structures semblab<strong>les</strong><br />

en convection de Rayleigh-Bénard [30], dans l’instabilité de Faraday [31], <strong>et</strong> dans<br />

l’instabilité des colonnes liquides [163].<br />

Fig.4.1 – Structures stationnaires schématiques de solidification eutectique : a) lamellaire ; b)<br />

fibreuse. α, β : phases cristallines ; z : axe du gradient thermique.<br />

En échantillon mince, on observe des structures dont la périodicité ne peut<br />

s’étendre que dans une seule direction. La symétrie des structures lamellaires y<br />

est respectée, mais <strong>les</strong> lamel<strong>les</strong> sont contraintes (condition de flux diffusif nul aux<br />

parois) de rester perpendiculaires au plan de l’échantillon : leur degré de liberté de<br />

rotation autour de l’axe z est bloqué. Dans <strong>les</strong> alliages transparents, le contraste<br />

55


d’indice optique entre <strong>les</strong> deux phases solides rend visib<strong>les</strong> <strong>les</strong> joints d’interphase,<br />

qui donnent un enregistrement naturel du diagramme spatio-temporel –trajectoires<br />

des lignes de jonction triphasée, ou "trijonctions". La taille λ d’une paire de lamel<strong>les</strong><br />

α-β (la longueur d’onde spatiale d’une structure périodique) est appelée espacement<br />

(inter)lamellaire.<br />

Fig.4.2 – Structure eutectique lamellaire en échantillon mince. Représentation schématique <strong>et</strong><br />

micrographie (CBr 4 -C 2 Cl 6 ; concentration eutectique ; V = 0.5 µm ; G = 80 Kcm −1 .<br />

Les investigations experimenta<strong>les</strong> de la croissance eutectique lamellaire présentées<br />

ici ont pour but principal de répondre à la question de la stabilité des morphologies<br />

stationnaires en fonction de V , G <strong>et</strong> λ, pour un alliage de concentration donnée.<br />

L’étude fondamentale de la croissance eutectique, importante en métallurgie, a commencé<br />

il y a plus d’un demi-siècle [164]. A partir de l’observation systématique de<br />

coupes métallographiques de lingots solidifiés d’alliages eutectiques métalliques, des<br />

questions variées ont été formulées, entre autres, celle (i) de l’influence des eff<strong>et</strong>s<br />

cristallographiques (anisotropie interfaciale) sur la solidification eutectique, celle (ii)<br />

de l’apparente variation de l’espacement interlamellaire moyen λ av en V −1/2 , <strong>et</strong> celle<br />

(iii) de l’origine des grandes variations de la distribution de λ, observées à conditions<br />

de solidification données, à l’intérieur d’un intervalle s’étendant souvent sur plus de<br />

±20% autour de λ av [169, 170]. Notre étude, en conjonction avec cel<strong>les</strong>, par simulations<br />

numériques, de Karma and Sarkissian [166] <strong>et</strong> de Plapp <strong>et</strong> Karma (incluses<br />

dans <strong>les</strong> Réfs. [42, 43]), répond assez complètement à ces deux dernières questions,<br />

<strong>et</strong> à leur contradiction apparente, dans des situations où l’influence de l’anisotropie<br />

interfaciale est faible, voire négligeable, <strong>et</strong> en géométrie 2D.<br />

Faisant suite aux études de SDEM d’alliages eutectiques CBr 4 -C 2 Cl 6 de Trivedi<br />

<strong>et</strong> coll. [16], puis de Faivre <strong>et</strong> coll. [20, 68], nos travaux révèlent l’existence de phénomènes<br />

dynamiques dans <strong>les</strong> fronts eutectiques lamellaires totalement insoupçonnés<br />

auparavant. Ils établissent clairement l’existence <strong>et</strong> la stabilité de fronts eutectiques<br />

lamellaires stationnaires (2D) sur un continuum de valeurs de λ. Contredisant la<br />

conjecture de la sélection de l’espacement lamellaire émise par Jackson <strong>et</strong> Hunt,<br />

<strong>les</strong> expériences (<strong>et</strong> le calcul numérique) montrent que l’état stationnaire de base<br />

est stable, à V fixée, à l’intérieur d’un intervalle de valeurs de λ dont la largeur<br />

dépend de la concentration [40, 43, 165, 166]. La valeur moyenne de l’espacement<br />

<strong>et</strong> la distribution spatiale λ(x) dans un échantillon dépendent des conditions aux<br />

limites <strong>et</strong> de l’histoire de l’expérience. Nos observations perm<strong>et</strong>tent de localiser <strong>les</strong><br />

limites de stabilité des structures de solidification eutectique lamellaires 2D de l’alliage<br />

CBr 4 -C 2 Cl 6 , <strong>et</strong> de déterminer la nature des instabilités correspondantes. Le cas<br />

des échantillons massifs, ainsi que la question de l’influence de l’anisotropie interfa-<br />

56


ciale, proches des préoccupations industriel<strong>les</strong> en métallurgie, seront examinées dans<br />

le Chapitre 5.1.<br />

Le plan de ce chapitre est le suivant. Nous aborderons d’abord le problème de<br />

la stabilité des états stationnaires aux p<strong>et</strong>ites valeurs de l’espacement, ce qui nous<br />

conduira à étudier le phénomène de diffusion de la phase <strong>et</strong> à la mesure directe de coefficients<br />

phénoménologiques associés, <strong>et</strong> à localiser le seuil de l’instabilité d’Eckhaus<br />

(§4.3). Nous montrons que l’instabilité d’élimination de lamel<strong>les</strong> peut se produire de<br />

manière propagative en dessous de ce seuil, ce qui donne un nouvel exemple de<br />

sélection "marginale" de la propagation d’une paroi d’un état stable dans un état<br />

instable. Dans la partie 4.4, nous présenterons le diagramme des morphologies des<br />

eutectiques lamellaires <strong>et</strong> <strong>les</strong> différents types de structures dynamiques localisées associées<br />

aux bifurcations secondaires par brisure de symétrie identifiées. Dans deux<br />

études indépendantes, pour <strong>les</strong>quel<strong>les</strong> nous renvoyons le lecteur à deux artic<strong>les</strong> placés<br />

en Annexe, nous avons détaillé 1-la production de grands grains eutectiques (régions<br />

homogènes du point de vue cristallographique) de très faible anisotropie contenant<br />

plusieurs centaines d’unités de répétition, par un mécanisme d’"invasion" durant<br />

<strong>les</strong> stades initiaux de la croissance eutectique [45], <strong>et</strong> 2-la formation de "cellu<strong>les</strong><br />

eutectiques" (ou "colonies") par eff<strong>et</strong> de troisième constituant (impur<strong>et</strong>é résiduelle).<br />

4.2 Bases théoriques - Calcul de Jackson <strong>et</strong> Hunt<br />

On considère ici un alliage strictement binaire (pas d’impur<strong>et</strong>é résiduelle). La<br />

coexistence des trois phases α-β-liquide au front impose que la température des trijonctions<br />

soit égale à la température eutectique T E , aux eff<strong>et</strong>s de courbure près (Fig.<br />

4.3). L’amplitude de la déformation de l’interface solide-liquide est p<strong>et</strong>ite devant λ<br />

<strong>et</strong> le front de croissance couplée, plan à grande échelle (>> λ), est parallèle à une<br />

isotherme proche de T E . L’équilibre des tensions de surface aux trijonctions (condition<br />

de Young) détermine <strong>les</strong> ang<strong>les</strong> de raccordement des trois interfaces, <strong>et</strong> impose<br />

l’existence d’une courbure moyenne permanente non nulle du front, proportionnelle à<br />

λ −1 . A l’échelle de λ, la déformation des interfaces solide-liquide qu’induit le champ<br />

de diffusion du soluté entre <strong>les</strong> deux phases est d’autant plus forte que l’espacement<br />

est grand. Ces deux phénomènes contribuent au sous-refroidissement moyen<br />

du front ∆T ≡ T E − T av , où T av est la température du front moyennée sur λ. La<br />

variation de ∆T (> 0) en λ, tous <strong>les</strong> autres paramètres étant fixés, passe par un<br />

minimum ∆T m , atteint pour une valeur de l’espacement λ m appelé espacement de<br />

sous-refroidissement minimum (minimum undercooling spacing).<br />

Dans des conditions ordinaires de solidification, 1-<strong>les</strong> valeurs observab<strong>les</strong> de λ<br />

sont très inférieures à la longueur de diffusion l d = D/V (limite des p<strong>et</strong>its nombres<br />

de Pécl<strong>et</strong> : λ/l d


Fig.4.3 – Vue à fort grossissement d’une front eutectique lamellaire. ζ(x) <strong>et</strong> ¯ζ(x) : forme <strong>et</strong><br />

position moyenne du front. La température moyenne du front T av est proche de la température<br />

eutectique T E . Le sous-refroidissement moyen est ∆T = T E − T av > 0.<br />

trope, dans des conditions où <strong>les</strong> eff<strong>et</strong>s de cinétique interfaciale sont négligeab<strong>les</strong>. Les<br />

équations sont <strong>les</strong> mêmes que cel<strong>les</strong> de la solidification monophasée (§2.1.3), auxquel<strong>les</strong><br />

s’ajoute la condition d’Young aux trijonctions –<strong>les</strong> équations de conservation<br />

<strong>et</strong> de Gibbs-Thomson doivent être écrites séparément pour chaque phase solide. La<br />

concentration moyenne de l’alliage C 0 est comprise entre <strong>les</strong> bornes C α <strong>et</strong> C β du<br />

plateau eutectique (cf. Fig. 2.4). Il est pertinent, à la place du paramètre C 0 , d’utiliser<br />

la fraction volumique η de la phase β, qui, dans le cas de CBr 4 -C 2 Cl 6 (parce<br />

que α <strong>et</strong> β ont approximativement le même volume spécifique), est aussi à peu<br />

près égale à C 0−C α<br />

C β −C α<br />

[on note η E (≈ 0.3 pour CBr 4 -C 2 Cl 6 ) sa valeur à l’eutectique].<br />

On appelle alliage hypoeutectique (hypereutectique) un alliage tel que 0 < η < η E<br />

(1 > η > η E ). On trouve que le champ de concentration se compose d’un profil de<br />

diffusion (homogène en x) en exp−z/l d , <strong>et</strong> d’un terme modulé en x à l’échelle de λ.<br />

Au premier ordre d’un développement en λ/l d , <strong>les</strong> concentrations des phases α <strong>et</strong> β<br />

prennent leurs valeurs d’équilibre (C α <strong>et</strong> C β ), ainsi que η. La résolution de l’équation<br />

de Gibbs-Thomson, en tenant compte de la condition d’Young, perm<strong>et</strong> de remonter<br />

à la forme du front. A ce point du calcul, Jackson <strong>et</strong> Hunt supposent, de manière<br />

arbitraire, que <strong>les</strong> interfaces α-liquide <strong>et</strong> β-liquide ont le même sous-refroidissement<br />

moyen ("simplification JH"). Sous c<strong>et</strong>te hypothèse simplificatrice, on trouve :<br />

∆T (λ, V ) = K 1 V λ + K 2 /λ. (4.1)<br />

où K 1 and K 2 ne dépendent que des constantes physiques de l’alliage <strong>et</strong> de η<br />

[16]. On peut récrire c<strong>et</strong>te équation en faisant apparaître explicitement le minimum<br />

(∆T JH , λ JH ) :<br />

∆T (λ, V ) = ∆T JH<br />

2<br />

( λ<br />

+ λ )<br />

JH<br />

λ JH λ<br />

, (4.2)<br />

ce qui donne : {<br />

λJH = √ K 2 /K 1 V −1/2<br />

∆T JH = 2 √ K 1 K 2 V 1/2 .<br />

(4.3)<br />

Pour un alliage donné, λ 2 JHV est une constante. En pratique, l’erreur faite en adoptant<br />

la simplification JH, c’est-à-dire en posant λ JH ≈ λ m <strong>et</strong> ∆T JH ≈ ∆T m est<br />

négligeable, la plupart du temps, par rapport aux erreurs expérimenta<strong>les</strong> [165, 43].<br />

Pour c<strong>et</strong>te raison, nous ne garderons dans la suite qu’une seule notation, λ m . En<br />

toute généralité, la forme du front dépend légèrement de la variable G/V , mais ce<br />

de manière négligeable pour des valeurs habituel<strong>les</strong> (pas trop grandes) de G/V [165].<br />

On établit alors l’existence d’une loi de similitude en vertu de laquelle la forme des<br />

58


structures eutectiques lamellaires ne dépend pas de G, mais d’un paramètre unique<br />

proportionnel à λV 1/2 [165] –habituellement la variable réduite Λ = λ/λ JH ≈ λ/λ m .<br />

C<strong>et</strong>te loi de similitude est particulière aux eutectiques lamellaires –il n’en existe<br />

pas d’équivalent, par exemple, dans <strong>les</strong> fronts cellulaires. L’autre paramètre important,<br />

rappelons-le, est la fraction de phase η. Dans la réf. [165], Kassner <strong>et</strong> Misbah<br />

étendent également la loi de similitude à des valeurs remarquab<strong>les</strong> de l’espacement,<br />

notamment <strong>les</strong> seuils des instabilités secondaires. Notons, finalement, que le calcul<br />

théorique peut se faire sans la simplification JH. Il a été utile de recourir au calcul<br />

non simplifié pour la caractérisation in situ (mesures de constantes physiques) des<br />

alliages eutectiques CBr 4 -C 2 Cl 6 [68] <strong>et</strong> succinonitrile-camphre [48].<br />

4.3 Diffusion de la phase <strong>et</strong> élimination de lamel<strong>les</strong><br />

4.3.1 Mesure expérimentale de la courbe de surfusion<br />

Nous nous intéressons à la dynamique des fronts eutectiques lamellaires en géométrie<br />

2D au voisinage de leur limite de stabilité inférieure, qui correspond à une<br />

instabilité d’élimination de lamel<strong>les</strong>. C<strong>et</strong>te instabilité modifie la valeur de λ moy . Nous<br />

établissons expérimentalement qu’elle est une conséquence ultime d’une instabilité<br />

de type Eckhaus [2], ce qui nous amènera à étudier le phénomène de diffusion de la<br />

phase. Le terme de diffusion de la phase désigne le mode de relaxation d’une modulation<br />

à grande échelle de la distribution de λ vers une valeur uniforme. C’est un<br />

phénomène général, qui s’applique aux structures stationnaires périodiques qui possèdent<br />

un intervalle continu d’espacements stab<strong>les</strong>. Les fronts eutectiques lamellaires<br />

en font partie –la proximité d’une bifurcation primaire n’est pas requise.<br />

Par des arguments qualitatifs, Jackson <strong>et</strong> Hunt ont avancé l’idée que le seuil λ c de<br />

l’instabilité d’élimination de lamel<strong>les</strong> devait coïncider avec λ m . Pour cela, ils ont fait<br />

l’hypothèse (attribuée à J. Cahn) que, sous l’eff<strong>et</strong> d’une modulation à grande échelle,<br />

<strong>les</strong> lamel<strong>les</strong> croissent en restant localement perpendiculaires à l’enveloppe moyenne<br />

du front. Combinée avec l’équation donnant la variation du sous-refroidissement<br />

moyen du front ∆T en fonction de λ (eq. 4.2), on en déduit facilement que toute<br />

valeur de l’espacement inférieure à λ m doit être instable [167, 168].<br />

Nos observations contredisent c<strong>et</strong>te conjecture. Plus précisément, l’analyse de nos<br />

expériences mène à la conclusion que λ c est toujours inférieur à λ m , mais dépend<br />

de la valeur du gradient G, de sorte que λ c ne converge vers λ m que quand le<br />

rapport V/G devient suffisamment grand. Pour arriver à c<strong>et</strong>te conclusion, nous avons<br />

d’abord mesuré la valeur de λ m directement, de la manière qui suit. La Fig. 4.4a<br />

montre un front eutectique lamellaire stationnaire aussi uniforme que possible (<strong>et</strong> la<br />

distribution d’espacement λ(x) associée), obtenu par des méthodes expérimenta<strong>les</strong><br />

qui ne seront pas décrites ici. Ces mêmes méthodes perm<strong>et</strong>tent aussi d’obtenir des<br />

structures quasi stationnaires mais modulées à grande échelle : dans la Fig. 4.4b,<br />

<strong>les</strong> valeurs extrêmes de λ encadrent λ m . Nous mesurons le sous-refroidissement local<br />

en notant que T av (x) = G¯ζ(x) + T 0 , où T av est la température locale du front<br />

moyennée sur une paire de lamel<strong>les</strong> (T 0 est une constante inconnue). En éliminant<br />

x entre T 0 − T av (x) <strong>et</strong> λ(x), nous obtenons <strong>les</strong> points T 0 − T av (λ) de la Figure 4.5<br />

59


Fig.4.4 – Micrographies : Fronts lamellaires eutectiques (CBr 4 -C 2 Cl 6 ). Graphes : Espacement λ<br />

(lignes fines) <strong>et</strong> position moyenne ¯z du front (lignes épaisses) en fonction de x. a) Front stationnaire<br />

(V = 0.5 µms −1 ; G = 80 Kcm −1 ). b) Front modulé (V = 0.25 µms −1 ; G = 48 Kcm −1 ).<br />

[42]. L’ensemble des points présente clairement un minimum pour λ ≈ 27µm. Nous<br />

ajustons alors l’expression T 0 − T av (λ) = ∆T m (λ/λ m + λ m /λ)/2 − ∆T 0 à ces points,<br />

c’est-à-dire l’éq. (4.2) à un terme additionnel constant ∆T 0 = T E −T 0 , en utilisant λ m ,<br />

∆T m <strong>et</strong> ∆T 0 comme paramètres adjustab<strong>les</strong>. Nous avons pu faire de tel<strong>les</strong> mesures<br />

pour V = 0.125−0.5 µms −1 . Nous en déduisons λ 2 mV = K 2 /K 1 = 193±16 µm 3 s −1 <strong>et</strong><br />

∆T 2 m/V = 4K 1 K 2 = (2.7±1.3)×10 −3 K 2 sµm −1 . Les valeurs trouvées antérieurement<br />

[68] sont λ 2 mV = 185±20 µm 3 s −1 <strong>et</strong> ∆T 2 m/V = (1.2±0.5)×10 −3 K 2 sµm −1 . L’accord<br />

est excellent pour λ 2 mV , raisonnable pour ∆T 2 m/V .<br />

L’intervalle de valeurs de λ observées dans l’expérience de la Fig. 4.4b s’étend<br />

bien en deçà de λ m . Ceci indique que le seuil inférieur de stabilité λ c de l’état de<br />

base est inférieur à λ m , ce que nous avons vérifié en étudiant la diffusion de la phase<br />

<strong>et</strong> l’instabilité d’Eckhaus dans ces structures.<br />

2.5<br />

2.0<br />

T o<br />

-T av<br />

(mK)<br />

1.5<br />

1.0<br />

0.5<br />

0.0<br />

15 20 25 30 35 40<br />

λ (µm)<br />

Fig.4.5 – Sous-refroidissement T 0 − T av vs espacement λ (d’après Fig. 4.4b). Ligne épaisse :<br />

ajustement de l’éq.4.2 aux points expérimentaux (cerc<strong>les</strong>). Barre : incertitude. L’intervalle de λ<br />

s’étend jusque vers 0.7λ m (λ m ≈ 27 µm), qui est à peu près le seuil λ c prédit par l’éq. 4.8.<br />

L’ajustement est très bon pour λ < 1.25λ m . Au-delà, l’écart est compatible avec celui qui existe<br />

entre <strong>les</strong> valeurs calculées numériquement <strong>et</strong> l’approximation JH [166, 165].<br />

60


4.3.2 Diffusion de la phase - Instabilité d’Eckhaus<br />

On considère un front de période moyenne λ 0 , perturbé par une modulation de<br />

la distribution d’espacement de vecteur d’onde k tel que kλ 0 /2π


(µm)<br />

26<br />

24<br />

22<br />

A (µm)<br />

1.2<br />

0.8<br />

0.4<br />

0<br />

0 200 400 600 800 1000<br />

t (s)<br />

20<br />

18<br />

t = 0 s<br />

t = 36 s<br />

t = 215 s<br />

t = 535 s<br />

t = 974 s<br />

t = 2115 s<br />

0 100 200 300 400 500 600 700<br />

x (µm)<br />

Fig.4.7 – Micrographie : Front lamellaire modulé à grande échelle. Deux particu<strong>les</strong> responsab<strong>les</strong><br />

de l’élimination de deux lamel<strong>les</strong> ont été emprisonnées dans le solide. C ≈ C E ; G = 80 Kcm −1 ;<br />

V = 0.5 µms −1 ; Λ 0 ≈ 1. Graphe : espacement λ vs x à différents temps. Insert : amplitude A du<br />

mode dominant (145µm) vs t <strong>et</strong> ajustement d’une loi exponentielle A ∼ e −t/τ (τ = 410s).<br />

D E ≈ 1.3×10 −12 m 2 s −1 . Nous avons répété c<strong>et</strong>te analyse dans un cas où λ 0 ≈ 0.88λ m<br />

avec, encore une fois, une valeur positive D E = 7.2 × 10 −13 m 2 s −1 .<br />

Ces résultats nous perm<strong>et</strong>tent de conclure que λ c < λ m . D’après l’ensemble de<br />

nos observations, nous estimons que 0.7 < Λ c < 0.8. L’étude numérique conjointe<br />

(modèle de champ de phase ; alliage modèle symétrique) des Réfs. [42, 43] a confirmé<br />

ces résultats.<br />

Nous sommes conduits, en définitive, à écrire une nouvelle expression du coefficient<br />

de diffusion de la phase sous la forme :<br />

D E = D ⊥ + D ‖ . (4.7)<br />

Cela revient à remplacer l’hypothèse de Langer (eq. 4.5) sur le mouvement des trijonctions<br />

par ∂ t y(x, t) = −V ∂ x ¯ζ(x, t) + D‖ ∂ x λ(x, t)/λ 0 , où D ‖ est une constante<br />

qui a effectivement la dimension d’un coefficient de diffusion. La valeur de D ‖ est<br />

extraite des mesures expérimenta<strong>les</strong> <strong>et</strong> numériques en soustrayant l’expression théorique<br />

de D ⊥ (éq. 4.6) de la valeur mesurée de D E . Les résultats du calcul numérique<br />

perm<strong>et</strong>tent de conclure que le rapport D ‖ /(V λ 0 ) varie approximativement comme<br />

Λ 0 , mais est indépendant de G and V . On peut écrire simplement D ‖ sous la forme<br />

D ‖ /(V λ 0 ) ≈ A Λ 0 , où A est une constante empirique. En prenant A = 0.15, nous<br />

avons vérifié que c<strong>et</strong>te expression de D ‖ prédit bien <strong>les</strong> valeurs extraites des calculs<br />

<strong>et</strong> des expériences, bien que <strong>les</strong> alliages <strong>et</strong> <strong>les</strong> paramètres de contrôle utilisés soient<br />

différents. La valeur de A ne semble donc pas dépendre très sensiblement du système<br />

ni des conditions de solidification.<br />

On trouve une expression de la valeur Λ c = λ c /λ m de l’espacement réduit au<br />

seuil de l’instabilité d’Eckhaus en écrivant D E = D ⊥ + D ‖ = 0 en combinaison avec<br />

l’expression 4.3 de λ m . On trouve une équation cubique :<br />

1 − 1 Λ 2 c<br />

+ AG<br />

K 1 V Λ c = 0, (4.8)<br />

qui montre clairement que Λ c dépend du rapport G/V . Nous avons vérifié la pertinence<br />

de c<strong>et</strong>te loi semi-empirique dans de nombreuses expériences (par exemple, la<br />

valeur la plus basse de l’espacement observée dans la Fig. 4.5 se place juste au-dessus<br />

62


de la valeur prédite du seuil). Il est tentant aussi de se reporter aux mesures faites<br />

dans <strong>les</strong> lingots métalliques par Trivedi <strong>et</strong> coll. [170] <strong>et</strong> Lesoult <strong>et</strong> coll. [169] qui ont,<br />

<strong>les</strong> premières, attiré l’attention sur l’existence d’un intervalle continu de valeurs observab<strong>les</strong><br />

de l’espacement à V donnée, dont <strong>les</strong> bornes ne suivent pas la loi en λ 2 V .<br />

Malheureusement, la comparaison entre ces résultats (obtenus dans des échantillons<br />

massifs) <strong>et</strong> <strong>les</strong> nôtres est rendue difficile par la grande différence des paramètres (le<br />

rapport V/G dans <strong>les</strong> expériences métallurgiques est souvent d’un ordre de grandeur<br />

plus fort que dans nos études). Enfin, deux études théoriques détaillées des<br />

Réfs [171, 172] montrent bien un eff<strong>et</strong> stabilisant du gradient de température sur<br />

<strong>les</strong> structures eutectiques lamellaires, mais dans des approximations tel<strong>les</strong> que la<br />

différence entre λ m <strong>et</strong> λ c qui est trouvée est négligeable pour <strong>les</strong> valeurs standard de<br />

G.<br />

En conclusion, nous avons mis en évidence la stabilité des eutectiques lamellaires<br />

pour des valeurs de l’espacement inférieures à l’espacement de sous-refroidissement<br />

minimum. Ces structures obéissent comme prévu au phénomène de diffusion de la<br />

phase pour des perturbations de grande longueur d’onde. Le résultat nouveau est<br />

que le coefficient de diffusion de la phase ne se déduit pas de la conjecture de croissance<br />

perpendiculaire. Les calculs numériques montrent que le seuil λ c de l’instabilité<br />

d’Eckhaus, approximativement égal à 0.7λ m dans <strong>les</strong> conditions ordinaires de nos<br />

expériences de SDEM d’alliages CBr 4 -C 2 Cl 6 , décroît quand le gradient de température<br />

augmente. Il est important de noter que nous avons mené une étude semblable,<br />

<strong>et</strong> concordante, sur un autre alliage eutectique binaire non-fac<strong>et</strong>té, le succinonitriled,camphre<br />

(SCN-DC) [48].<br />

4.3.3 Elimination de lamel<strong>les</strong><br />

En dessous du seuil d’Eckhaus, on observe des éliminations de lamel<strong>les</strong> successives.<br />

Lorsque l’écart au seuil Λ c − Λ 0 est grand, on observe l’élimination d’un grand<br />

nombre de lamel<strong>les</strong> (plus d’une sur deux) dans un temps bref. Nous nous attacherons<br />

ici au régime observé à faible "trempe", c’est-à-dire lorsque Λ c − Λ 0 est assez<br />

faible (Λ 0 > 0.6). La procédure expérimentale consiste à créer un front eutectique<br />

lamellaire avec une distribution d’espacement λ i (x) aussi homogène que possible à<br />

une vitesse V 1 (en pratique, 2 < V 1 < 3µms −1 ), puis de diminuer la vitesse vers<br />

une valeur V 2 (< 1µms −1 ) telle que λ i < λ c (V 2 ). Dans l’expérience de la Fig. 4.8,<br />

on peut voir que, pendant un transitoire après le saut de vitesse, la front reste à<br />

peu près stationnaire. Puis une première élimination survient, à partir de laquelle<br />

l’instabilité se propage dans la partie instable du front. Les éliminations successives<br />

laissent derrière el<strong>les</strong> un front lamellaire d’espacement plus grand, dans l’intervalle<br />

de stabilité.<br />

Des informations plus quantitatives peuvent être obtenues en étudiant l’évolution<br />

de la distribution d’espacement λ(x). En bref, la première élimination se produit à<br />

l’endroit où la valeur de λ c − λ est maximum (Fig. 4.9). C<strong>et</strong> événement est luimême<br />

précédé d’une modulation spatiale de l’espacement à p<strong>et</strong>ite échelle (de 2,8λ 0<br />

à 3.2λ 0 dans <strong>les</strong> cas étudiés) qui se développe spontanément <strong>et</strong> s’amplifie sur place<br />

jusqu’à l’élimination d’une lamelle. C<strong>et</strong>te structure survit en se propageant dans<br />

63


Fig.4.8 – A gauche : Image binarisée (faible grossissement) d’un grain eutectique, équivalente à un<br />

diagramme spatio-temporel (temps vers le haut). C ≈ C E . Pour plus de clarté, seule une interface<br />

α-β sur deux est affichée. Le saut de vitesse initial va de 2 à 0.75 µms −1 (Λ 0 ≈ 0.6). Dimension<br />

horizontale : 985 µm. L’échelle verticale est contractée d’un facteur 0.3. A droite : même type<br />

d’image obtenue par calcul numérique (M. Plapp).<br />

la région instable du front. Ainsi, à un temps donné, le front se divise en deux<br />

régions, l’une instable, l’autre stable, séparées par une "paroi" complexe faite de<br />

l’oscillation spatiale amortie qui déclenche l’élimination de lamelle à interval<strong>les</strong> de<br />

temps réguliers.<br />

20<br />

18<br />

t = 1600 s<br />

t = 2200 s<br />

t = 2800 s<br />

14<br />

16<br />

12<br />

λ (µm)<br />

14<br />

λ (µm)<br />

10<br />

16<br />

12<br />

8<br />

12<br />

10<br />

0 400 800 1200 1600<br />

x (µm)<br />

8<br />

600 700 800 900<br />

6<br />

600 650 700 750 800 850 900 950 1000<br />

x (µm)<br />

Fig.4.9 – Micrographie : Elimination propagative de lamel<strong>les</strong> après un saut de vitesse de 2.25 à<br />

0.5 µms −1 . Au milieu : distribution λ(x) à t = 0 (saut de vitesse), à t = 2200 s (amplification<br />

d’une oscillation spatiale), <strong>et</strong> à t = 2800 s (fin de l’expérience). A droite : Evolution de λ(x) en<br />

fonction du temps près de la première élimination (temps entre deux courbes : 200 s) ; insert :<br />

ajustement d’une sinusoïde avec amortissement exponentiel de pseudo période ≈ 3.2λ 0 .<br />

Ces caractéristiques –oscillation précurseur, élimination de lamel<strong>les</strong>, propagation<br />

régulière d’un front– sont bien reproduites par le calcul numérique, à partir d’un<br />

état initial parfaitement uniforme avec une perturbation initiale localisée (Fig. 4.8).<br />

La propagation d’un front à vitesse constante, apparemment sélectionnée, lors de la<br />

pénétration d’un état stable dans un état instable peut être qualitativement décrite<br />

par un critère de stabilité marginale [173, 174]. Enfin, nous constatons que ce scénario<br />

d’invasion, qui peut démarrer à différents endroits de l’échantillon plus ou moins<br />

simultanément, délivre une distribution d’espacement finale qui n’est pas régulière<br />

<strong>et</strong> dépend des conditions initia<strong>les</strong>. En général, cependant, la valeur finale moyenne<br />

de λ est très proche de λ c . Ceci doit provenir de la dépendance de la fréquence des<br />

éliminations de lamel<strong>les</strong> en fonction de la distance initiale au seuil.<br />

64


4.4 Instabilités secondaires des fronts eutectiques<br />

lamellaires<br />

4.4.1 Diagramme des morphologies<br />

L’étude de la dynamique des morphologies à symétrie brisée des fronts eutectiques<br />

lamellaires en géométrie 2D a fait l’obj<strong>et</strong> de deux thèses successives au laboratoire<br />

–celle de J. Mergy (1993), puis celle de M. Ginibre (1997). Nous en faisons<br />

ici un bref compte-rendu. La première instabilité secondaire identifiée en SDEM<br />

d’alliages eutectiques CBr 4 -C 2 Cl 6 a été l’instabilité de dérive, ou d’inclinaison (tilt).<br />

Depuis, nous avons identifié au moins cinq autres types de morphologies issues d’instabilités<br />

secondaires (bifurcations homogènes) par brisure de symétrie du côté des<br />

fortes valeurs de Λ (Figs. 4.10 à 4.12).<br />

Fig.4.10 – Oscillation d’oscillation à doublement de période (2λO). A gauche : vue à grande<br />

échelle. A droite : détails pour un échantillon hyper- (en haut) <strong>et</strong> hypoeutectique (en bas).<br />

Fig.4.11 – Instabilités d’oscillation à période spatiale préservée (1λO), <strong>et</strong> d’inclinaison (T).<br />

Fig.4.12 – Morphologies "mixtes" : 1λO-T, 2λO-1λO-T.<br />

Ces morphologies sont obtenues expérimentalement à partir de l’état stationnaire<br />

en effectuant des sauts de vitesse vers le haut. A nombre de lamel<strong>les</strong>, donc à<br />

65


λ av , constants (le mécanisme de branchement est bloqué), augmenter V revient à<br />

augmenter Λ d’un même facteur (loi de similitude). Nos observations ont conduit à<br />

l’établissement d’un diagramme des morphologies eutectiques lamellaires (Fig. 4.13)<br />

dans <strong>les</strong> coordonnées (η,Λ). En parallèle, Karma <strong>et</strong> Sarkissian ont construit ce même<br />

diagramme par des simulations numériques [166].<br />

Fig.4.13 – Diagramme (η,Λ) des morphologies de solidification eutectique lamellaires (CBr 4 -<br />

C 2 Cl 6 ; η E = 0.3). A gauche : calculs numériques de Karma <strong>et</strong> Sarkissian [166]. Régions hachurées<br />

: états stab<strong>les</strong> (un type de hachure est associé à chaque mode élémentaire de brisure de<br />

symétrie ; voir cartouche). Superposition de deux types de hachure : états résultant de l’association<br />

de deux modes élémentaires. Lignes en trait continu : seuils de bifurcation. Ligne en tir<strong>et</strong>és : valeur<br />

calculée (différente de λ JH ) de l’espacement de sous-refroidissement minimum. La limite de<br />

stabilité inférieure (éliminations) n’est pas tracée. Dans <strong>les</strong> régions laissées blanches –mis à part<br />

le domaine de stabilité des états stationnaires– aucun état permanent n’a été trouvé. A droite :<br />

Données expérimenta<strong>les</strong> (<strong>les</strong> lignes sont le calque des résultats numériques).<br />

On constate l’excellent accord entre <strong>les</strong> expériences <strong>et</strong> le calcul. Par exemple,<br />

au voisinage de la concentration eutectique, on observe dans <strong>les</strong> deux cas la succession,<br />

en faisant croître Λ, des états stationnaires, puis 1λO, puis mixtes 1λO-2λO-T.<br />

Pour une fraction de phase proche de 0.5, on observe, toujours à Λ croissant, des<br />

états stationnaires, 2λO, 2λO-T, T, <strong>et</strong> enfin 1λO-T. L’accord n’est pas seulement<br />

qualitatif. On pourrait d’ailleurs faire coïncider <strong>les</strong> données expérimenta<strong>les</strong> <strong>et</strong> numériques<br />

par une simple dilatation de ces dernières d’environ 12% sur l’échelle des Λ,<br />

ce qui est compatible avec l’incertitude expérimentale sur λ m . Des différences apparentes<br />

comme le chevauchement de certains domaines de stabilité contigus peuvent<br />

s’expliquer, entre autres, par des erreurs sur la mesure de η. Le calcul numérique<br />

ne révèle l’existence d’une bifurcation sous-critique que dans le cas de la bifurcation<br />

2λO, avec un domaine de bistabilité bien plus étroit que ce que <strong>les</strong> données<br />

expérimenta<strong>les</strong> laissent supposer.<br />

66


4.4.2 Défauts dynamiques<br />

Au voisinage des bifurcations associées aux limites de stabilité du diagramme de<br />

morphologies, on peut observer la formation de défauts dynamiques, se présentant<br />

souvent sous la forme d’obj<strong>et</strong>s localisés, dont l’existence est également prévue, dans<br />

le cadre de la phénoménologie générale des instabilités secondaires, par l’analyse des<br />

équations d’amplitude pertinentes. Un premier phénomène remarquable, associé à<br />

la bifurcation d’inclinaison, est la décomposition spatiale d’un front eutectique en<br />

domaines d’inclinaison (Fig. 4.14). C<strong>et</strong>te décomposition spatiale est le résultat d’une<br />

instabilité générique de la structure inclinée homogène près du seuil : on n’observe<br />

jamais (sauf transitoirement) de structures à p<strong>et</strong>it angle d’inclinaison (voir, par<br />

exemple, [175]). Les parois des domaines d’inclinaison dérivent à vitesse constante,<br />

dans le sens opposé à l’inclinaison elle-même (conservation du nombre de lamel<strong>les</strong>).<br />

On vérifie une prédiction théorique de Coull<strong>et</strong> <strong>et</strong> al [34], selon laquelle la dérive de<br />

ces parois stationnaires entraîne un phénomène (le seul de ce type connu à ce jour)<br />

de sélection dynamique de la longueur d’onde (thèse de J. Mergy [20]).<br />

Fig.4.14 – Domaines d’inclinaison dans un front eutectique lamellaire (SDEM ; CBr 4 -C 2 Cl 6 ).<br />

Fig.4.15 – Défauts de phase dans l’oscillation 2λO : sauts de phase stationnaires (à gauche) <strong>et</strong><br />

"bouffée" d’oscillation (à droite).<br />

Dans <strong>les</strong> structures oscillantes de type 2λO, on observe des "défauts de phase".<br />

Des structures d’apparence homogène (Fig. 4.10) possèdent souvent une modulation<br />

continue à grande échelle de la phase temporelle des oscillations. Dans certains<br />

cas, le déphasage se concentre dans des sauts de phase localisés stationnaires (Fig.<br />

4.15). Deux autres types de défauts associés à la bifurcation 2λO peuvent apparaître<br />

au sein d’une structure stationnaire, légèrement en dessous du seuil de l’oscillation<br />

λ osc : 1-des "bouffées" d’oscillation (Fig. 4.15), similaires à un type de défaut (compatible<br />

avec le caractère légèrement sous-critique de la bifurcation) prédit dans un<br />

modèle phénoménologique par Fauve <strong>et</strong> Thual [176] ; 2-des ondes solitaires localisées<br />

sur deux paires de lamel<strong>les</strong> (Fig. 4.16), qui dérivent à vitesse constante (dans un<br />

environnement uniforme). Ces trois types de défauts (sauts de phase, "bouffées" <strong>et</strong><br />

ondes solitaires) sont observés dans des systèmes expérimentaux variés, par exemple<br />

67


<strong>les</strong> allées de colonnes liquides. Ils ont été trouvés à partir d’équations d’amplitude<br />

par L. Gil.<br />

Fig.4.16 – Onde solitaire. Dimension horizontale : 2.1 mm (l’insert est à l’échelle double).<br />

Les ondes solitaires, remarquablement stab<strong>les</strong>, peuvent balayer le front sur de<br />

grandes distances. La distribution laissée à l’arrière est plus uniforme. Produites en<br />

grand nombre, el<strong>les</strong> peuvent être à l’origine de régimes très désordonnés, en particulier<br />

pour des valeurs de la concentration proche d’un bord du plateau eutectique (λ osc<br />

décroît <strong>et</strong> se rapproche de la limite d’Eckhaus). On observe alors un certain type<br />

de dynamique instationnaire (Fig. 4.17) caractérisé par l’apparition de régions "turbulentes",<br />

avec de nombreuses éliminations <strong>et</strong> créations de lamel<strong>les</strong>, qui se ferment<br />

lorsqu’on réduit la vitesse. Ce type de dynamique rappelle certains régimes d’intermittence<br />

spatio-temporelle dominés par la propagation d’ondes solitaires tels que<br />

ceux mis en évidence, par exemple par van Hecke (voir [177] <strong>et</strong> références incluses).<br />

L’étude détaillée de ces régimes désordonnés reste à faire.<br />

4.5 Conclusion <strong>et</strong> questions ouvertes<br />

En conclusion, nous avons mis en évidence expérimentalement une grande variété<br />

de phénomènes dynamiques dans <strong>les</strong> fronts de solidification eutectique lamellaires<br />

en SDEM d’un alliage non-fac<strong>et</strong>té (CBr 4 -C 2 Cl 6 ). Nous avons dressé un diagramme<br />

des morphologies stationnaires <strong>et</strong> permanentes en géométrie 2D. Le très bon accord<br />

quantitatif entre observations expérimenta<strong>les</strong> <strong>et</strong> calcul numérique sans paramètre<br />

ajustable perm<strong>et</strong> de valider <strong>les</strong> hypothèses de base de la théorie de la solidification<br />

eutectique. Nos résultats s’analysent clairement dans le cadre général de la physique<br />

non-linéaire des systèmes conduisant à la formation de structures hors d’équilibre,<br />

68


Fig.4.17 – Vue à grande échelle d’une dynamique de type intermittence spatio-temporelle, dominée<br />

par des ondes solitaires. Echantillon hypoeutectique (η ≈ 0.2). Le bas correspond au moment d’un<br />

saut de vitesse vers le haut à partir d’une structure 2λO. On a fait un saut de vitesse vers le bas<br />

pour fermer <strong>les</strong> zones turbulentes.<br />

y compris la formation de structures localisées (fronts d’éliminations de lamel<strong>les</strong>,<br />

défauts de phase, parois de domaines, ondes solitaires). Des zones d’ombre subsistent<br />

encore naturellement, <strong>et</strong> certaines études, en prolongement de cel<strong>les</strong> qui ont été<br />

exposées, peuvent être envisagées (<strong>les</strong> suj<strong>et</strong>s de recherche à plus long terme, comme<br />

la question de l’influence de l’anisotropie interfaciale dans la formation des structures<br />

eutectiques <strong>et</strong> celle de la solidification eutectique en géométrie 3D sont abordées dans<br />

le chapitre suivant) :<br />

1-Oscillations géantes. L’hypothèse du front plan devient irréaliste dans le cas des<br />

"oscillations géantes", ainsi nommées parce qu’el<strong>les</strong> correspondent à structures oscillantes<br />

où l’excursion latérale des trijonctions est à peu près égale à λ. El<strong>les</strong> sont<br />

observées en SDEM pour de faib<strong>les</strong> vitesses (de l’ordre de 0.1 µms −1 ) <strong>et</strong> pour de<br />

fortes valeurs de λ (de l’ordre de 0.1 à 1 mm), proches de la longueur de diffusion l d<br />

L’oscillation s’accompagne ou non d’un doublement de période spatiale. Elle conduit<br />

à des régimes spatio-temporels très désordonnés (Fig. 4.18). Des simulations numériques<br />

non publiées de R. Folch <strong>et</strong> M. Plapp ont montré la possibilité de régimes<br />

chaotiques. Ceci reste à prouver expérimentalement.<br />

2-Transition eutectique/dendrite. L’apparition de dendrites de la phase majoritaire<br />

noyées dans la structure eutectique est très importante du point de vue industriel.<br />

Elle peut être observée dans des alliages de concentration proche d’un bord du plateau<br />

eutectique. Elle est aussi favorisée par de p<strong>et</strong>ites valeurs du paramètre G/V<br />

(selon un critère de type "surfusion constitutionnelle") [178]. Nos observations préliminaires<br />

semblent montrer que l’apparition des dendrites à partir d’un front eutectique<br />

présente une forte hystérésis (en échantillons minces). Ce phénomène reste<br />

cependant encore mal connu –<strong>et</strong> l’on s’attend à de grandes différences entre systèmes<br />

2D <strong>et</strong> 3D.<br />

3-Alliages eutectiques ternaires. En présence d’une impur<strong>et</strong>é résiduelle (en très faible<br />

quantité), il se forme des cellu<strong>les</strong> eutectiques ou "colonies" (Fig. 4.19). Nous avons<br />

69


Fig.4.18 – Oscillations géantes. SDEM (V = 0.19 µms −1 ) d’un alliage eutectique CBr 4 -C 2 Cl 6<br />

(interfaces α − β binarisées). Dimension horizontale : ≈ 2 mm.<br />

Fig.4.19 – Cellu<strong>les</strong> eutectiques (à gauche) <strong>et</strong> "doigts biphasés"<br />

(à droite, dimension horizontale : 190 µm) ; SDEM<br />

d’un alliage CBr 4 -C 2 Cl 6 eutectique avec impur<strong>et</strong>és résiduel<strong>les</strong><br />

[44].<br />

Fig.4.20 – Structure stationnaire<br />

de SDEM (V = 0.014 µms −1 ) d’un<br />

alliage SCN-DC-NPG(-AMPD) ;<br />

barre : 20µm.<br />

étudié en détail <strong>les</strong> stades initiaux de ce phénomène, comparable à une instabilité<br />

cellulaire à grande échelle d’un front "effectif" structuré à p<strong>et</strong>ite échelle (celle de<br />

λ) [44]. La dynamique à grande échelle des cellu<strong>les</strong> eutectiques <strong>et</strong> celle des "doigts<br />

biphasés" (Fig. 4.19), qui présentent une certaine similarité avec le doublon en solidification<br />

monophasée, restent à étudier. Dans des alliages ternaires concentrés, en<br />

se plaçant au voisinage de la composition d’un point invariant, le solide est triphasé.<br />

Une structure stationnaire de type ABACABAC (Fig. 4.20) a été observée dans<br />

des expériences préliminaires en SDEM d’un alliage pseudo-ternaire succinonitriled,Camphre-Néopentylglycol(-AMPD)<br />

[179]. L’étude des stades initiaux de la formation<br />

de c<strong>et</strong>te structure <strong>et</strong> de sa stabilité n’a pas été faite. Il serait intéressant aussi<br />

d’étudier la solidification d’alliages binaires péritectiques (voir [180]) ou monotectiques.<br />

70


Chapitre 5<br />

Etudes en cours - Proj<strong>et</strong>s de<br />

recherche<br />

5.1 Observation directe des fronts de solidification<br />

eutectique en échantillons massifs<br />

Résultats récents<br />

L’une des avancées expérimenta<strong>les</strong> récentes dans le domaine des microstructures<br />

de solidification concerne l’étude par observation directe des fronts de solidification<br />

directionnelle en containers massifs. Ce suj<strong>et</strong> présente de grandes difficultés théoriques<br />

<strong>et</strong> expérimenta<strong>les</strong>. En géométrie tridimensionnelle (3D), la convection dans le<br />

liquide, bloquée en échantillon mince, peut perturber la dynamique diffusive à plus<br />

ou moins grande échelle. On peut aussi s’attendre à une influence plus prononcée<br />

de l’anisotropie interfaciale. Ces eff<strong>et</strong>s sont importants notamment d’un point de<br />

vue industriel. Plus fondamentalement, la pléthore <strong>et</strong> la complexité des phénomènes<br />

dynamiques que l’on découvre est liée au grand nombre de degrés de liberté, en<br />

particulier la rotation autour de l’axe du gradient thermique, qui perm<strong>et</strong> l’existence<br />

de plusieurs types de microstructures stationnaires de degré de symétrie plus ou<br />

moins élevé (des arrangements de symétrie hexagonale aux structures lamellaires<br />

ou "en rouleaux"), <strong>et</strong> la production de certains défauts topologiques inexistants en<br />

géométrie 2D [2, 29].<br />

Nous nous intéressons à la solidification eutectique. Nous avons commencé à<br />

m<strong>et</strong>tre au point il y a quelques années un dispositif original (Fig. 5.1) qui perm<strong>et</strong><br />

l’observation directe du front de solidification en contraste de fond noir, en visée<br />

oblique –angle θ– à travers le liquide <strong>et</strong> une paroi en verre de l’échantillon, au moyen<br />

d’un microscope optique longue-distance (Questar QM100 ; distance de travail : 15<br />

cm). Ceci n’avait jamais été réalisé auparavant. L’appareil, opérationnel en 2003, a<br />

subi depuis de nombreuses améliorations [49]. Les échantillons semi-épais à parois de<br />

verre planes sont de section utile rectangulaire (5×0.4 mm 2 ). Le banc de solidification<br />

est de conception similaire à ceux utilisés pour <strong>les</strong> échantillons minces. Il est placé en<br />

position verticale, le bloc chaud en haut, pour éviter la convection thermique. Nous<br />

avons utilisé l’alliage CBr 4 -C 2 Cl 6 à différentes concentrations (proches de l’eutectique)<br />

pour l’étude des structures lamellaires, <strong>et</strong> l’alliage eutectique succinonitrile-<br />

71


d,camphre (SCN-DC) pour celle des structures fibreuses. Dans <strong>les</strong> deux cas, nous<br />

avons veillé à rester dans des conditions stab<strong>les</strong> vis-à-vis de la convection thermosolutale.<br />

La partie optique <strong>et</strong> l’expérience de solidification sont découplées (méthode<br />

non-invasive) –<strong>les</strong> appareils développés dans d’autres groupes pour l’observation de<br />

fronts cellulaires <strong>et</strong> dendritiques sont, eux, de type endoscopique à optique (semi-)<br />

immergée [183, 184].<br />

Fig.5.1 – Principe de l’observation oblique des fronts de solidification eutectique (géométrie 3D).<br />

Une analyse des performances optiques de notre dispositif conduit aux conclusions<br />

suivantes. Du fait de la réfraction des rayons lumineux dans le liquide, l’image<br />

est contractée d’un facteur f dans la direction y, qui adm<strong>et</strong> un maximum pour une<br />

valeur optimale θ ⊥ (dépendant de l’indice optique du liquide). On vérifie de plus<br />

qu’à l’angle θ ⊥ , l’image du front de solidification se situe dans le plan frontal du<br />

microscope : l’image est alors entièrement au point. Nous travaillons donc toujours<br />

au voisinage de θ = θ ⊥ . Pour <strong>les</strong> deux alliages (eutectiques) CBr 4 -C 2 Cl 6 <strong>et</strong> SCN-DC,<br />

on trouve θ ⊥ ≈ 50˚<strong>et</strong> f ≈ 0.4. Un fort contraste lumineux –d’origine "chimique"–<br />

entre <strong>les</strong> deux phases solides est, par ailleurs, obtenu en inclinant la source de lumière<br />

selon un certain angle θ L , différent de θ, qui perm<strong>et</strong> l’extinction de l’une des<br />

phases solide, <strong>et</strong> que l’on peut calculer connaissant la valeur des indices optiques des<br />

trois phases en présence (le liquide <strong>et</strong> <strong>les</strong> deux solides). Ceci nécessite la réduction de<br />

l’angle d’ouverture du microscope par un diaphragme à iris, ce qui perm<strong>et</strong> du même<br />

coup de limiter <strong>les</strong> aberrations d’astigmatisme. En pratique, l’image des fibres <strong>et</strong> des<br />

lamel<strong>les</strong> est celle d’une caustique liée, outre ce contraste chimique, à la topographie<br />

(courbure des interfaces) du front. La résolution spatiale effective du dispositif pour<br />

ce type d’images est d’environ 3 µm.<br />

Avec c<strong>et</strong> appareil (appelé DIRSOL dans le cadre du programme d’expérience en<br />

micropesanteur de l’ESA), nous avons obtenu <strong>les</strong> résultats suivants :<br />

-Bifurcation "zigzag" des eutectiques lamellaires. La bifurcation zigzag est liée, dans<br />

<strong>les</strong> systèmes à structuration hors équilibre en géométrie 3D, au phénomène de diffusion<br />

de la phase [2]. En toute généralité, un front périodique de vecteur d’onde<br />

k 0 (<strong>les</strong> caractères gras sont des vecteurs) de norme 2π/λ 0 parallèle à l’axe x est représenté<br />

par U 0 (k 0 .r), où U 0 est une fonction 2π-périodique <strong>et</strong> r le vecteur position.<br />

Le front perturbé est écrit sous la forme U(r, t) = U 0 [φ(r, t)], où φ(r, t) est appelée<br />

fonction de phase. La valeur locale de l’espacement, supposée lentement variable en<br />

temps <strong>et</strong> espace, est donnée par ∇φ(r, t) = k(r, t), où k = 2π/λ. Les variations de<br />

la forme du front (à p<strong>et</strong>ite échelle) associées en réalité à la modulation de λ ne sont<br />

72


Fig.5.2 – Structures de solidification eutectique d’alliages CBr 4 -<br />

C 2 Cl 6 en échantillons d’épaisseur 400 µm (hauteur des images) :<br />

structures lamellaire symétrique <strong>et</strong> "zigzag" (on distingue une ligne<br />

de faute en bas à droite).<br />

Fig.5.3 – Structure en labyrinthe<br />

<strong>et</strong> étirement d’une structure<br />

lamellaire symétrique par<br />

eff<strong>et</strong> de biais thermique.<br />

pas prises en compte, <strong>et</strong> la dynamique est décrite par une seule équation qui s’écrit,<br />

au premier ordre :<br />

∂φ<br />

∂t = D E(k 0 ) ∂2 φ<br />

∂x + D Z(k 2 0 ) ∂2 φ<br />

, (5.1)<br />

∂y 2<br />

où D E and D Z sont <strong>les</strong> coefficients de diffusion de la phase. La modulation relaxe,<br />

donc le front est stable, dans l’intervalle de k où ces deux coefficients sont positifs.<br />

L’instabilité d’Eckhaus correspond à D E < 0, l’instabilité zigzag à D Z < 0. L’instabilité<br />

d’Eckhaus ne change pas l’espacement moyen, tandis que le mode zigzag le<br />

fait décroître. En régime non-linéaire, l’instabilité zigzag sature <strong>et</strong> correspond à une<br />

bifurcation à partir de l’état symétrique, alors que l’instabilité d’Eckhaus conduit<br />

à l’élimination de lamel<strong>les</strong>. Dans nos expériences (Fig. 5.2 ; alliage CBr 4 -C 2 Cl 6 ;<br />

η ≈ 0.4) [46], nous situons le seuil des zigzag vers λ zigzag = 0.85λ m environ, ce<br />

qui est remarquablement bas. Les simulations numériques de Parisi <strong>et</strong> Plapp [185]<br />

(voir aussi [186]) confirment ces résultats, mis à part un désaccord sur la valeur de<br />

λ zigzag qu’ils trouvent légèrement supérieur à λ m . L’origine de c<strong>et</strong>te incertitude reste<br />

à éclaircir.<br />

-Eff<strong>et</strong>s de biais thermique. En général, à l’issue du transitoire initial (voir la Réf. [47]<br />

pour des observations préliminaires), on observe d’abord des structures désordonnées,<br />

dites "labyrinthes". Ces structures relaxent très lentement, <strong>et</strong> peuvent perdurer<br />

sur de longs temps (ce que confirment des simulations numériques, non publiées, de<br />

M. Plapp). Comment se fait-il alors que l’on puisse observer des structures lamellaires<br />

ordonnées, <strong>et</strong> comment se forment-el<strong>les</strong> à partir des structures-labyrinthes ?<br />

La réponse tient à l’existence, dans <strong>les</strong> expériences, d’un léger biais thermique : <strong>les</strong><br />

isothermes, idéalement planes <strong>et</strong> perpendiculaires à l’axe z de solidification, ont, en<br />

fait, une inclinaison moyenne φ (rotation autour de l’axe x) par rapport au plan (horizontal)<br />

xy. Notre dispositif de solidification perm<strong>et</strong> de corriger ce biais, avec une<br />

précision d’environ 2˚, au début d’une expérience, mais, sans précaution particulière,<br />

ou si <strong>les</strong> conditions thermiques évoluent, l’angle φ peut dépasser 5˚. Dans ce cas, on<br />

observe une dérive globale de la structure, dans le sens de la pente des isothermes,<br />

<strong>et</strong> l’étirement des lamel<strong>les</strong> à partir du bord supérieur, qui joue le rôle de précurseur<br />

de la structure finale (Fig. 5.3). Nous poursuivons l’étude de ces phénomènes <strong>et</strong> de<br />

leurs conséquences pratiques.<br />

73


Fig.5.4 – Structure eutectique fibreuse (SCN-DC ; V = 0.053 µms −1 ). On distingue en bas à<br />

gauche une région hexagonale bien ordonnée. Les "lignes" sinueuses à travers la structure correspondent<br />

à des joints de grains<br />

-Dynamique <strong>et</strong> stabilité de fronts eutectiques fibreux. Nos observations dans l’alliage<br />

SCN-DC (travail de thèse de M. Perrut [187]) montrent que <strong>les</strong> structures<br />

fibreuses ont une symétrie de base hexagonale (Fig. 5.4), mais contiennent souvent<br />

de nombreux défauts topologiques. Les principaux résultats de notre étude sont <strong>les</strong><br />

suivants : 1-aux longs temps de solidification, l’espacement interfibre moyen λ moy<br />

suit approximativement une variation en V −1/2 , <strong>et</strong> reste très proche de la valeur<br />

de λ m estimée à partir d’expériences de SDEM [48] ; 2-la dynamique des structures<br />

fibreuses est forcée (étirée) par une légère courbure des isothermes dans la direction<br />

transverse y (le front est "bombé" vers le liquide), qui impose un accroissement<br />

permanent de l’espace interfibres (c<strong>et</strong> eff<strong>et</strong>, qu’on ne sait pas corriger, est différent<br />

de l’eff<strong>et</strong> de dérive lié au biais thermique évoqué ci-dessus) ; 3-durant un transitoire<br />

d’étirement sans branchement, la structure se réarrange lentement, en formant des<br />

domaines ordonnés séparés par des parois de défauts ; 4-après un long temps de solidification,<br />

ce forçage (voir aussi [194]) conduit la structure à un régime permanent<br />

d’étirement-branchement dans lequel λ moy sature à la valeur du seuil λ br de branchement<br />

(création de nouvel<strong>les</strong> fibres), en bordure du domaine de stabilité, dont nous<br />

montrons directement qu’elle est approximativement égale à λ m ; 5-l’instabilité d’élimination<br />

de fibres, qui est liée à une instabilité d’Eckhaus <strong>et</strong> limite le domaine de<br />

stabilité des structures fibreuses vers <strong>les</strong> p<strong>et</strong>ites valeurs de l’espacement, se produit<br />

pour une valeur-seuil λ c sensible à la variable V/G, <strong>et</strong> suit apparemment la même loi<br />

semi-empirique que celle établie en géométrie 2D (§4.3.2). Nous obtenons des informations<br />

quantitatives tant en ce qui concerne la courbure des isothermes (de rayon<br />

variant entre 5 <strong>et</strong> 8 mm) que pour la position des limites de stabilité. Nous montrons<br />

que le domaine de stabilité s’étend entre environ λ c ≈ 0.8λ m <strong>et</strong> λ br ≈ λ m pour <strong>les</strong><br />

plus basses vitesses explorées (inférieures à 0.01 µms −1 ). Nous établissons aussi qu’il<br />

se réduit fortement quand V augmente, car λ c suit à peu près la loi semi-empirique<br />

obtenue en échantillons minces pour la limite d’Eckhaus (éq. 4.8). Notons enfin que<br />

la phénoménologie des structures hexagona<strong>les</strong> reste peu étudiée expérimentalement<br />

<strong>et</strong> théoriquement (voir, par exemple, [29, 163, 191, 190, 192, 193]).<br />

74


Etudes à venir<br />

1-Stabilité des structures lamellaires. Nous n’avons pas encore pu mesurer précisément<br />

le seuil de stabilité inférieur des structures lamellaires symétriques. Le mode<br />

d’instabilité correspondant ne se présente pas sous la forme d’un mode d’Eckhaus<br />

simple, qui conduirait à l’élimination de lamel<strong>les</strong> entières. On assiste plutôt à la<br />

formation <strong>et</strong> à la migration de défauts topologiques de type "dislocation" (lamella<br />

termination). L’étude de la dynamique de ces défauts reste à faire. Leur propagation<br />

peut se faire dans <strong>les</strong> deux sens, selon que la dynamique évolue vers une augmentation<br />

ou une réduction de l’espacement lamellaire moyen. Le deuxième cas s’observe<br />

au-dessus de la limite supérieure de stabilité des structures zigzag. C<strong>et</strong>te limite, qui<br />

coïncide avec une instabilité de branchement lamellaire, est difficile à localiser. Elle<br />

ne semble pas excéder 1.1λ m . Les seuils remarquab<strong>les</strong> (élimination, zigzag, branchement)<br />

sont donc proches <strong>les</strong> uns des autres. Ceci a certainement des conséquences<br />

importantes sur la dynamique des structures lamellaires, en particulier lors des transitoires<br />

suivant des sauts de vitesse de solidification.<br />

2-Production des "lignes de faute". On trouve presque toujours, au sein des structures<br />

(quasi)stationnaires, différents types de défauts subsistant sur de longs temps,<br />

en plus ou moins grand nombre –par exemple des "dislocations" dans <strong>les</strong> structures<br />

lamellaires (on en voit un exemple en haut de la première micrographie de la<br />

Fig. 5.2). Dans <strong>les</strong> eutectiques lamellaires, on observe aussi des fautes d’empilement<br />

(conduisant à la formation de défauts plans dans le solide) de la structure lamellaire<br />

(Fig. 5.2). On observe une forte densité de ces lignes de fautes dans <strong>les</strong> lingots<br />

métalliques, quel que soit l’alliage, à des distances d’environ 10λ [7]. El<strong>les</strong> ont souvent<br />

été considérées comme une conséquence de la présence de sous-joints de grains.<br />

C<strong>et</strong>te assertion semble être infirmée par des études récentes par EBSD dans des<br />

alliages pseudo-binaires Al-Cu(-Ag) (U. Hecht <strong>et</strong> coll., résultats non publiés). Nous<br />

soupçonnons fortement, en fait, que <strong>les</strong> lignes de faute sont des défauts d’origine<br />

fondamentalement dynamique. Des observations supplémentaires seront nécessaires<br />

pour conclure fermement.<br />

3-Transition lamel<strong>les</strong>-fibres. Il s’agit de trouver <strong>les</strong> critères perm<strong>et</strong>tant de prédire la<br />

morphologie –lamellaire ou fibreuse– de solidification d’un alliage eutectique donné.<br />

On sait qu’une concentration proche d’un bord de plateau (faible fraction de phase)<br />

favorise la formation de fibres. La concentration n’est cependant pas le seul paramètre<br />

en jeu. Certaines de nos observations montrent qu’il peut y avoir coexistence<br />

des deux structures le long d’un même front de solidification. Des simulations numériques<br />

(M. Plapp, résultats non publiés) tendent effectivement à prouver qu’il peut<br />

exister un large domaine de bistabilité entre lamel<strong>les</strong> <strong>et</strong> fibres. Une étude détaillée<br />

reste à faire.<br />

4-Eff<strong>et</strong>s de l’anisotropie interfaciale. Les eff<strong>et</strong>s cristallographiques dans <strong>les</strong> structures<br />

eutectiques sont d’une grande importance industrielle –le suj<strong>et</strong> des "textures" de<br />

solidification, c’est-à-dire la "sélection", au bout d’un long temps de solidification,<br />

d’un nombre restreint, toujours <strong>les</strong> mêmes, d’orientations relatives (voire par rapport<br />

à l’axe de solidification) particulières (en relation d’épitaxie) des cristaux des deux<br />

phases solides reste largement ouvert. C<strong>et</strong>te question concerne en premier lieu <strong>les</strong><br />

75


eff<strong>et</strong>s de l’anisotropie des joints d’interphase sur la dynamique des structures de<br />

solidification eutectique (voir §5.2).<br />

5-Eff<strong>et</strong>s de la convection. Nous avons utilisé des alliages stab<strong>les</strong> vis-à-vis de la convection<br />

thermosolutale. Il y a toujours une convection "résiduelle" à l’échelle de l’échantillon<br />

(rendue visible par le brassage de poussières dans le liquide, <strong>et</strong> dont la vitesse<br />

ne dépasse pas 10 µms −1 près de l’interface), due aux gradients thermiques horizontaux<br />

(biais), par exemple près des espaceurs, ou à la présence de bul<strong>les</strong> (défaut de<br />

remplissage) en haut de l’échantillon (convection Marangoni). Cependant elle induit<br />

peu de perturbations sur la dynamique de solidification. En dehors de ces conditions,<br />

favorab<strong>les</strong> à une étude de phénomènes dynamiques généraux, on peut assister<br />

à la formation de rouleaux de convection de taille typique 100 µm, parfois dérivants<br />

(alliages hypoeutectiques ou dilués –phase α– de CBr 4 -C 2 Cl 6 ), s’accompagnant d’un<br />

creusement de l’interface. Ceci justifie notre participation à un programme d’expériences<br />

en micropesanteur ESA/CNES (DIRSOL/SEBA ; "phase B"). Nous voulons<br />

de plus m<strong>et</strong>tre à profit la micropesanteur pour une expérience originale de solidification<br />

d’échantillons à fort gradient de concentration longitudinal (le long de z),<br />

irréalisable au sol. L’idée est de pouvoir faire évoluer continûment une structure fibreuse<br />

vers une structure lamellaire. Ce proj<strong>et</strong> donnera aussi l’occasion de tenter des<br />

expériences dans des échantillons de plus grande épaisseur (1 mm), dans <strong>les</strong>quels on<br />

s’attend à observer des changements significatifs, par exemple sur la courbure des<br />

isothermes (<strong>et</strong> ses eff<strong>et</strong>s sur la dynamique des structures de solidification).<br />

6-Etudes exploratoires. On peut envisager d’étendre l’utilisation du dispositif DIR-<br />

SOL aux eutectiques non-fac<strong>et</strong>tés-fac<strong>et</strong>tés ou aux alliages monotectiques. Nous tenterons<br />

d’adapter notre méthode à l’observation de fronts de solidification monophasés<br />

(alliages dilués) cellulaires (près du seuil). Un premier but serait d’observer la<br />

formation des trois types de structures, identifiées par Morris <strong>et</strong> Winegard [152],<br />

<strong>et</strong> r<strong>et</strong>rouvées numériquement [188] : <strong>les</strong> cellu<strong>les</strong> "ordinaires" en arrangement hexagonal,<br />

<strong>les</strong> cellu<strong>les</strong> allongées (de même symétrie en "rouleaux" que <strong>les</strong> eutectiques<br />

lamellaires) <strong>et</strong> <strong>les</strong> cellu<strong>les</strong> inversées (appelées nodes dans la ref. [152]). Ces dernières<br />

n’existent qu’en géométrie 3D (parce que l’arrangement hexagonal n’est alors pas<br />

invariant par la transformation z → −z). La stabilité de ces structures dépend<br />

de l’anisotropie interfaciale (donc de l’orientation du cristal). Les cellu<strong>les</strong> inversées<br />

<strong>et</strong> <strong>les</strong> cellu<strong>les</strong> allongées n’ont pas été observées in situ, à notre connaissance. Par<br />

ailleurs, des simulations numériques ont montré l’existence d’une instabilité secondaire<br />

des cellu<strong>les</strong> ordinaires, sous la forme d’une oscillation des trois sous-réseaux<br />

de l’arrangement hexagonal [189]. Il serait intéressant d’observer c<strong>et</strong>te oscillation<br />

expérimentalement.<br />

5.2 Solidification d’alliages métalliques en échantillons<br />

minces<br />

Alliages dilués – Il est bien établi que l’on peut observer des différentes morphologies<br />

de croissance cristalline en géométrie 2D, en solidification directionnelle<br />

en échantillons minces d’un alliage non-fac<strong>et</strong>té dilué, en fonction de l’anisotropie<br />

76


interfaciale. Le Chap. 3 a clairement démontré, dans le cas de la SDEM de l’alliage<br />

CBr 4 -C 2 Cl 6 , que de tel<strong>les</strong> observations pouvaient donner des informations quantitatives<br />

sur <strong>les</strong> fonctions d’anisotropie capillaire <strong>et</strong> cinétique. Nous avons cependant<br />

souligné que nos mesures pouvaient souffrir d’une assez grande marge d’incertitude<br />

en particulier faute de bien l’orientation des cristaux sélectionnés. Pour c<strong>et</strong>te raison,<br />

nous avons amorcé une étude des morphologies des fronts de solidification directionnelle<br />

en échantillons minces d’alliages métalliques. Pour orienter <strong>les</strong> monocristaux, on<br />

pourra alors avoir recours à des méthodes d’analyse cristallographique par diffraction<br />

de rayonnement X ou électronique. Nous espérons obtenir suffisamment d’informations<br />

pour pouvoir statuer, selon l’alliage, sur l’importance relative de l’anisotropie<br />

capillaire <strong>et</strong> cinétique, <strong>et</strong> déterminer, dans un alliage donné, s’il est, ou non, nécessaire<br />

de tenir compte de facteurs d’anisotropie d’ordre supérieur à 4. L’enjeu est,<br />

entre autres, d’introduire des valeurs réalistes des coefficients d’anisotropie dans des<br />

simulations numériques à but prédictif [109] ou de vérifier expérimentalement <strong>les</strong><br />

résultats de calculs à l’échelle atomique [126, 127, 141].<br />

Nous avons pu fabriquer des échantillons d’environ 15 µm d’épaisseur d’un alliage<br />

métallique à bas point de fusion (In-Bi) –nous avons, pour cela, bénéficié de<br />

conseils de nos collègues d’Access e.V. (Aix-la-Chapelle) [61]. Nous <strong>les</strong> avons testés<br />

dans un banc de SDEM. L’observation se fait en lumière réfléchie. On ne voit alors<br />

que <strong>les</strong> structures en contact avec une des parois de verre (dans <strong>les</strong> échantillons<br />

minces d’alliages transparents, observés en lumière transmise, l’image est intégrée<br />

sur l’épaisseur). A l’arrêt (équilibre), <strong>et</strong> en régime de front plan stationnaire (faible<br />

valeur deV ), le contraste entre le solide <strong>et</strong> le liquide est faible, mais suffisant pour<br />

localiser l’interface solide-liquide. En revanche, nous n’avons pas pu observer de<br />

structures cellulaires. Nous nous heurtons selon toute vraisemblance à la formation<br />

d’un film liquide, probablement submicrométrique, mais opaque, entre <strong>les</strong> parois de<br />

verre <strong>et</strong> le solide. La partie modulée du front de solidification est, du coup, "invisible".<br />

Nous ne connaissons pas, pour l’instant, l’origine de ce film.<br />

Alliages eutectiques binaires –<br />

Fig.5.5 – Grain eutectique accroché ; SDEM de CBr 4 -C 2 Cl 6 .<br />

Pour l’analyse cristallographique par diffraction (ou microdiffraction) de rayons<br />

X, ou par EBSD, nous devrons contourner des difficultés assez habituel<strong>les</strong> liées, par<br />

exemple, au contact aux plaques de verre. L’avantage de ces analyses serait aussi<br />

de pouvoir montrer la formation de défauts de réseau (si leur densité n’est pas trop<br />

importante) <strong>et</strong> progresser dans la compréhension du mécanisme de polygonisation<br />

dynamique (§3.4). Ces analyses se feront post mortem dans un premier temps. Nous<br />

proposerons, dans un second temps, de m<strong>et</strong>tre au point un banc de solidification<br />

adapté à des observations in situ par radiographie ou topographie de rayonnement<br />

X synchrotron (voir, par exemple, [62, 63, 64]).<br />

77


5.3 Alliages fac<strong>et</strong>tés<br />

Dans la nature, la plupart des cristaux sont fac<strong>et</strong>tés. Contrairement aux interfaces<br />

"rugueuses", <strong>les</strong> fac<strong>et</strong>tes (qui correspondent, à l’équilibre, à un point singulier<br />

du diagramme de Wulff) ont presque toujours une cinétique très lente <strong>et</strong> possèdent<br />

un domaine de métastabilité (intervalle de température en dessous de la température<br />

d’équilibre sur lequel la croissance de la fac<strong>et</strong>te paraît "bloquée" sur des temps<br />

expérimentaux ordinaires) important [54, 55, 56]. Leur croissance est tributaire de la<br />

présence de sources de marches élémentaires, qui peuvent être des défauts de réseau<br />

(dislocations) affleurant à l’interface. La solidification d’un matériau fac<strong>et</strong>té conduit<br />

la plupart du temps à des régimes très instationnaires. Malgré tout, la question de<br />

l’interaction, dans des alliages, entre une dynamique de solidification fac<strong>et</strong>tée <strong>et</strong> la<br />

diffusion d’un soluté peut se poser. Nous avons amorcé l’étude des fronts de solidification<br />

fac<strong>et</strong>tés il y a quelques années, avec T. Börzsönyi, en utilisant une molécule<br />

mésomorphe de type cyanobicyclohexane notée CCH4, présentant une transition de<br />

phase entre un liquide nématique <strong>et</strong> un Smectique B –phase cristalline plastique de<br />

structure lamellaire. Ces cristaux possèdent une seule direction de fac<strong>et</strong>te (le plan<br />

des couches), toutes <strong>les</strong> autres étant rugueuses, dont l’intervalle de métastabilité<br />

est remarquablement p<strong>et</strong>it (environ 0.1 K) [50, 51]. Grâce à des traitements des<br />

surfaces internes des échantillons [51], on peut orienter des monocristaux de manière<br />

"planaire" (lamel<strong>les</strong> moléculaires perpendiculaires au plan de l’échantillon), en<br />

contrôlant aussi l’angle azimutal des couches. Ceci nous a conduit, entre autres, à la<br />

découverte d’obj<strong>et</strong>s dérivants localisés, que nous appelons "fac<strong>et</strong>tons" (Fig. 5.6).<br />

Fig.5.6 – SDEM de CCH4. a) Front cellulaire. b) Fac<strong>et</strong>ton.<br />

Un des buts poursuivis est l’étude de la formation <strong>et</strong> de la dynamique de cellu<strong>les</strong><br />

<strong>et</strong> dendrites fac<strong>et</strong>tées stationnaires. Les travaux théoriques des références [195, 196,<br />

197, 198] ont montré l’existence de tels obj<strong>et</strong>s dans le cas, non standard, où <strong>les</strong><br />

fac<strong>et</strong>tes d’équilibre ne sont pas singulières vis-à-vis de la cinétique interfaciale. Il<br />

semble, d’après la réf. (expérimentale) [199], que ce soit le cas des dendrites de<br />

NH 4 Br (précipitation à partir d’une solution aqueuse). Nous avons eu la surprise d’en<br />

observer aussi dans des expériences de SDEM de cristaux monocliniques (biphényl<br />

<strong>et</strong> naphtalène). Ces résultats sont en cours d’analyse. La question la plus difficile<br />

est celle de la nature des sources de marches, actives en permanence le long des<br />

fac<strong>et</strong>tes. Plusieurs hypothèses peuvent être émises, mais rien ne nous perm<strong>et</strong> de<br />

trancher à partir de nos observations à grande échelle. Par ailleurs, la formation<br />

de structures de solidification stationnaires dans le biphényl sont limitées par un<br />

phénomène de déformation plastique des grands monocristaux en cours de croissance<br />

dans le gradient thermique. Ce phénomène, spectaculaire à faible vitesse (en dessous<br />

du seuil cellulaire) <strong>et</strong> dans des échantillons purifiés, est remarquable parce que la<br />

déformation fait apparaître une longueur caractéristique dont l’origine est inconnue<br />

pour l’instant (interaction entre croissance <strong>et</strong> défauts de réseau, contact aux <strong>les</strong><br />

78


parois, contraintes thermiques). Dans nos perspectives figure aussi l’étude de la<br />

solidification des matériaux vitrifiab<strong>les</strong> <strong>et</strong> de la croissance sphérolitique en interaction<br />

avec un soluté.<br />

5.4 Maîtrise des microstructures de solidification par<br />

micromanipulation<br />

Dans de nombreuses situations industriel<strong>les</strong> ou de laboratoire, on souhaiterait<br />

pouvoir former des microstructures avec une valeur déterminée à l’avance de l’espacement<br />

λ. Il y a quelques années, Lee <strong>et</strong> Losert [200] ont développé dans ce but<br />

une méthode de micromanipulation, utilisant un chauffage local par des microspots<br />

lasers à proximité de l’interface solide-liquide, perm<strong>et</strong>tant de forcer l’arrangement<br />

de cellu<strong>les</strong> (alliages dilués) dès <strong>les</strong> stades initiaux, pourvu que λ soit stable. Si λ<br />

est inférieur au seuil λ c de l’instabilité de doublement de période 2λS (§3.3.2), des<br />

événements d’élimination de cellu<strong>les</strong> se répètent jusqu’à ce que la valeur moyenne<br />

de l’espacement dépasse λ c [149, 201].<br />

Fig.5.7 – Forme d’un front cellulaire sous contrôle à différents temps d’une simulation en champ<br />

de phase. Paramètres <strong>et</strong> constantes du matériau : voir Refs. [53] <strong>et</strong> [22]. Graphe (e) : énergie E q<br />

injectée devant chaque cellule q en fonction du temps.<br />

Une étude récente [53], combinant des travaux numériques (A. Pons <strong>et</strong> A. Karma)<br />

<strong>et</strong> des expériences (dans le laboratoire de W. Losert, University of Maryland) a montré<br />

l’efficacité d’une méthode de maîtrise de fronts cellulaires instab<strong>les</strong> d’espacement<br />

bien inférieur à λ c (structures "à p<strong>et</strong>it espacement") par boucle de rétroaction (feedback<br />

control). C<strong>et</strong>te méthode consiste à ralentir la croissance de toute cellule qui<br />

dépasse la position moyenne au sein de la structure dans la direction z. Dans <strong>les</strong> simulations,<br />

on introduit le champ de température T (˜x, t) suivant (modification d’un<br />

modèle récent de champ de phase [24, 202, 203]) :<br />

p(˜x, t) = ∑ q<br />

T (˜x, t) = T 0 + G(y − V t) + p(˜x, t), (5.2)<br />

( )<br />

−(˜x − ˜xq (t)) 2<br />

gλ 0 H [y q (t) − ȳ(t) − δ] , exp<br />

,<br />

ξ 2 (5.3)<br />

79


˜x étant le vecteur position 2D, p(˜x, t) la perturbation thermale ("spot") imposée <strong>et</strong><br />

H[...] la fonction de Heaviside ; la somme s’étend sur <strong>les</strong> N cellu<strong>les</strong> contrôlées. Chaque<br />

spot prend la forme d’une gaussienne (superposée au gradient linéaire). A partir<br />

d’une structure très instable (λ 0 ≈ λ c /2 ; Fig. 5.7a), un grand désordre transitoire<br />

(Fig. 5.7b) s’instaure dans un premier temps (tant que |y q − ȳ| reste grand), mais<br />

cesse assez rapidement, <strong>et</strong> <strong>les</strong> cellu<strong>les</strong> se stabilisent (Fig. 5.7c). C<strong>et</strong>te procédure, d’une<br />

surprenante efficacité, perm<strong>et</strong> de stabiliser une structure d’espacement très uniforme<br />

(Fig. 5.7d). La puissance, ou, de manière équivalente, la fréquence f q = dE q /dt à<br />

laquelle la cellule q est visée par un spot (E q : énergie injectée devant un cellule q),<br />

assez forte durant le transitoire de mise en ordre, devient beaucoup plus faible en<br />

régime permanent (Fig. 5.7).<br />

Les expériences sont faites en SDEM d’un alliage de succinonitrile (SCN) <strong>et</strong> d’un<br />

colorant laser, coumarin 152 (C152), en faible concentration (< 0.5 mol%), dans<br />

des échantillons (de section 2000 × 100 µm 2 ) monocristallins, d’orientation axiale.<br />

Les spots laser sont issus d’un système de pince optique holographique (BioRyx200,<br />

Arryx Inc) : un faisceau laser (2 W ; NdYAG ; λ=532 nm) focalisé sur un modulateur<br />

de phase spatiale ; la figure de diffraction est proj<strong>et</strong>ée sur la région observée au travers<br />

de l’objectif du microscope. Le système perm<strong>et</strong> de positionner plusieurs dizaines de<br />

spots indépendants avec une résolution du pixel (≈ 3 µm dans le cas présent) avec<br />

un temps de réponse de moins de 0.1 s. Chaque spot produit de la chaleur dans<br />

une région d’environ 10µm de diamètre par absorption partielle de la lumière par<br />

<strong>les</strong> molécu<strong>les</strong> de C152 [52]. Pendant la procédure de contrôle, le front cellulaire est<br />

observé avec une caméra numérique <strong>et</strong>, après traitement d’image en temps réel (H.<br />

Singer), la liste des pointes de cellu<strong>les</strong> détectées est envoyée au boîtier de commande<br />

du système pour positionner <strong>les</strong> spots laser devant <strong>les</strong> cellu<strong>les</strong> devant être freinées.<br />

Ces opérations se répètent à une fréquence d’environ 0.7 Hz (N = 8 cellu<strong>les</strong>).<br />

La Figure 5.8 donne un exemple d’une structure cellulaire instable contrôlée.<br />

A partir d’une structure initiale créée à assez forte vitesse (Fig. 5.8a), on fait décroître<br />

V tout en activant la boucle de contrôle (λ c croît quand V diminue). Après<br />

un transitoire assez court (Fig. 5.8b <strong>et</strong> 5.8e), on maintient une structure cellulaire<br />

de p<strong>et</strong>it espacement sur des temps longs (Fig. 5.8c <strong>et</strong>5.8f). Plusieurs cellu<strong>les</strong> sont<br />

éliminées dans la région "libre", alors que le nombre de cellu<strong>les</strong> reste constant dans<br />

la zone contrôlée (limitée à 1 mm). Lorsqu’on éteint le laser, l’instabilité s’amplifie<br />

clairement (Fig. 5.8d <strong>et</strong> 5.8g). La similitude des observations expérimenta<strong>les</strong> (Fig.<br />

5.8d à 5.8h) <strong>et</strong> des simulations (Fig. 5.7) est remarquable. Nous avons pu répéter ce<br />

type d’expérience, en nous assurant de la permanence du régime. Comme dans <strong>les</strong><br />

simulations, on voit l’uniformisation de la structure <strong>et</strong> la décroissance de l’énergie<br />

injectée. Le second point est illustré par la Fig. 5.9. Nous pouvons conclure que<br />

des structures de solidification directionnelle cellulaires très instab<strong>les</strong> peuvent être<br />

stabilisées par boucle de rétroaction (voir aussi [204]).<br />

Nous prévoyons d’implanter un dispositif de ce type à l’INSP. Ceci nécessitera un<br />

effort de mise au point. Des perfectionnements peuvent être apportés, en particulier<br />

sur <strong>les</strong> algorithmes de contrôle. Nous prévoyons d’adapter c<strong>et</strong>te méthode aux structures<br />

eutectiques lamellaires en échantillon mince [52]. De nombreuses utilisations<br />

sont envisageab<strong>les</strong>, que ce soit pour des mesures de paramètres phénoménologiques<br />

80


Fig.5.8 – Contrôle d’un front cellulaire instable. Barre<br />

(1 mm) : région contrôlée. Cadres : contrôle actif. (a)<br />

Structure initiale (t = 0 ; V = 18.4 µms −1 ) ; (b) Transitoire<br />

suivant le début du contrôle (1100 s ; 6.1 µms −1 ) ;<br />

(c) Structure stabilisée (1520 s ; 4.6 µms −1 ) ; d) Instabilité<br />

2λS hors contrôle (1610 s) ; (e) Transitoire après<br />

rétablissement du contrôle (1620 s) ; (f) Structure restabilisée<br />

(2100 s) ; (g) Instabilité (2300 s) ; (h) Structure<br />

naturellement stable.<br />

Fig.5.9 – Energie E q injectée en face<br />

de chaque cellule q en fonction de t ;<br />

l’insert donne la valeur finale de la fréquence<br />

f d’illumination pour chaque<br />

cellule q. Les ruptures de pente simultanées<br />

sont le signe d’une réaction du système<br />

à des perturbations accidentel<strong>les</strong>.<br />

Expérience différente de la Fig. 5.8.<br />

comme <strong>les</strong> coefficients de diffusion de la phase, ou pour des applications plus expérimenta<strong>les</strong><br />

(sélection de grains).<br />

81


Chapitre 6<br />

Conclusion<br />

Nous avons présenté nos travaux de recherche expérimentale sur la dynamique<br />

des fronts de solidification directionnelle monophasés <strong>et</strong> eutectiques en échantillons<br />

minces. D’une part, des progrès notab<strong>les</strong> ont été faits dans la compréhension de la<br />

phénoménologie générale de la dynamique de formation des microstructures de solidification<br />

en géométrie bidimensionnelle dans le cadre de la physique non-linéaire<br />

de la structuration des systèmes étendus maintenus hors d’équilibre. D’autre part,<br />

des preuves expérimenta<strong>les</strong> claires ont été apportées qui montrent l’influence déterminante<br />

de l’anisotropie interfaciale sur l’existence <strong>et</strong> la stabilité de certaines<br />

microstructures de solidification. Sur le plan théorique, à la conjecture ancienne<br />

d’une "sélection" des microstructures on substitue l’étude des branches de solutions<br />

stationnaires, de leur domaine de stabilité, des instabilités <strong>et</strong> bifurcations qu’el<strong>les</strong> subissent<br />

vers des morphologies à symétrie brisée, <strong>et</strong> de la dynamique spatio-temporelle<br />

à base d’obj<strong>et</strong>s localisés. Du point de vue expérimental, le suivi de la dynamique par<br />

observation en temps réel <strong>et</strong> l’utilisation des échantillons minces sont deux élémentsclé<br />

sans <strong>les</strong>quels <strong>les</strong> progrès évoqués n’auraient probablement pas pu être réalisés.<br />

Ces travaux font maintenant partie d’un socle assez compl<strong>et</strong> sur lequel peuvent<br />

s’appuyer des études plus approfondies encore, ou plus complexes. L’ouverture de nos<br />

recherches à l’étude des fronts de solidification en géométrie 3D, à des expériences de<br />

solidification directionnelle en échantillons minces d’alliages métalliques avec analyse<br />

cristallographique, à la croissance fac<strong>et</strong>tée <strong>et</strong> à des méthodes actives de maîtrise des<br />

microstructures offre des perspectives intéressantes sur le plan fondamental, grâce au<br />

support offert par le calcul numérique, <strong>et</strong> innovantes du point de vue des applications.<br />

82


Bibliographie<br />

[1] J. Langer, Rev. of Mod. Phys., 52, 1 (1980).<br />

[2] P. Manneville, "Dissipatives structures and weak turbulence", Academic Press,<br />

Boston (1990).<br />

[3] Solids Far from Equilibrium, C. Godrèche ed., Cambridge University Press<br />

(1992).<br />

[4] M. Cross, P. Hohenberg, Rev. of Mod. Phys., 65, 851 (1993).<br />

[5] J.P. Gollub, J.S. Langer, Rev. Mod. Phys., 71, S396 (1999).<br />

[6] J.W. Rutter, B. Chalmers, Can. J. Phys., 31, 15 (1953).<br />

[7] L.M. Hogan, R.W. Kraft, F.D. Lemkey, Adv. Mater. Res., 5,83 (1971).<br />

[8] W. Kurz, D. J. Fisher, Fundamentals of solidification, 4th ed. (Enfield Publishing<br />

& Distribution Company, Enfield, NH, 2001).<br />

[9] W.A. Tiller, K.A. Jackson, J.W. Rutter, B. Chalmers, Acta M<strong>et</strong>., 1 428 (1953).<br />

[10] W.W Mullins, R.F. Sekerka, J. Appl. Phys., 34, 323 (1963).<br />

[11] W.W Mullins, R.F. Sekerka, J. Appl. Phys., 35, 444 (1964).<br />

[12] D.P. Woodruff, " The solid-liquid interface",Cambridge University Press (1973).<br />

[13] K. A. Jackson, J. D. Hunt, Acta M<strong>et</strong>., 13, 1212 (1965).<br />

[14] K. A. Jackson, J. D. Hunt, Trans. M<strong>et</strong>all. Soc. AIME 236, 1129 (1966).<br />

J. D. Hunt, K.A. Jackson, Trans. M<strong>et</strong>all. Soc. AIME 236, 843 (1966).<br />

[15] H. Esaka, W. Kurz, J. Cryst. Growth, 72, 5 (1985).<br />

[16] V. Se<strong>et</strong>haraman, R. Trivedi, M<strong>et</strong>all. Trans., 19A, 2955 (1988).<br />

[17] X. W. Qian, H. Z. Cummins, Phys. Rev. L<strong>et</strong>., 64, 3038 (1990).<br />

[18] A.J. Simon, J. Bechhoefer, A. Libchaber, Phys. Rev. L<strong>et</strong>t., 61, 2574 (1988).<br />

[19] S. de Cheveigné, C. Guthmann, J. Phys. I 2, 193 (1992).<br />

[20] G. Faivre and J. Mergy, Phys. Rev. A, 45, 7320 (1992) ; 46, 963 (1992).<br />

[21] P. Oswald, M. Moulin, P. M<strong>et</strong>z, J. C. Géminard, P. Sotta, L. Sallen, J. Phys.<br />

III, 3, 1891 (1993).<br />

[22] M. Georgelin, A. Pocheau, Phys. Rev. E, 57, 3189 (1998).<br />

[23] K. Kassner, Pattern Formation in Diffusion-Limited Crystal Growth, World<br />

Scientific (1996), ISBN 9810215320.<br />

[24] A. Karma, W. J. Rappel, Phys. Rev. E, 57, 4323 (1998). W. J. Rappel, A.<br />

Karma, Phys. Rev. E, 60, 3614 (1999).<br />

83


[25] W.J. Bo<strong>et</strong>tinger, S. R. Coriell, A. L. Greer, A. Karma, W. Kurz, M. Rappaz <strong>et</strong><br />

R. Trivedi, Acta Mater., 48, 43 (2000).<br />

[26] H. Emmerich, The Diffuse Interface Approach in Materials Science. Thermodynamic<br />

Concepts and Applications of Phase-Field Models, Lecture Notes in<br />

Physics, M 73, Springer (2003) ISBN : 3-540-00416-5.<br />

[27] M.R.E. Proctor, C.A. Jones, J. Fluid Mech., 188, 301 (1988)<br />

[28] P. Coull<strong>et</strong>, G. Iooss, Phys. Rev. L<strong>et</strong>t., 64, 866 (1990).<br />

[29] L. Gil, Thèse d’Habilitation à diriger <strong>les</strong> recherches, Nice (2004).<br />

[30] F. H. Busse, Rep. Prog. Phys,. 41, 1929 (1978).<br />

[31] M. Faraday, Philos. Trans. R. Soc. London, 121, 299 (1831). A. Kudrolli, J.P.<br />

Gollub, Physica D 97, 133 (1996).<br />

[32] M. Rabaud, S. Michalland , Y. Couder, Phys. Rev. L<strong>et</strong>t., 64, 184 (1990).<br />

[33] F. Giorgiutti, A. Bl<strong>et</strong>on, L. Limat, J.E. Weisfried, Phys. Rev. L<strong>et</strong>t., 74, 538<br />

(1995).<br />

[34] P. Coull<strong>et</strong>, R.E. Goldstein, G.H. Gunaratne, Phys. Rev. L<strong>et</strong>t., 63, 195 (1989).<br />

[35] S. Akamatsu, G. Faivre, T. Ihle, Phys. Rev. E, 51, 4751 (1995).<br />

[36] S. Akamatsu, G. Faivre, J. Phys. I France, 6, 503 (1996).<br />

[37] S. Akamatsu, T. Ihle, Phys. Rev. E, .56, 4479 (1997).<br />

[38] S. Akamatsu, G. Faivre, Phys. Rev. E, 58, 3302 (1998).<br />

[39] S. Bottin-Rousseau, S. Akamatsu, G. Faivre, Phys. Rev. B, 66, 054102 (2002).<br />

[40] M. Ginibre, S. Akamatsu, G. Faivre, Phys. Rev. E, 56 780 (1997)<br />

[41] M. Ginibre, Thèse de Doctorat de l’Université <strong>Pierre</strong> <strong>et</strong> Marie Curie, Paris,<br />

France, 1997.<br />

[42] S. Akamatsu, M. Plapp, G. Faivre, A. Karma, Phys. Rev. E, 66, 030501(R)<br />

(2002).<br />

[43] S. Akamatsu, M. Plapp, G. Faivre, A. Karma, M<strong>et</strong>al. Mater. Trans. A, 35, 1815<br />

(2004).<br />

[44] S. Akamatsu, G. Faivre, Phys. Rev. E, 61, 3757 (2000).<br />

[45] S. Akamatsu, S. Moulin<strong>et</strong>, G. Faivre, M<strong>et</strong>. Mater. Trans A, 32A, 2039 (2001).<br />

[46] S. Akamatsu, S. Bottin-Rousseau, G. Faivre, Phys. Rev. L<strong>et</strong>t., 93, 175701<br />

(2004).<br />

[47] A. Parisi, M. Plapp, S. Akamatsu, S. Bottin-Rousseau, M. Perrut, <strong>et</strong> G. Faivre,<br />

in "Modeling of Casting, Welding, and Advanced Solidification Processes - XI",<br />

pp. 417-424, ed. C.-A. Gandin <strong>et</strong> M. Bell<strong>et</strong>, The Minerals, M<strong>et</strong>al and Materials<br />

Soci<strong>et</strong>y, Warrendale, PA (2006).<br />

[48] S. Akamatsu, S. Bottin-Rousseau, M. Perrut, G. Faivre, V.T. Witusiewicz, L.<br />

Sturz, J. Cryst. Growth, 299 418 (2007).<br />

[49] S. Bottin-Rousseau, M. Perrut, C. Picard, S. Akamatsu, G. Faivre, J. Cryst.<br />

Growth, 306, 465 (2007).<br />

[50] T. Börzsönyi, S. Akamatsu, G. Faivre, Phys. Rev. E, 65, 011702 (2002).<br />

84


[51] T. Börzsönyi, S. Akamatsu, Phys. Rev. E, 66, 051709 (2002).<br />

[52] S. Akamatsu, K. Y. Lee, W. Losert, J. Cryst. Growth, 289, 331 (2006).<br />

[53] A. J. Pons, A. Karma, S. Akamatsu, M. Newey, A. Pomerance, H. Singer, W.<br />

Losert, Phys. Rev. E, 75, 021602 (2007).<br />

[54] W. K. Burton, N. Cabrera, F. C. Frank, Phil. Roy. Soc. London, 243, 299<br />

(1951).<br />

[55] K.A. Jackson, Liquid m<strong>et</strong>als and Solidification, ASM Cleveland (1958) ; in<br />

Growth and perfection of crystals (eds : R.H. Doremus, B.W. Roberts, D. Turnbull)<br />

Wiley, New York (1958).<br />

[56] I. Markov, Crystal growth for beginners, World Scientific, Singapore (1996).<br />

[57] H.A. Wilson, Philos. Mag., 50, 238 (1909).<br />

[58] U. Hecht <strong>et</strong> al., Materials Science and Engineering Reports, 46, 1 (2004).<br />

[59] R. Racek, G. Lesoult, M. Turpin, J. Cryst. Growth, 22, 210 (1974).<br />

[60] J.J. Favier, M. Turpin, C. P<strong>et</strong>ipas, B. Labulle, J. Cryst. Growth, 38, 109 (1977).<br />

[61] V.T. Witusiewicz, U. Hecht, S. Rex, M. Apel, Acta Mater, 53, 3663 (2005).<br />

[62] G. Grange, C. Jourdan, J. Gastaldi, B. Billia, J. Phys. III France, 4, 293 (1994).<br />

[63] Mathiesen RH, Arnberg L, Mo F, Weitkamp T, Snigirev A, Phys. Rev. L<strong>et</strong>t.,<br />

83, 5062 (1999).<br />

[64] B. Billia, N. Bergeon, H. Nguyen Thi, H. Jamgotchian, J. Gastaldi, G. Grange,<br />

Phys. Rev. L<strong>et</strong>ters, 93, 126105 (2004). H. Nguyen-Thi, G. Reinhart, N.<br />

Mangelinck-Noel, H. Jung, B. Billia, T. Schenk, J. Gastaldi, J. Hartwig, J.<br />

Baruchel, M<strong>et</strong>al Mater Trans A, 38A, 1458 (2007).<br />

[65] E. Rolley, S. Balibar, F. Gall<strong>et</strong>, Europhys. L<strong>et</strong>t., 2, 247 (1986).<br />

[66] U. Bisang, J. H. Bilgram, Phys. Rev. L<strong>et</strong>t., 21, 3898 (1995) ; Phys. Rev. E, 54,<br />

5309 (1996).<br />

[67] van Suchteleen, communication privée à G. Faivre.<br />

[68] J. Mergy, G. Faivre, C. Guthmann, R. Mell<strong>et</strong>, J. Cryst. Growth, 134, 353<br />

(1993).<br />

[69] J. Mergy, Thèse de Doctorat de l’Université <strong>Pierre</strong> <strong>et</strong> Marie Curie, Paris, France,<br />

1992.<br />

[70] H. Jamgotchian, R. Trivedi, B. Billia, Phys. Rev. E, 47, 4313 (1993).<br />

[71] M. Georgelin, A. Pocheau, Phys. Rev. L<strong>et</strong>t. ,79, 2698 (1997).<br />

[72] V.T. Witusiewicz, L. Sturz , U. Hecht, S. Rex, Acta Materialia, 52, 4561 (2004).<br />

[73] H.M. Hawthorne, J.N. Sherwood, Trans. Faraday Soc., 66, 1783 (1970).<br />

[74] V.T. Witusiewicz, L. Sturz, U. Hecht, S. Rex, Acta Mater, 52, 5071 (2004) ;<br />

52, 5519 (2004) ; 53, 173 (2005). L. Sturz, V.T. Witusiewicz, U. Hecht, S. Rex,<br />

J. Cryst. Growth, 270, 273 (2004).<br />

[75] Guggenheim E.A., Thermodynamics, North-Holland Publishing, Amsterdam,<br />

1967.<br />

[76] V.G. Smith, W.A. Tiller, J.W. Rutter, Can. J. Phys., 33, 723 (1955).<br />

85


[77] J.A. Warren, J.S. Langer, Phys. Rev. E, 47, 2702 (1993).<br />

[78] W. Losert, B.Q. Shi, H.Z. Cummins, Proc. Nat. Acad. Sci. USA, 95, 431 (1998) ;<br />

95, 439 (1998).<br />

[79] B. Caroli, C. Caroli, L. Ramirez-Piscina, J. Cryst. Growth, 132, 377 (1993).<br />

[80] D. J. Wollkind, L. A. Segel, Philos. Trans. R. Soc. London, Ser. A, 268, 351<br />

(1970).<br />

[81] K. Brattkus, C. Misbah, Phys. Rev. L<strong>et</strong>t. 64, 1925 (1990).<br />

[82] M. More, J. Lefèbvre, R. Four<strong>et</strong>, Acta Cryst. B, 33, 3681 <strong>et</strong> 3862 (1977).<br />

[83] J.-M. Laherrère, R. Mell<strong>et</strong>, H. Savary, C. Licoppe, J.F. Scott, J.-C. Tolédano,<br />

Phys. Rev. A, 42 3634 (1990). J.-M. Laherrère, J.-C. Tolédano, H. Savary, R.<br />

Mell<strong>et</strong>, Europhys. L<strong>et</strong>t., 15, 197 (1991).<br />

[84] L.M. Williams, M.R. Srinavasan, H.Z. Cummins, Phys. Rev. L<strong>et</strong>t., 64, 1526<br />

(1990).<br />

[85] P.E. Cladis, J.T. Gleeson, P.L. Finn, in Patterns and Instabilities, eds D. Walgraef<br />

<strong>et</strong> N. Ghoniem (Kluwer, Dordrecht, 1990).<br />

[86] H. Jamgotchian, R. Trivedi, B. Billia, J. Cryst. Growth, 134, 181 (1993).<br />

[87] W.R. Wilcox, Kuo V.H.S., J. Cryst. Growth, 19, 221 (1973).<br />

[88] A-A Chernov, Temkin D-E, in "Crystal Growth and Materials, E. Kaldis and<br />

H.-J. Scheel Eds. (North-Holland, Amsterdam, 1977) p. 1.<br />

[89] J. E. Gegusin, A.S. Dzjuba, Krystallografiya, 22, 348 (1977).<br />

[90] Y. Wang, L. L. Regel, W. R. Wilcox, Crystal Growth and Design, 2, 453 (2002).<br />

[91] I. Farup, J.M. Drez<strong>et</strong>, M. Rappaz, Acta Mater., 49, 1261 (2001).<br />

[92] G. Couturier, M. Rappaz, Modelling Simul. Mater. Sci. Eng., 14, 253 (2006).<br />

[93] W.-J. Rappel, H. Riecke, Phys. Rev. A, 45 846 (1992).<br />

[94] J. Weeks, W. Van Saarloos, 42, 5056 (1990).<br />

[95] S. C. Huang, M.E. Glicksmann, Acta M<strong>et</strong>all., 29, 701, (1981) ; 29, 717, (1981).<br />

[96] A. Dougherty, P.D. Kaplan, J.P. Gollub, Phys Rev. L<strong>et</strong>t., 58, 1652 (1987).<br />

[97] A. Dougherty, J. P. Gollub, Phys. Rev. A, 38, 3043 (1988).<br />

[98] J.C. LaCombe, M.B. Koss, V. E. Fradkov, M.E. Glicksman, Phys. Rev. E, 52,<br />

2778 (1995).<br />

[99] Ivantsov G.P., Doklady Akad. Nauk. SSR, 58, 567 (1947).<br />

[100] M. Ben Amar, Y. Pomeau, Europhys. L<strong>et</strong>t., 2 307 (1986).<br />

[101] A. Barbieri, D. C. Hong, J. S. Langer, Phys. Rev. A 35, 1802 (1987).<br />

[102] D.A. Kessler, J. Koplik, H. Levine, Adv. Phys., 37, 255 (1988).<br />

[103] E. A. Brener, J. Cryst. Growth, 99, 165 (1990).<br />

[104] M. Ben Amar, E. Brener, Phys. Rev. L<strong>et</strong>t., 71, 589 (1993).<br />

[105] E. Brener, Phys. Rev. L<strong>et</strong>t., 71, 3653 (1993)<br />

86


[106] J. S. Langer, in Chance and Matter, ed. J. Soul<strong>et</strong>ie, J. Vannimenus and R.<br />

Stora, (Elsevier, Amsterdam, 1987). H. Müller-Krumbhaar <strong>et</strong> W. Kurz, Phase<br />

Transformation in Materials, ed. P. Haasen (VCH-Verlag, Weinheim, 1991).<br />

[107] Y. Saito, G. Goldbeck-Wood, H. Müller-Krumbhaar, Phys. Rev. L<strong>et</strong>t., 58,<br />

1541 (1987) ; Phys. Rev. A, 38, 2148 (1988).<br />

[108] A. Karma, Y. H. Lee, M. Plapp, Phys. Rev. E, 61, 3996 (2000).<br />

[109] J. Bragard, A. Karma, Y.H. Lee, M. Plapp, Interface Science, 10 (2-3), 121<br />

(2002).<br />

[110] E. Brener, H. Müller-Krumbhaar, Y. Saito, D. Temkin, Phys. Rev., E 47, 1151<br />

(1993)<br />

[111] T. Ihle, H. Müller-Krumbhaar, Phys. Rev. L<strong>et</strong>t., 70, 3083 (1993) ; Phys. Rev.<br />

E, 49, 2972 (1994).<br />

[112] M. Ben Amar, E. Brener, Phys. Rev. L<strong>et</strong>t., 75, 561 (1995).<br />

[113] E. A. Brener, H. Müller-Krumbhaar, D.E. Temkin, Europhys. L<strong>et</strong>t., 17 535<br />

(1992).<br />

[114] R. Kupferman, D. A. Kessler, E. Ben-Jacob, Physica A, 213, 451 (1995).<br />

[115] P. Molho, A.J. Simon, A. Libchaber, Phys. Rev. A, 42, 904 (1990).<br />

[116] E. Ben-Jacob, G. Deutscher, P. Garik, N.D. Goldenfeld, Y. Lareah, Phys. Rev.<br />

L<strong>et</strong>t., 57, 1903 (1986). E. Ben-Jacob, P. Garik, T. Müller, D. Grier, Phys. Rev.<br />

A, 38, 1370 (1988).<br />

[117] Y. Saito, C. Misbah, H. Müller-Krumbhaar, Phys. Rev. L<strong>et</strong>t., 63, 2377 (1989).<br />

[118] A. Classen, C. Misbah, H. Müller-Krumbhaar, Y. Saito, Phys. Rev. A, 43,<br />

6920 (1991).<br />

[119] P. Oswald, J. Malthête, P. Pelcé, J. Phys. France 50, 2121 (1989).<br />

[120] W. Losert, D. A. Stillman, H. Z. Cummins, P. Kopczynski, W.-J. Rappel, A.<br />

Karma, Phys. Rev. E, 58, 7492 (1998).<br />

[121] B. Utter, R. Ragnarsson, E. Bodenschatz, Phys. Rev. L<strong>et</strong>ters, 86, 4604 (2001).<br />

[122] B. Utter, E. Bodenschatz, Phys. Rev. E, 66, 051604 (2002).<br />

[123] K.A. Jackson, Phil. Mag., 7, 1615 (1962).<br />

[124] A. Pavlovska, D. Nemow, J. Cryst. Growth, 39, 346 (1977).<br />

[125] Fehlner <strong>et</strong> al, Can. J. Phys., 54, 2159 (1976).<br />

[126] J. J. Hoyt, M. Asta, A. Karma, Phys. Rev. L<strong>et</strong>t., 86, 5530 (2001).<br />

[127] J. J. Hoyt, M. Asta, A. Karma, Phys. Rev. B., 66, 100101 (2002).<br />

[128] R.E. Napolitano, S. Liu, Phys. Rev. B, 70, 214103 (2004).<br />

[129] R. Trivedi, J.A. Sekhar, V. Se<strong>et</strong>haraman, M<strong>et</strong>all. Trans., 20A, 769 (1989).<br />

[130] P. Kurowski, C. Guthmann, S. de Cheveigné, Phys. Rev. A, 42, 7368 (1990).<br />

[131] K. Somboonsuk, J. T. Mason, R. Trivedi, M<strong>et</strong>all. Trans. A, 15A, 967 (1984).<br />

[132] M. Muschol, D. Liu, H. Cummins, Phys. Rev. A, 46, 1038 (1992).<br />

[133] S.-K. Chan, H.-H. Reimer, M. Kahlweit, J. Cryst. Growth, 32, 303 (1976).<br />

87


[134] T. Ihle, Eur. Phys. J. B, 16, 337 (2000).<br />

[135] R. Kupferman, D.A. Kessler, Phys. Rev. E, 51, R20 (1995).<br />

[136] J. Deschamps, M. Georgelin, A. Pocheau, Europhys. L<strong>et</strong>ters, 76, 291(2006) ;<br />

J. Deschamps, Thèse de Doctorat, Université de Provence, Marseille (2007).<br />

[137] T. Okada, Y. Saito, Phys. Rev. E, 54, 650 (1996).<br />

[138] F. Heslot, A. Libchaber, Physica Scripta, T9, 126 (1985).<br />

[139] K. Dragnevski, R. F. Cochrane, A. M. Mullis, Phys. Rev. L<strong>et</strong>t., 89, 215502-1<br />

(2002).<br />

[140] W. L. Kaukler, J. W. Rutter, in In Situ Composites, F. D. Lemkey, H. E. Cline<br />

<strong>et</strong> M. McLean eds, Elsevier, Amsterdam, 1982, p. 305.<br />

[141] X.B. Feng, B. Laird, J. Chem. Phys., 124, 044707 (2006).<br />

[142] J.C. Géminard, Thèse de Doctorat, Université Claude Bernard - Lyon I (1993).<br />

[143] L .Sallen, P. Oswald, J.C. Géminard, J. Malthête, Journal de Physique II, 5,<br />

937 (1995).<br />

[144] Des dendrites fendues ont été observées, par exemple, dans des cristaux de<br />

glace [K.K. Koo, R. Ananth, W.N. Gill, Phys. Rev. A, 44, 3782 (1991)], <strong>et</strong> dans<br />

un cristal liquide discotique (cf réf. [142]).<br />

[145] T. Abel, E. Brener, H. Müller–Krumbhaar, Phys. Rev. E, 55, 7789 (1997).<br />

[146] H. Singer, I. Singer-Loginova, J. Bilgram, G. Amberg, J. Cryst. Growth, 296,<br />

58 (2006).<br />

[147] M. Georgelin, S. Bodea, A. Pocheau, Europhys. L<strong>et</strong>t., 77, 46001 (2007).<br />

[148] P. Kopczynski, W-J. Rappel, A. Karma, Phys. Rev. L<strong>et</strong>t., 77, 3387 (1996).<br />

[149] P. Kopczynski, W-J. Rappel, A. Karma, Phys. Rev. L<strong>et</strong>t., 79, 2698 (1997).<br />

[150] C. Misbah, A. Valance, Phys. Rev. E, 49, 166 (1994).<br />

[151] S. R. Coriell, R. F. Sekerka, J. Cryst. Growth, 34, 157 (1976).<br />

[152] L. R. Morris, W.C. Winegard, J. Cryst. Growth, 5, 361 (1969).<br />

[153] Fiore.<br />

[154] K. Kassner, C. Misbah, J. Müller, J. Kappey, P. Kohlert, 63, 036117 (2001).<br />

[155] R. Kikuchi, J.W. Cahn, Phys. Rev., 21, 1893, (1985).<br />

[156] R.J. Schaefer, M.E. Glicksman, J.D. Ayers, Philos. Mag., 32, 725 (1975).<br />

[157] M. Gündüz, J.D. Hunt, Acta M<strong>et</strong>all., 33, 1651 (1985).<br />

[158] B. Chalmers, "Princip<strong>les</strong> of Solidification", R.E. Krieger Pub. Co (1977).<br />

[159] For an introduction to the physics of lattice defects, see D. Hull, "Introduction<br />

to dislocations", Pergamon Press, Oxford, 1975.<br />

[160] S.R. Coriell, R.F. Sekerka, J. Cryst. Growth, 19, 285 (1973).<br />

[161] L.H. Ungar and R.A. Brown, Phys. Rev. B, 30, 3993 (1984).<br />

[162] E. Billig, Proc. R. Soc. London, Ser. A, 235, 37 (1956).<br />

[163] C. Pirat, A. Naso, J.L. Meunier, P. Maissa, C. Mathis, Phys. Rev. L<strong>et</strong>t., 94,<br />

134502 (2005).<br />

88


[164] G.A. Chadwick, Prog. in Mater. Sci., 12, 97 (1963).<br />

[165] K. Kassner, C. Misbah, Phys. Rev. A, 44, 6513 (1991).<br />

[166] A. Karma, A. Sarkissian, M<strong>et</strong>. Trans. A, 27, 635 (1996).<br />

[167] J. S. Langer, Phys. Rev. L<strong>et</strong>t., 44, 1023 (1980).<br />

[168] V. Datye, J. S. Langer, Phys. Rev. B, 24, 4155 (1981).<br />

[169] G. Lesoult, Ann. Chim. Fr., 5 154 (1980), <strong>et</strong> references incluses.<br />

[170] R. Trivedi, J. T. Mason, J. D. Verhoeven, W. Kurz, M<strong>et</strong>all. Trans. , 22A,<br />

252(1991).<br />

[171] K. Brattkus, B. Caroli, C. Caroli, B. Roul<strong>et</strong>, J. Phys. (France), 51, 1847<br />

(1990) ; B. Caroli, C. Caroli, B. Roul<strong>et</strong>, ibid. 51, 1865 (1990).<br />

[172] Y.-J. Chen, S. H. Davis, Acta Mater., 49, 1363 (2001).<br />

[173] G. Dee, J.S. Langer, Phys. Rev. L<strong>et</strong>t., 50, 383 (1983).<br />

[174] W. van Saarloos, Front Propagation into Unstable States : Some Recent Developments<br />

and Surprises, in Nonlinear Evolution of Spatio-Temporal Structures<br />

in Continuous Systems, F. H. Busse and L. Kramer, Eds. (Plenum, New York,<br />

1990), p. 499-508.<br />

[175] S. Fauve, S. Douady, O. Thual, Phys. Rev. L<strong>et</strong>t., 65, 385 (1990) ; S. Fauve, S.<br />

Douady, O. Thual, J. Phys. II, 1, 311 (1991).<br />

[176] S. Fauve, O. Thual, Phys. Rev. L<strong>et</strong>t., 64, 282 (1990).<br />

[177] T. Bohr, M. van Hecke, R. Mikkelsen, M. Ipsen, Phys. Rev. L<strong>et</strong>t., 86, 5482<br />

(2001).<br />

[178] F.R. Mollard, M.C. Flemings, Trans. AIME, 239, 1534 (1967).<br />

[179] V.T. Witusiewicz, L. Sturz, U. Hecht, S. Rex, J. Cryst. Growth, 297, 117<br />

(2006).<br />

[180] T.S. Lo, S. Dobler, M. Plapp, A. Karma, W. Kurz, Acta Mater, 51, 599 (2003).<br />

[181] B. Caroli, C. Caroli, G. Faivre, J. Mergy, J. Cryst. Growth, 118, 135 (1992).<br />

[182] J. C. La Combe, M. B. Koss, L. A. Tennenhouse, E. A. Winsa, M. E. Glicksman,<br />

J. Cryst. Growth, 194, 143 (1998).<br />

[183] Kauerauf B, Zimmermann G, Murmann L, Rex S, J. Cryst. Growth, 193, 701<br />

(1998).<br />

[184] H. Jamgotchian, N. Bergeon, D. Benielli, Ph. Voge, B. Billia, R. Guérin, Phys.<br />

Rev. L<strong>et</strong>t., 87, 166105 (2001).<br />

[185] A. Paris, M. Plapp, Acta Mat., à paraître (2008).<br />

[186] A. Karma, M. Plapp, JOM, 56 (4), 28 (2004).<br />

[187] M. Perrut, Thèse de Doctorat de l’Université Paris VI, octobre 2007.<br />

[188] E. Meca, M. Plapp, M<strong>et</strong>all. Mater. Trans. A, 38, 1407 (2007).<br />

[189] M. Plapp, M. Dejmek, Europhys. L<strong>et</strong>t., 65, 276 (2004).<br />

[190] B. Caroli, C. Caroli, B. Roul<strong>et</strong>, J. Cryst. Growth 68, 677 (1984).<br />

[191] I. Daumont, K. Kassner, C. Misbah, A. Valance, Phys. Rev. E, 55, 6902 (1997).<br />

89


[192] M.M. Sushchik, L.S. Tsimring, Physica D, 74, 90 (1994).<br />

[193] E. Bodenschatz, D.S. Cannell, J.R. de Bruyn, R. Ecke, Y. Hu, K. Lerman, G.<br />

Ahlers, Physica D, 61, 77 (1992).<br />

[194] S. Bottin-Rousseau, A. Pocheau, Phys. Rev. L<strong>et</strong>t., 87, 076101 (2001).<br />

[195] M. Ben Amar, Y. Pomeau, Europhys. L<strong>et</strong>t., 6, 609 (1988).<br />

[196] M. Adda Bedia, V. Hakim, J. Phys. I France, 4, 383 (1994).<br />

[197] M. Adda Bedia, M. Ben Amar, Phys. Rev. E, 51, 1268 (1995).<br />

[198] J.M. Debierre, A. Karma, F. Ce<strong>les</strong>tini, R. Guérin, Phys. Rev. E, 68, 041604<br />

(2003).<br />

[199] J. Maurer, P. Bouissou, B. Perrin, P. Tabeling, Europhys. L<strong>et</strong>t, 8, 67 (1989).<br />

[200] K. Lee, W. Losert J. Cryst. Growth, 269, 592 (2004).<br />

[201] J. A. Warren, J. S. Langer, Phys. Rev. E, 47, 2702 (1993).<br />

[202] A. Karma, Phys. Rev. L<strong>et</strong>t., 87, 115701 (2001).<br />

[203] B. Echebarria, R. Folch, A. Karma, M. Plapp, Phys. Rev. E, 70, 061604 (2004).<br />

[204] T.V. Savina, A.A. Nepomnyashchy, S. Brandon, A.A. Golovin, D.R. Lewin, J.<br />

Cryst. Growth, 237-239, 178 (2002).<br />

90


Annexe<br />

91


PHYSICAL REVIEW E, VOLUME 65, 011702<br />

Dynamics of a fac<strong>et</strong>ed nematic–smectic-B front in thin-sample directional solidification<br />

T. Börzsönyi, 1,2 S. Akamatsu, 1 and G. Faivre 1<br />

1 Groupe de Physique des Solides, CNRS UMR 75-88, Universités Denis Diderot and <strong>Pierre</strong> <strong>et</strong> Marie Curie, Tour 23, 2 place Jussieu,<br />

75251 Paris Cedex 05, France<br />

2 Research Institute for Solid State Physics and Optics, Hungarian Academy of Sciences, H-1525 Budapest, P.O. Box 49, Hungary<br />

Received 5 June 2001; published 13 December 2001<br />

We present an experimental study of the directional-solidification patterns of a nematic–smectic-B front.<br />

The chosen system is C 4 H 9 (C 6 H 10 ) 2 CN in short, CCH4 in 12 m-thick samp<strong>les</strong>, and in the planar<br />

configuration director parallel to the plane of the sample. The nematic–smectic-B interface presents a fac<strong>et</strong> in<br />

one direction—the direction parallel to the smectic layers—and is otherwise rough and devoid of forbidden<br />

directions. We measure the Mullins-Sekerka instability threshold and establish the morphology diagram of the<br />

system as a function of the solidification rate V and the angle 0 b<strong>et</strong>ween the fac<strong>et</strong> and the isotherms. We focus<br />

on the phenomena occurring immediately above the instability threshold when 0 is neither very small nor<br />

close to 90°. Under these conditions, we observe drifting shallow cells and a type of solitary wave, called<br />

‘‘fac<strong>et</strong>on,’’ which consists essentially of an isolated macroscopic fac<strong>et</strong> traveling laterally at such a velocity that<br />

its growth rate with respect to the liquid is small. Fac<strong>et</strong>ons may propagate either in a stationary or an<br />

oscillatory way. The d<strong>et</strong>ailed study of their dynamics casts light on the microscopic growth mechanisms of the<br />

fac<strong>et</strong>s in this system.<br />

DOI: 10.1103/PhysRevE.65.011702<br />

PACS numbers: 64.70.Md, 81.10.Aj, 64.70.Dv, 68.70.w<br />

I. INTRODUCTION<br />

A crystal growing from an undercooled melt rejects heat<br />

and chemical species, which must diffuse away in the liquid<br />

for the process to continue. The thus-generated thermal and<br />

solutal gradients tend to destabilize the advancing solidliquid<br />

interface. This effect is counterbalanced by the surface<br />

tension and the so-called interfacial kin<strong>et</strong>ics, which tends to<br />

slow down the progression of the interface, and hence, stabilize<br />

it. As a result of the comp<strong>et</strong>ition b<strong>et</strong>ween these conflicting<br />

factors, solidification fronts may assume a large vari<strong>et</strong>y<br />

of nonlinear patterns, the characteristics of which<br />

depend on the control param<strong>et</strong>ers, and the initial and boundary<br />

conditions of the process.<br />

The study of solidification patterns has been an active<br />

field of research for several decades 1–3. Most of the existing<br />

studies are devoted to fully nonfac<strong>et</strong>ed systems. In<br />

such systems, the surface tension and the kin<strong>et</strong>ic coefficient<br />

defined as the ratio of the kin<strong>et</strong>ic undercooling to the<br />

growth velocity are nonsingular functions of the orientation<br />

of the interface with respect to the crystal lattice. On a molecular<br />

scale, this corresponds to the fact that the interface is<br />

rough in all orientations. Familiar aspects of the dynamics of<br />

fully nonfac<strong>et</strong>ed systems in directional solidification, i.e.,<br />

when the system is pulled at a constant velocity V toward the<br />

cold side of an applied unidirectional thermal gradient G see<br />

Fig. 1, are the existence of a stable planar front at low values<br />

of V, the primary cellular or Mullins-Sekerka instability<br />

occurring at a threshold velocity V c , the quasiperiodic arrays<br />

of rounded cells at V slightly above V c , and of dendrites at V<br />

much higher than V c . Many dynamical features of these patterns<br />

e.g., stability limits, modes of instability are not y<strong>et</strong><br />

fully understood, but some of their fundamental properties<br />

are now clear, among which the crucial role played by interfacial<br />

anisotropy 1,4,5. In fact, a certain minimum degree<br />

of interfacial anisotropy is a necessary condition for cellular<br />

and dendritic arrays to be stable, or even to exist. In thin<br />

samp<strong>les</strong>—i.e., quasibidimensional 2D systems— and <br />

are functions of a single variable, say, the tilt angle of the<br />

interface with respect to the isotherms. The functions ()<br />

and (), and thus the solidification patterns, depend on the<br />

orientation of the crystal with respect to the solidification<br />

s<strong>et</strong>up 6–8.<br />

In contrast with the case of fully nonfac<strong>et</strong>ed systems, little<br />

is y<strong>et</strong> known about the directional-solidification dynamics of<br />

fac<strong>et</strong>ed crystals. The few existing experimental studies on<br />

this subject first of all show that a distinction must be made<br />

b<strong>et</strong>ween fully and partly fac<strong>et</strong>ed systems 6,9–12. Growth<br />

fac<strong>et</strong>s which most generally, although not necessarily, coincide<br />

with equilibrium fac<strong>et</strong>s 13 correspond to planes of the<br />

crystal containing several directions of strong binding. Fully<br />

fac<strong>et</strong>ed crystals have numerous fac<strong>et</strong> directions, and their<br />

directional-solidification fronts consist of a succession of<br />

fac<strong>et</strong>s limited by sharp edges. The dynamics of such fronts<br />

does not give rise to any stationary state, in general, and<br />

bears no obvious relation with that of nonfac<strong>et</strong>ed fronts.<br />

Partly fac<strong>et</strong>ed systems only have a few fac<strong>et</strong> directions connected<br />

to one another by large rounded regions. In lamellar<br />

crystals, the solid-liquid interface may be rough in all but<br />

one direction, namely, that of the molecular layers. In this<br />

case, when the tilt angle 0 of the layers with respect to the<br />

FIG. 1. Sk<strong>et</strong>ch of a thin-sample directional-solidification s<strong>et</strong>up.<br />

z: axis of the thermal gradient; x: axis parallel to the isotherms; V:<br />

pulling velocity. After a transient, the front advances in average at<br />

the imposed velocity V with respect to the liquid, and thus remains<br />

essentially immobile in the laboratory reference frame. It can then<br />

be continuously observed with an optical microscope.<br />

1063-651X/2001/651/01170211/$20.00<br />

65 011702-1<br />

©2001 The American Physical Soci<strong>et</strong>y


T. BÖRZSÖNYI, S. AKAMATSU, AND G. FAIVRE PHYSICAL REVIEW E 65 011702<br />

isotherms is large, the dynamics of the front must obviously<br />

be that of a nonfac<strong>et</strong>ed crystal as long as the deformation of<br />

the front remains small, that is, below V c and in a small<br />

range of V above V c . Fac<strong>et</strong>s only appear at higher V when<br />

the deformation of the interface is large. A relatively smooth<br />

transition from the nonfac<strong>et</strong>ed to partly fac<strong>et</strong>ed dynamics<br />

may then be observed. This is the experimental configuration<br />

considered in this study.<br />

In this paper, we study the directional-solidification dynamics<br />

of the front associated to the nematic–smectic-B<br />

transition of the liquid-crystal C 4 H 9 (C 6 H 10 ) 2 CN in short,<br />

CCH4. A long-range order exists in the direction perpendicular<br />

to the molecular layers in the smectic-B phase, so<br />

that this phase actually is a lamellar crystal. Previous freegrowth<br />

studies have indeed shown that the nematic–smectic-<br />

B fronts of the n3,4,5 members of the series CCHn where<br />

n stands for the number of carbon atoms in the aliphatic<br />

chain have a single fac<strong>et</strong> direction parallel to the molecular<br />

layers of the smectic phase, and are rough in all other directions<br />

14–16. Moreover, they have no unstable orientations<br />

in a direction perpendicular to the molecular layers, contrary<br />

to the smectic-A –smectic-B fronts previously studied in directional<br />

solidification by Melo and Oswald and Oswald<br />

<strong>et</strong> al. 6,11,12. The present study is performed in thin<br />

(12 m-thick samp<strong>les</strong> and in the planar configuration director<br />

parallel to the plane of the sample, in order for the<br />

front—including the fac<strong>et</strong>s, if any—to remain perpendicular<br />

to the sample plane. Practically, the system is thus a 2D one.<br />

We shall mostly focus on a type of solitary wave appearing<br />

near the Mullins-Sekerka threshold, called ‘‘fac<strong>et</strong>on’’ because<br />

it contains a single small fac<strong>et</strong> traveling along the front<br />

at such a velocity that the normal growth rate of the fac<strong>et</strong>,<br />

i.e., its growth rate with respect to the liquid, is generally<br />

much smaller than V. Such a phenomenon, which has never<br />

been observed before, to the best of our knowledge, is obviously<br />

highly specific to fac<strong>et</strong>ed directional solidification, and<br />

therefore particularly interesting from our present viewpoint.<br />

A preliminary comment about the nematic–smectic-B fac<strong>et</strong>s<br />

in the CCHn series is in order. The growth rate of a fac<strong>et</strong> is<br />

controlled by the dynamics of the molecular steps flowing<br />

along it. Therefore, it crucially depends on wh<strong>et</strong>her or not the<br />

fac<strong>et</strong> contains, or is connected with, step sources 13,17.<br />

When no step source is available, the fac<strong>et</strong> grows through<br />

nucleation and spreading of terraces surface nucleation,<br />

which is a very slow process at low undercooling. In fact, the<br />

growth rate of a perfect fac<strong>et</strong> is totally negligible when the<br />

undercooling is lower than some finite value. Such a behavior<br />

‘‘blocked’’ fac<strong>et</strong>s at low undercoolings has clearly been<br />

observed during the solidification of many, but not all the<br />

studied fac<strong>et</strong>ed systems. What concerns us here is that it was<br />

not observed during the free growth of the smectic-B phase<br />

of CCH3, despite the strongly fac<strong>et</strong>ed aspect of the growing<br />

crystals 14–16. Numerical simulations in which a cusplike<br />

minimum of () but no anisotropy of was taken into<br />

account satisfactorily reproduced the observed growth<br />

shapes. Thus, the observation of a fac<strong>et</strong> on a macroscopic<br />

scale would not necessarily mean the presence of a singularity<br />

in . In order to clarify this point in the case of CCH4,<br />

we report, in Sec. III, preliminary observations in free<br />

FIG. 2. The nematic–smectic-B equilibrium temperature in a<br />

CCH4 sample as a function of time. T NS was measured by controlling<br />

the temperature of a free-growth stage in order to keep a small<br />

smectic-B crystal in quasiequilibrium with the nematic. The relatively<br />

low initial value of T NS indicates that the sample was rather<br />

impure at the outs<strong>et</strong>.<br />

growth showing that the nematic–smectic-B fac<strong>et</strong> of CCH4<br />

is capable of remaining immobile at undercoolings lower<br />

than 0.1 K. Thus, in CCH4 at least, the nematic–smectic-B<br />

fronts can form growth fac<strong>et</strong>s.<br />

II. EXPERIMENT<br />

The relevant material param<strong>et</strong>ers of the liquid-crystal<br />

CCH4 MERCK IS-0558 may be found in Ref. 15. The<br />

residual impurities, the chemical nature of which is unknown,<br />

were characterized as regards solidification by the<br />

usual m<strong>et</strong>hods see below. We found that the impurity content<br />

at the outs<strong>et</strong> of the experiments was reproducible, but<br />

slowly increased during the experiments, indicating that the<br />

product was undergoing a decomposition in the nematic<br />

phase, as previously noticed and analyzed for the case of<br />

CCH3 18. The nematic–smectic-B transition temperature<br />

T NS was generally of about 53.1 K in fresh samp<strong>les</strong>. Figure 2<br />

shows T NS measured as a function of time in one sample. It<br />

can be seen that the decomposition rate is sufficiently slow<br />

not to severely perturb a solidification run, but sufficiently<br />

rapid to prevent us to carry out several successive runs with<br />

the same sample. Outgasing the as-received product resulted<br />

in a significant slowing down of the decomposition process.<br />

We have studied the crystal structure of the smectic-B<br />

phase of CCH4 by low-angle x-ray diffraction 19. As expected,<br />

this phase is basically an AB-type stacking of hexagonal<br />

layers. The param<strong>et</strong>ers are approximately a5.9 Å<br />

and c29 Å, which is in accordance with the data available<br />

for the other members of the homologue series 20. The<br />

hexagonal layers however appear to be slightly distorted,<br />

which may entail the existence of superstructures in the layers.<br />

A schematic view of a thin-sample directionalsolidification<br />

experiment is shown in Fig. 1. A d<strong>et</strong>ailed description<br />

of our s<strong>et</strong>up is given elsewhere 7,8. In this study,<br />

the samp<strong>les</strong> were made of two parallel glass plates separated<br />

by 12-m-thick plastic spacers. Their useful width was of 9<br />

mm and their length of 60 mm. They were filled under an<br />

argon atmosphere at a temperature higher than T NS , and then<br />

011702-2


DYNAMICS OF A FACETED NEMATIC–SMECTIC-B. .. PHYSICAL REVIEW E 65 011702<br />

FIG. 3. Free growth. Successive snapshots of a smectic-B crystal<br />

of CCH4 growing from the nematic phase at T0.07 K.<br />

cooled down to room temperature. Numerous smectic crystals<br />

appeared by h<strong>et</strong>erogeneous nucleation during cooling.<br />

The samp<strong>les</strong> were placed in the thermal gradient, and a<br />

smectic-B crystal of known orientation d<strong>et</strong>ermined through<br />

the observed value of 0 ) was selected by a m<strong>et</strong>hod to be<br />

explained shortly. The sample was annealed at rest for about<br />

30 minutes in order to homogenize the concentration in the<br />

liquid. V was then switched to a chosen value, left for a<br />

given time at this value, and then increased step by step. The<br />

temperature gradient at the growth front was of 53 K cm 1 ,<br />

un<strong>les</strong>s otherwise mentioned. The pulling velocity was in the<br />

range of 0.3–30 ms 1 . The observations were made with<br />

a polarizing microscope Leica equipped with a chargecoupled<br />

device camera. The video signal was analyzed with<br />

digital image processing.<br />

It is obviously crucial for our experiments that large<br />

smectic-B crystals of arbitrary orientation might be selected.<br />

To this aim, we have studied the influence of various treatments<br />

of the inner sides of the glass plates. Three types of<br />

plates were used: untreated plates, plates covered with a monooriented<br />

thin film of polyt<strong>et</strong>rafluor<strong>et</strong>hylene prepared by<br />

friction transfer at T200 °C 21, or with a 100 Å-thick<br />

layer of Al or In deposited by oblique evaporation. In the<br />

nematic phase, the orientation of the director was essentially<br />

planar in all samp<strong>les</strong>. The director was more or <strong>les</strong>s aligned<br />

along the direction of friction, or deposition, in treated<br />

samp<strong>les</strong>, but domains corresponding to small a few degrees<br />

variations in the orientation of the director, still existed see<br />

Fig. 4 below. This inhomogeneity of the nematic phase<br />

caused but minor perturbations in our experiments, since the<br />

phenomena of interest turned out to be essentially independent<br />

of the orientation of the nematic director. In all samp<strong>les</strong>,<br />

the smectic-B phase had a planar orientation, but was divided<br />

into different crystals or grains corresponding to a<br />

different value of 0 . The surface treatment gave a pronounced<br />

preferential distribution of 0 among the various<br />

grains, facilitating the selection of the desired value of 0 .<br />

The size of the selected smectic-B grain was increased by a<br />

m<strong>et</strong>hod consisting of forcing the crystal to grow through a<br />

funnel-shaped obstacle 7. By this m<strong>et</strong>hod, smectic grains of<br />

a millim<strong>et</strong>ric width, and arbitrary values of 0 were obtained.<br />

III. CHARACTERIZATION OF THE SYSTEM<br />

A. Free growth at small undercoolings<br />

The observations reported in this section were performed<br />

with a free-growth s<strong>et</strong>up similar to the one described in Ref.<br />

FIG. 4. Directional solidification in this, and all the following<br />

micrographs, growth is upwards. N: nematic; Sm1: smectic-B;<br />

Sm2: smectic-B oriented differently from Sm1. a Sample at rest<br />

(V0); b sample in the process of solidification at V<br />

0.9 ms 1 . Note the domains in the nematic. Sm1 is a single<br />

crystal, but Sm2 is a polycrystal, as shown by the presence of cusps<br />

on the Sm1-Sm2 front.<br />

22, in which the changes in the undercooling are produced,<br />

via the Clausius-Clapeyron effect, by a sudden pressure<br />

change at constant temperature, and are therefore quasiinstantaneous<br />

22,23. The samp<strong>les</strong> were the same as those<br />

used in directional solidification. At the beginning of the experiments,<br />

the samp<strong>les</strong> were heated step by step until only<br />

one small smectic-B crystal was left in the nematic. The<br />

sample was maintained at constant temperature until the<br />

changes in the shape of the crystal became very slow this<br />

took about 20 minutes. Admittedly, this shape is not the<br />

exact equilibrium shape of the crystal, but it exhibits clear<br />

reproducible features, namely, long fac<strong>et</strong>s parallel to the<br />

smectic layers and rounded ends in the perpendicular direction<br />

Fig. 3 a, which is enough for our present purpose. It<br />

should be noted that the observed near-equilibrium shape<br />

clearly shows the absence of a forbidden orientation range<br />

around 90°, where is the deviation of the interface<br />

from the direction of the molecular layers, but suggests that<br />

the fac<strong>et</strong> might actually be limited by a sharp edge, i.e., the<br />

interface might be unstable at small values of . The fact<br />

that we have not observed the Herring instability 24,11 in<br />

directional solidification at the lowest-explored value of 0<br />

indicates that this forbidden orientation range is very small<br />

(2°), if it exists at all.<br />

A sudden increase of the undercooling T was applied at<br />

time t0, and the subsequent growth of the crystal recorded<br />

Fig. 3. The growth process, which is governed by the anisotropic<br />

interfacial properties and diffusivities, is very complicated.<br />

Its study is beyond the scope of this paper. Here, we<br />

limit ourselves to the following observation: the fac<strong>et</strong>s of the<br />

smectic-B crystals remained blocked within experimental<br />

uncertainty their growth rate was lower than about<br />

0.01 ms 1 ) at undercoolings lower than 0.1 K Fig. 3.At<br />

higher undercoolings, they generally grow at a measurable<br />

rate. The apparent threshold undercooling T nucl for growth<br />

by surface nucleation of our system is thus larger than 0.1 K<br />

and probably not much larger than this value. This estimate<br />

of T nucl is small compared to what it is in ordinary solid-<br />

011702-3


T. BÖRZSÖNYI, S. AKAMATSU, AND G. FAIVRE PHYSICAL REVIEW E 65 011702<br />

liquid systems, but this may be explained by the small value<br />

of in our system 25,26. It is also possible that in our thin<br />

samp<strong>les</strong>, surface nucleation is in fact h<strong>et</strong>erogeneous, i.e.,<br />

takes place preferentially along the line of contact with the<br />

glass plates. The nucleation rate would then depend on the<br />

treatment of the glass plates.<br />

B. Directional solidification: Instability threshold<br />

The Mullins-Sekerka instability threshold was found to lie<br />

b<strong>et</strong>ween approximately 2 and 3ms 1 in all the studied<br />

fresh samp<strong>les</strong>. No influence of the orientation of the smectic,<br />

or the nematic was observed within experimental uncertainty.<br />

However, it should be noted that this uncertainty was<br />

large (1 ms 1 ) for the reason to be explained presently.<br />

Figure 4 shows a sample at rest, and pulled at a rate lower<br />

than V c . Two isothermal fronts are visible, namely, a front<br />

separating the nematic N phase from a smectic-B domain<br />

Sm1, and at a lower temperature, a front separating Sm1<br />

from a second smectic-B domain Sm2. The nature of the<br />

transition from Sm1 to Sm2 is not y<strong>et</strong> clear. This transition<br />

was observed in most, but not all experiments. Observations<br />

not reported here incline us to think that Sm2 is the same<br />

phase as Sm1, but with a different orientation, thus, a different<br />

interaction energy with the glass plates. In any case, we<br />

need not take into account the Sm1-Sm2 front here since this<br />

front, when present, does not perturb the dynamics of the<br />

N-Sm1 front.<br />

It can be seen in Fig. 4 that the nematic–smectic-B front<br />

remains planar during solidification at VV c , except for<br />

small, long-wavelength distortions due to the presence of<br />

domains in the nematic phase. These distortions are larger<br />

during solidification than at rest, and undergo sudden<br />

changes each time the front leaves a nematic domain for<br />

another. This phenomenon has thus an equilibrium as well as<br />

a kin<strong>et</strong>ic origin. In our experiments, it plays the role of a<br />

relatively strong, long-wavelength, low-frequency noise,<br />

which blurs some of the morphological-transition thresholds<br />

of the system. This is the main origin of the aforementioned<br />

large uncertainty on the measured values of V c . However,<br />

we may state with certainty that V c was higher than<br />

2 ms 1 since the distortions caused by nematic domains,<br />

or any other source of perturbation e.g., dust partic<strong>les</strong> did<br />

not amplify below this velocity.<br />

C. Solute redistribution transient<br />

When V is smaller than V c , the front reaches a stationary<br />

planar state through the so-called solute redistribution transient.<br />

A recoil curve—i.e., the curve representing the variation<br />

of the position or temperature of the planar front as a<br />

function of time during the initial transient of a particular<br />

run—is reproduced in Fig. 5. It is well known that information<br />

about the relevant properties of the solute diffusion coefficient<br />

D in the liquid, partition coefficient K, thermal gap<br />

T o may be gained from the characteristics of the transient,<br />

and the value of V c . We have utilized this m<strong>et</strong>hod in order to<br />

characterize the unknown impurity playing the role of solute<br />

in our system.<br />

FIG. 5. Recoil curve at V0.9 ms 1 . Same run as in Fig. 4.<br />

Continuous line: best fit according to the Warren-Langer approximation.<br />

The rapid decrease at the ons<strong>et</strong> of the recoil is an instrumental<br />

effect.<br />

The threshold velocity, and the amplitude of the solute<br />

redistribution transient are given by V c (1KD s /<br />

D)DG/T o (D s : diffusion coefficient in the solid and T o ,<br />

respectively, 27. By fitting the recoil data using the Warren-<br />

Langer approximate theory 28 Fig. 5 and assuming V c<br />

2.5 ms 1 and KD s /D1, we obtained K0.12. This<br />

gives D80 m 2 s 1 and T o 0.2 K. These data give us<br />

no information about D s , but there is good reason to believe<br />

that our system is a two-sided one—i.e., that D s is not much<br />

smaller than D 12.<br />

IV. RESULTS<br />

A. Morphology diagram<br />

A diagram displaying the observed morphologies as a<br />

function of the pulling velocity and the orientation of the<br />

smectic-B crystal is shown in Fig. 6.<br />

It can be seen that the sequence of morphologies observed<br />

as a function of V for a fixed value of 0 is the same for all<br />

FIG. 6. Morphology diagram. Measurement points: waves and<br />

fac<strong>et</strong>ons (), fac<strong>et</strong>ons and unstationary fac<strong>et</strong>ed fingers (), unstationary<br />

fac<strong>et</strong>ed fingers (), stationary fac<strong>et</strong>ed fingers (), and<br />

unstable fac<strong>et</strong>s x. Heavy dashed line: Mullins-Sekerka instability<br />

threshold. Ins<strong>et</strong> micrographs: see Fig. 7.<br />

011702-4


DYNAMICS OF A FACETED NEMATIC–SMECTIC-B. .. PHYSICAL REVIEW E 65 011702<br />

FIG. 7. The different growth morphologies observed as a function<br />

of V for 0 25°. a Planar front; b drifting shallow cells; c<br />

drifting fac<strong>et</strong>on stationary mode; d drifting fac<strong>et</strong>ons oscillatory<br />

mode at different stages of their oscillation cycle; see Fig. 18 below;<br />

e nonstationary array of fac<strong>et</strong>ed fingers ; f stationary array<br />

of fac<strong>et</strong>ed fingers.<br />

values of 0 , except for those close to 0° fac<strong>et</strong>s parallel to<br />

the growth front or 90° fac<strong>et</strong>s perpendicular to the growth<br />

front. This generic sequence is illustrated in Fig. 7.<br />

Small-amplitude, nearly sinusoidal traveling waves appear<br />

near the instability threshold Fig. 7b, in accordance<br />

with previous observations in two-sided anisotropic systems<br />

6. Such weakly nonlinear waves are commonly called<br />

‘‘shallow cells.’’<br />

We observed drifting shallow cells in a broad range of V<br />

around the threshold (1 ms 1 V8 ms 1 ). In the<br />

same range of V, we also observed ‘‘fac<strong>et</strong>ons’’ Fig. 7c.<br />

These solitary waves may propagate in a stationary or an<br />

oscillatory way. They appear when the amplitude of the cells<br />

is so large that the tilt angle of the front locally reaches the<br />

value 0 corresponding to the fac<strong>et</strong>s. Most generally, this<br />

occurs under the effect of perturbations due to the nematic<br />

domains. The frequency of creation of fac<strong>et</strong>ons, and thus,<br />

their average number by unit length of the front increases as<br />

V increases. When the average spacing of the fac<strong>et</strong>ons becomes<br />

smaller than their width (200 m), they cease to<br />

behave as non-interacting objects. In fact, they disappear altog<strong>et</strong>her,<br />

giving way to arrays of much narrower objects,<br />

called fac<strong>et</strong>ed fingers Fig. 7e. This occurs at about<br />

8 ms 1 . However, this transition is strongly noise dependent,<br />

and thus, relatively ill defined from an experimental<br />

viewpoint. Shallow cells and fac<strong>et</strong>ons are studied in d<strong>et</strong>ail in<br />

the next section.<br />

The arrays of fac<strong>et</strong>ed fingers, which are observed above<br />

8 ms 1 exhibit a relatively sharp transition from an unstationary<br />

Figs. 7e and 8 to a stationary dynamics as V<br />

increases Figs. 7f and 9; the dispersion appearing in Fig. 6<br />

is mostly due to the aging of the samp<strong>les</strong>. The spatiotemporal<br />

diagrams of the unstationary arrays shown in Fig. 8 reveal<br />

the transitory or local existence of well-defined oscillatory<br />

modes. These modes become more and more apparent as V<br />

increases because the oscillation period T osc is a rapidly decreasing<br />

function of V Fig. 10a. This strongly suggests<br />

the existence of a homogeneous oscillatory bifurcation of the<br />

FIG. 8. Transition from isolated fac<strong>et</strong>ons to fac<strong>et</strong>ed fingers for<br />

0 70° in an aged sample. a V3.1 ms 1 snapshot of the<br />

front; b corresponding spatiotemporal diagram time series of the<br />

intensity distribution along a line located 20 m below the front;<br />

c V6.5 ms 1 ; d corresponding spatiotemporal diagram; e<br />

V13.5 ms 1 . Note that another grain ( 0 73°) appears in the<br />

leftmost part of the figure; f corresponding spatiotemporal diagram.<br />

high-V stationary patterns as V decreases within some narrow<br />

range of spacing.<br />

We now turn to the particular orientations corresponding<br />

to the bounds of the scanned interval of 0 . When 0 90°,<br />

the system is reflection symm<strong>et</strong>ric. Shallow cells no longer<br />

drift, and fac<strong>et</strong>ons cease to exist. The shallow cells break up<br />

into narrow fac<strong>et</strong>ed fingers as V is increased above threshold<br />

Fig. 11a. The widest fac<strong>et</strong>ed fingers, which are the majority<br />

ones, are not reflection symm<strong>et</strong>ric, whereas the narrowest<br />

ones are reflection symm<strong>et</strong>ric. The two opposite but equivalent<br />

directions of symm<strong>et</strong>ry breaking are equally populated.<br />

The resulting arrays were nonstationary even at the highestexplored<br />

values of V Fig. 11b. This is very different from<br />

what was observed by Oswald <strong>et</strong> al. in smectic-A-smectic-B<br />

fronts for a similar orientation of the fac<strong>et</strong> 6. In that system,<br />

because of the existence of forbidden directions, the finger<br />

tips exhibited pointed triangular shapes, and formed stationary<br />

arrays.<br />

When 0 is sufficiently close to zero, the growth front of<br />

smectic-B grains is entirely occupied by a fac<strong>et</strong> at any value<br />

011702-5


T. BÖRZSÖNYI, S. AKAMATSU, AND G. FAIVRE PHYSICAL REVIEW E 65 011702<br />

FIG. 9. Stationary array of fac<strong>et</strong>ed fingers at V13.5 ms 1<br />

and 0 24°. a Snapshot of the front; b spatiotemporal diagram<br />

piling up of skel<strong>et</strong>onized images of the growth front.<br />

of V. This may be considered as a finite-size effect resulting<br />

from the following fact: fac<strong>et</strong>s are always present in the<br />

grooves attached to grain boundaries for whatever values of<br />

0 and V; the stationary size of these fac<strong>et</strong>s is more or <strong>les</strong>s<br />

proportional to 1/(tan 0 ); they thus occupy the whole grain<br />

when 0 is lower than a certain value, which is of about 2°<br />

for a grain size of 500 m. At sufficiently high V, these long<br />

fac<strong>et</strong>s break up through the mechanism illustrated in Fig.<br />

11c. It is not necessary to repeat here the description of this<br />

process, which has been presented by other authors 12. We<br />

simply note that, in our fresh samp<strong>les</strong>, this instability was<br />

observed to result from the occasional collisions of the front<br />

with defects domain walls, dust partic<strong>les</strong> present in the<br />

nematic. In the <strong>les</strong>s pure samp<strong>les</strong>, it was superseded by another<br />

well-known process, namely, the nucleation of crystals<br />

in the undercooled melt ahead of the front 6. Both mechanisms<br />

give rise to more or <strong>les</strong>s permanently cyclic growth<br />

regimes.<br />

B. Near-threshold patterns<br />

1. Drifting shallow cells<br />

Most generally shallow cells appeared in the form of a<br />

noise-induced wave pack<strong>et</strong>. A spontaneous homogeneous<br />

growth of the cells was never observed with certainty. We<br />

FIG. 10. Oscillation period a as a function of V for 0 56°<br />

b as a function of 0 for three values of V. The leftmost point in<br />

a corresponds to an isolated oscillatory fac<strong>et</strong>on.<br />

attribute this fact to the interplay b<strong>et</strong>ween shallow cells and<br />

fac<strong>et</strong>ons see below. At, or below 2 ms 1 , noise-induced<br />

wave pack<strong>et</strong>s systematically disappeared when the source of<br />

noise disappeared, as already mentioned. At higher V, they<br />

evolved as illustrated in Figs. 12 and 13.<br />

A careful analysis of the spatiotemporal diagram of Fig.<br />

12 has shown that i the cells are initially sinusoidal; ii<br />

they grow in amplitude with a uniform amplification rate of<br />

0.002 s 1 ; iii the amp<strong>les</strong>t cells are no longer sinusoidal<br />

at the end of the time sequence, iv the spacing and the<br />

drift velocity V d are uniform in space and constant in time.<br />

V d is thus amplitude independent. This is in keeping with the<br />

idea that this sequence is the initial stage of the usual amplification<br />

process leading from a linearly unstable state to a<br />

stationary weakly nonlinear regime. The final regime was not<br />

observed because the process was interrupted by an external<br />

perturbation giving rise to a fac<strong>et</strong>on.<br />

The traces on the lefthand side of Fig. 13 are the trajectories<br />

of three oscillatory fac<strong>et</strong>ons. These objects are studied<br />

below. For now, the point of interest is that the rearmost<br />

fac<strong>et</strong>on leaves behind a region of the front that is free of<br />

d<strong>et</strong>ectable shallow cells see also Figs. 15 and 18 below.<br />

The cells reappear at 200 m from the fac<strong>et</strong>on, and then<br />

amplify following a process entirely similar to the above<br />

FIG. 11. a Array of<br />

symm<strong>et</strong>ry-broken fac<strong>et</strong>ed fingers<br />

at 0 90° and V10 ms 1 ;<br />

b corresponding spatiotemporal<br />

diagram; c instability of a fac<strong>et</strong><br />

at 0 2° and V10 ms 1 .<br />

011702-6


DYNAMICS OF A FACETED NEMATIC–SMECTIC-B. .. PHYSICAL REVIEW E 65 011702<br />

FIG. 12. Spatiotemporal diagram of a drifting wave pack<strong>et</strong>; V<br />

3.1 ms 1 , 0 25°.<br />

one, except for two points: i in the present case, the amplification<br />

rate (0.02 s 1 ) is much larger than in the preceding<br />

case, since V is higher, and ii a stationary regime of<br />

nonlinear shallow cells is reached. This confirms clearly, although<br />

only semiquantitatively, that the system admits stationary<br />

weakly nonlinear cellular states within a measurable<br />

range of V above V c . These states are m<strong>et</strong>astable with respect<br />

to the formation of fac<strong>et</strong>ons. Also, we note that the<br />

direction of drift of the cells is opposite to that of fac<strong>et</strong>ons.<br />

This is somewhat of a surprise since, in other systems, shallow<br />

cells and fac<strong>et</strong>s have been found to drift in the same<br />

direction 6.<br />

The measured values of V d and are plotted in Fig. 14<br />

as a function of 0 for a given value of V. The data are<br />

compatible with the fact that V d ( 0 ) must go to zero at 0<br />

0° and 90° for symm<strong>et</strong>ry reasons. The maximum is at<br />

about 70°, and corresponds to a relatively large value of<br />

V d /V, indicating that the system is strongly anisotropic even<br />

in the orientation range in which the interface is rough.<br />

We have noted above that V d seems to be independent of<br />

the amplitude of the cells. It is thus legitimate to admit but<br />

not certain that the measured value of V d is the same as in<br />

the linear regime. We have performed a linear stability analysis<br />

of the planar front of a two-sided system taking into account<br />

the anisotropy of the diffusion in the two bulk phases<br />

nematic and smectic-B), and that of the linear kin<strong>et</strong>ic coefficient<br />

the anisotropy of does not come into play in a<br />

linear calculation 4. We have solved the dispersion equation<br />

numerically under various assumptions concerning the<br />

orientation dependences of D, D s , and , which are not<br />

known. Qualitatively, the results may be summed up as follows<br />

27. We find that the observed sign and absolute value<br />

of V d ( 0 ) could be ascribed to diffusion anisotropy only if,<br />

in the smectic-B phase, the impurities diffused much faster<br />

through the smectic layers than parallel to them, which is<br />

very unlikely to be true. Thus, the observed drift of the shallow<br />

cells is most probably due to kin<strong>et</strong>ic anisotropy. In such<br />

FIG. 13. Spatiotemporal diagram. The three traces on the lefthand<br />

side are the trajectories of oscillatory fac<strong>et</strong>ons drifting leftwards.<br />

Note the disappearance of the cells which drift rightwards<br />

in the wake of the rearmost fac<strong>et</strong>on. A temporary exception to this<br />

rule is visible near the end of the recording, when the fac<strong>et</strong>on emits<br />

a pack<strong>et</strong> of three or four cells. This exception is only apparent,<br />

however, since this occurs during a period of time when the fac<strong>et</strong>on<br />

no longer exists it is drifting rightwards. V6.5 ms 1 , 0<br />

55°, recording time 250 s.<br />

a case, the sign of V d is given by d/d 4. In conclusion,<br />

the observed direction of drift of the shallow cells if it<br />

is really the same as in the linear regime indicates that, in<br />

our system, increases as increases. This result poses no<br />

particular problem except for the vicinal domain, in which <br />

is expected to be more or <strong>les</strong>s proportional to the reciprocal<br />

of the step density, and hence, to the reciprocal of 29.<br />

The crossover from the vicinal to the rough domains as <br />

increases should thus manifest itself through a change in the<br />

sign of V d . It is tempting to assume that this crossover cor-<br />

011702-7


T. BÖRZSÖNYI, S. AKAMATSU, AND G. FAIVRE PHYSICAL REVIEW E 65 011702<br />

FIG. 16. Normal growth velocity of fac<strong>et</strong>s belonging to fac<strong>et</strong>ons<br />

or arrays of fac<strong>et</strong>ed fingers as a function of the tilt angle of the fac<strong>et</strong><br />

for the indicated values of the pulling velocity. In the case of oscillatory<br />

fac<strong>et</strong>ons, the minimum value of V n has been plotted. The data<br />

point at 0 2° corresponds to the fac<strong>et</strong> shown in Fig. 11c prior to<br />

its destabilization.<br />

FIG. 14. Drift velocity a and wavelength b of the cells as a<br />

function of the tilt angle of the fac<strong>et</strong> at V6.5 ms 1 .<br />

responds to the zero of V d ( 0 ), which perhaps appears near<br />

12° in Fig. 14a. However, the observation of macroscopic<br />

fac<strong>et</strong>s drifting in the same direction as the shallow cells disproves<br />

this assumption, and indicates that the vicinal domain<br />

is actually very narrow in our system see below.<br />

2. Stationary fac<strong>et</strong>ons<br />

The spatiotemporal diagram of a stationary fac<strong>et</strong>on is<br />

shown in Fig. 15. Clearly, a fac<strong>et</strong>on is a solitary wave consisting<br />

of a macroscopic fac<strong>et</strong> and a broad rounded finger<br />

separated from each other by a very thin liquid groove. The<br />

regularity of the spatiotemporal diagram shows that fac<strong>et</strong>ons,<br />

once formed, are quite stable. In particular, they absorb the<br />

shallow cells that they may encounter ahead of themselves<br />

without being modified, and seem to be insensitive to the<br />

perturbations caused by the nematic domains. The depth of<br />

the fac<strong>et</strong>—i.e., the distance z f b<strong>et</strong>ween the two edges of the<br />

fac<strong>et</strong> along the z axis—is difficult to measure with accuracy<br />

because the lower edge, located near the bottom of the<br />

groove, is generally not resolved. However, it is certain that<br />

z f is in the 3050 m range the difference of temperature<br />

T f b<strong>et</strong>ween the two edges is thus in the range 0.15–<br />

0.25 K, and decreases as 0 increases. The upper edge of the<br />

fac<strong>et</strong> corresponds to a small pointed maximum of the front<br />

shape, but it is not possible to decide wh<strong>et</strong>her, or not, this<br />

edge is sharp on a molecular scale. The width of the rounded<br />

finger—i.e., the extension of the deformed region of the front<br />

behind the finger tip—is of about 200 m. As mentioned,<br />

shallow cells do not develop in this region of the front. The<br />

trajectory of the fac<strong>et</strong>on makes a small angle with the direction<br />

of the macroscopic fac<strong>et</strong>, indicating that the normal<br />

growth rate of the fac<strong>et</strong> is small but finite. Thus, the fac<strong>et</strong> is<br />

not blocked, and the question arises as to its microscopic<br />

growth mechanisms.<br />

Figure 16 displays a large number of values of the normal<br />

velocity of fac<strong>et</strong>s V n measured in isolated fac<strong>et</strong>ons as well as<br />

in arrays of fac<strong>et</strong>ed fingers for various values of V and 0 .In<br />

spite of a large dispersion of the data, it is clear that V n is<br />

essentially a nonzero quantity that decreases as 0 increases,<br />

and increases as V increases. The regularity of the stationary<br />

fac<strong>et</strong>ons or arrays see Figs. 15 and 9, and the fact that V n is<br />

very close to zero when 0 is large allow us to exclude screw<br />

dislocation growth as the dominant mechanism. Moreover,<br />

the fact that both V n and z f are decreasing functions of 0<br />

suggests that V n is essentially d<strong>et</strong>ermined by events occurring<br />

near the lower edge of the fac<strong>et</strong>. One may imagine either<br />

that surface nucleation takes place at a relatively high rate at<br />

this point, or that the fac<strong>et</strong> is supplied with steps coming<br />

from the bottom of the groove where the interface is necessarily<br />

rough. In both cases, V n would be very sensitive to the<br />

d<strong>et</strong>ails of the conformation of the interface in this region.<br />

These d<strong>et</strong>ails may depend on the treatment of the glass<br />

plates, which could explain the dispersion b<strong>et</strong>ween values<br />

measured in different samp<strong>les</strong>.<br />

FIG. 15. Stationary fac<strong>et</strong>on. 0 25°, V6.5 ms 1 . a<br />

Snapshot of the front. The faint dark line appearing in the solid in<br />

the continuation of the fac<strong>et</strong> is a thin liquid groove; see Fig. 17. b<br />

Spatiotemporal diagram. The normal growth rate of the fac<strong>et</strong> is<br />

V n 0.9 ms 1 .<br />

3. Oscillating fac<strong>et</strong>ons<br />

Figure 17 shows a process of formation of fac<strong>et</strong>ons in<br />

response to a perturbation. Macroscopic fac<strong>et</strong>s progressively<br />

develop on one side of the shallow cells as the amplitude of<br />

the latter increases. These fac<strong>et</strong>s first drift with the same<br />

velocity as the shallow cells, and then change their direction<br />

of drift. This change is not accompanied by any modification<br />

in the orientation of the fac<strong>et</strong>s within experimental uncer-<br />

011702-8


DYNAMICS OF A FACETED NEMATIC–SMECTIC-B. .. PHYSICAL REVIEW E 65 011702<br />

often adopt an oscillatory mode of propagation Fig. 18.<br />

Obviously, this oscillation consists of a more or <strong>les</strong>s ample<br />

cycle b<strong>et</strong>ween the aforementioned rapid and slow regimes.<br />

The conditions under which fac<strong>et</strong>ons are stationary, or oscillatory,<br />

could not be d<strong>et</strong>ermined. In fact, stationary fac<strong>et</strong>ons<br />

were observed much <strong>les</strong>s frequently than, and always in coexistence<br />

with oscillating fac<strong>et</strong>ons. Moreover, some oscillating<br />

fac<strong>et</strong>ons were regular Fig. 18, but most of them were<br />

irregular Figs. 13 or 19. It is possible that the system intrinsically<br />

admits stationary, periodic, and more or <strong>les</strong>s, chaotic<br />

fac<strong>et</strong>ons. However, the following explanation is also<br />

possible.<br />

A careful inspection of Fig. 18 reveals that the transition<br />

of the oscillating fac<strong>et</strong>ons from a slow to a rapid regime<br />

corresponds to a sudden pinching off of the liquid groove,<br />

whereas the reverse transition from a rapid to a slow regime<br />

consists of a progressive deepening of the groove. If we focus<br />

on the sole groove, this behavior is strongly reminiscent<br />

of the periodic pinching off called cusp instability of the<br />

intercell grooves in nonfac<strong>et</strong>ed cellular fronts 30. This in-<br />

FIG. 17. Fac<strong>et</strong>ons appearing in response to a perturbation. Spatiotemporal<br />

diagram. V6.5 ms 1 , 0 25°, recording time: 60<br />

s. Note the opposite signs of the drift velocities of the cells and the<br />

fac<strong>et</strong>s.<br />

tainty (0.5°). Thus, the same macroscopic fac<strong>et</strong> may be in<br />

two different microscopic states, or growth regimes. One of<br />

these the ‘‘slow’’ regime is that of the stationary state, discussed<br />

in the preceding section, while the other the ‘‘rapid’’<br />

regime corresponds to a rough interface. As announced, we<br />

are thus led to assume that the crossover from vicinal to<br />

rough interfaces occurs at values of lower than 0.5° in<br />

our system. This is indeed surprising since this disorientation<br />

corresponds to a very low density of steps <strong>les</strong>s than 1 per<br />

m), but not impossible. We also note that the persistence of<br />

a macroscopic fac<strong>et</strong> while the interface is rough on a microscopic<br />

scale is explainable by the sole singularity of the <br />

plot 16.<br />

The existence of two different growth regimes of a macroscopic<br />

fac<strong>et</strong> is confirmed by the fact that fac<strong>et</strong>ons most<br />

FIG. 18. Oscillating fac<strong>et</strong>on. 0 42°, V6.5 ms 1 . a<br />

Snapshots of the front at different stages of an oscillation period. b<br />

Spatiotemporal diagram.<br />

011702-9


T. BÖRZSÖNYI, S. AKAMATSU, AND G. FAIVRE PHYSICAL REVIEW E 65 011702<br />

FIG. 19. Spatiotemporal diagram showing shallow cells, oscillating<br />

fac<strong>et</strong>ed solitary waves, and microfac<strong>et</strong>s arrow, 0 36°, V<br />

3.1 ms 1 , and G25 K cm 1 .<br />

stability, we recall, is most probably of a capillary origin<br />

Rayleigh instability 31, and very sensitive to the lattice<br />

defects that, in the nonfac<strong>et</strong>ed systems, are often attached to<br />

the groove—in fact, the grooves to which subboundaries<br />

low-angle grain boundaries are attached are not subject to<br />

the cusp instability 32. If, by analogy, we assume that the<br />

intercell groove of fac<strong>et</strong>ons, similar to that of nonfac<strong>et</strong>ed<br />

cells, is intrinsically subject to an oscillatory Rayleigh instability,<br />

we are led to the conclusion that the transition of the<br />

fac<strong>et</strong> from a slow to a rapid regime is a secondary effect due<br />

to changes occurring in the configuration of the interface<br />

near the lower edge of the fac<strong>et</strong>. The presence of lattice<br />

defects e.g., sub-boundaries emerging into the liquid at the<br />

bottom of the groove may hinder these changes, suppressing<br />

the oscillation. This would explain that fac<strong>et</strong>ons are much<br />

more often oscillatory than stationary.<br />

4. Lattice defects<br />

Some lattice defects mostly, grain boundaries may be<br />

d<strong>et</strong>ected with the optical microscope thanks to the fact that<br />

they create macroscopic depressions grooves of the growth<br />

front around the point at which they emerge into the liquid.<br />

In our system, these grooves must be partly fac<strong>et</strong>ed during<br />

solidification. We have lowered the applied thermal gradient<br />

in some experiments in order to facilitate the observation of<br />

such grooves. This allowed us to reveal that the growth front<br />

of our system is often swept by very small fac<strong>et</strong>s, called<br />

microfac<strong>et</strong>s, certainly attached to lattice defects emerging<br />

into the liquid.<br />

We observed several types of microfac<strong>et</strong>s, corresponding<br />

probably to different types of lattice defects. The microfac<strong>et</strong>s<br />

of the type shown in Fig. 19 were relatively easy to identify<br />

because they travel at a perfectly constant velocity, catching<br />

up, and running through all the other structures of the front,<br />

in particular, fac<strong>et</strong>ons. Their drift velocity has thus most<br />

probably the maximum possible value, i.e., the value corresponding<br />

to totally blocked fac<strong>et</strong>s. They must be attached to<br />

lattice defects—stacking faults, or twist subboundaries—<br />

strongly locked onto the lamella plane of the smectic. However,<br />

these microfac<strong>et</strong>s seem to have but little effect on the<br />

dynamics of the front. They indeed provoke an instantaneous<br />

slowing down of the macroscopic ‘‘rapid’’ fac<strong>et</strong>s when they<br />

collide with them see Fig. 19, but do not trigger a durable<br />

transition to the slow regime. So this observation, whatever<br />

its intrinsic interest may be, does not cast light on the question<br />

of the possible role played by lattice defects in the dynamics<br />

of the fac<strong>et</strong>ons.<br />

V. DISCUSSION<br />

We have shown that the directional solidification of a<br />

nematic–smectic-B front in the planar configuration gives<br />

rise to a wealth of interesting nonlinear phenomena, the most<br />

striking of which are the stationary or oscillatory ‘‘fac<strong>et</strong>ons’’<br />

encountered in the vicinity of the Mullins-Sekerka threshold.<br />

These observations raise numerous unsolved problems concerning<br />

the microscopic growth mechanisms of the fac<strong>et</strong>s, as<br />

well as the nonlinear dynamics of the observed macroscopic<br />

patterns. An important question is wh<strong>et</strong>her these phenomena<br />

are specific of the nematic–smectic-B fronts, or are of frequent<br />

occurrence in partly fac<strong>et</strong>ed fronts. In order to clarify<br />

this point, we are currently searching for similar phenomena<br />

in more conventional, partly fac<strong>et</strong>ed solidification fronts.<br />

Also, numerical simulations based on a phase-field m<strong>et</strong>hod<br />

are in progress in order to test the consistency of the numerous<br />

conjectures that we have been led to make in order to<br />

explain the peculiar dynamical features of the fac<strong>et</strong>ons.<br />

ACKNOWLEDGMENTS<br />

The authors wish to thank A-M. Levelut for the help in<br />

characterizing CCH4 with x-ray diffraction, T. Tóth-Katona<br />

and Á. Buka for many useful discussions, and A. Fleury and<br />

C. Picard for their technical assistance. We are also grateful<br />

to MERCK Darmstadt for kindly providing us with CCH4.<br />

T.B. would like to thank the European Commission for financial<br />

support.<br />

1 Solids Far from Equilibrium, edited by C. Godrèche Cambridge<br />

University Press, Cambridge, England, 1992.<br />

2 M. Cross and P. Hohenberg, Rev. Mod. Phys. 65, 851 1993.<br />

3 J. Langer. Rev. Mod. Phys. 52, 11980.<br />

4 S.R. Coriell and R.F. Sekerka, J. Cryst. Growth 34, 157 1976.<br />

5 P. Kopczynski, W.-J. Rappel, and A. Karma, Phys. Rev. L<strong>et</strong>t.<br />

77, 3387 1996.<br />

6 F. Melo and P. Oswald, Phys. Rev. L<strong>et</strong>t. 64, 1381 1990.<br />

7 S. Akamatsu, G. Faivre, and T. Ihle, Phys. Rev. E 51, 4751<br />

1995.<br />

011702-10


DYNAMICS OF A FACETED NEMATIC–SMECTIC-B. .. PHYSICAL REVIEW E 65 011702<br />

8 S. Akamatsu and G. Faivre, Phys. Rev. E 58, 3302 1998.<br />

9 D.K. Shangguan and J.D. Hunt, M<strong>et</strong>all. Trans. A 22A, 941<br />

1991.<br />

10 L.M. Fabi<strong>et</strong>ti and R. Trivedi, J. Cryst. Growth 182, 185 1997.<br />

11 P. Oswald, F. Melo, and C. Germain, J. Phys. France 50,<br />

3527 1989.<br />

12 F. Melo and P. Oswald, J. Phys. II 1, 353 1991.<br />

13 W.K. Burton, N. Cabrera, and F.C. Frank, Philos. Trans. R.<br />

Soc. London 243, 299 1951.<br />

14 Á. Buka, T. Tóth-Katona, and L. Kramer, Phys. Rev. E 51, 571<br />

1995.<br />

15 T. Tóth-Katona, T. Börzsönyi, Z. Váradi, J. Szabon, Á. Buka,<br />

R. González-Cinca, L. Ramirez-Piscina, J. Casademunt, and A.<br />

Hernández-Machado, Phys. Rev. E 54, 1574 1996.<br />

16 R. González-Cinca, L. Ramirez-Piscina, J. Casademunt, A.<br />

Hernández-Machado, T. Tóth-Katona, T. Börzsönyi, and Á.<br />

Buka, Physica D 99, 359 1996.<br />

17 A.A. Chernov, Contemp. Phys. 30, 251 1989.<br />

18 T. Tóth-Katona, N. Éber, and Á. Buka, Mol. Cryst. Liq. Cryst.<br />

Sci. Technol., Sect. A 328, 467 1999.<br />

19 A.M. Levelut unpublished.<br />

20 R. Brownsey and A. Leadb<strong>et</strong>ter, J. Phys. France L<strong>et</strong>t. 42, 135<br />

1981.<br />

21 P. Damman, M. Dosière, M. Brunel, and J.C. Wittmann, J. Am.<br />

Chem. Soc. 119, 4633 1997.<br />

22 T. Börzsönyi, T. Tóth-Katona, Á. Buka, and L. Gránásy, Phys.<br />

Rev. E 62, 7817 2000.<br />

23 J.C. La Combe, M.B. Koss, L.A. Tennenhouse, E.A. Winsa,<br />

and M.E. Glicksman, J. Cryst. Growth 194, 143 1998.<br />

24 C. Herring, Phys. Rev. 82, 871951.<br />

25 D.R. Ulhmann, in Advances in Nucleation and Crystallization<br />

in Glasses American Ceramic Soci<strong>et</strong>y, Columbus, OH, 1971.<br />

26 D.R. Ulhmann, in Nucleation and Crystallization in Glasses<br />

American Ceramic Soci<strong>et</strong>y, Columbus, OH, 1982.<br />

27 T. Börzsönyi, S. Akamatsu, and G. Faivre unpublished.<br />

28 J.A. Warren and J.S. Langer, Phys. Rev. E 47, 2702 1993.<br />

29 A.A. Chernov, S.R. Coriell, and B.T. Murray, J. Cryst. Growth<br />

132, 405 1993.<br />

30 P. Kurowski, S. de Cheveigné, G. Faivre, and C. Guthmann, J.<br />

Phys. France 50, 3007 1989.<br />

31 K. Brattkus, J. Phys. France 50, 2999 1989.<br />

32 S. Bottin-Rousseau, S. Akamatsu, and G. Faivre unpublished.<br />

011702-11


The Formation of Lamellar-Eutectic Grains in Thin Samp<strong>les</strong><br />

SILVÈRE AKAMATSU, SÉBASTIEN MOULINET, and GABRIEL FAIVRE<br />

We present an experimental study of the formation of lamellar-eutectic grains in directional solidification<br />

of thin hypereutectic samp<strong>les</strong> of the transparent nonfac<strong>et</strong>ed alloy CBr 4 -C 2 Cl 6 . We start solidification<br />

from a partly stabilized solid residue. This solid is in a single phase ( phase) along the solidliquid<br />

interface. The successive stages of the transient leading to the final lamellar structure are (1)<br />

the solute redistribution transient of the -liquid front; (2) the appearance, without nucleation, of<br />

seeds of the other solid phase ( phase) onto the front; (3) the growth of the phase along the -<br />

liquid front (primary invasion); (4) the secondary invasion of the newly formed -liquid front by the<br />

phase; and (5) the oscillatory instability, called periodic lamellar branching, occurring during the<br />

secondary invasion. We study stages (2) through (5) in d<strong>et</strong>ail. Stages (2) through (4) are similar to<br />

those leading to banded microstructures in peritectics. Stage (5) is specifically responsible for the<br />

ons<strong>et</strong> of two-phase growth and the formation of eutectic grains.<br />

I. INTRODUCTION The formation of eutectic grains is thus an important aspect<br />

of the transient, which needs to be clarified.<br />

THE fine microstructure of directionally solidified lamel-<br />

The experimental investigation of the processes occurring<br />

lar eutectics is the trace left behind in the solid by the<br />

during the transient is made difficult by the broad ranges<br />

periodic stationary pattern that the solid-liquid interface took<br />

of characteristic lengths (from <strong>les</strong>s than 1 m to several<br />

on during growth. Such stationary growth patterns are the<br />

millim<strong>et</strong>ers) and times (from <strong>les</strong>s than 1 second to several<br />

result of a dynamical balance b<strong>et</strong>ween the comp<strong>et</strong>ing effects<br />

minutes) involved. This difficulty can be surmounted by<br />

of solutal diffusion in the liquid and capillary forces at the<br />

using the m<strong>et</strong>hod of thin-sample directional solidification<br />

interface. [1] During the last decade, much progress has been<br />

of a transparent analogue of m<strong>et</strong>allic eutectics (CBr 4 -C 2 Cl 6 ),<br />

made in the understanding of the dynamical characteristics<br />

which allows a continuous follow-up of dynamical processes<br />

of lamellar eutectic solidification patterns (stability range<br />

with the adequate spatiotemporal resolution. We report here a<br />

and modes of instability) thanks to numerous experimental<br />

d<strong>et</strong>ailed experimental study of the initial transient of lamellar<br />

and numerical studies, [2–11] but the mechanisms through<br />

eutectic growth carried out with this m<strong>et</strong>hod.<br />

which they appear during the early stages of directional<br />

Concerning the relevance of this study to bulk samp<strong>les</strong>,<br />

solidification are still largely unclear, in spite of many valuthe<br />

following remark is in order. The thickness of the samp<strong>les</strong><br />

able experimental investigations. [12,13] Usually, the solid in<br />

used in this study (12 m) is comparable to lamellar<br />

contact with the liquid is in a single phase at the ons<strong>et</strong> of<br />

spacing. In such very thin samp<strong>les</strong>, the lamellar plane is<br />

directional solidification. Before the final lamellar two-phase<br />

constrained to remain perpendicular to the plane of the sampattern<br />

becomes established, a complex transient, including<br />

ple by confinement and capillary effects in whatever eutectic<br />

a series of distinct processes (appearance of a seed of the<br />

grain (Figs. 1(b) and 2). This may indeed cause the dynamiinitially<br />

absent phase, spreading of this phase onto the front,<br />

cal processes observed in our samp<strong>les</strong> to be somewhat differand<br />

rearrangement in order to form lamellae), has to take<br />

ent from what they would be in bulk samp<strong>les</strong>. We r<strong>et</strong>urn to<br />

place. A good understanding of these processes is the key<br />

this point in the discussion of the results.<br />

to a fine control of the eutectic microstructure.<br />

Lamellar eutectics are most generally composed of large<br />

regions (large compared to lamellar spacing), called eutectic II. EXPERIMENTAL METHODS<br />

grains, in which all the lamellae of either of the two phases<br />

The phase diagram of the binary system CBr 4 -C 2 Cl 6 is<br />

have the same orientation (Fig. 1). [12] Recent studies have<br />

reproduced in Figure 3. Some useful material constants are<br />

shown that there is a strong, although not obvious, connecgiven<br />

in Table I. The CBr 4 -rich phase is face-centered<br />

tion b<strong>et</strong>ween this crystallographic feature, and the regularity<br />

of the lamellar microstructure. Most of the often observed<br />

cubic, and the C 2 Cl 6 -rich phase is body-centered cubic.<br />

irregularities can be ascribed to the existence of eutectic<br />

The alloys are prepared by mixing zone-refined and outgased<br />

CBr 4 and C 2 Cl 6 . [7] grain boundaries. [4] Thus, the larger the eutectic-grain size<br />

The uncertainty on the molar fraction C <br />

is the more perfect the lamellar microstructure. Perfectly<br />

of C 2 Cl 6 in the samp<strong>les</strong> is of about 0.002. Most of the<br />

periodic lamellar patterns were obtained on a macroscopic<br />

experiments were performed at the slightly hypereutectic<br />

concentration of 0.122. This concentration corresponds to a<br />

scale in (thin) samp<strong>les</strong> containing a single eutectic grain. [8]<br />

-phase fraction in the solid of about 0.35 and to a minimumundercooling<br />

spacing [1] of about 16.4 m atV 1 m s 1<br />

(V: growth velocity).<br />

SILVÈRE AKAMATSU and GABRIEL FAIVRE, Senior Researchers The samp<strong>les</strong> are made of two parallel glass plates sepa-<br />

(CNRS), are with the Groupe de Physique des Solides, CNRS UMR 7588, rated by plastic spacers, delimiting an empty space about 8-<br />

Universités Denis-Diderot <strong>et</strong> <strong>Pierre</strong>-<strong>et</strong>-Marie-Curie, 75251 Paris, Cedex<br />

05, France. SÉBASTIEN MOULINET, PhD Student, is with the Laboratoire mm wide, 70-mm long, and 12-m thick. They are filled<br />

de Physique Statistique, Ecole Normale Supérieure, 75005 Paris, France. by capillarity at about 100 C and then quenched to room<br />

Manuscript submitted May 15, 2000.<br />

temperature in order to avoid macrosegregation. The<br />

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 32A, AUGUST 2001—2039


Fig. 1—Sk<strong>et</strong>ch of the eutectic grain structure of directionally solidified<br />

lamellar eutectics. The different shades symbolize the different orientations<br />

of one phase (the lamellae of the other phase are left blank). (a) Bulk<br />

sample. The lamellar plane is free to rotate and takes on different orientations<br />

in different grains. (b) Thin samp<strong>les</strong>. The lamellar plane is constrained to<br />

remain perpendicular to the plates enclosing the sample. : lamellar spacing,<br />

t: sample thickness, and z: pulling direction.<br />

Fig. 3—Phase diagram of the CBr 4 -C 2 Cl 6 alloy. [7] Also see Table I.<br />

the microscope stage of amplitude 3w/4, allowing a nearly<br />

automatic linking of the images.<br />

Because one of the aims of this study was to provide<br />

precise quantitative data, which might be compared with<br />

numerical simulations in the near future, we prepared samp<strong>les</strong><br />

in a specific type of initial state (by initial state, we<br />

mean the state at the ons<strong>et</strong> of solidification) following a<br />

m<strong>et</strong>hod, which we will be able to explain only at the end<br />

of this article. About 20 solidification runs of this type<br />

were used. The characteristics of four typical runs, labeled<br />

1 through 4, are given in Table II. Runs 2 through 4 were<br />

performed in one sample.<br />

Fig. 2—Stationary lamellar-eutectic pattern in a thin sample of a CBr 4 -<br />

C 2 Cl 6 alloy (G 80Kcm 1 , and V 0.5 m s 1 ). In this micrograph,<br />

as in the following ones, the growth direction is upward.<br />

III. INITIAL STATE<br />

When the samp<strong>les</strong> are inserted into the solidification<br />

device, they undergo directional melting. The coldest part<br />

quenched samp<strong>les</strong> have a very fine two-phase microstructure.<br />

of the samp<strong>les</strong> is not melted and keeps the pre-existing two-<br />

The mixing process and the filling of the samp<strong>les</strong> are phase microstructure. The samp<strong>les</strong> are maintained at rest in<br />

carried out under a low pressure of argon. The amount of the thermal gradient for a certain time, which may range<br />

residual gas in the samp<strong>les</strong> is low (5 10 4 ) [14] and has from a few minutes to several hours. During this annealing<br />

no noticeable effect on the phenomena studied here.<br />

treatment, the system relaxes toward a partial equilibrium<br />

The samp<strong>les</strong> are pulled at constant V in an externally state (no diffusion in the solid has time to take place), in<br />

imposed thermal gradient G with a d.c. motor via a microm<strong>et</strong>ric<br />

which (1) the concentration in the liquid is uniform, and<br />

screw. The d<strong>et</strong>ails of the s<strong>et</strong>up are given elsewhere. [5–9] equal to C ; (2) the solid residue is in contact with the liquid<br />

Two values of G (70 and 110 K cm 1 ) have been used. through a continuous layer of the phase, which extends<br />

Absolute temperatures are not known precisely, but the tem- b<strong>et</strong>ween the liquidus temperature T l (Figure 3) and the<br />

perature difference b<strong>et</strong>ween points of the same field of view eutectic temperature T E ; (3) below T E , the solid remains in<br />

is known with a precision that is only limited by the uncer- a two-phase state; and (4) the grain boundaries of the -<br />

tainty of G (5 pct). The used range of V is 0.12 to 0.9 m phase layer, if any, lie perpendicular to the front and are<br />

s 1 . The error on V is 0.01 m s 1 . We define a xyz only manifested by small cusps appearing along the front<br />

reference frame, as explained in Figure 1. The images of at T l (Figure 4(a)). In our samp<strong>les</strong>, annealing times of several<br />

the growth front are continuously observed with an optical hours must be applied in order to reach such a state. When<br />

microscope (Leica). They are recorded with the help of a this is done, the transient follows the scenario that is briefly<br />

CCD camera and a videotape recorder and are analyzed studied in Section VI. Except in that section, we focus on<br />

numerically. [15] Two magnifications are used, corresponding another type of scenario, which is observed after much<br />

to values of the width w of the field of view of 190 and 625 shorter annealing times (typically, 30 minutes).<br />

m, respectively. In order to follow the lateral invasions<br />

across the entire width of the sample, the front is scanned<br />

along the x direction by means of rapid stepwise shifts of<br />

After only a short stay at rest, the initial state of the system<br />

generally presents two types of deviations from equilibrium:<br />

(1) the concentration in the liquid is not perfectly uniform,<br />

2040—VOLUME 32A, AUGUST 2001<br />

METALLURGICAL AND MATERIALS TRANSACTIONS A


Table I. Physical Constants of the CBr 4 -C 2 Cl 6 Alloy [8] *<br />

T E C C E mol% C mol% m K/mol% K d nm C mol% m K/mol% K d nm D cm 2 s 1<br />

84.4 11.6 8.8 0.8 0.75 11 18.5 1.65 1.5 3.5 5 10 6<br />

*T E ,C E : temperature and C 2 Cl 6 mole fraction at the eutectic point; C , m , K , d ( , ): C 2 Cl 6 mole fraction, slope of the liquidus,<br />

partition coefficient, and capillary length (defined as d a /m (C C ), where a is the Gibbs–Thomson coefficient) at the edge<br />

of the eutectic plateau; and D: diffusion coefficient in the liquid.<br />

Table II. Characteristics of Runs 1 through 4<br />

Control Param<strong>et</strong>ers<br />

C G V<br />

Run Label (Mole Fraction) (K cm 1 ) (m s 1 )<br />

1 0.123 70 0.5<br />

2 0.128 110 0.25<br />

3 0.128 110 0.25<br />

4 0.128 110 0.5<br />

Fig. 5—Sample after a 10-min annealing at rest (G 70 K cm 1 ). :<br />

thickness of the layer.<br />

Fig. 4—The first five stages of the transient. Each sk<strong>et</strong>ch shows the entire<br />

sample. (a) Sample at rest (thin vertical lines: grain boundaries), (b) appearance<br />

of invasion seeds, (c) primary invasion; (d ) secondary invasion,<br />

and (e) periodic lamellar branching.<br />

(1) The temperature at the -liquid front decreases under<br />

the effect of the accumulation of the rejected species<br />

(CBr 4 ) in the liquid. [17] We call this process “front recoil”<br />

(this, and the other terms put in quotation marks below,<br />

is nonstandard) because one observes a displacement of<br />

the planar front toward low temperatures in the labora-<br />

tory reference frame. Simultaneously, very thin fingers<br />

of the phase grow toward the front along grain bound-<br />

aries. This is explained as follows. The tip of these<br />

fingers is in contact with the liquid and the phase and<br />

must, therefore, remain at a fixed temperature and, thus,<br />

a fixed position in the laboratory reference frame, while<br />

the front is recoiling. The tip temperature of the fingers<br />

is close to, but significantly lower than, T E because of<br />

the strong curvature of the -liquid interface;<br />

(2) When the front reaches the tip temperature of the<br />

fingers, these fingers grow into small protruding <br />

crystals, called “invasion seeds” (Figure 4(b)). Invasion<br />

seeds can also appear by other mechanisms, as will be<br />

seen subsequently.<br />

(3) After a certain lag time, the crystals begin to grow<br />

so that the front is not perfectly planar, and its average<br />

temperature T 0 is somewhat lower than T l and (2) the liquid<br />

remains in contact with the partic<strong>les</strong> located below T E<br />

through various defects crossing the layer, in particular,<br />

shallow liquid channels running along grain boundaries.<br />

Broadly speaking, these out-of-equilibrium features are<br />

stronger the shorter is the annealing time. The annealing<br />

time can thus be used as a rough control param<strong>et</strong>er in order<br />

to vary the initial conditions of the runs. However, the d<strong>et</strong>ail<br />

of the residual imperfections (in particular, their spatial distribution)<br />

is generally not controlled.<br />

The aspect of a solid residue after a short stay at rest is<br />

shown in Figure 5. Numerous grains boundaries are running<br />

across the -phase layer. Since grain boundaries alone do<br />

not give rise to any optical contrast, the faint dark lines which<br />

reveal them originate from a certain type of h<strong>et</strong>erogeneity<br />

attached to them, namely, shallow liquid channels lying<br />

b<strong>et</strong>ween the glass plates and the solid (hence, their weak<br />

contrast). The origin of such channels is obvious (the <br />

phase layer is formed by the coa<strong>les</strong>cence of crystals initially<br />

dispersed in the liquid), but their long persistence time<br />

is surprising. We shall not consider this problem here, but<br />

simply note that it is part of the more general problem of<br />

the annealing process of the solid residue (which also<br />

includes, for instance, the problem of the migration of liquid<br />

and gaseous inclusions [16] ) and that it involves the differential<br />

w<strong>et</strong>ting of the container walls by the various phases.<br />

Incidentally, the theor<strong>et</strong>ical value of the thickness of<br />

the phase layer is m (C C E )/G (m : slope of the<br />

liquidus). Measuring provides a means of cross-checking<br />

the value of C in situ (in the case of Figure 5, one finds<br />

C 0.120, in good agreement with the nominal concentra-<br />

tion of 0.122) and testing the uniformity of the concentration<br />

along the front. In our samp<strong>les</strong>, the relative variation of C <br />

in the lateral direction was <strong>les</strong>s than 2 pct.<br />

IV.<br />

THE SUCCESSIVE STAGES OF THE<br />

TRANSIENT<br />

Starting from an incompl<strong>et</strong>ely stabilized initial state (Figure<br />

4(a)), the following transient is generally observed.<br />

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 32A, AUGUST 2001—2041


Fig. 6—A primary invasion tongue propagating rightward (G 110 K<br />

cm 1 , and V 0.5 m s 1 ). The invasion rate at the time of the snapshot<br />

is T 18 ms 1 . The cusps in the - interface are former grain boundary<br />

emergence points.<br />

Fig. 7—(a) A secondary invasion tongue propagating rightward (G <br />

110Kcm 1 , and V 0.13 m s 1 ). T 9 m s 1 . The initial layer<br />

is visible below the layer resulting from the primary invasion. (b) Fully<br />

developed lamellar-branching invasion regime (G 110 K cm 1 , and<br />

V 0.25 m s 1 ); T 30 m s 1 .<br />

Fig. 9—Recoil curves in reduced coordinates (text). Open symbols: sample<br />

with a concentration out of the eutectic plateau (C 0.2, G 80 K<br />

cm 1 , and V 0.5 m s 1 ). Filled symbols: slightly hypereutectic sample<br />

(C 0.13, G 70Kcm 1 , and V 0.5 m s 1 ).<br />

spacing) and is, therefore, unstable against lamella termination.<br />

[1,8,11] The final stage of the transient consists<br />

Fig. 8—Alternate layers of and phases (“banding”) resulting from of repeated lamella termination events, which stop when<br />

successive invasions (G 110Kcm 1 , and V 0.13 ms 1 ). An invasion the spacing has become larger than the instability threshof<br />

a front by an tongue is in progress.<br />

old all over the front. The system is then in a stationary<br />

state, although the lamellar spacing distribution is generally<br />

laterally, triggering the process called “primary invasion.”<br />

nonuniform.<br />

Since their lateral growth velocity (“invasion In the solidification runs selected for this study, the primary<br />

rate”) is much larger than V, these crystals rapidly take<br />

invasion started from a single invasion seed (this was<br />

the form of thin layers with pointed tips, called “tongues” obtained by applying a series of short preliminary solidification<br />

(Figures 4(c) and 6). The primary invasion ends when<br />

runs and annealing treatments), so that the primary and<br />

the layer covers the front entirely. The solid in secondary invasions could cover long distances (2 to 6 mm).<br />

contact with the liquid is then essentially a single-phase<br />

solid. We stress that primary invasion takes only a<br />

few tenths of seconds.<br />

V. RECOIL OF THE FRONT<br />

(4) After another lag time, the “secondary invasion” of the When the front recoil is not interrupted by any instability,<br />

newly formed front by the phase takes place: it lasts until the front reaches the solidus temperature<br />

invasion seeds emerge onto the front and then grow<br />

laterally, forming secondary invasion tongues (Figures<br />

T s . Defining the recoil as T i (t) T l<br />

T i (t), or (t) <br />

T i (t)/G, where T i (t) is the temperature of the front at<br />

4(d) and 7(a)).<br />

time t, and t 0 the ons<strong>et</strong> of solidification, the final value<br />

(5) The nature of the next step depends on the value of V.<br />

In this study, we focus on a V range (approximately<br />

of the recoil is T T l<br />

T s ,orl th T /G. We define<br />

the reduced undercooling u T i /T /l th (undercooling<br />

0.2 V 1 m s 1 ), in which the secondary invasion<br />

of a liquid of concentration C and temperature T i with<br />

tongues undergo a dynamical instability, called “periodic respect to the liquidus), which increases from 0 to 1 during<br />

lamellar branching” (Figures 4(e) and 7(b)). This instability,<br />

the recoil.<br />

which was noted previously by Lemaignan, [13] is It is well known that the terminal part of the (noninter-<br />

the mechanism through which lamellar patterns appear. rupted) recoil is exponential with a characteristic time d <br />

Incidentally, we note that, at lower V, periodic lamellar D/V 2 , where D is the diffusion coefficient in the liquid (the<br />

branching does not occur. In this case, alternate invasions<br />

partition coefficient does not appear here because it is close<br />

by and recur indefinitely, giving rise to banded to 1 in our system). [17,19] In this study, d varied from 600<br />

patterns (Figure 8). This permanent oscillatory regime, to 35,000 seconds and was always much longer than the<br />

which bears a striking similarity with the so-called band- time at which a primary invasion occurred. We are thus only<br />

ing of directionally solidified peritectics, [18] is not studied<br />

concerned by the early stages of the recoil, during which<br />

here.<br />

the recoil time must not be scaled with d but with th <br />

(6) The lamellar pattern delivered by the periodic lamellar l th /V (this follows from the fact that d/dt V at t 0). [19]<br />

branching mechanism has a very small spacing (typi- Figure 9 shows two reduced recoil curves u (t/ th ). One<br />

cally, 4 times smaller than the minimum-undercooling of them was measured in a sample with a concentration<br />

2042—VOLUME 32A, AUGUST 2001<br />

METALLURGICAL AND MATERIALS TRANSACTIONS A


Fig. 11—Invasion subsequent to nucleation in a sample subject to a prolonged<br />

annealing prior to solidification (C 0.138, G 80 K cm 1 , and<br />

V 0.65 m s 1 ).<br />

Fig. 10—Recoil curves in reduced coordinates for runs 1 through 4 (Table<br />

II). t 1 : ons<strong>et</strong> of invasion, and u 1 : value of u at t t 1 .<br />

outside the eutectic plateau (C 0.2 C ), and the other<br />

in a hypereutectic sample (C 0.13). It can be seen that<br />

the aforementioned scaling is verified since the slope of the<br />

curves at t 0 is close to 1, and the two curves perfectly<br />

match. In the case of the hypereutectic sample, u Fig. 12—Three types of invasion seeds: (a) emerging from a liquid channel,<br />

stops to<br />

(b) conveyed by a gas bubble, and (c) pre-existing as a remnant of a<br />

increase when the lamellar-eutectic structure appears. Inci- previous lamellar microstructure.<br />

dentally, it can be verified that, beyond this time, T i remains<br />

constant and equal to T l<br />

T E , as it should, indicating that<br />

the sample was free of measurable residual-impurity effects. of an intercell groove and is followed by a very rapid inva-<br />

During the primary invasion, the not y<strong>et</strong> invaded region sion of the undercooled part of the liquid by a finely disof<br />

the front continues to recoil in the same way as before persed two-phase solid (Figure 11). The measured value of<br />

the ons<strong>et</strong> of the invasion. Figure 10 shows the recoil curves the nucleation undercooling is of a few Kelvin (2.5 K in<br />

measured during runs 1 through 4 while the primary invasion Figure 11), a value typical for h<strong>et</strong>erogeneous nucleation.<br />

was in progress. The quantity (t) was recorded along a The invasion rate is in the 10 4 ms 1 range, which corresvertical<br />

line intersecting the front at a point located far ahead ponds to a spacing in the 10 1 m range, according to the<br />

of the invasion tongue. The position of the line was shifted V 1/2 scaling law. [1] The lamellar structure coarsens by<br />

along x tog<strong>et</strong>her with the field of view each time the micro- lamellar termination as the eutectic front approaches T E ,<br />

scope stage was moved in order to follow the invasion. It until a steady, but nonuniform, spacing distribution is<br />

can be seen that u (t) remains essentially linear during the obtained, as explained previously.<br />

invasion. The slight variations of slope from one run to It is striking that the eutectic grains resulting from this<br />

another are explainable by small differences in the initial process are small, and of the particular crystallographic type<br />

conditions of the different runs.<br />

called “locked” eutectic grains in a previous study. [4] This<br />

appellation refers to the fact that, in such grains, the direction<br />

VI. INVASION SUBSEQUENT TO NUCLEATION of the lamella plane remains essentially locked onto a preferred<br />

direction. Most probably, this direction is that of a<br />

In this section, we study the transient in samp<strong>les</strong> having<br />

low-energy orientation of the - interface, implying that<br />

undergone a prolonged annealing at rest. This is a digression the two phases have a singular orientation relationship. Thus,<br />

with respect to the main purpose of this work, which allows the prevalence of locked grains in the microstructure sugus<br />

to emphasize the crucial role played by the treatments gests that repeated epitaxial nucleation of one phase onto<br />

preparatory to solidification. We consider the case when the other occurred during the invasion process. However,<br />

there is no liquid channel allowing the phase to reach the<br />

this remains to be confirmed by further studies.<br />

front by continuous growth. The phase must appear by<br />

nucleation from the liquid, which requires a relatively large<br />

value of the undercooling of the liquid with respect to the<br />

VII. PRIMARY INVASION<br />

liquidus T l<br />

T i (T l : temperature of the liquidus).<br />

Accordingly, it is observed that the front recoil continues<br />

A. Invasion Seeds<br />

for a long time after T i has reached T E . On the other hand, Figure 12 illustrates three possible modes of initiation of<br />

the experiments are performed at values of V substantially a primary invasion. The corresponding solidification runs<br />

higher than the cellular-instability threshold velocity of the are those labelled 1 through 3 in Table II. In run 1 (Figure<br />

front. The formation of deep cells takes place before the 12(a)), the invasion seed grew out of a liquid channel associ-<br />

phase nucleation. Nucleation then occurs at the bottom ated with a grain boundary at t 25 seconds. In run 2<br />

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 32A, AUGUST 2001—2043


Fig. 13—Invasion rate T as a function of the reduced time t/ th for runs<br />

1 through 4. Ins<strong>et</strong>: same T data as a function of real time t.<br />

Fig. 14—Invasion rate T and front shape O as a function of x for run 3.<br />

The curves have been smoothed on a length scale of about 0.1 mm.<br />

(Figure 12(b)), the phase was conveyed to the front by a<br />

gas inclusion migrating in the solid at t 244 s. (Note that<br />

such events may occur at any time when V is smaller than<br />

the migration rate of the bubb<strong>les</strong>). Run 3 (Figure 12(c)) was<br />

preceded by a short pulling at low V and a very short annealing<br />

at rest. Consequently, at the ons<strong>et</strong> of this run, the lamellar<br />

pattern inherited from the preliminary low-V run was still<br />

in contact with the liquid over a large portion of the front.<br />

As the pulling was started, the single-phase region recoiled<br />

while the temperature of the lamellar region remained close<br />

to T E (the calculated average undercooling of the lamellar<br />

pattern was 0.025 K [1] ). Figure 12(c) shows the invasion<br />

starting from an lamella adjacent to a region at t <br />

90 seconds.<br />

B. Invasion Rate<br />

Fig. 15—Primary invasion tongue. Top: experiment (G 80 K cm 1 ,<br />

and V 0.25 m s 1 ); T 0.4 m s 1 . A skel<strong>et</strong>onized version of the<br />

image is superimposed onto the micrograph. Bottom: perspective sk<strong>et</strong>ch.<br />

The lateral growth velocity, or invasion rate, of invasions<br />

was measured as follows. On a time series of photographs, of the front (Section II). In the case of run 3, Figure 14<br />

we measured the coordinates x T and z T (where T stands for shows T as a function of x T tog<strong>et</strong>her with the shape o (x)<br />

Trijunction point) of the tongue tip as a function of time by of the front before the invasion. The shape o (x) exhibits<br />

pointing the foremost point of the trijunction line on the a deep depression b<strong>et</strong>ween x 0 and x 200 m corres-<br />

computer screen. The scatter in the thus obtained measure- ponding to the boundary b<strong>et</strong>ween the lamellar pattern and<br />

ments is relatively large (1.5 m). We smoothed the x T the front in Figure 12(c), and then undulations of much<br />

data on a time scale of about 1 second and calculated the smaller amplitude (2 m) and larger extension (1 mm)<br />

invasion rate T dx T /dt from the smoothed x T (t) data. resulting from the long-range variations of the concentration<br />

In Figure 13, the invasion rates measured during runs 1 in the liquid. It can be seen that there is a clear correlation<br />

through 4 are plotted as a function of the reduced time b<strong>et</strong>ween the variations of T and those of o . The invasion<br />

t/ th . The differences of lag time b<strong>et</strong>ween the runs are clearly experiences an extra acceleration when the slope of o is<br />

visible. However, T globally increases as t/ th increases negative and a deceleration (or a decrease of the acceleration)<br />

following the same law in all the runs (the apparent discrep- when it is positive.<br />

ancy b<strong>et</strong>ween curve 1 and the other curves can be eliminated<br />

by an appropriate choice of the zero time). Note that the<br />

different curves merge at long times in the (t/ th , T ) coordinates,<br />

C. Shape of the Invasion Tongue<br />

but not in the (t, T ) ones (ins<strong>et</strong> in Figure 13). A There is clear evidence that the shape of the tongues<br />

closer examination reveals that the increase in T is regular is actually tridimensional (3-D) near the tip in spite of the<br />

(but not linear) in runs 1 and 4, but irregular, and even small thickness of our samp<strong>les</strong>. Figure 15 shows an tongue<br />

nonmonotonous, in runs 2 and 3.<br />

at an early stage of an invasion and, thus, at a small value<br />

The ample modulations of T (t) observed during runs 2 of the invasion rate. The relatively large curvature of the<br />

and 3 can clearly be ascribed to the slightly nonplanar shape - interface along the transverse direction is revealed by<br />

2044—VOLUME 32A, AUGUST 2001<br />

METALLURGICAL AND MATERIALS TRANSACTIONS A


Fig. 16—Contours of six invasion tongues for the values of T given in<br />

ins<strong>et</strong> (C 0.124, V 0.5 m s 1 , and G 70 K cm 1 ). The foremost<br />

points of the tips have been made to coincide. Ins<strong>et</strong>: Tongue thickness at<br />

5 m from the tip as a function of T . The curve is the best-fit power law<br />

(exponent 0.17).<br />

Fig. 17— T as a function of T T T 1 T T for runs 1 through 4. T 1 has<br />

been adjusted in order to obtain a good superposition of the terminal part<br />

of the curves.<br />

the two lines of strong contrast appearing behind the tip.<br />

This curvature is negative, i.e., the - interface bulges<br />

toward the hot temperatures. A sk<strong>et</strong>ch of the inferred shape D. Discussion<br />

of an tongue is given in Figure 15. For simplicity, symm<strong>et</strong>ry<br />

about the median plane of the sample has been assumed, From a fundamental viewpoint, the central aspect of the<br />

although there are indications that the tongue is actually not findings in Section C is that the invasion rate does not tend<br />

symm<strong>et</strong>ric with respect to this plane, either for instrumental toward a steady value, but increases continually during the<br />

reasons (e.g., because of a slight transverse bias in the temof<br />

the formation of bands in peritectic alloys performed by<br />

invasion. It is worth noting that recent numerical simulations<br />

perature field) or as a result of a nonlinear symm<strong>et</strong>ry breaking<br />

effect.<br />

Lo <strong>et</strong> al. lead to a similar conclusion. [20] No theor<strong>et</strong>ical<br />

However, the dynamics of a tongue tip is essentially conbest<br />

explanation of this fact is available for the moment, to our<br />

trolled by the excess of solute concentration, which runs<br />

knowledge. Some remarks, however, can be made. The<br />

ahead of it along the x-axis. The thickness of this excess driving force for the invasion is the undercooling of the<br />

concentration layer is of a few l T D/ T , and, although it liquid with respect to the liquidus at some distance ahead<br />

decreases as T increases, it remains much larger than the of the invasion tongue. In reduced units, this undercooling<br />

thickness of the sample during most of the process (l is u [T l T i ]/[T l T s ](T s T <br />

: temperature of the sol-<br />

50 m in the case of Figure 15). The dynamics of the idus). Since u increases during the invasion as a conseinvasion<br />

is thus essentially two-dimensional (2-D), except quence of the front recoil, it is tempting to conclude that<br />

near the tip. It is therefore not a surprise that the 2-D contour the continual acceleration of the invasion is simply due to<br />

of the tongue is a function of the sole variable T , within the front recoil. The sensitivity of the invasion rate to the<br />

experimental uncertainty, as was verified by comparing conin<br />

the same way, since a positive (negative) slope of o<br />

irregular shape of the front could apparently be explained<br />

tours observed at the same rate T for different values of V<br />

(data not shown). Figure 16 shows the contours of six corresponds to a decrease (increase) of the value of u experitongues<br />

observed at the same values of the control parame- enced by the invasion tongue as it advances along the front.<br />

ters but at different stages of the invasion process and, thus, In practice, this view amounts to assuming that T is a<br />

at various values of T . Note the following.<br />

function of the sole variable z T , since the difference b<strong>et</strong>ween<br />

and z T is generally negligible. A close examination of the<br />

(1) The radius of curvature at the trijunction point is very data, however, reveals that it is not the case. Figure 17 shows<br />

small ( 0.8 m for T 4 m s 1 ). It cannot be the T data of runs 1 through 4 plotted as a function of<br />

measured accurately, but clearly decreases when T T T Gz T . It can be seen that the curves that were nonmonot-<br />

increases.<br />

onous in the ( T , t) variab<strong>les</strong> remain nonmonotonous in the<br />

(2) The front just ahead of the tip has a positive curvature. ( T , z T ) variab<strong>les</strong>. This demonstrates that the sole variations<br />

The temperature T T of the trijunction point is thus of u encountered by an invasion tongue sweeping a nonplanar<br />

slightly higher than the temperature T i of the invaded<br />

front do not account for the observed variations of T .<br />

front far ahead of the tongue. The length over which In other words, the invasion rate depends on the local value<br />

the front is curved is comparable to l T , as it should of the undercooling, and on some additional, perhaps nonlo-<br />

be, and thus decreases as T increases.<br />

cal, effects, which remain to be identified.<br />

(3) The global shape of the tongue can be roughly characterized<br />

Neverthe<strong>les</strong>s, these additional effects remain quite moderthe<br />

by the distance h b<strong>et</strong>ween the - interface and ate and, thus, are negligible from a practical viewpoint, as<br />

-liquid interface at a given distance from the tip. long as the growth front does not present sharp bumps or<br />

The value of h decreases monotonously as T increases troughs. This is clearly shown by the merging of the curves<br />

(ins<strong>et</strong> in Figure 16).<br />

T (t/ th ) at long times in Figure 13 and is further illustrated<br />

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 32A, AUGUST 2001—2045


Fig. 20—Formation of a invasion seed following the me<strong>et</strong>ing of two<br />

primary invasion tongues.<br />

Fig. 21—Secondary invasion tongue (G 80Kcm 1 , and V 0.25<br />

m s 1 ).<br />

Fig. 18— T as a function of T i (text) for 13 different runs.<br />

Fig. 19—Average invasion acceleration d 2 x T /dt 2 as a function of average<br />

recoil velocity dT i /dt for the same runs as in Fig. 18.<br />

in Figure 18, in which the T data of 13 experiments (including<br />

runs 1 through 4) are plotted vs the average temperature<br />

T i of the front. Another simplified, but instructive, presentation<br />

of the data consists of plotting the average acceleration<br />

d 2 x T /dt 2 , calculated from a linear fit of T (t) over the entire<br />

invasion time, as a function of the average recoil velocity<br />

dT i /dt (Figure 19). The thus obtained representation is<br />

independent of the origins of the time and temperature axes.<br />

Again, it can be seen that all the data points roughly fall on<br />

a single curve. Incidentally, we note that one of the data<br />

points has negative values of dT i /dt and d 2 x T /dt 2 . This<br />

point originates from a run during which the pulling was<br />

stopped when the invasion was in progress. As soon as V<br />

was s<strong>et</strong> to zero, the front stopped recoiling and began to<br />

advance toward the equilibrium temperature T l<br />

. The invasion<br />

rate also began immediately to decrease, confirming that<br />

T principally depends on the local value of the undercooling<br />

provided that the front is reasonably flat.<br />

Concerning the nature of the variab<strong>les</strong> controlling the<br />

invasion rate in addition to the local value of the undercooling,<br />

the following remarks may be useful. Morphologically<br />

speaking, there is an obvious resemblance b<strong>et</strong>ween an invasion<br />

tongue and a surface dendrite (i.e., a dendrite growing<br />

along a wall). However, it is well known that the shape near<br />

the tip and the growth velocity of dendrites, free or attached<br />

to a wall, is a unique function of the undercooling of the<br />

melt, contrary to what we find for the invasion tongues.<br />

Moreover, the growth velocity of dendrites in the system<br />

CBr 4 -C 2 Cl 6 has been measured. [21] These measurements<br />

were performed at values of the undercooling different from<br />

those encountered here, but the laws relating growth velocity<br />

to undercooling are known. It is thus possible to calculate<br />

the growth velocity that would have free dendrites at the<br />

values of u encountered in this study. The calculation, which<br />

it is not useful to reproduce here, leads to values of the<br />

dendritic velocity several orders of magnitude smaller than<br />

the measured invasion rates. Thus, it seems that there are<br />

important quantitative and qualitative differences, the nature<br />

of which is still unclear, b<strong>et</strong>ween invasion tongues and surface<br />

dendrites.<br />

VIII.<br />

SECONDARY INVASION AND PERIODIC<br />

LAMELLAR BRANCHING<br />

A primary invasion tongue stops growing laterally when<br />

it me<strong>et</strong>s an edge of the sample or when it approaches another<br />

tongue growing in the opposite direction. In the last case,<br />

the tongues do not merge, whatever the maximum value of<br />

T reached before the collision is. This is explained by the<br />

excess concentration layer running ahead of the tongue tips,<br />

as mentioned in Section VII–C. The tongues slow down<br />

when they are at a distance of about l T from each other, and<br />

eventually stop, leaving a gap in the newly formed layer of<br />

crystal. The underlying phase rapidly rises up through<br />

this gap, forming a secondary invasion seed (Figure 20).<br />

The first stages of the subsequent invasion are qualitatively<br />

similar to those of a primary invasion. Figure 21 shows<br />

that the morphology of a tongue presents only two minor<br />

differences with that of an tongue: the - interface is<br />

curved in the opposite direction (positive curvature), and<br />

the deformation of the front ahead of the tip is much more<br />

important. The first fact probably signals a b<strong>et</strong>ter w<strong>et</strong>ting<br />

of the glass plates by the phase. The second stems from a<br />

smaller value of the capillary length (Table I). No systematic<br />

study of the dependence of T on the undercooling has<br />

been performed for the tongues because of the instability<br />

presently described.<br />

2046—VOLUME 32A, AUGUST 2001<br />

METALLURGICAL AND MATERIALS TRANSACTIONS A


Fig. 23—Sk<strong>et</strong>ch of the periodic lamellar-branching instability close to<br />

ons<strong>et</strong>.<br />

Fig. 22—Successive stages of a secondary invasion (G 80 K cm 1 ,<br />

and V 0.25 m s 1 ). The field of view has been shifted rightward (4) In Figures 22(c) and 23, there is a clear separation<br />

b<strong>et</strong>ween the successive snapshots. (a) Stable tongue (t 0, and T b<strong>et</strong>ween the tip (defined as the region in which the<br />

4.5 m s 1 ). (b) Ons<strong>et</strong> of an oscillatory instability (t 18.5 s, and T interface of the tongue is noticeably curved) and the<br />

5 m s 1 ), (c) Ons<strong>et</strong> of a periodic lamellar-branching instability (t region behind the tip where branching takes place. At<br />

46 s, and T 10 ms 1 ); (d ) Fully developed lamellar-branching invasion<br />

regime (t 56 s, and still higher values of T , such a separation is no longer<br />

T 20 m s 1 ).<br />

possible (Figures 22(d) and (b)). In other words, the <br />

front seems to be directly invaded by a lamellar twophase<br />

We observed four successive stages in the destabilization<br />

solid. In this fully developed lamellar-branching<br />

process of tongues (Figure 22).<br />

regime, the envelope of the two-phase invasion front is<br />

curved on a large scale, and the lamellae grow roughly<br />

(1) At the beginning of the invasion, T is small, and the<br />

perpendicular to the nonplanar envelope of the front.<br />

tongue is stable (Figure 22(a)).<br />

This is a new example of the behavior known as Cahn’s<br />

(2) Above a certain invasion rate TO (where O stands for<br />

rule, which is often observed in lamellar eutectics when<br />

oscillation), the line left behind by the tongue tip<br />

a weak curvature is imposed to the lamellar front.<br />

becomes wavy (Figure 22(b)). A closer examination<br />

[1,9]<br />

shows that the entire contour of the tongue is slightly<br />

wavy, with a relatively well-defined wavelength O , and<br />

that the value of T oscillates with a time period equal<br />

IX. CONCLUDING REMARKS<br />

to O / T . Clearly, the tongue, which is now a fully 3-D The periodic lamellar branching mechanism, which we<br />

structure, is undergoing a global oscillatory instability. have shown to be responsible for the ons<strong>et</strong> of two-phase<br />

The value of TO is of about 5 m s 1 , for V 0.25 growth, also explains the formation of eutectic grains. It is<br />

ms 1 , and decreases when V decreases (for a comparison,<br />

a purely morphological branching, in which the original<br />

see Figure 5(a)). The wavelength O decreases and crystals and their branches form two interpen<strong>et</strong>rating single<br />

the amplitude of the oscillation increases as T increases crystals. Thus, an invasion tongue subject to that instability<br />

(in the run corresponding to Figure 22, O decreased delivers one eutectic grain per grain of the underlying single<br />

from 16 to 9 m as T increased from 6 to 11 m s 1 ). phase front.<br />

(3) When T exceeds a second characteristic value TB Based on this remark, we can explain the m<strong>et</strong>hod of<br />

(where B stands for branching), the amplitude of the growing single, or at least large, eutectic grains used in this<br />

oscillation becomes large enough for the phase to and former [8] studies. Basically, the idea is to reduce the<br />

periodically overgrow the tongue, leading to an alternate,<br />

number of primary invasion seeds. The seeds generally<br />

roughly periodic emission of and lamellae appear, we recall, by growth of the phase along small<br />

(Figure 22(d)). This is the aforementioned periodic residual liquid channels in the solid. These channels can be<br />

lamellar branching mechanism through which two- eliminated either by maintaining the sample at rest for a<br />

phase growth is established. A sk<strong>et</strong>ch of the branching sufficiently long time or by masking them with an layer<br />

mechanism is given in Figure 23. Note that it is a fully resulting from a primary invasion. The m<strong>et</strong>hod therefore<br />

3-D process. The shape of the tongue is indeed remi-<br />

consists of applying a series of short runs at low V alternating<br />

niscent of that of a dendrite with its sidebranches, as<br />

noted by Lemaignan. [13] However, we have shown previously<br />

that this analogy should not be pushed too far.<br />

with stays at rest until, ideally, only one open active channel<br />

is left. At the end of the next invasion, the sample is then<br />

an single crystal and, at the end of the subsequent <br />

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 32A, AUGUST 2001—2047


invasion, is composed of two eutectic grains. If, moreover,<br />

REFERENCES<br />

the active channel is close to an edge of the sample, one of<br />

1. K.A. Jackson and J.D. Hunt: Trans. AIME, 1966, vol. 236, pp. 1129-42.<br />

the eutectic grains occupies the largest part of the sample. 2. V. Se<strong>et</strong>haraman and R. Trivedi: M<strong>et</strong>all. Trans. A, 1988, vol. 19A, pp.<br />

This m<strong>et</strong>hod was successfully applied to thin samp<strong>les</strong> of 2955-64.<br />

CBr 4 -C 2 Cl 6 within a relatively broad concentration range 3. R. Trivedi, J.T. Mason, J.D. Verhoeven, and W. Kurz: M<strong>et</strong>all. Trans.<br />

around the eutectic point. [8] It is worth noting that the<br />

A, 1991, vol. 22A, pp. 2523-33.<br />

4. B. Caroli, C. Caroli, G. Faivre, and J. Mergy: J. Cryst. Growth, 1992,<br />

obtained eutectic grains are generally not of the locked type; vol. 118, pp. 135-50.<br />

i.e., they have no special orientation relationship b<strong>et</strong>ween 5. G. Faivre and J. Mergy: Phys. Rev. A, 1992, vol. 45, pp. 7320-29;<br />

the phases, which simply means that no long-range preferred 1992, vol. 46, pp. 963-72.<br />

orientation existed in the seed.<br />

6. J. Mergy: Thèse de l’Université Paris VII, Paris, 1992.<br />

Our last remark concerns the question of wh<strong>et</strong>her the 7. J. Mergy, G. Faivre, C. Guthmann, and R. Mell<strong>et</strong>: J. Cryst. Growth,<br />

1993, vol. 134, pp. 353-68.<br />

results of this study are relevant to bulk samp<strong>les</strong>. The main 8. M. Ginibre, S. Akamatsu, and G. Faivre: Phys. Rev. E, 1997, vol.<br />

argument in favor of a positive answer to this question is<br />

56, pp. 780-96; also M. Ginibre: Thèse de l’Université Paris VI,<br />

that periodic lamellar branching, which is the key stage of Paris, 1997.<br />

the transient, is already fully 3-D in our thin samp<strong>les</strong>. We 9. S. Akamatsu and G. Faivre: Phys. Rev. E, 2000, vol. 61, pp. 3757-70.<br />

10. K. Kassner and C. Misbah: Phys. Rev. A, 1991, vol. 44, pp. 6513-22<br />

are therefore inclined to think that the differences b<strong>et</strong>ween and 6533-54.<br />

thin and bulk samp<strong>les</strong> are only quantitative as far as this 11. A. Karma and A. Sarkissian: M<strong>et</strong>all. Mater. Trans. A, 1996, vol. 27A,<br />

mechanism is concerned. More precisely, it is to be expected pp. 635-56.<br />

that branching occurs more easily, i.e., at a lower value of 12. L.M. Hogan, R.W. Kraft, and F.D. Lemkey: Adv. Mater. Res., 1971,<br />

the invasion rate, in bulk than in thin samp<strong>les</strong>. An extension<br />

vol. 5, pp. 83-126; also R.H. Hopkins and R.W. Kraft: Trans. AIME,<br />

1968, vol. 242, pp. 1627-33.<br />

of our m<strong>et</strong>hod of observation to thicker samp<strong>les</strong>, which 13. C. Lemaignan: Acta M<strong>et</strong>all., 1981, vol. 29, pp. 1379-84.<br />

should allow us to confirm, or disprove, this prediction, is 14. S. Akamatsu and G. Faivre: J. Phys. I France, 1996, vol. 6, pp. 503-27.<br />

currently under study.<br />

15. We use the public domain NIH Image program (developed at the U.S.<br />

National Institutes of Health and available from Intern<strong>et</strong> by anonymous<br />

FTP from zippy.nimh.nih.gov or on floppy disk from the National<br />

Technical Information Service, Springfield, VA, part number PB95-<br />

ACKNOWLEDGMENTS<br />

500195GEI).<br />

16. W.G. Pfann and R.S. Wagner: Trans. TMS-AIME, 1962, vol. 224, p.<br />

We gratefully acknowledge many stimulating discussions 1139; also W.A. Tiller: J. Appl. Phys., 1963, vol. 34, pp. 2757-62.<br />

17. W.A. Tiller, K.A. Jackson, J.W. Rutter, and B. Chalmers: Acta M<strong>et</strong>all.,<br />

with M. Plapp. We thank T.S. Lo, M. Plapp, and A. Karma<br />

1953, vol. 1, pp. 428-37; also V.G. Smith, W.A. Tiller, and J.W. Rutter:<br />

for communicating their results concerning the peritectic Can. J. Phys., 1955, vol. 33, pp. 723-45.<br />

banded structure prior to publication. Thanks are due to A. 18. W.J. Bo<strong>et</strong>tinger: M<strong>et</strong>all. Trans., 1974, vol. 5, 2023-31; also J.S. Park<br />

Fleury for her kind technical help and to H. Savary and A.-M. and R. Trivedi: J. Cryst. Growth, 1988, vol. 187, pp. 511-15.<br />

Pougn<strong>et</strong>, Centre National d’Etudes des Télécommunications, 19. B. Caroli, C. Caroli, and L. Ramirez-Piscina: J. Cryst. Growth, 1993,<br />

vol. 132, pp. 377-88.<br />

France-Telecom (Bagneux, France), for providing us with 20. T.S. Lo, A. Karma, and M. Plapp: Phys. Rev. E, 63, 031504 (2001).<br />

zone-refined chemicals. This research was financially sup- 21. S. Akamatsu, G. Faivre, and T. Ihle: Phys. Rev. E, 1995, vol. 51, pp.<br />

ported by the Centre National d’Etudes Spatia<strong>les</strong>, France.<br />

4751-73.<br />

2048—VOLUME 32A, AUGUST 2001<br />

METALLURGICAL AND MATERIALS TRANSACTIONS A


PHYSICAL REVIEW E VOLUME 61, NUMBER 4<br />

APRIL 2000<br />

Traveling waves, two-phase fingers, and eutectic colonies in thin-sample directional solidification<br />

of a ternary eutectic alloy<br />

Silvère Akamatsu and Gabriel Faivre<br />

Groupe de Physique des Solides, CNRS UMR No. 7588, Universités Denis Diderot <strong>et</strong> <strong>Pierre</strong> <strong>et</strong> Marie Curie,<br />

Tour 23, 2 place Jussieu, 75251 Paris Cedex 05, France<br />

Received 9 September 1999<br />

We present an experimental investigation of the morphological transition of lamellar eutectic growth fronts<br />

called ‘‘formation of eutectic colonies’’ by the m<strong>et</strong>hod of thin-sample directional solidification of a transparent<br />

model alloy, CBr 4 -C 2 Cl 6 . This morphological transition is due to the presence in the melt of traces of chemical<br />

components other than those of the base binary alloy impurities. In this study, we use naphthalene as an<br />

impurity. The formation of eutectic colonies has generally been viewed as an impurity-driven Mullins-Sekerka<br />

instability of the envelope of the lamellar front. This traditional view neglects the strong interaction existing<br />

b<strong>et</strong>ween the Mullins-Sekerka process and the dynamics of the lamellar pattern. This investigation brings to<br />

light several original features of the formation of eutectic colonies, in particular, the emission of longwavelength<br />

traveling waves, and the appearance of dendritelike structures called two-phase fingers, which are<br />

connected with this interaction. We study the part played by these phenomena in the transition to eutectic<br />

colonies as a function of the impurity concentration. Recent theor<strong>et</strong>ical results on the linear stability of ternary<br />

lamellar eutectic fronts Plapp and Karma, Phys. Rev. E 60, 6865 1999 shed light on some aspects of the<br />

observed phenomena.<br />

PACS numbers: 81.30.Fb, 05.70.Ln<br />

I. INTRODUCTION<br />

The solidification microstructure of directionally solidified<br />

nonfac<strong>et</strong>ed binary eutectic alloys usually consists of a<br />

regular stacking of lamellae of the two eutectic crystal<br />

phases 1,2. This microstructure is the trace left behind in<br />

the solid by the stationary periodic pattern assumed by the<br />

growth front during solidification 3. The order of magnitude<br />

of the pattern wavelength, or interlamellar spacing , is<br />

fixed by the comp<strong>et</strong>ition b<strong>et</strong>ween solute diffusion in the liquid<br />

and capillarity, and is of 10 m for growth velocities in<br />

the ms 1 range. An example observed in thin-sample directional<br />

solidification of the transparent eutectic alloy<br />

CBr 4 -C 2 Cl 6 is shown in Fig. 1. Necessary conditions for such<br />

a growth pattern to be observed are that the alloy concentration<br />

C be sufficiently close to the center of the eutectic plateau,<br />

and the solidification rate V smaller than a limit value,<br />

which is a function of C and the applied thermal gradient G<br />

1–3. These conditions define the zone of the param<strong>et</strong>er<br />

space in which the growth front is lamellar and planar<br />

planar-growth zone, the word ‘‘planar’’ applying not to the<br />

front itself, but to its envelope, i.e., its shape smoothed over<br />

a distance of a few . The growth pattern of Fig. 1 is not<br />

only planar, but also stationary, symm<strong>et</strong>rical, and bidimensional,<br />

or nearly so. This type of pattern is called ‘‘basic’’ in<br />

contradistinction to other types of planar patterns with a<br />

lower symm<strong>et</strong>ry tilted and oscillatory patterns 4–10.<br />

In this paper, we report the results of an experimental<br />

study of the morphological instability called ‘‘formation of<br />

eutectic colonies,’’ which corresponds to the upper bound of<br />

the planar-growth zone when C is close to the eutectic concentration<br />

C E of the alloy. This instability consists of the<br />

appearance of large cells—‘‘eutectic colonies’’ EC’s—<br />

superimposed on the lamellar pattern Fig. 2, and is due to<br />

the presence of impurities with a small solubility in the solid<br />

phases 11. M<strong>et</strong>allurgical studies performed around 1960<br />

led to the following conclusion, at least implicitly 12–15:<br />

the formation of EC’s is an impurity-driven Mullins-Sekerka<br />

instability occurring on a scale much larger than , and<br />

therefore essentially insensitive to the dynamics of the lamellar<br />

pattern. This conclusion seemed to be supported by the<br />

fact that, like the usual Mullins-Sekerka bifurcation 16,17,<br />

the transition from planar fronts to EC’s generally occurs at<br />

values of V close to the theor<strong>et</strong>ical constitutional supercooling<br />

velocity V cs of the system 18. It was noted in these old<br />

studies that the lamellar pattern adapts itself to the presence<br />

of EC’s in such a way that the trajectories of the lamellae<br />

remain locally perpendicular to the distorted front Cahn’s<br />

rule 19, without the consequences of this fact being drawn.<br />

The similarity b<strong>et</strong>ween EC’s and dilute-alloy cells is unquestionable,<br />

but leaves aside the most interesting aspects of<br />

the formation of EC’s, namely, those involving the interaction<br />

b<strong>et</strong>ween this process and the dynamics of the underlying<br />

lamellar pattern. Cahn’s rule means that the distortions of the<br />

FIG. 1. Planar lamellar eutectic front in a basic i.e., stationary<br />

symm<strong>et</strong>rical state. Undoped CBr 4 -C 2 Cl 6 alloy (0.02; V<br />

1.5 ms 1 ). and are the two terminal solid solutions of the<br />

alloy. In this photograph, as in all the following ones, the growth<br />

direction is upward. Ins<strong>et</strong>: enlarged view of the front.<br />

1063-651X/2000/614/375714/$15.00 PRE 61 3757 © 2000 The American Physical Soci<strong>et</strong>y


3758 SILVÈRE AKAMATSU AND GABRIEL FAIVRE<br />

PRE 61<br />

FIG. 2. Eutectic colonies in a CBr 4 -C 2 Cl 6 -naphthalene alloy<br />

(C naph 510 4 ; V31 ms 1 ). At this small value of 4<br />

m, the lamellae cannot be resolved, and appear as single dark<br />

lines in the image.<br />

front associated with a modulation of the impurity concentration<br />

field entail modifications of the distribution. The<br />

existence of a feedback of the distribution on the front<br />

shape follows from the fact that the average temperature of<br />

the front depends on Jackson-Hunt’s law 3. Because of<br />

this interaction, the formation of EC’s must exhibit a much<br />

richer dynamics than the formation of dilute-alloy cells for a<br />

d<strong>et</strong>ailed experimental study of the latter, see Ref. 20. This<br />

conclusion was recently asserted by Plapp and Karma in conclusion<br />

of a linear stability analysis of the basic lamellar<br />

patterns of ternary eutectics based on Cahn’s hypothesis<br />

21. It is experimentally substantiated by this study for the<br />

first time, to our best knowledge.<br />

The process of formation of EC’s is sensitive to relatively<br />

small changes in the impurity concentration, the deviation of<br />

the base binary alloy from C E , and the initial value of , as<br />

will be seen below. A d<strong>et</strong>ailed quantitative study of this process<br />

would require a more accurate control of these param<strong>et</strong>ers<br />

than is currently achievable. In this paper, we only<br />

claim to present a clear, semiquantitative characterization of<br />

some robust new aspects of the process. Even to this limited<br />

aim, the standard experimental m<strong>et</strong>hods had to be improved<br />

considerably. In thin-sample directional solidification, we recall,<br />

a layer of a transparent model alloy enclosed b<strong>et</strong>ween<br />

two glass plates is placed in an externally imposed unidirectional<br />

temperature gradient, and pulled at an imposed velocity<br />

toward the cold side of the gradient. The front—which<br />

remains immobile, or nearly so, in the laboratory reference<br />

frame—is continuously observed with an optical microscope.<br />

This m<strong>et</strong>hod has often been applied to the fundamental study<br />

of lamellar eutectic growth 3,9,10,22–26. The principal<br />

nonstandard aspects of our experimental m<strong>et</strong>hods are that we<br />

employ very thin 10 m thick samp<strong>les</strong>, and grow large<br />

1 mm ‘‘floating’’ eutectic grains see Sec. II C. This<br />

endows the system with three important characteristics—a<br />

compl<strong>et</strong>e crystallographic homogeneity, a low capillary anisotropy,<br />

and a pronounced two-dimensional 2D<br />

character—without which the dynamics of lamellar eutectics<br />

are of a disconcerting complexity. With very thin, largegrained<br />

samp<strong>les</strong>, it was possible to d<strong>et</strong>ermine the full 2D<br />

stability diagram of CBr 4 -C 2 Cl 6 inside the planar-growth<br />

zone experimentally 10. The experimental results were in<br />

quantitative agreement with the numerical results obtained<br />

for the same system i.e., bidimensional CBr 4 -C 2 Cl 6 without<br />

capillary anisotropy by Karma and Sarkissian 6, confirming<br />

that our samp<strong>les</strong> really have the above characteristics.<br />

Another feature of the samp<strong>les</strong> used in this study is that their<br />

residual-gas content was much lower than in most previous<br />

studies. A ternary chemical component had to be added in<br />

order that impurity-induced phenomena appear within the<br />

explored range in V/G. The ternary component used in this<br />

study is naphthalene. EC’s were only found to form in<br />

CBr 4 -C 2 Cl 6 -naphthalene alloys with a molar fraction of<br />

naphthalene C naph larger than about 510 4 . Other<br />

impurity-induced phenomena, but no standard EC’s, are observed<br />

at lower concentrations of naphthalene. This relatively<br />

sharp transition as a function of the impurity concentration<br />

is one of the unexpected findings of this study.<br />

The plan of the paper is as follows. Previous results concerning<br />

the dynamics of lamellar eutectics, in particular, the<br />

results of the linear stability analysis of ternary eutectics by<br />

Plapp and Karma, are recalled in Sec. II. Section III is devoted<br />

to experimental m<strong>et</strong>hods. In Sec. IV we present preliminary<br />

observations about phase diffusion in the absence of<br />

an impurity, serving as a reference for the observations made<br />

in the presence of impurity. The results concerning the<br />

impurity-driven effects are reported in Sec. V. This section<br />

includes a comparison of the experimental observations with<br />

Plapp and Karma’s theor<strong>et</strong>ical results. A general conclusion<br />

is given in Sec. VI.<br />

II. BACKGROUND<br />

A. Binary eutectic alloys<br />

Two essential dynamical features of binary lamellar eutectics<br />

are the absence of a mechanism of selection of the<br />

spacing, and the ineffectiveness of phase diffusion at low<br />

pulling velocities. These properties explain the important experimental<br />

fact that basic patterns never have a uniform <br />

distribution, but contain smooth, relatively ample spatial<br />

modulations of inherited from the early stages of the<br />

growth 9,25. L<strong>et</strong> us briefly sum up the present state of our<br />

knowledge on these subjects.<br />

The basic patterns of binary lamellar eutectics are stable<br />

over a finite range in at fixed values of the control param<strong>et</strong>ers<br />

6–10. The bounds of this range essentially depend on<br />

two variab<strong>les</strong>, namely, the reduced deviation from eutectic<br />

concentration (CC E )/(C C ), where C and C are<br />

the concentrations of the and solid phases bounding the<br />

eutectic plateau, respectively, and the reduced spacing <br />

/ JH where JH the Jackson-Hunt minimumundercooling<br />

spacing is a scaling length proportional to<br />

V 1/2 .At0, the basic patterns are stable for ranging<br />

from about 1 to 2. Below 1, they are unstable against<br />

lamella termination. Above 2, they undergo a homogenous<br />

bifurcation to a period-preserving oscillatory (1O)<br />

pattern, followed by a secondary bifurcation to a perioddoubling<br />

oscillatory (2O) pattern at a slightly higher value<br />

of . At slightly off-eutectic concentrations (0.04), the<br />

bifurcation sequence is reversed the primary bifurcation is<br />

the 2O one. At still higher values of and/or , a tilt<br />

bifurcation also comes into play.<br />

The average temperature of a binary lamellar eutectic<br />

front, or, equivalently, the position of the envelope along<br />

the axis of the thermal gradient z axis, depend on the value<br />

of . When is uniform, the relationship b<strong>et</strong>ween and is<br />

approximately given by the Jackson-Hunt equation, which<br />

reads


PRE 61 TRAVELING WAVES, TWO-PHASE FINGERS, AND . . .<br />

3759<br />

T JH<br />

2G 1 ,<br />

where the origin is taken at the eutectic temperature, and<br />

T JH the Jackson-Hunt minimum undercooling is a scaling<br />

quantity proportional to V 1/2 3. In the experiments, the spatial<br />

distribution of is generally not uniform, is associated<br />

with a nonplanar envelope of the front, and is nonstationary.<br />

The linear stability analysis of a modulated lamellar eutectic<br />

front of average spacing 0 was performed by Datye and<br />

Langer 4 under the assumption that Cahn’s rule is obeyed.<br />

This rule states that the motion of each trijunction point is<br />

submitted to the constraint t (n) V x 0, where t is the time,<br />

x is the coordinate parallel to the front, and (n) is the x<br />

coordinate of the nth trijunction point. From this, it is easily<br />

derived that<br />

t 0 V xx 0,<br />

where is considered as a continuous function of x. Equations<br />

1 and 2 are the equations governing the spatiotemporal<br />

evolution of the two functions (x,t) and (x,t). Following<br />

the usual linear-stability analysis procedure, one finds<br />

that the linear growth coefficient (k) of a small perturbation<br />

of wave vector k obeys a second-order equation. Only<br />

one of the solutions of this equation is relevant in the conditions<br />

of the experiments to be described in Sec. IV, and<br />

corresponds to a diffusive mode, i.e., a mode such that<br />

(x,t) follows a diffusion equation t D ph xx , where D ph<br />

is the so-called phase diffusion coefficient. D ph reads<br />

1<br />

2<br />

D ph T JH0V<br />

2G 0 0 1 3<br />

the Cahn-Datye-Langer equation, where the subscript 0 refers<br />

to the nonperturbed stationary state of uniform spacing<br />

0 . The most striking aspect of this equation is that the sign<br />

of D ph changes at 0 1, which means that the lamellar<br />

pattern grows unstable below this value of . As mentioned,<br />

lamella termination events, which are probably the ultimate<br />

outcome of such an instability, have indeed been experimentally<br />

observed to occur at, or at least a little below, 1. In<br />

Sec. IV, our interest will not be in this instability, but in the<br />

phase diffusion process that occurs at values of noticeably<br />

larger than 1. It will be kept in mind that the Cahn-Datye-<br />

Langer is a linearized equation, which is valid only when the<br />

amplitude of the gradients are vanishingly small. When<br />

this condition is not fulfilled, Eq. 3 can only be utilised to<br />

estimate the order of magnitude of the damping rate of the <br />

modulations.<br />

B. Multicomponent eutectic alloys<br />

The results of the Plapp-Karma stability analysis of ternary<br />

eutectics can be summed up as follows 21. The simplifying<br />

assumptions made in the calculation are the same as<br />

in Datye and Langer’s previous work, i.e., they mostly<br />

amount to Cahn’s hypothesis. It is found that the Cahn-<br />

Datye-Langer diffusive branch is not significantly modified<br />

by the presence of an impurity as long as V is much smaller<br />

than V cs . However, on approaching V cs , this branch mixes<br />

with the Mullins-Sekerka branch of the envelope. Like a<br />

standard Mullins-Sekerka branch, the mixed branch presents<br />

a broad maximum at a wavelength much longer than , but,<br />

contrary to what occurs in dilute alloys, this branch may be<br />

complex near the maximum—in other words the critical, or<br />

nearly critical, long-wavelength perturbations may be drifting<br />

or oscillating. Plapp and Karma carried out a d<strong>et</strong>ailed<br />

study of the conditions under which (k) is real, or, on the<br />

contrary, complex, in the region of the long-wavelength<br />

maximum. Unfortunately, our experiments cannot crosscheck<br />

their predictions in d<strong>et</strong>ail. We can only state that, in<br />

the conditions of our experiments 1.5, VV cs , is<br />

probably always complex in the ranges in k and V of interest.<br />

Plapp and Karma calculated the critical velocity for the<br />

long-wavelength mode of ternary eutectics i.e., the velocity<br />

V c above which the maximum of the real part of is positive,<br />

and established that V c is given by an approximate<br />

formula quite similar to the Mullins-Sekerka formula for the<br />

critical velocity of dilute alloys. For the sake of clarity, l<strong>et</strong> us<br />

reproduce this well-known formula in the case of a dilute<br />

binary alloy CBr 4 -naphthalene. The partition coefficient of<br />

naphthalene is denoted K naph , the liquidus slope m naph , and<br />

the diffusion coefficient in the liquid D naph . The same symbols<br />

with a subscript X, will be used in the case of the system<br />

CBr 4 -X, where X is the residual gas. The thermal gap of the<br />

alloy is T naph m naph (K 1 naph 1)C naph , the thermal length<br />

1 th T naph /G, and the capillary length d 0 a 0 /T naph<br />

where a 0 is the Gibbs-Thomson capillary coefficient. To a<br />

good approximation, V c is given by the constitutionalsupercooling<br />

velocity V cs D naph 1 th multiplied by a<br />

‘‘capillary-correction factor’’ denoted 1V cl /V cs . The<br />

capillary-correction term is V cl /V cs 1.5(2K naph d 0 /1 th ) 1/3 .In<br />

most experiments, d 0 /1 th 1, and the capillary-correction<br />

term is relatively small. However, it must be kept in mind<br />

that d 0 /1 th increases rapidly as the solute concentration decreases.<br />

The system becomes absolutely stable when V cl /V cs<br />

reaches a value close to 1.<br />

In the case of ternary eutectic alloys, Plapp and Karma<br />

showed that T naph must be replaced by an effective thermal<br />

gap T eff , which depends on the partition coefficients and<br />

the liquidus slopes of the impurity with respect to the two<br />

eutectic crystal phases. These quantities are not known in our<br />

case, so we shall assume that they are equal to the known<br />

quantities pertaining to the dilute CBr 4 -X and<br />

CBr 4 -naphthalene systems see Sec. III A. The formulas for<br />

V cs are then the same as in a dilute alloy. In the case of a<br />

ternary dilute alloy CBr 4 -naphthalene-X, the constitutionalsupercooling<br />

velocity is V cs G/(T naph /D naph T X /D X ).<br />

Taking for C X the highest value found in our samp<strong>les</strong> (4<br />

10 4 ), and G110 K cm 1 , we obtain V cs 3/(14<br />

10 3 C naph ) ms 1 . The thus obtained values of V cs are<br />

compared with the measured instability threshold velocities<br />

of planar fronts below. It will be seen that the calculated and<br />

the measured quantities are within a factor of 2 or 3 of each<br />

other, which is satisfactory given the rough approximations<br />

made in the calculations, and the large experimental uncertainty.<br />

We also note that, according to the above formula, the<br />

influence of the highest residual-gas concentration is comparable<br />

to that of a molar fraction of 2.510 4 of naphthalene.<br />

This explains the variability of the threshold velocities observed<br />

in the low impurity concentration range.


3760 SILVÈRE AKAMATSU AND GABRIEL FAIVRE<br />

PRE 61<br />

Concerning the leading-order correction term to V cs ,<br />

Plapp and Karma showed that d 0 must be replaced by an<br />

effective value d eff which includes a term which is not of a<br />

capillary origin, but represents the stabilizing effect of the<br />

interlamellar diffusion field. In our case, Plapp and Karma’s<br />

formula gives values of V cl /V cs which are of a few 10 2 at<br />

C naph 510 4 , but larger than 10 1 at C naph 2.510 4 .<br />

This large value of V cl /V cs is perhaps the origin of the surprisingly<br />

high instability threshold velocities found in low<br />

impurity concentration samp<strong>les</strong>.<br />

The long-wavelength Mullins-Sekerka-like mode is not<br />

the only mode of instability of the planar lamellar eutectic<br />

fronts. As already stated, in the absence of impurity, the<br />

lamellar pattern is unstable to various oscillatory or tilt<br />

modes for certain values of V, , and . Of these modes,<br />

only the 2O one is compatible with Cahn’s rule, at least<br />

approximately, and was found by Datye and Langer. Plapp<br />

and Karma showed that the presence of an impurity greatly<br />

enhances the 2O mode, suggesting that, in some cases, the<br />

destabilization of the planar front may occur through the<br />

2O mode rather than the long-wavelength mode. Observations<br />

supporting this view are reported below.<br />

C. Anisotropy effects. Eutectic grains<br />

Eutectic fronts are composed of crystallographic domains<br />

called eutectic grains 27. The crystal orientation in the<br />

lamellae of the two solid phases is uniform within a eutectic<br />

grain, and varies from a eutectic grain to another. The capillary<br />

anisotropy of the system i.e., the orientation dependence<br />

of the surface tensions of the -liquid, -liquid, and<br />

- interfaces is different in different grains. In a previous<br />

study, a classification of the eutectic grains of our system<br />

according to their capillary anisotropy was established 25.<br />

In the so-called locked grains, the direction of growth of the<br />

lamellae is locked onto a preferential orientation, corresponding,<br />

most probably, to a sharp minimum of the surface<br />

tension of the - interface. The grains in which no such<br />

locking is observed were called ‘‘floating’’ grains. It was<br />

shown that, in these grains, the growth dynamics is similar to<br />

that of a system without capillary anisotropy as calculated<br />

numerically 6, except for the small capillary-anisotropy effects<br />

to be described shortly. A m<strong>et</strong>hod of obtaining large<br />

floating grains was applied in this study see Sec. III B.<br />

If no capillary anisotropy at all was present, the system<br />

would be symm<strong>et</strong>rical with respect to the z axis, and the<br />

lamellae would run parallel to this axis when the system is in<br />

a basic pattern. However, since some capillary anisotropy is<br />

actually present, the mirror symm<strong>et</strong>ry of the system is broken<br />

in most eutectic grains, and, even in the basic patterns,<br />

the lamellae run at a small angle the anisotropy-driven tilt<br />

angle from the z axis. This angle is a function of the orientation<br />

of the grain, and, in a given grain, increases as <br />

increases. In the floating grains of CBr 4 -C 2 Cl 6 , the value m<br />

of at the reference value of the spacing JH is generally<br />

<strong>les</strong>s than 2°, but increases relatively rapidly as increases<br />

because of the spontaneous tilt bifurcation undergone by the<br />

system at large values of . Since is different in different<br />

grains, the two patterns on either side of a eutectic grain<br />

boundary have different values of . Consequently, lamellae<br />

are repeatedly terminated, or created at eutectic grain boundaries,<br />

which are therefore permanently surrounded by steep<br />

local gradients. These gradients may act as sources of traveling<br />

dynamical defects. In particular, eutectic grain boundary<br />

gradients are permanent sources of traveling waves in<br />

the presence of impurities at sufficiently high V see Sec.<br />

VC1.<br />

III. EXPERIMENTAL METHODS<br />

A. Products and samp<strong>les</strong><br />

The samp<strong>les</strong> are made of two parallel glass plates separated<br />

by plastic spacers, delimiting an empty space about 8<br />

mm wide, 70 mm long, and 12 m thick. The alloys are<br />

prepared by mixing zone-refined CBr 4 ,C 2 Cl 6 , and naphthalene.<br />

The mixing process is carried out under a low pressure<br />

of argon. A fragment of the solidified mixture deposited at<br />

one end of a heated empty sample remelts, and fills the<br />

sample by capillarity. The filled samp<strong>les</strong> are placed in an<br />

externally imposed thermal gradient, and pulled with a dc<br />

motor via a microm<strong>et</strong>ric screw. During the pulling, the<br />

growth front is continuously observed with an optical microscope<br />

over the whole width of the selected eutectic grain.<br />

The images are recorded with the help of a CCD camera and<br />

a videotape recorder, and then analyzed with a computer<br />

28. In the present study, the value of G is 110 K cm 1<br />

un<strong>les</strong>s otherwise stated. The scanned range of V is 0.9–31<br />

ms 1 .<br />

In this study, the concentration of the base binary alloy<br />

CBr 4 -C 2 Cl 6 is C E to within experimental uncertainty<br />

0.003, un<strong>les</strong>s otherwise mentioned. The above scaling<br />

quantities are JH V 1/2 14 m 3/2 s 1/2 and T JH V 1/2<br />

0.033 K m 1/2 s 1/2 to within about 20% 26. Alloys without<br />

naphthalene undoped alloys, and alloys doped with molar<br />

fractions of naphthalene of 2.510 4 , 510 4 , and<br />

10 3 are used. The samp<strong>les</strong> also contain residual gases,<br />

which we admit to consist of a single component X, introduced<br />

during the mixing and filling processes. The partition<br />

coefficient (K X 0.02), the liquidus slope m X 50 K per<br />

molar fraction, and the liquid-phase diffusion coefficient<br />

(D X 300 m 2 s 1 ) of X in CBr 4 have been d<strong>et</strong>ermined<br />

29. C X was found to vary from a sample to another depending<br />

on the care with which the outgasing, mixing, and filling<br />

processes were carried out, but was at most of 410 4 .In<br />

the present study, C X was generally markedly lower than this<br />

values, as shown by the weakness of the impurity-induced<br />

effects at high V in most undoped samp<strong>les</strong>. A rough experimental<br />

d<strong>et</strong>ermination of the partition coefficient (K naph<br />

0.07), the liquidus slope (m naph 300 K per molar fraction,<br />

and the liquid-phase diffusion coefficient (D naph<br />

300 m 2 s 1 ) of naphthalene in CBr 4 has been performed<br />

by the same m<strong>et</strong>hods as in Ref. 29.<br />

B. Grain boundary effects<br />

Considering the great complexity of the dynamics of<br />

lamellar eutectics, it is obviously desirable that the samp<strong>les</strong><br />

be as homogeneous as possible. The two main sources of<br />

nonhomogeneity are the presence of several eutectic grains,<br />

and the nonuniform distribution of the initial state see Sec.<br />

IV. In a previous study, a m<strong>et</strong>hod of obtaining samp<strong>les</strong> containing<br />

a single floating eutectic grain was successfully ap-


PRE 61 TRAVELING WAVES, TWO-PHASE FINGERS, AND . . .<br />

3761<br />

plied to hypereutectic samp<strong>les</strong> 10. In brief, this m<strong>et</strong>hod<br />

consists of preceding the experimental runs by a short preliminary<br />

pulling at a very low value of V during which the<br />

lateral invasion of the front through which the lamellar pattern<br />

is created is controlled. The same m<strong>et</strong>hod was applied in<br />

this study, but turned out to be <strong>les</strong>s efficient near the eutectic<br />

concentration than far from it. We could not grow singlegrain<br />

samp<strong>les</strong>, but were neverthe<strong>les</strong>s able to obtain grains of<br />

millim<strong>et</strong>ric sizes.<br />

C. V jump m<strong>et</strong>hod<br />

Basically, the experimental procedure that we follow is<br />

that of upward V jumps. First, the growth front is put in a<br />

stationary planar basic state by pulling a long time at a low<br />

velocity V 1 , and then the pulling velocity is suddenly<br />

switched to a higher velocity V 2 at which some morphological<br />

instability is expected to occur. V jumps of moderate<br />

amplitude do not change the average spacing , generally,<br />

and thus change the average reduced spacing by a factor<br />

of V 2 /V 1 . Since the instability threshold of the planar front<br />

can only be reached through upward V jumps, is always<br />

relatively high 1.5 in this study.<br />

V jumps are followed by long solute redistribution transients<br />

30. During the initial solute redistribution transient,<br />

the naphthalene concentration ahead of the front increases<br />

from C naph to the steady-state value C naph /K naph . The characteristic<br />

time of the transient is naph D naph /(K naph V 2 ), and<br />

the corresponding solidified length l naph D naph /(K naph V).<br />

After a V jump from V 1 to V 2 , the concentration ahead of<br />

the front rises above the steady-state value, and then r<strong>et</strong>urns<br />

to this value with a characteristic time D naph /(K naph V 1 V 2 ).<br />

The amplitude of the post-jump concentration overshoot depends<br />

on V 2 /V 1 , but may be quite large even for moderate<br />

values of V 2 /V 1 when the partition coefficient of the solute<br />

is small, which is our case.<br />

If no V jump is applied, i.e., if the pulling is entirely<br />

performed at constant V the short preliminary stage at a<br />

lower V can be neglected as far as solute redistribution is<br />

concerned, the destabilization of the planar front can only<br />

be observed in the course of initial transients at values of V<br />

higher than the instability threshold. The order of magnitude<br />

of the time after which the destabilization is observed is<br />

naph . It will be seen below that the order of magnitude of<br />

the instability thresholds is of 1 ms 1 in our system. The<br />

corresponding transients are very long typically naph 3h<br />

and l naph 1cm. Therefore, only very few different values<br />

of V could be explored with a given sample, limiting the<br />

accuracy with which instability thresholds could be measured.<br />

The values given below are the lowest value of V at<br />

which a destabilization was observed, and are probably substantially<br />

higher than the corresponding theor<strong>et</strong>ical transition<br />

velocities.<br />

On the other hand, large upward V jumps can be used to<br />

provoke a transitory increase of the impurity concentration at<br />

the front, and hence a transient destabilization of the planar<br />

front. This technique was used in order to establish the m<strong>et</strong>astable<br />

character of planar fronts at low impurity concentration.<br />

FIG. 3. Lamellar pattern with a strongly nonuniform distribution<br />

(C naph 0; V4.1 ms 1 ; G0.005 K m 1 ).<br />

IV. PRELIMINARY OBSERVATIONS<br />

Before considering the effect of impurities on the dynamical<br />

properties of lamellar eutectic fronts, we go back over the<br />

dynamics in the absence of impurities. The initial basic patterns<br />

i.e., the patterns obtained at the end of the preliminary<br />

stage of the pulling contain a spatial modulation of with a<br />

characteristic wavelength of a few tenths of , and an amplitude<br />

practically equal to the width of the basic-state stability<br />

range. In previous studies performed at V1 ms 1 and<br />

G10 2 K m 1 , the distortions of the front associated<br />

with this nonuniform grown-in distribution were not d<strong>et</strong>ected,<br />

and phase diffusion was found to be ineffective<br />

9,25. At the relatively high values of V which we are now<br />

able to reach without impurity-driven effects thanks to the<br />

higher purity of our samp<strong>les</strong>, these effects can be measured.<br />

A d<strong>et</strong>ailed study is currently in progress. In view of what<br />

follows, it is useful to present the following preliminary observation<br />

here.<br />

Figure 3 shows an undoped sample pulled at V<br />

4.1 ms 1 under a thermal gradient of about<br />

510 3 K m 1 . The snapshot was taken a few seconds<br />

after the ons<strong>et</strong> of the pulling, when residual-gas effects if<br />

any are small. The distortion of the front, although hardly<br />

visible in Fig. 3, was easily measurable on the computer<br />

screen. Figure 4 shows (x) and (x) measured along the<br />

front, and the function calc (x) calculated by inserting the<br />

measured values of (x) into the Jackson-Hunt equation. It<br />

FIG. 4. Measured spacing distribution (x) thick line and<br />

front shape (x) thin line for Fig. 3. Dotted line: calculated value<br />

of (x). For the latter quantity, the origin is taken at the average<br />

position of the front.


3762 SILVÈRE AKAMATSU AND GABRIEL FAIVRE<br />

PRE 61<br />

FIG. 6. Snapshots taken during the transient leading to the eutectic<br />

colonies of Fig. 2 at times a t11.9 s, b 13.0 s; c 14.8 s,<br />

and d 18.9 s. A velocity jump from 0.9 to 31 ms 1 was applied<br />

at time t0 bottom of a.<br />

FIG. 5. a (x) measured from Fig. 3 at time intervals of 6.1 s.<br />

The last plot is the same as in Fig. 4. For clarity, the successive<br />

plots have been shifted along the ordinate axis by an arbitrary quantity<br />

1 m. b Standard deviation of (x) as a function of time.<br />

can be seen that the ratio (x)/ calc (x) is constant and of the<br />

order of unity. Figure 5a shows a time series of plots<br />

extracted from Fig. 3. The distribution is progressively<br />

damped out without any lateral drift. Figure 5b shows the<br />

time evolution of the standard deviation of the distribution.<br />

An exponential law fitted to the terminal part of<br />

(t) gives a damping coefficient of about 710 3 s 1 .<br />

Noting that, near the end of the process, the Fourier transform<br />

of the distribution is peaked around 25 , we have<br />

calculated the phase-diffusion damping coefficient D ph k 2 for<br />

the mode of wavelength 2/k25 see Eq. 3. The<br />

value found is of about 210 2 s 1 , in relatively good<br />

agreement with the measured value. These results strongly<br />

suggest that the terminal stage of the damping process is<br />

essentially due to the Cahn-Datye-Langer phase-diffusion<br />

mechanism. Similar observations were made in samp<strong>les</strong> containing<br />

naphthalene, showing that even relatively large impurity<br />

concentrations do not qualitatively modify the dynamics<br />

of lamellar eutectic fronts at sufficiently low V. A more<br />

extensive comparison with the predictions of the Cahn-<br />

Datye-Langer equation based on a larger s<strong>et</strong> of experimental<br />

data is currently in progress.<br />

V. IMPURITY-DRIVEN DYNAMICAL PHENOMENA<br />

A. Overview<br />

We begin with a brief enumeration of our findings, which<br />

will be used to introduce some nonstandard terms. As already<br />

stated, the phenomena observed in the samp<strong>les</strong> with<br />

C naph 0 or 2.510 4 are different from those observed in<br />

the samp<strong>les</strong> with C naph 510 4 or 10 3 . For the sake of<br />

simplicity, we call these two concentration ranges the ‘‘low’’<br />

and ‘‘high’’ impurity concentration ranges, respectively.<br />

High impurity concentration<br />

1 At high impurity concentration, EC’s always appear<br />

above a velocity V EC which is comparable to the calculated<br />

value of V cs , in conformity with the observations reported in<br />

the m<strong>et</strong>allurgical literature.<br />

2 Two unexpected phenomena occur during the transients<br />

prior to the formation of EC’s. The first of them is the<br />

appearance of traveling waves TW’s, i.e., small-amplitude<br />

long-wavelength deformations drifting laterally at a finite velocity<br />

Fig. 6. Surprisingly, these TW’s do not grow in amplitude<br />

in the course of time, and play no essential part in the<br />

transition to EC’s.<br />

3 The second new phenomenon is the appearance of<br />

protruding local structures baptized ‘‘two-phase fingers’’<br />

Figs. 6b and 6c. This turns out to be an essential intermediate<br />

stage of the transition to EC’s. It occurs after the<br />

excitation of a short-wavelength mode of instability of the


PRE 61 TRAVELING WAVES, TWO-PHASE FINGERS, AND . . .<br />

3763<br />

FIG. 9. Shallow eutectic colonies (C naph 2.510 4 ; V<br />

31 ms 1 ).<br />

FIG. 7. -- two-phase fingers and eutectic colonies in an<br />

undoped hypoeutectic (0.14) sample. V4.1 ms 1 .<br />

lamellar pattern, the nature of which depends on the initial <br />

distribution. This mode is som<strong>et</strong>imes, but not always, the<br />

2O mode.<br />

4 Large-amplitude, quasistable two-phase fingers are<br />

observed to coexist with EC’s at slightly off-eutectic concentrations<br />

Fig. 7, suggesting that periodic arrays of two-phase<br />

fingers may be stationary states of the system.<br />

5 Once formed, the EC’s are subjected to tip-splitting<br />

and squeezing-off mechanisms, leading to a rough selection<br />

of the average EC size. No steady state is reached on the<br />

scale of the individual EC’s after long solidification times,<br />

suggesting that EC patterns are intrinsically unsteady.<br />

Low impurity concentration<br />

6 At low impurity concentration, no spontaneous transition<br />

to EC’s is observed within the explored range in V/G<br />

Fig. 8. The value of V EC if such a velocity still exists is<br />

thus much larger than V cs .<br />

7 Above a velocity close to V cs , planar fronts are m<strong>et</strong>astable<br />

against the formation of small-amplitude ‘‘shallow’’<br />

EC’s Fig. 9. The dynamics of the shallow-EC patterns is<br />

characterized by a very large dispersion of the EC size, and<br />

rectilinear trajectories of the inter-EC grooves, strongly suggesting<br />

that the latter are localized entities separated from<br />

each other by large portions of m<strong>et</strong>astable planar or wavy<br />

front.<br />

8 TW’s are emitted near the eutectic grain boundaries<br />

above a velocity V TW , which is also close to V cs Fig. 10.<br />

The TW’s are progressively damped out as they propagate<br />

along the front, but their damping time is long compared to<br />

the known characteristic times of the system, especially, the<br />

phase diffusion time see Sec. V C 3. At high V, two different<br />

situations are observed at a sufficiently large distance<br />

from the eutectic grain boundaries, depending on the efficiency<br />

of the latter as sources of TW’s, namely, a perfectly<br />

planar front Fig. 8, or a dense population of rightward and<br />

leftward TW’s Fig. 11. By ‘‘wavy front’’ we define any<br />

dynamical state in which an essentially planar front is continually<br />

swept by finite-amplitude TW’s.<br />

B. High impurity concentration<br />

1. Transition to eutectic colonies<br />

A series of long-duration runs at constant V were performed<br />

in samp<strong>les</strong> with C naph 510 4 and 10 3 . In some<br />

of these runs, the front remained planar up to the end of the<br />

run, while in others EC’s appeared in the course of the initial<br />

transient. We obtained 2.5V EC 3.2 ms 1 at C naph <br />

510 4 and 1V EC 1.8 ms 1 at C naph 10 3 , where<br />

the lower bounds are the highest value of V at which no EC’s<br />

were observed and the higher bounds are the lowest value of<br />

V at which EC’s appeared. The calculated values of V cs for<br />

these concentrations are 1 and 0.6 ms 1 , respectively. The<br />

order of magnitude of V cs is the same as that of V EC , and<br />

both quantities vary with C naph in the same way, as they<br />

should. The fact that the values of V cs fall outside the experimental<br />

margin for V EC is not significant since rough approximations<br />

were made in the calculation of V cs .<br />

EC patterns formed at VV EC were observed to disappear<br />

when V was decreased below V EC . No m<strong>et</strong>astability<br />

range was positively observed. The range of m<strong>et</strong>astability<br />

FIG. 8. Planar lamellar eutectic front at high velocity (C naph<br />

2.510 4 ; V31 ms 1 ; 1.9).<br />

FIG. 10. Traveling waves. A grain boundary is located on the<br />

leftmost side of the figure (C naph 2.510 4 ; V13.5 ms 1 ).


3764 SILVÈRE AKAMATSU AND GABRIEL FAIVRE<br />

PRE 61<br />

FIG. 11. Dense population of traveling waves at high velocity<br />

(V31 ms 1 ). Undoped sample. The loss of contrast in the gray<br />

regions is due to three-dimensional instabilities.<br />

b<strong>et</strong>ween planar fronts and EC patterns is thus narrow at<br />

most a few ms 1 wide at high impurity concentration, if it<br />

exists at all.<br />

2. Process of formation of eutectic colonies<br />

FIG. 12. Transient subsequent to a V jump from 3.1 to 5 ms 1<br />

in an undoped sample (0.09). a t41 s, b 49 s, and c 59 s.<br />

As already stated, V EC was probably noticeably higher<br />

than the instability threshold velocity of the planar front in<br />

all the observed cases. In order to gain insight into the critical<br />

instabilities of the planar front, experimental runs consisting<br />

of subcritical V jumps jumps from V 1 to V 2 with V 1<br />

V 2 V EC , provoking a slight transitory overstepping of<br />

the threshold were carried out. The result of such a V jump<br />

performed in an undoped sample containing a relatively<br />

large concentration of residual gas is shown in Fig. 12. Similar<br />

observations were obtained in samp<strong>les</strong> doped with naphthalene.<br />

Figure 12 clearly shows that the destabilization of<br />

the planar front rapidly gives rise to two-phase fingers consisting<br />

of a protruding lamella sandwiched b<strong>et</strong>ween to<br />

strongly deformed lamellae ‘‘-- fingers’’. Concerning<br />

the linear stage of the destabilization process, its short<br />

duration time, and the complications due to the nonuniform<br />

initial distribution prevented us from identifying the critical<br />

mode with certainty. It can only be stated that it is probably<br />

a nonoscillatory short-wavelength mode. Such modes as<br />

well as the two-phase fingers are incompatible with Cahn’s<br />

rule.<br />

The typical course of the initial transients at C naph <br />

510 4 and VV EC is illustrated in Fig. 6. Three distinct<br />

stages can be noted, corresponding to the successive appearance<br />

of TW’s, two-phase fingers, and EC’s, respectively.<br />

Each stage corresponds to a relatively large, sudden increase<br />

of the average temperature of the front, suggesting that the<br />

wavy front, the two-phase fingers, and the EC pattern are all<br />

stationary or quasistationary states of the system. The TW’s<br />

appear a short time after the ons<strong>et</strong> of the run, when the impurity<br />

concentration ahead of the front is still low, and do not<br />

substantially grow in amplitude, nor change their drift velocity<br />

afterwards, in agreement with the observations performed<br />

in low impurity concentration samp<strong>les</strong>. The creation of twophase<br />

fingers occurs when the transient is nearly compl<strong>et</strong>ed,<br />

and follows a scenario which is not unique, but depends on<br />

the local distribution. The formation of EC’s begins with<br />

the occurrence of lamella branching on the sides of the twophase<br />

fingers. The transient is compl<strong>et</strong>ed when the EC tips<br />

have reached their steady-state undercooling and the EC<br />

sides have come into contact with their neighbors.<br />

Two different scenarios of creation of two-phase fingers<br />

appear in Fig. 6. In the largest part of the front, the twophase<br />

fingers arise from the same destabilization process as<br />

in Fig. 12. In a small region of the front, in which the local<br />

value of was particularly large, the 2O mode of the<br />

lamellar pattern is excited, causing two-phase fingers and<br />

EC’s to appear at a much shorter time than in the remainder<br />

of the front. The 2O mode is thus an efficient, but not a<br />

necessary, precursor of the formation of EC’s. More importantly,<br />

the 2O mode leads to EC’s only via the creation of<br />

two-phase fingers.<br />

3. Two-phase fingers at off-eutectic concentrations<br />

Figure 7 shows quasistable -- fingers coexisting with<br />

EC’s in a slightly hypoeutectic sample. The tip undercooling<br />

of the fingers is smaller than that of the EC’s, contrary to<br />

what occurs at 0. Such well-developed -- fingers bear<br />

striking morphological and dynamical similarities with the<br />

symm<strong>et</strong>ry-broken double fingers ‘‘doublons’’ of the lowanisotropy<br />

dilute-alloy solidification fronts 31,32. Like the<br />

latter, they are constituted by two symm<strong>et</strong>rically disposed<br />

broad fingers of one phase separated by a thin lamella of<br />

another phase, and can change their growth direction without<br />

being destroyed except by collision with another object. At<br />

still larger deviations from C E ,-- fingers have been observed<br />

to comp<strong>et</strong>e with single-phase dendrites.<br />

A type of two-phase finger qualitatively different from the<br />

above -- fingers is predominantly observed in slightly<br />

hypereutectic samp<strong>les</strong> Fig. 13. The sides of these fingers<br />

are occupied by the phase, and the tips by several lamellae<br />

separated from each other by lamellae. The dynamical<br />

properties of these ‘‘-- multipl<strong>et</strong>s’’ are similar to those<br />

of the -- fingers. Like the latter, they are reminiscent of<br />

structures observed in thin-sample directional solidification<br />

of dilute alloys 31–35.<br />

4. Long-time dynamics of the eutectic-colony patterns<br />

The dynamics of the permanent EC patterns as a function<br />

of V and has not y<strong>et</strong> been studied in d<strong>et</strong>ail. However, all


PRE 61 TRAVELING WAVES, TWO-PHASE FINGERS, AND . . .<br />

3765<br />

FIG. 13. Eutectic colonies and -- multipl<strong>et</strong>s. (C naph <br />

510 4 ; V9.9 ms 1 ).<br />

the observations converge to show that EC patterns are unsteady.<br />

This is obviously the case at off-eutectic concentrations,<br />

as illustrated in Figs. 7 and 13, and also near the eutectic<br />

concentration. Figure 14 displays the spatiotemporal<br />

diagram of a near-eutectic EC pattern over a period of time<br />

much longer than naph . For clarity, only the trajectories of<br />

the inter-EC grooves are represented. It can be seen that<br />

tip-splitting and squeezing-off events are taking place at a<br />

constant average frequency. However, it should be noted that<br />

most of these events concern short-lived grooves. The longlived<br />

grooves are in small number 9, and undergo only<br />

rare annihilation or creation events. The observations show<br />

that the short-lived grooves are shallow, while the long-lived<br />

ones are deep. This distinction b<strong>et</strong>ween two types of grooves<br />

is further illustrated in Fig. 15, which shows the average EC<br />

spacing measured at different values of V in a given sample.<br />

Two largely different average values of the EC spacing are<br />

obtained depending on wh<strong>et</strong>her the shallow grooves are<br />

taken into account, or not.<br />

Unsteady ‘‘seaweed’’ and shallow-cell patterns are observed<br />

in thin-sample directional solidification of dilute alloys<br />

when the anisotropy of the interfacial properties is low<br />

31–36. Thus the unsteadiness of the EC patterns is not an<br />

exceptional phenomenon. At off-eutectic concentrations, it<br />

certainly has to do with the peculiar dynamical properties of<br />

the two-phase fingers and multipl<strong>et</strong>s, but this is <strong>les</strong>s clear at<br />

near-eutectic concentrations.<br />

5. Comparison with theory<br />

Broadly speaking, the above observations are compatible<br />

with Plapp and Karma’s theor<strong>et</strong>ical results. We observe that<br />

the planar front is unstable above a velocity of the same<br />

order of magnitude as V cs , and that both long-wavelength<br />

drifting modes and short-wavelength modes are involved in<br />

the destabilization process. The latter actually play a largely<br />

predominant part, a possibility which was not rejected by<br />

Plapp and Karma. The only apparent discrepancy concerns<br />

the nature of the critical mode. We find that this mode is not<br />

the 2O mode, but a nonoscillatory mode which had not<br />

FIG. 14. Spatiotemporal diagram of an EC pattern same run as<br />

in Fig. 2. Only the trajectories of the inter-EC liquid grooves are<br />

shown.<br />

been noted previously, to our best knowledge. That such a<br />

discrepancy arises is not a compl<strong>et</strong>e surprise, however, since<br />

all the modes incompatible with Cahn’s rule are eliminated<br />

from Plapp and Karma’s calculation.<br />

On the other hand, the transition from planar front to EC’s<br />

is clearly not a continuous one. It is mediated by at least two<br />

FIG. 15. Average EC spacing as a function of V measured taking<br />

account of deep liquid grooves only open symbols; error bars:<br />

minimum and maximum measured values, or both deep and shallow<br />

grooves filled symbols. C naph 510 4 .


3766 SILVÈRE AKAMATSU AND GABRIEL FAIVRE<br />

PRE 61<br />

phenomena, namely, the formation of two-phase fingers, and<br />

the occurrence of lamella branching. The intermediate transition<br />

from planar front to two-phase fingers is perhaps continuous,<br />

i.e., it may correspond to a bifurcation of the system.<br />

In this case, the ons<strong>et</strong> of the above-mentioned nonoscillatory<br />

mode would correspond to the upper bifurcation threshold.<br />

The existence of an alternative way of reaching the twophase<br />

finger branch via the 2O mode pleads for a subcritical<br />

character of this bifurcation. This is not proven, however.<br />

It should also be noted that lamella branching results from<br />

3D processes which are not y<strong>et</strong> fully understood, but are<br />

certain to depend on the sample thickness, and the w<strong>et</strong>ting<br />

conditions at the glass plates 9,10. It is therefore conceivable<br />

that lamella branching would occur earlier in samp<strong>les</strong><br />

thicker than ours, or in another alloy than CBr 4 -C 2 Cl 6 , and<br />

would hide the other stages of the process.<br />

C. Low impurity concentration<br />

1. Transition to wavy fronts<br />

FIG. 16. Spacing distribution (x) thick line and front profile<br />

(x) thin line of the growth front of Fig. 10.<br />

At C naph 2.510 4 as well as in samp<strong>les</strong> without naphthalene<br />

but with residual impurities, TW’s are observed to<br />

appear near eutectic grain boundaries above a certain velocity<br />

V TW . The emitted TW’s are not isolated, but are part of<br />

more or <strong>les</strong>s continuous wave trains Fig. 10. Given that<br />

long-wavelength perturbations are continually generated in<br />

the vicinity of eutectic grain boundaries see Sec. II B 3, the<br />

obvious meaning of this observation is that long-wavelength<br />

perturbations spontaneously drift when VV TW . Some eutectic<br />

grain boundaries are much more efficient than others<br />

as sources of TW’s. This is connected with the frequency of<br />

the lamella termination, or branching events occurring at the<br />

boundary, which depends on the misorientation of the adjacent<br />

grains. Incidentally, the outer boundaries of the sample,<br />

i.e., the lines of contact of the alloy with the plastic spacers,<br />

also are active sources of TW’s.<br />

Within experimental uncertainty, the transition is abrupt<br />

no TW’s are observed below V TW . The value of the drift<br />

angle tan 1 W/V, where W is the drift velocity, is finite at<br />

V TW and remains essentially constant as V increases. However,<br />

it must be kept in mind that V TW is certainly larger than<br />

the actual transition velocity, so that the dependence of on<br />

V in the vicinity of the transition is unknown. The observed<br />

values of range from about 25° to 30°.<br />

Substantially different values of the transition velocity<br />

were found in different samp<strong>les</strong> with the same nominal concentration<br />

of naphthalene. This large dispersion is not entirely<br />

attributable to the small number of values of V explored<br />

in each sample. It is mostly due to the fact that V TW<br />

increases very rapidly as the impurity concentration goes to<br />

zero, and is therefore very sensitive to the unwanted variations<br />

of C X from a sample to another. In some well outgased<br />

undoped samp<strong>les</strong>, no TW’s were observed at any V, except<br />

during post-jump transients, showing that V TW was higher<br />

than 31 ms 1 . On the other hand, the lowest value of V TW<br />

found in the samp<strong>les</strong> with C naph 2.510 4 was of 5<br />

ms 1 . This value is comparable to the calculated value of<br />

V cs for this concentration of naphthalene, and C X <br />

410 4 .<br />

The planar fronts remain robust against the formation of<br />

structures of larger amplitude than the TW’s up to the maximum<br />

explored velocity, i.e., at least 6V cs . This was tested by<br />

applying upward V jumps to wavy fronts. Low amplitude V<br />

jumps had no effect. Stronger V jumps resulted in the appearance<br />

of shallow EC patterns, i.e., patterns in which the<br />

depth of the inter-EC grooves was only of a few Fig. 9.<br />

These shallow EC patterns spontaneously disappeared below<br />

a value of V comparable with V TW . The whole range in V<br />

above V TW can thus be considered as a m<strong>et</strong>astability range<br />

b<strong>et</strong>ween planar, or wavy fronts and shallow EC’s.<br />

2. Wavy fronts and Cahn’s rule<br />

In the presence of TW’s, the tilting of the lamellae and the<br />

distortion of the front are sufficiently large for the validity of<br />

Cahn’s rule to be directly studied. We define the local<br />

lamella tilt angle as the angle b<strong>et</strong>ween the trajectory of the<br />

lamellae and the normal to the envelope of the front. Cahn’s<br />

rule states that 0. In fact, the lamella tilt angle is generally<br />

different from zero even when the spacing is uniform<br />

and the front perfectly planar because of capillary anisotropy<br />

see Sec. II B 3. Furthermore, this ‘‘homogeneous’’ tilt<br />

angle generally is an increasing function of . It is thus natural<br />

to recast Cahn’s rule as<br />

x hom x,<br />

where hom () is the lamella tilt angle of a homogeneous<br />

pattern of spacing . This equation means that the lamellar<br />

pattern is locally the same as it would be if the spacing was<br />

uniform and the front was normal to the gradient. The two<br />

questions to be considered are: Is Eq. 4 verified in the<br />

experiment? Does the nonzero term on the right-hand side of<br />

Eq. 4 make a difference as concerns the dynamics of the<br />

TW’s?<br />

The (x) and (x) plots measured along the front of Fig.<br />

10 are displayed in Fig. 16. The azimuthal ang<strong>les</strong> ang<strong>les</strong><br />

with respect to z of the trajectories of the lamellae and<br />

the normal to the envelope of the front are plotted in Fig.<br />

17. The plot of as a function of obtained by eliminating<br />

the variable x b<strong>et</strong>ween (x) and (x)(x)(x) is<br />

shown in Fig. 18. A second-order polynomial fitted to the<br />

data of Fig. 18 is similar to the previously found hom ()<br />

curves 9,37. Similar results were obtained with other<br />

samp<strong>les</strong>, allowing us to conclude that wavy fronts obey the<br />

generalized form of Cahn’s rule represented by Eq. 4<br />

within experimental uncertainty. Other observations not re-<br />

4


PRE 61 TRAVELING WAVES, TWO-PHASE FINGERS, AND . . .<br />

3767<br />

FIG. 17. Azimuthal angle of the lamellae and of the normal<br />

to the envelope of the growth front of Fig. 10.<br />

produced have shown that noticeable deviations from this<br />

equation are observed in the presence of large gradients.<br />

Within a given grain, no marked difference is observed<br />

b<strong>et</strong>ween the properties drift velocity, profile of rightward<br />

and leftward TW’s. The specific effect of capillary anisotropy<br />

is to break the right-left symm<strong>et</strong>ry, as shows up in the<br />

tilt of the lamellae. Since no such symm<strong>et</strong>ry breaking is observed<br />

in the properties of the TW’s, we conclude that the<br />

influence of a small capillary anisotropy—thus of the term<br />

hom () in Eq. 4—on the dynamics of the wavy fronts is<br />

negligible.<br />

3. Isolated traveling waves<br />

Figure 19 shows a time series of plots measured through<br />

the wave train of Fig. 10. The plots are represented in the<br />

reference frame attached to the wave (24°). The amplitudes<br />

of the secondary maxima decrease much more rapidly<br />

than that of the principal maximum, so that only an isolated<br />

TW finally survives. This isolated wave has an asymm<strong>et</strong>ric<br />

profile with a width of about 16 . Such a profile is a<br />

reproducible feature of the isolated TW’s, as will be shown<br />

shortly. Near the end of the recording, the standard deviation<br />

of the distribution decreases with a damping rate of about<br />

0.02 s 1 . This rate is low compared to the phase diffusion<br />

damping rate calculated for the characteristic wavelength of<br />

16 0.4 s 1 .<br />

Figure 20 shows an isolated TW emitted a mechanism to<br />

be described below see Fig. 23b, and Fig. 21 the corresponding<br />

plots. The profile of this wave is quite similar<br />

to the one appearing in Fig. 19. The standard deviation of the<br />

distribution first decreases with a damping rate of about<br />

0.05 s 1 , and then reaches a plateau. The corresponding<br />

FIG. 19. plots measured at time intervals of 3.95 s represented<br />

in the reference frame drifting at the velocity W6 ms 1 . Same<br />

run as in Fig. 10. Each plot has been shifted upwards with respect to<br />

the preceding one by 0.5 m.<br />

phase diffusion damping rate would be of about 3 s 1 . The<br />

plateau signals the existence of a permanent background of<br />

low amplitude waves, the d<strong>et</strong>ailed structure of which is not<br />

resolved.<br />

The reproducible features of the isolated TW’s drift velocity,<br />

profile, and slow damping rate raise the question of<br />

the possible existence of solitary waves. The observed TW’s<br />

are clearly not solitary waves, but are perhaps close to a<br />

solitary wave at the moment of their creation. In any case,<br />

they are certain to be strongly nonlinear objects, as shown by<br />

the complex evolution of the profi<strong>les</strong> of the wave trains.<br />

4. Dense populations of traveling waves<br />

The TW’s are a boundary condition dependent phenomenon.<br />

This is clearly illustrated in Fig. 8, which shows the<br />

central part of a large grain free of TW’s at a high V in a<br />

sample doped with naphthalene. In smaller grains, however,<br />

the front is generally continually swept by rightward and<br />

FIG. 18. Lamella tilt angle vs spacing . Symbols:<br />

experimental values from Figs. 16 and 17. Broken line: second<br />

order polynomial fit. JH 4.0 m.<br />

FIG. 20. Isolated traveling wave. V31 ms 1 . C naph <br />

2.510 4 .


3768 SILVÈRE AKAMATSU AND GABRIEL FAIVRE<br />

PRE 61<br />

FIG. 21. plots measured at time intervals of 1.05 s represented<br />

in the reference frame drifting at W19.5 ms 1 . Same run as in<br />

Fig. 20. The data points of one of the plots are shown to illustrate<br />

the smoothing procedure. Each plot has been shifted upwards with<br />

respect to the preceding one by 1 m.<br />

leftward wave trains originating from the two boundaries.<br />

The resulting pattern resemb<strong>les</strong> but is not really standing<br />

waves Fig. 11. In fact, the interactions b<strong>et</strong>ween TW’s are<br />

highly nonlinear. This is illustrated in Fig. 22, which shows<br />

two different interaction processes b<strong>et</strong>ween TW’s observed<br />

during the same experimental run. The corresponding time<br />

series of plots are displayed in Fig. 23. In one case Fig.<br />

23a, two TW’s of slightly different amplitudes me<strong>et</strong>. The<br />

amplitudes and the trajectories of the TW’s are changed during<br />

the crossing process, the weakest of the two being further<br />

damped down. In the other case Fig. 23b, a lowamplitude<br />

leftward TW is strongly amplified through the<br />

crossing with a large-amplitude rightward TW. The amplification<br />

mechanism involves an oscillatory mode of instability<br />

of the lamellar pattern and a lamella branching event. Similar<br />

short-frequency oscillations are observed in Fig. 23a. This<br />

type of oscillation was not observed previously, and is probably<br />

a 3D mode of instability occurring when is much<br />

smaller than the sample thickness, thus at high V.<br />

5. Shallow eutectic colonies<br />

A typical feature of the shallow EC patterns is the existence<br />

of extremely wide long-lived EC’s Fig. 9. These<br />

large EC’s do not seem subjected to tip splitting in the usual<br />

sense of the term the front profile shows no d<strong>et</strong>ectable<br />

trough near the center of the EC’s. This suggests that one<br />

should consider the shallow inter-EC grooves as localized<br />

entities, and the tip of the wide EC’s as portions of an extended<br />

m<strong>et</strong>astable wavy front. Thus considered, each groove<br />

FIG. 22. Two different examp<strong>les</strong> of TW crossings same run as<br />

in Fig. 20. Each image is 113 m wide.<br />

is a defect containing a lamella sink and two lamella sources<br />

operating at the same average frequency so that the average<br />

number of lamellae remains constant, and bound tog<strong>et</strong>her by<br />

the Cahn distortion of the front. Symm<strong>et</strong>rical, stationary<br />

inter-EC grooves as well as asymm<strong>et</strong>rical, traveling ones are<br />

observed. The traveling velocity of the latter does not seem<br />

to be unique, which obviously implies that the pattern as a<br />

whole is unsteady.<br />

Shallow inter-EC grooves are active sources of TW’s<br />

Fig. 9. The tips of the shallow EC’s are therefore permanently<br />

occupied by a particularly dense population of TW’s.<br />

This som<strong>et</strong>imes results in the creation of a new groove at the<br />

tip of an EC. The d<strong>et</strong>ailed mechanism of these rare, nonstandard<br />

tip-splitting events has not been elucidated. It can<br />

only be stated that lamella termination and lamella branching<br />

events are not rare in wavy fronts, but are most generally<br />

healed up by the emission or absorption of a TW See Fig.<br />

23b, and are therefore not the only cause of the occurrence<br />

of tip splitting events.<br />

The same nonelucidated mechanism may give rise to a<br />

process of nucleation and invasion of wavy fronts by shallow<br />

EC’s starting from eutectic grain boundaries. In this process,<br />

a first inter-EC groove is created at some distance of a eutectic<br />

grain boundary by the emitted TW’s, a second<br />

inter-EC groove is created by the TW’s emitted by the first<br />

inter-EC groove, and so on. This process was actually observed<br />

in a few experimental runs, and was extremely slow.<br />

6. Comparison with theory<br />

The observed TW’s share many important features with<br />

the critical long-wavelength modes found by Plapp and<br />

FIG. 23. Time series of plots for Fig. 22a<br />

a and Fig. 22b b. Time interval: 0.54 s. Upward<br />

shift: 1 m. Dotted lines: frames of Figs.<br />

22a and 22b. Filled circle: lamella branching.


PRE 61 TRAVELING WAVES, TWO-PHASE FINGERS, AND . . .<br />

3769<br />

Karma: the TW’s obey Cahn’s rule, appear at values of V<br />

comparable to V cs , and have the same order of magnitude of<br />

their characteristic width and drift velocity as the calculated<br />

long-wavelength modes. Moreover, as already stated, (k),<br />

as calculated by Plapp and Karma, is probably complex in<br />

the conditions in which the TW’s are observed. Thus the<br />

basic mechanism responsible for the drift of the TW’s most<br />

probably is the one pointed out by these authors, namely, the<br />

interaction b<strong>et</strong>ween the distortions of the envelope and those<br />

of the lamellar pattern embodied in Cahn’s rule. The fact that<br />

the TWs are only emitted by strong external perturbations,<br />

and fade out progressively afterward, shows that the experimentally<br />

measured transition velocity V TW is not the critical<br />

velocity for the Plapp-Karma long-wavelength mode, but a<br />

lower velocity, below which the imaginary part of is zero,<br />

and above which it is nonzero.<br />

On the other hand, several aspects of the observations<br />

remain unexplained, in particular, the fact that the TW’s do<br />

not amplify even at velocities much higher than V TW . As far<br />

as we can see, the only theor<strong>et</strong>ical clue to this spectacular<br />

stabilization of the planar front is the rapid rise of the effective<br />

capillary correction term as the impurity concentration<br />

decreases from 5 to 2.510 4 see Sec. II B.<br />

VI. CONCLUSION<br />

This study has brought to light the specific features of the<br />

transition to eutectic colonies in ternary lamellar eutectics<br />

compared to the transition to cells in dilute alloys. These<br />

features stem from the interaction b<strong>et</strong>ween the dynamics of<br />

the lamellar pattern and the large-scale impurity-driven cellulation<br />

process of the front. As far as long-wavelength distortions<br />

of the front are concerned, this interaction is well<br />

described by Cahn’s rule, which must, however, be amended<br />

in order to take into account the tilt of the lamellae due to<br />

capillary anisotropy and spontaneous symm<strong>et</strong>ry breaking. A<br />

consequence of Cahn’s rule recently asserted by Plapp and<br />

Karma is that long-wavelength distortions of the front drift<br />

laterally above a certain velocity when the interlamellar<br />

spacing is sufficiently large. We have observed traveling<br />

waves, which, although highly nonlinear, are most probably<br />

due to this effect.<br />

However, the main contribution of this study probably<br />

consists of a series of unexpected facts, the most important<br />

of which are the following: the transition to EC’s is not<br />

continuous, but mediated by a first transition to a new type of<br />

dynamical structure called two-phase finger, similar to the<br />

doublon and multipl<strong>et</strong> observed in directionally solidified<br />

low-anisotropy dilute alloys; the transition to EC’s disappears<br />

relatively abruptly as the impurity concentration decreases;<br />

at low impurity concentration, strongly m<strong>et</strong>astable<br />

wavy fronts i.e., planar front swept by traveling waves are<br />

observed over a large range in velocity above a certain<br />

threshold value. Further studies are in progress concerning,<br />

in particular, the properties of the two-phase fingers at<br />

slightly off-eutectic concentrations.<br />

ACKNOWLEDGMENTS<br />

We gratefully acknowledge many stimulating discussions<br />

with A. Karma and M. Plapp, and thank them for communicating<br />

their results prior to publication. Thanks are also due<br />

to H. Savary and A.-M. Pougn<strong>et</strong>, of the Centre National<br />

d’Etudes des Télécommunications, France-Telecom, Bagneux,<br />

France, for providing us with zone-refined chemicals.<br />

This research was financially supported by the Center National<br />

d’Etudes Spatia<strong>les</strong>, France.<br />

1 W. Kurz and D. J. Fischer, in Fundamentals of solidification<br />

Trans. Tech, Aedermannsdorf, 1984, p.93.<br />

2 J. D. Hunt and K. A. Jackson, Trans. AIME 236, 843 1966.<br />

3 K. A. Jackson and J. D. Hunt, Trans. AIME 236, 1129 1966.<br />

4 W. Datye and J. S. Langer, Phys. Rev. B 24, 4155 1981; see<br />

also J. S. Langer, Phys. Rev. L<strong>et</strong>t. 44, 1023 1980.<br />

5 A. Karma, Phys. Rev. L<strong>et</strong>t. 59, 711987.<br />

6 A. Karma and A. Sarkissian, M<strong>et</strong>all. Mater. Trans. A 27, 635<br />

1996.<br />

7 K. Kassner and C. Misbah, Phys. Rev. A 44, 6533 1991.<br />

8 K. Kassner, C. Misbah, and R. Bauman, Phys. Rev. E 51,<br />

R2751 1995.<br />

9 G. Faivre and J. Mergy, Phys. Rev. A 45, 7320 1992; 46, 963<br />

1992; also see J. Mergy, Thèse de l’ Université Paris VII,<br />

France, 1992 unpublished.<br />

10 M. Ginibre, S. Akamatsu, and G. Faivre, Phys. Rev. E 56, 780<br />

1997; also see M. Ginibre, Thèse de l’ Université Paris VI,<br />

France, 1997 unpublished.<br />

11 For simplicity, we ignore here the distinction b<strong>et</strong>ween impurities<br />

unwanted or uncontrolled chemical components and addition<br />

components intentionally added species.<br />

12 W. H. Weart and J. D. Mack, Trans. M<strong>et</strong>all. Soc. AIME 236,<br />

1129 1958.<br />

13 J. P. Chilton and W. C. Winegard, J. Inst. M<strong>et</strong>. 89, 161961.<br />

14 J. E. Gruz<strong>les</strong>ki and W. C. Winegard, J. Inst. M<strong>et</strong>. 96, 304<br />

1968; Trans. AIME 242, 1785 1968.<br />

15 M. D. Rinaldi, R. M. Sharp, and M. C. Flemings, M<strong>et</strong>all.<br />

Trans. 3, 3133 1972.<br />

16 J. W. Rutter and B. Chalmers, Can. J. Phys. 31, 151953; see<br />

also W. A. Tiller, K. A. Jackson, J. W. Rutter, and B. Chalmers,<br />

Acta M<strong>et</strong>all. 1, 498 1953.<br />

17 W. W. Mullins and R. F. Sekerka, J. Appl. Phys. 35, 444<br />

1964.<br />

18 In a binary dilute alloy, V cs is defined as the value of V above<br />

which the liquid ahead of the front is supersaturated with respect<br />

to the solute 16. This definition can be extended to<br />

multicomponent eutectic alloys by averaging the concentrations<br />

along the direction perpendicular to the lamellae 15.<br />

19 The experimental fact that, in the EC patterns, the trajectories<br />

of the lamellae run approximately perpendicular to the envelope<br />

of the deformed front was apparently first noted by Weart<br />

and Mack 12. On the other hand, according to Jackson and<br />

Hunt 3, Cahn was the first to make the assumption that this<br />

rule was verified in all cases i.e., also in the absence of EC’s,<br />

and point out the consequences of this hypothesis. Here we<br />

primarily use the term of Cahn’s rule in reference to experimental<br />

observations.<br />

20 W. Losert, B. Q. Shi, and H. Z. Cummins, Proc. Natl. Acad.


3770 SILVÈRE AKAMATSU AND GABRIEL FAIVRE<br />

PRE 61<br />

Sci. USA 95, 431 1998; 95, 439 1998.<br />

21 M. Plapp and A. Karma, Phys. Rev. E 60, 6865 1999.<br />

22 W. L. Kaukler and J. W. Rutter, Mater. Sci. Eng. 65, L1<br />

1984.<br />

23 V. Se<strong>et</strong>haraman and R. Trivedi, M<strong>et</strong>all. Trans. A 19, 2955<br />

1988.<br />

24 R. Trivedi, J. T. Mason, J. D. Verhoeven, and W. Kurz, M<strong>et</strong>all.<br />

Trans. A 22, 2523 1991.<br />

25 B. Caroli, C. Caroli, G. Faivre, and J. Mergy, J. Cryst. Growth<br />

118, 135 1992.<br />

26 J. Mergy, G. Faivre, C. Guthmann, and R. Mell<strong>et</strong>, J. Cryst.<br />

Growth 134, 353 1993.<br />

27 L. M. Hogan, R. W. Kraft, and F. D. Lemkey, Adv. Mater.<br />

Res. N.Y. 5, 831971.<br />

28 We use the public domain NIH Image program developed at<br />

the U.S. National Institutes of Health and available from Intern<strong>et</strong><br />

by anonymous FTP from zippy.nimh.nih.gov or on floppy<br />

disk from the National Technical Information Service, Springfield,<br />

Virginia, part No. PB95-500195GEI.<br />

29 S. Akamatsu and G. Faivre, J. Phys. I 6, 503 1996.<br />

30 V. G. Smith, W. A. Tiller, and J. W. Rutter, Can. J. Phys. 33,<br />

723 1955; also see B. Caroli, C. Caroli, and L. Ramirez-<br />

Piscina, J. Cryst. Growth 132, 377 1993.<br />

31 T. Ihle and H. Müller-Krumbhaar, Phys. Rev. E 49, 2972<br />

1994.<br />

32 S. Akamatsu, G. Faivre, and T. Ihle, Phys. Rev. E 51, 4751<br />

1995.<br />

33 H. Jamgotchian, R. Trivedi, and B. Billia, Phys. Rev. E 47,<br />

4313 1993.<br />

34 W. Losert, D. A. Stillman, H. Z. Cummins, P. Kopczynski,<br />

W.-J. Rappel, and A. Karma, Phys. Rev. E 58, 7492 1998.<br />

35 P. Kopczynski, W. J. Rappel, and A. Karma, Phys. Rev. L<strong>et</strong>t.<br />

77, 3387 1996; 79, 2698 1997.<br />

36 S. Akamatsu and G. Faivre, Phys. Rev. E 58, 3302 1998.<br />

37 K. Kassner and C. Misbah, Phys. Rev. A 45, 7372 1992.


PHYSICAL REVIEW E, VOLUME 65, 011702<br />

Dynamics of a fac<strong>et</strong>ed nematic–smectic-B front in thin-sample directional solidification<br />

T. Börzsönyi, 1,2 S. Akamatsu, 1 and G. Faivre 1<br />

1 Groupe de Physique des Solides, CNRS UMR 75-88, Universités Denis Diderot and <strong>Pierre</strong> <strong>et</strong> Marie Curie, Tour 23, 2 place Jussieu,<br />

75251 Paris Cedex 05, France<br />

2 Research Institute for Solid State Physics and Optics, Hungarian Academy of Sciences, H-1525 Budapest, P.O. Box 49, Hungary<br />

Received 5 June 2001; published 13 December 2001<br />

We present an experimental study of the directional-solidification patterns of a nematic–smectic-B front.<br />

The chosen system is C 4 H 9 (C 6 H 10 ) 2 CN in short, CCH4 in 12 m-thick samp<strong>les</strong>, and in the planar<br />

configuration director parallel to the plane of the sample. The nematic–smectic-B interface presents a fac<strong>et</strong> in<br />

one direction—the direction parallel to the smectic layers—and is otherwise rough and devoid of forbidden<br />

directions. We measure the Mullins-Sekerka instability threshold and establish the morphology diagram of the<br />

system as a function of the solidification rate V and the angle 0 b<strong>et</strong>ween the fac<strong>et</strong> and the isotherms. We focus<br />

on the phenomena occurring immediately above the instability threshold when 0 is neither very small nor<br />

close to 90°. Under these conditions, we observe drifting shallow cells and a type of solitary wave, called<br />

‘‘fac<strong>et</strong>on,’’ which consists essentially of an isolated macroscopic fac<strong>et</strong> traveling laterally at such a velocity that<br />

its growth rate with respect to the liquid is small. Fac<strong>et</strong>ons may propagate either in a stationary or an<br />

oscillatory way. The d<strong>et</strong>ailed study of their dynamics casts light on the microscopic growth mechanisms of the<br />

fac<strong>et</strong>s in this system.<br />

DOI: 10.1103/PhysRevE.65.011702<br />

PACS numbers: 64.70.Md, 81.10.Aj, 64.70.Dv, 68.70.w<br />

I. INTRODUCTION<br />

A crystal growing from an undercooled melt rejects heat<br />

and chemical species, which must diffuse away in the liquid<br />

for the process to continue. The thus-generated thermal and<br />

solutal gradients tend to destabilize the advancing solidliquid<br />

interface. This effect is counterbalanced by the surface<br />

tension and the so-called interfacial kin<strong>et</strong>ics, which tends to<br />

slow down the progression of the interface, and hence, stabilize<br />

it. As a result of the comp<strong>et</strong>ition b<strong>et</strong>ween these conflicting<br />

factors, solidification fronts may assume a large vari<strong>et</strong>y<br />

of nonlinear patterns, the characteristics of which<br />

depend on the control param<strong>et</strong>ers, and the initial and boundary<br />

conditions of the process.<br />

The study of solidification patterns has been an active<br />

field of research for several decades 1–3. Most of the existing<br />

studies are devoted to fully nonfac<strong>et</strong>ed systems. In<br />

such systems, the surface tension and the kin<strong>et</strong>ic coefficient<br />

defined as the ratio of the kin<strong>et</strong>ic undercooling to the<br />

growth velocity are nonsingular functions of the orientation<br />

of the interface with respect to the crystal lattice. On a molecular<br />

scale, this corresponds to the fact that the interface is<br />

rough in all orientations. Familiar aspects of the dynamics of<br />

fully nonfac<strong>et</strong>ed systems in directional solidification, i.e.,<br />

when the system is pulled at a constant velocity V toward the<br />

cold side of an applied unidirectional thermal gradient G see<br />

Fig. 1, are the existence of a stable planar front at low values<br />

of V, the primary cellular or Mullins-Sekerka instability<br />

occurring at a threshold velocity V c , the quasiperiodic arrays<br />

of rounded cells at V slightly above V c , and of dendrites at V<br />

much higher than V c . Many dynamical features of these patterns<br />

e.g., stability limits, modes of instability are not y<strong>et</strong><br />

fully understood, but some of their fundamental properties<br />

are now clear, among which the crucial role played by interfacial<br />

anisotropy 1,4,5. In fact, a certain minimum degree<br />

of interfacial anisotropy is a necessary condition for cellular<br />

and dendritic arrays to be stable, or even to exist. In thin<br />

samp<strong>les</strong>—i.e., quasibidimensional 2D systems— and <br />

are functions of a single variable, say, the tilt angle of the<br />

interface with respect to the isotherms. The functions ()<br />

and (), and thus the solidification patterns, depend on the<br />

orientation of the crystal with respect to the solidification<br />

s<strong>et</strong>up 6–8.<br />

In contrast with the case of fully nonfac<strong>et</strong>ed systems, little<br />

is y<strong>et</strong> known about the directional-solidification dynamics of<br />

fac<strong>et</strong>ed crystals. The few existing experimental studies on<br />

this subject first of all show that a distinction must be made<br />

b<strong>et</strong>ween fully and partly fac<strong>et</strong>ed systems 6,9–12. Growth<br />

fac<strong>et</strong>s which most generally, although not necessarily, coincide<br />

with equilibrium fac<strong>et</strong>s 13 correspond to planes of the<br />

crystal containing several directions of strong binding. Fully<br />

fac<strong>et</strong>ed crystals have numerous fac<strong>et</strong> directions, and their<br />

directional-solidification fronts consist of a succession of<br />

fac<strong>et</strong>s limited by sharp edges. The dynamics of such fronts<br />

does not give rise to any stationary state, in general, and<br />

bears no obvious relation with that of nonfac<strong>et</strong>ed fronts.<br />

Partly fac<strong>et</strong>ed systems only have a few fac<strong>et</strong> directions connected<br />

to one another by large rounded regions. In lamellar<br />

crystals, the solid-liquid interface may be rough in all but<br />

one direction, namely, that of the molecular layers. In this<br />

case, when the tilt angle 0 of the layers with respect to the<br />

FIG. 1. Sk<strong>et</strong>ch of a thin-sample directional-solidification s<strong>et</strong>up.<br />

z: axis of the thermal gradient; x: axis parallel to the isotherms; V:<br />

pulling velocity. After a transient, the front advances in average at<br />

the imposed velocity V with respect to the liquid, and thus remains<br />

essentially immobile in the laboratory reference frame. It can then<br />

be continuously observed with an optical microscope.<br />

1063-651X/2001/651/01170211/$20.00<br />

65 011702-1<br />

©2001 The American Physical Soci<strong>et</strong>y


T. BÖRZSÖNYI, S. AKAMATSU, AND G. FAIVRE PHYSICAL REVIEW E 65 011702<br />

isotherms is large, the dynamics of the front must obviously<br />

be that of a nonfac<strong>et</strong>ed crystal as long as the deformation of<br />

the front remains small, that is, below V c and in a small<br />

range of V above V c . Fac<strong>et</strong>s only appear at higher V when<br />

the deformation of the interface is large. A relatively smooth<br />

transition from the nonfac<strong>et</strong>ed to partly fac<strong>et</strong>ed dynamics<br />

may then be observed. This is the experimental configuration<br />

considered in this study.<br />

In this paper, we study the directional-solidification dynamics<br />

of the front associated to the nematic–smectic-B<br />

transition of the liquid-crystal C 4 H 9 (C 6 H 10 ) 2 CN in short,<br />

CCH4. A long-range order exists in the direction perpendicular<br />

to the molecular layers in the smectic-B phase, so<br />

that this phase actually is a lamellar crystal. Previous freegrowth<br />

studies have indeed shown that the nematic–smectic-<br />

B fronts of the n3,4,5 members of the series CCHn where<br />

n stands for the number of carbon atoms in the aliphatic<br />

chain have a single fac<strong>et</strong> direction parallel to the molecular<br />

layers of the smectic phase, and are rough in all other directions<br />

14–16. Moreover, they have no unstable orientations<br />

in a direction perpendicular to the molecular layers, contrary<br />

to the smectic-A –smectic-B fronts previously studied in directional<br />

solidification by Melo and Oswald and Oswald<br />

<strong>et</strong> al. 6,11,12. The present study is performed in thin<br />

(12 m-thick samp<strong>les</strong> and in the planar configuration director<br />

parallel to the plane of the sample, in order for the<br />

front—including the fac<strong>et</strong>s, if any—to remain perpendicular<br />

to the sample plane. Practically, the system is thus a 2D one.<br />

We shall mostly focus on a type of solitary wave appearing<br />

near the Mullins-Sekerka threshold, called ‘‘fac<strong>et</strong>on’’ because<br />

it contains a single small fac<strong>et</strong> traveling along the front<br />

at such a velocity that the normal growth rate of the fac<strong>et</strong>,<br />

i.e., its growth rate with respect to the liquid, is generally<br />

much smaller than V. Such a phenomenon, which has never<br />

been observed before, to the best of our knowledge, is obviously<br />

highly specific to fac<strong>et</strong>ed directional solidification, and<br />

therefore particularly interesting from our present viewpoint.<br />

A preliminary comment about the nematic–smectic-B fac<strong>et</strong>s<br />

in the CCHn series is in order. The growth rate of a fac<strong>et</strong> is<br />

controlled by the dynamics of the molecular steps flowing<br />

along it. Therefore, it crucially depends on wh<strong>et</strong>her or not the<br />

fac<strong>et</strong> contains, or is connected with, step sources 13,17.<br />

When no step source is available, the fac<strong>et</strong> grows through<br />

nucleation and spreading of terraces surface nucleation,<br />

which is a very slow process at low undercooling. In fact, the<br />

growth rate of a perfect fac<strong>et</strong> is totally negligible when the<br />

undercooling is lower than some finite value. Such a behavior<br />

‘‘blocked’’ fac<strong>et</strong>s at low undercoolings has clearly been<br />

observed during the solidification of many, but not all the<br />

studied fac<strong>et</strong>ed systems. What concerns us here is that it was<br />

not observed during the free growth of the smectic-B phase<br />

of CCH3, despite the strongly fac<strong>et</strong>ed aspect of the growing<br />

crystals 14–16. Numerical simulations in which a cusplike<br />

minimum of () but no anisotropy of was taken into<br />

account satisfactorily reproduced the observed growth<br />

shapes. Thus, the observation of a fac<strong>et</strong> on a macroscopic<br />

scale would not necessarily mean the presence of a singularity<br />

in . In order to clarify this point in the case of CCH4,<br />

we report, in Sec. III, preliminary observations in free<br />

FIG. 2. The nematic–smectic-B equilibrium temperature in a<br />

CCH4 sample as a function of time. T NS was measured by controlling<br />

the temperature of a free-growth stage in order to keep a small<br />

smectic-B crystal in quasiequilibrium with the nematic. The relatively<br />

low initial value of T NS indicates that the sample was rather<br />

impure at the outs<strong>et</strong>.<br />

growth showing that the nematic–smectic-B fac<strong>et</strong> of CCH4<br />

is capable of remaining immobile at undercoolings lower<br />

than 0.1 K. Thus, in CCH4 at least, the nematic–smectic-B<br />

fronts can form growth fac<strong>et</strong>s.<br />

II. EXPERIMENT<br />

The relevant material param<strong>et</strong>ers of the liquid-crystal<br />

CCH4 MERCK IS-0558 may be found in Ref. 15. The<br />

residual impurities, the chemical nature of which is unknown,<br />

were characterized as regards solidification by the<br />

usual m<strong>et</strong>hods see below. We found that the impurity content<br />

at the outs<strong>et</strong> of the experiments was reproducible, but<br />

slowly increased during the experiments, indicating that the<br />

product was undergoing a decomposition in the nematic<br />

phase, as previously noticed and analyzed for the case of<br />

CCH3 18. The nematic–smectic-B transition temperature<br />

T NS was generally of about 53.1 K in fresh samp<strong>les</strong>. Figure 2<br />

shows T NS measured as a function of time in one sample. It<br />

can be seen that the decomposition rate is sufficiently slow<br />

not to severely perturb a solidification run, but sufficiently<br />

rapid to prevent us to carry out several successive runs with<br />

the same sample. Outgasing the as-received product resulted<br />

in a significant slowing down of the decomposition process.<br />

We have studied the crystal structure of the smectic-B<br />

phase of CCH4 by low-angle x-ray diffraction 19. As expected,<br />

this phase is basically an AB-type stacking of hexagonal<br />

layers. The param<strong>et</strong>ers are approximately a5.9 Å<br />

and c29 Å, which is in accordance with the data available<br />

for the other members of the homologue series 20. The<br />

hexagonal layers however appear to be slightly distorted,<br />

which may entail the existence of superstructures in the layers.<br />

A schematic view of a thin-sample directionalsolidification<br />

experiment is shown in Fig. 1. A d<strong>et</strong>ailed description<br />

of our s<strong>et</strong>up is given elsewhere 7,8. In this study,<br />

the samp<strong>les</strong> were made of two parallel glass plates separated<br />

by 12-m-thick plastic spacers. Their useful width was of 9<br />

mm and their length of 60 mm. They were filled under an<br />

argon atmosphere at a temperature higher than T NS , and then<br />

011702-2


DYNAMICS OF A FACETED NEMATIC–SMECTIC-B. .. PHYSICAL REVIEW E 65 011702<br />

FIG. 3. Free growth. Successive snapshots of a smectic-B crystal<br />

of CCH4 growing from the nematic phase at T0.07 K.<br />

cooled down to room temperature. Numerous smectic crystals<br />

appeared by h<strong>et</strong>erogeneous nucleation during cooling.<br />

The samp<strong>les</strong> were placed in the thermal gradient, and a<br />

smectic-B crystal of known orientation d<strong>et</strong>ermined through<br />

the observed value of 0 ) was selected by a m<strong>et</strong>hod to be<br />

explained shortly. The sample was annealed at rest for about<br />

30 minutes in order to homogenize the concentration in the<br />

liquid. V was then switched to a chosen value, left for a<br />

given time at this value, and then increased step by step. The<br />

temperature gradient at the growth front was of 53 K cm 1 ,<br />

un<strong>les</strong>s otherwise mentioned. The pulling velocity was in the<br />

range of 0.3–30 ms 1 . The observations were made with<br />

a polarizing microscope Leica equipped with a chargecoupled<br />

device camera. The video signal was analyzed with<br />

digital image processing.<br />

It is obviously crucial for our experiments that large<br />

smectic-B crystals of arbitrary orientation might be selected.<br />

To this aim, we have studied the influence of various treatments<br />

of the inner sides of the glass plates. Three types of<br />

plates were used: untreated plates, plates covered with a monooriented<br />

thin film of polyt<strong>et</strong>rafluor<strong>et</strong>hylene prepared by<br />

friction transfer at T200 °C 21, or with a 100 Å-thick<br />

layer of Al or In deposited by oblique evaporation. In the<br />

nematic phase, the orientation of the director was essentially<br />

planar in all samp<strong>les</strong>. The director was more or <strong>les</strong>s aligned<br />

along the direction of friction, or deposition, in treated<br />

samp<strong>les</strong>, but domains corresponding to small a few degrees<br />

variations in the orientation of the director, still existed see<br />

Fig. 4 below. This inhomogeneity of the nematic phase<br />

caused but minor perturbations in our experiments, since the<br />

phenomena of interest turned out to be essentially independent<br />

of the orientation of the nematic director. In all samp<strong>les</strong>,<br />

the smectic-B phase had a planar orientation, but was divided<br />

into different crystals or grains corresponding to a<br />

different value of 0 . The surface treatment gave a pronounced<br />

preferential distribution of 0 among the various<br />

grains, facilitating the selection of the desired value of 0 .<br />

The size of the selected smectic-B grain was increased by a<br />

m<strong>et</strong>hod consisting of forcing the crystal to grow through a<br />

funnel-shaped obstacle 7. By this m<strong>et</strong>hod, smectic grains of<br />

a millim<strong>et</strong>ric width, and arbitrary values of 0 were obtained.<br />

III. CHARACTERIZATION OF THE SYSTEM<br />

A. Free growth at small undercoolings<br />

The observations reported in this section were performed<br />

with a free-growth s<strong>et</strong>up similar to the one described in Ref.<br />

FIG. 4. Directional solidification in this, and all the following<br />

micrographs, growth is upwards. N: nematic; Sm1: smectic-B;<br />

Sm2: smectic-B oriented differently from Sm1. a Sample at rest<br />

(V0); b sample in the process of solidification at V<br />

0.9 ms 1 . Note the domains in the nematic. Sm1 is a single<br />

crystal, but Sm2 is a polycrystal, as shown by the presence of cusps<br />

on the Sm1-Sm2 front.<br />

22, in which the changes in the undercooling are produced,<br />

via the Clausius-Clapeyron effect, by a sudden pressure<br />

change at constant temperature, and are therefore quasiinstantaneous<br />

22,23. The samp<strong>les</strong> were the same as those<br />

used in directional solidification. At the beginning of the experiments,<br />

the samp<strong>les</strong> were heated step by step until only<br />

one small smectic-B crystal was left in the nematic. The<br />

sample was maintained at constant temperature until the<br />

changes in the shape of the crystal became very slow this<br />

took about 20 minutes. Admittedly, this shape is not the<br />

exact equilibrium shape of the crystal, but it exhibits clear<br />

reproducible features, namely, long fac<strong>et</strong>s parallel to the<br />

smectic layers and rounded ends in the perpendicular direction<br />

Fig. 3 a, which is enough for our present purpose. It<br />

should be noted that the observed near-equilibrium shape<br />

clearly shows the absence of a forbidden orientation range<br />

around 90°, where is the deviation of the interface<br />

from the direction of the molecular layers, but suggests that<br />

the fac<strong>et</strong> might actually be limited by a sharp edge, i.e., the<br />

interface might be unstable at small values of . The fact<br />

that we have not observed the Herring instability 24,11 in<br />

directional solidification at the lowest-explored value of 0<br />

indicates that this forbidden orientation range is very small<br />

(2°), if it exists at all.<br />

A sudden increase of the undercooling T was applied at<br />

time t0, and the subsequent growth of the crystal recorded<br />

Fig. 3. The growth process, which is governed by the anisotropic<br />

interfacial properties and diffusivities, is very complicated.<br />

Its study is beyond the scope of this paper. Here, we<br />

limit ourselves to the following observation: the fac<strong>et</strong>s of the<br />

smectic-B crystals remained blocked within experimental<br />

uncertainty their growth rate was lower than about<br />

0.01 ms 1 ) at undercoolings lower than 0.1 K Fig. 3.At<br />

higher undercoolings, they generally grow at a measurable<br />

rate. The apparent threshold undercooling T nucl for growth<br />

by surface nucleation of our system is thus larger than 0.1 K<br />

and probably not much larger than this value. This estimate<br />

of T nucl is small compared to what it is in ordinary solid-<br />

011702-3


T. BÖRZSÖNYI, S. AKAMATSU, AND G. FAIVRE PHYSICAL REVIEW E 65 011702<br />

liquid systems, but this may be explained by the small value<br />

of in our system 25,26. It is also possible that in our thin<br />

samp<strong>les</strong>, surface nucleation is in fact h<strong>et</strong>erogeneous, i.e.,<br />

takes place preferentially along the line of contact with the<br />

glass plates. The nucleation rate would then depend on the<br />

treatment of the glass plates.<br />

B. Directional solidification: Instability threshold<br />

The Mullins-Sekerka instability threshold was found to lie<br />

b<strong>et</strong>ween approximately 2 and 3ms 1 in all the studied<br />

fresh samp<strong>les</strong>. No influence of the orientation of the smectic,<br />

or the nematic was observed within experimental uncertainty.<br />

However, it should be noted that this uncertainty was<br />

large (1 ms 1 ) for the reason to be explained presently.<br />

Figure 4 shows a sample at rest, and pulled at a rate lower<br />

than V c . Two isothermal fronts are visible, namely, a front<br />

separating the nematic N phase from a smectic-B domain<br />

Sm1, and at a lower temperature, a front separating Sm1<br />

from a second smectic-B domain Sm2. The nature of the<br />

transition from Sm1 to Sm2 is not y<strong>et</strong> clear. This transition<br />

was observed in most, but not all experiments. Observations<br />

not reported here incline us to think that Sm2 is the same<br />

phase as Sm1, but with a different orientation, thus, a different<br />

interaction energy with the glass plates. In any case, we<br />

need not take into account the Sm1-Sm2 front here since this<br />

front, when present, does not perturb the dynamics of the<br />

N-Sm1 front.<br />

It can be seen in Fig. 4 that the nematic–smectic-B front<br />

remains planar during solidification at VV c , except for<br />

small, long-wavelength distortions due to the presence of<br />

domains in the nematic phase. These distortions are larger<br />

during solidification than at rest, and undergo sudden<br />

changes each time the front leaves a nematic domain for<br />

another. This phenomenon has thus an equilibrium as well as<br />

a kin<strong>et</strong>ic origin. In our experiments, it plays the role of a<br />

relatively strong, long-wavelength, low-frequency noise,<br />

which blurs some of the morphological-transition thresholds<br />

of the system. This is the main origin of the aforementioned<br />

large uncertainty on the measured values of V c . However,<br />

we may state with certainty that V c was higher than<br />

2 ms 1 since the distortions caused by nematic domains,<br />

or any other source of perturbation e.g., dust partic<strong>les</strong> did<br />

not amplify below this velocity.<br />

C. Solute redistribution transient<br />

When V is smaller than V c , the front reaches a stationary<br />

planar state through the so-called solute redistribution transient.<br />

A recoil curve—i.e., the curve representing the variation<br />

of the position or temperature of the planar front as a<br />

function of time during the initial transient of a particular<br />

run—is reproduced in Fig. 5. It is well known that information<br />

about the relevant properties of the solute diffusion coefficient<br />

D in the liquid, partition coefficient K, thermal gap<br />

T o may be gained from the characteristics of the transient,<br />

and the value of V c . We have utilized this m<strong>et</strong>hod in order to<br />

characterize the unknown impurity playing the role of solute<br />

in our system.<br />

FIG. 5. Recoil curve at V0.9 ms 1 . Same run as in Fig. 4.<br />

Continuous line: best fit according to the Warren-Langer approximation.<br />

The rapid decrease at the ons<strong>et</strong> of the recoil is an instrumental<br />

effect.<br />

The threshold velocity, and the amplitude of the solute<br />

redistribution transient are given by V c (1KD s /<br />

D)DG/T o (D s : diffusion coefficient in the solid and T o ,<br />

respectively, 27. By fitting the recoil data using the Warren-<br />

Langer approximate theory 28 Fig. 5 and assuming V c<br />

2.5 ms 1 and KD s /D1, we obtained K0.12. This<br />

gives D80 m 2 s 1 and T o 0.2 K. These data give us<br />

no information about D s , but there is good reason to believe<br />

that our system is a two-sided one—i.e., that D s is not much<br />

smaller than D 12.<br />

IV. RESULTS<br />

A. Morphology diagram<br />

A diagram displaying the observed morphologies as a<br />

function of the pulling velocity and the orientation of the<br />

smectic-B crystal is shown in Fig. 6.<br />

It can be seen that the sequence of morphologies observed<br />

as a function of V for a fixed value of 0 is the same for all<br />

FIG. 6. Morphology diagram. Measurement points: waves and<br />

fac<strong>et</strong>ons (), fac<strong>et</strong>ons and unstationary fac<strong>et</strong>ed fingers (), unstationary<br />

fac<strong>et</strong>ed fingers (), stationary fac<strong>et</strong>ed fingers (), and<br />

unstable fac<strong>et</strong>s x. Heavy dashed line: Mullins-Sekerka instability<br />

threshold. Ins<strong>et</strong> micrographs: see Fig. 7.<br />

011702-4


DYNAMICS OF A FACETED NEMATIC–SMECTIC-B. .. PHYSICAL REVIEW E 65 011702<br />

FIG. 7. The different growth morphologies observed as a function<br />

of V for 0 25°. a Planar front; b drifting shallow cells; c<br />

drifting fac<strong>et</strong>on stationary mode; d drifting fac<strong>et</strong>ons oscillatory<br />

mode at different stages of their oscillation cycle; see Fig. 18 below;<br />

e nonstationary array of fac<strong>et</strong>ed fingers ; f stationary array<br />

of fac<strong>et</strong>ed fingers.<br />

values of 0 , except for those close to 0° fac<strong>et</strong>s parallel to<br />

the growth front or 90° fac<strong>et</strong>s perpendicular to the growth<br />

front. This generic sequence is illustrated in Fig. 7.<br />

Small-amplitude, nearly sinusoidal traveling waves appear<br />

near the instability threshold Fig. 7b, in accordance<br />

with previous observations in two-sided anisotropic systems<br />

6. Such weakly nonlinear waves are commonly called<br />

‘‘shallow cells.’’<br />

We observed drifting shallow cells in a broad range of V<br />

around the threshold (1 ms 1 V8 ms 1 ). In the<br />

same range of V, we also observed ‘‘fac<strong>et</strong>ons’’ Fig. 7c.<br />

These solitary waves may propagate in a stationary or an<br />

oscillatory way. They appear when the amplitude of the cells<br />

is so large that the tilt angle of the front locally reaches the<br />

value 0 corresponding to the fac<strong>et</strong>s. Most generally, this<br />

occurs under the effect of perturbations due to the nematic<br />

domains. The frequency of creation of fac<strong>et</strong>ons, and thus,<br />

their average number by unit length of the front increases as<br />

V increases. When the average spacing of the fac<strong>et</strong>ons becomes<br />

smaller than their width (200 m), they cease to<br />

behave as non-interacting objects. In fact, they disappear altog<strong>et</strong>her,<br />

giving way to arrays of much narrower objects,<br />

called fac<strong>et</strong>ed fingers Fig. 7e. This occurs at about<br />

8 ms 1 . However, this transition is strongly noise dependent,<br />

and thus, relatively ill defined from an experimental<br />

viewpoint. Shallow cells and fac<strong>et</strong>ons are studied in d<strong>et</strong>ail in<br />

the next section.<br />

The arrays of fac<strong>et</strong>ed fingers, which are observed above<br />

8 ms 1 exhibit a relatively sharp transition from an unstationary<br />

Figs. 7e and 8 to a stationary dynamics as V<br />

increases Figs. 7f and 9; the dispersion appearing in Fig. 6<br />

is mostly due to the aging of the samp<strong>les</strong>. The spatiotemporal<br />

diagrams of the unstationary arrays shown in Fig. 8 reveal<br />

the transitory or local existence of well-defined oscillatory<br />

modes. These modes become more and more apparent as V<br />

increases because the oscillation period T osc is a rapidly decreasing<br />

function of V Fig. 10a. This strongly suggests<br />

the existence of a homogeneous oscillatory bifurcation of the<br />

FIG. 8. Transition from isolated fac<strong>et</strong>ons to fac<strong>et</strong>ed fingers for<br />

0 70° in an aged sample. a V3.1 ms 1 snapshot of the<br />

front; b corresponding spatiotemporal diagram time series of the<br />

intensity distribution along a line located 20 m below the front;<br />

c V6.5 ms 1 ; d corresponding spatiotemporal diagram; e<br />

V13.5 ms 1 . Note that another grain ( 0 73°) appears in the<br />

leftmost part of the figure; f corresponding spatiotemporal diagram.<br />

high-V stationary patterns as V decreases within some narrow<br />

range of spacing.<br />

We now turn to the particular orientations corresponding<br />

to the bounds of the scanned interval of 0 . When 0 90°,<br />

the system is reflection symm<strong>et</strong>ric. Shallow cells no longer<br />

drift, and fac<strong>et</strong>ons cease to exist. The shallow cells break up<br />

into narrow fac<strong>et</strong>ed fingers as V is increased above threshold<br />

Fig. 11a. The widest fac<strong>et</strong>ed fingers, which are the majority<br />

ones, are not reflection symm<strong>et</strong>ric, whereas the narrowest<br />

ones are reflection symm<strong>et</strong>ric. The two opposite but equivalent<br />

directions of symm<strong>et</strong>ry breaking are equally populated.<br />

The resulting arrays were nonstationary even at the highestexplored<br />

values of V Fig. 11b. This is very different from<br />

what was observed by Oswald <strong>et</strong> al. in smectic-A-smectic-B<br />

fronts for a similar orientation of the fac<strong>et</strong> 6. In that system,<br />

because of the existence of forbidden directions, the finger<br />

tips exhibited pointed triangular shapes, and formed stationary<br />

arrays.<br />

When 0 is sufficiently close to zero, the growth front of<br />

smectic-B grains is entirely occupied by a fac<strong>et</strong> at any value<br />

011702-5


T. BÖRZSÖNYI, S. AKAMATSU, AND G. FAIVRE PHYSICAL REVIEW E 65 011702<br />

FIG. 9. Stationary array of fac<strong>et</strong>ed fingers at V13.5 ms 1<br />

and 0 24°. a Snapshot of the front; b spatiotemporal diagram<br />

piling up of skel<strong>et</strong>onized images of the growth front.<br />

of V. This may be considered as a finite-size effect resulting<br />

from the following fact: fac<strong>et</strong>s are always present in the<br />

grooves attached to grain boundaries for whatever values of<br />

0 and V; the stationary size of these fac<strong>et</strong>s is more or <strong>les</strong>s<br />

proportional to 1/(tan 0 ); they thus occupy the whole grain<br />

when 0 is lower than a certain value, which is of about 2°<br />

for a grain size of 500 m. At sufficiently high V, these long<br />

fac<strong>et</strong>s break up through the mechanism illustrated in Fig.<br />

11c. It is not necessary to repeat here the description of this<br />

process, which has been presented by other authors 12. We<br />

simply note that, in our fresh samp<strong>les</strong>, this instability was<br />

observed to result from the occasional collisions of the front<br />

with defects domain walls, dust partic<strong>les</strong> present in the<br />

nematic. In the <strong>les</strong>s pure samp<strong>les</strong>, it was superseded by another<br />

well-known process, namely, the nucleation of crystals<br />

in the undercooled melt ahead of the front 6. Both mechanisms<br />

give rise to more or <strong>les</strong>s permanently cyclic growth<br />

regimes.<br />

B. Near-threshold patterns<br />

1. Drifting shallow cells<br />

Most generally shallow cells appeared in the form of a<br />

noise-induced wave pack<strong>et</strong>. A spontaneous homogeneous<br />

growth of the cells was never observed with certainty. We<br />

FIG. 10. Oscillation period a as a function of V for 0 56°<br />

b as a function of 0 for three values of V. The leftmost point in<br />

a corresponds to an isolated oscillatory fac<strong>et</strong>on.<br />

attribute this fact to the interplay b<strong>et</strong>ween shallow cells and<br />

fac<strong>et</strong>ons see below. At, or below 2 ms 1 , noise-induced<br />

wave pack<strong>et</strong>s systematically disappeared when the source of<br />

noise disappeared, as already mentioned. At higher V, they<br />

evolved as illustrated in Figs. 12 and 13.<br />

A careful analysis of the spatiotemporal diagram of Fig.<br />

12 has shown that i the cells are initially sinusoidal; ii<br />

they grow in amplitude with a uniform amplification rate of<br />

0.002 s 1 ; iii the amp<strong>les</strong>t cells are no longer sinusoidal<br />

at the end of the time sequence, iv the spacing and the<br />

drift velocity V d are uniform in space and constant in time.<br />

V d is thus amplitude independent. This is in keeping with the<br />

idea that this sequence is the initial stage of the usual amplification<br />

process leading from a linearly unstable state to a<br />

stationary weakly nonlinear regime. The final regime was not<br />

observed because the process was interrupted by an external<br />

perturbation giving rise to a fac<strong>et</strong>on.<br />

The traces on the lefthand side of Fig. 13 are the trajectories<br />

of three oscillatory fac<strong>et</strong>ons. These objects are studied<br />

below. For now, the point of interest is that the rearmost<br />

fac<strong>et</strong>on leaves behind a region of the front that is free of<br />

d<strong>et</strong>ectable shallow cells see also Figs. 15 and 18 below.<br />

The cells reappear at 200 m from the fac<strong>et</strong>on, and then<br />

amplify following a process entirely similar to the above<br />

FIG. 11. a Array of<br />

symm<strong>et</strong>ry-broken fac<strong>et</strong>ed fingers<br />

at 0 90° and V10 ms 1 ;<br />

b corresponding spatiotemporal<br />

diagram; c instability of a fac<strong>et</strong><br />

at 0 2° and V10 ms 1 .<br />

011702-6


DYNAMICS OF A FACETED NEMATIC–SMECTIC-B. .. PHYSICAL REVIEW E 65 011702<br />

FIG. 12. Spatiotemporal diagram of a drifting wave pack<strong>et</strong>; V<br />

3.1 ms 1 , 0 25°.<br />

one, except for two points: i in the present case, the amplification<br />

rate (0.02 s 1 ) is much larger than in the preceding<br />

case, since V is higher, and ii a stationary regime of<br />

nonlinear shallow cells is reached. This confirms clearly, although<br />

only semiquantitatively, that the system admits stationary<br />

weakly nonlinear cellular states within a measurable<br />

range of V above V c . These states are m<strong>et</strong>astable with respect<br />

to the formation of fac<strong>et</strong>ons. Also, we note that the<br />

direction of drift of the cells is opposite to that of fac<strong>et</strong>ons.<br />

This is somewhat of a surprise since, in other systems, shallow<br />

cells and fac<strong>et</strong>s have been found to drift in the same<br />

direction 6.<br />

The measured values of V d and are plotted in Fig. 14<br />

as a function of 0 for a given value of V. The data are<br />

compatible with the fact that V d ( 0 ) must go to zero at 0<br />

0° and 90° for symm<strong>et</strong>ry reasons. The maximum is at<br />

about 70°, and corresponds to a relatively large value of<br />

V d /V, indicating that the system is strongly anisotropic even<br />

in the orientation range in which the interface is rough.<br />

We have noted above that V d seems to be independent of<br />

the amplitude of the cells. It is thus legitimate to admit but<br />

not certain that the measured value of V d is the same as in<br />

the linear regime. We have performed a linear stability analysis<br />

of the planar front of a two-sided system taking into account<br />

the anisotropy of the diffusion in the two bulk phases<br />

nematic and smectic-B), and that of the linear kin<strong>et</strong>ic coefficient<br />

the anisotropy of does not come into play in a<br />

linear calculation 4. We have solved the dispersion equation<br />

numerically under various assumptions concerning the<br />

orientation dependences of D, D s , and , which are not<br />

known. Qualitatively, the results may be summed up as follows<br />

27. We find that the observed sign and absolute value<br />

of V d ( 0 ) could be ascribed to diffusion anisotropy only if,<br />

in the smectic-B phase, the impurities diffused much faster<br />

through the smectic layers than parallel to them, which is<br />

very unlikely to be true. Thus, the observed drift of the shallow<br />

cells is most probably due to kin<strong>et</strong>ic anisotropy. In such<br />

FIG. 13. Spatiotemporal diagram. The three traces on the lefthand<br />

side are the trajectories of oscillatory fac<strong>et</strong>ons drifting leftwards.<br />

Note the disappearance of the cells which drift rightwards<br />

in the wake of the rearmost fac<strong>et</strong>on. A temporary exception to this<br />

rule is visible near the end of the recording, when the fac<strong>et</strong>on emits<br />

a pack<strong>et</strong> of three or four cells. This exception is only apparent,<br />

however, since this occurs during a period of time when the fac<strong>et</strong>on<br />

no longer exists it is drifting rightwards. V6.5 ms 1 , 0<br />

55°, recording time 250 s.<br />

a case, the sign of V d is given by d/d 4. In conclusion,<br />

the observed direction of drift of the shallow cells if it<br />

is really the same as in the linear regime indicates that, in<br />

our system, increases as increases. This result poses no<br />

particular problem except for the vicinal domain, in which <br />

is expected to be more or <strong>les</strong>s proportional to the reciprocal<br />

of the step density, and hence, to the reciprocal of 29.<br />

The crossover from the vicinal to the rough domains as <br />

increases should thus manifest itself through a change in the<br />

sign of V d . It is tempting to assume that this crossover cor-<br />

011702-7


T. BÖRZSÖNYI, S. AKAMATSU, AND G. FAIVRE PHYSICAL REVIEW E 65 011702<br />

FIG. 16. Normal growth velocity of fac<strong>et</strong>s belonging to fac<strong>et</strong>ons<br />

or arrays of fac<strong>et</strong>ed fingers as a function of the tilt angle of the fac<strong>et</strong><br />

for the indicated values of the pulling velocity. In the case of oscillatory<br />

fac<strong>et</strong>ons, the minimum value of V n has been plotted. The data<br />

point at 0 2° corresponds to the fac<strong>et</strong> shown in Fig. 11c prior to<br />

its destabilization.<br />

FIG. 14. Drift velocity a and wavelength b of the cells as a<br />

function of the tilt angle of the fac<strong>et</strong> at V6.5 ms 1 .<br />

responds to the zero of V d ( 0 ), which perhaps appears near<br />

12° in Fig. 14a. However, the observation of macroscopic<br />

fac<strong>et</strong>s drifting in the same direction as the shallow cells disproves<br />

this assumption, and indicates that the vicinal domain<br />

is actually very narrow in our system see below.<br />

2. Stationary fac<strong>et</strong>ons<br />

The spatiotemporal diagram of a stationary fac<strong>et</strong>on is<br />

shown in Fig. 15. Clearly, a fac<strong>et</strong>on is a solitary wave consisting<br />

of a macroscopic fac<strong>et</strong> and a broad rounded finger<br />

separated from each other by a very thin liquid groove. The<br />

regularity of the spatiotemporal diagram shows that fac<strong>et</strong>ons,<br />

once formed, are quite stable. In particular, they absorb the<br />

shallow cells that they may encounter ahead of themselves<br />

without being modified, and seem to be insensitive to the<br />

perturbations caused by the nematic domains. The depth of<br />

the fac<strong>et</strong>—i.e., the distance z f b<strong>et</strong>ween the two edges of the<br />

fac<strong>et</strong> along the z axis—is difficult to measure with accuracy<br />

because the lower edge, located near the bottom of the<br />

groove, is generally not resolved. However, it is certain that<br />

z f is in the 3050 m range the difference of temperature<br />

T f b<strong>et</strong>ween the two edges is thus in the range 0.15–<br />

0.25 K, and decreases as 0 increases. The upper edge of the<br />

fac<strong>et</strong> corresponds to a small pointed maximum of the front<br />

shape, but it is not possible to decide wh<strong>et</strong>her, or not, this<br />

edge is sharp on a molecular scale. The width of the rounded<br />

finger—i.e., the extension of the deformed region of the front<br />

behind the finger tip—is of about 200 m. As mentioned,<br />

shallow cells do not develop in this region of the front. The<br />

trajectory of the fac<strong>et</strong>on makes a small angle with the direction<br />

of the macroscopic fac<strong>et</strong>, indicating that the normal<br />

growth rate of the fac<strong>et</strong> is small but finite. Thus, the fac<strong>et</strong> is<br />

not blocked, and the question arises as to its microscopic<br />

growth mechanisms.<br />

Figure 16 displays a large number of values of the normal<br />

velocity of fac<strong>et</strong>s V n measured in isolated fac<strong>et</strong>ons as well as<br />

in arrays of fac<strong>et</strong>ed fingers for various values of V and 0 .In<br />

spite of a large dispersion of the data, it is clear that V n is<br />

essentially a nonzero quantity that decreases as 0 increases,<br />

and increases as V increases. The regularity of the stationary<br />

fac<strong>et</strong>ons or arrays see Figs. 15 and 9, and the fact that V n is<br />

very close to zero when 0 is large allow us to exclude screw<br />

dislocation growth as the dominant mechanism. Moreover,<br />

the fact that both V n and z f are decreasing functions of 0<br />

suggests that V n is essentially d<strong>et</strong>ermined by events occurring<br />

near the lower edge of the fac<strong>et</strong>. One may imagine either<br />

that surface nucleation takes place at a relatively high rate at<br />

this point, or that the fac<strong>et</strong> is supplied with steps coming<br />

from the bottom of the groove where the interface is necessarily<br />

rough. In both cases, V n would be very sensitive to the<br />

d<strong>et</strong>ails of the conformation of the interface in this region.<br />

These d<strong>et</strong>ails may depend on the treatment of the glass<br />

plates, which could explain the dispersion b<strong>et</strong>ween values<br />

measured in different samp<strong>les</strong>.<br />

FIG. 15. Stationary fac<strong>et</strong>on. 0 25°, V6.5 ms 1 . a<br />

Snapshot of the front. The faint dark line appearing in the solid in<br />

the continuation of the fac<strong>et</strong> is a thin liquid groove; see Fig. 17. b<br />

Spatiotemporal diagram. The normal growth rate of the fac<strong>et</strong> is<br />

V n 0.9 ms 1 .<br />

3. Oscillating fac<strong>et</strong>ons<br />

Figure 17 shows a process of formation of fac<strong>et</strong>ons in<br />

response to a perturbation. Macroscopic fac<strong>et</strong>s progressively<br />

develop on one side of the shallow cells as the amplitude of<br />

the latter increases. These fac<strong>et</strong>s first drift with the same<br />

velocity as the shallow cells, and then change their direction<br />

of drift. This change is not accompanied by any modification<br />

in the orientation of the fac<strong>et</strong>s within experimental uncer-<br />

011702-8


DYNAMICS OF A FACETED NEMATIC–SMECTIC-B. .. PHYSICAL REVIEW E 65 011702<br />

often adopt an oscillatory mode of propagation Fig. 18.<br />

Obviously, this oscillation consists of a more or <strong>les</strong>s ample<br />

cycle b<strong>et</strong>ween the aforementioned rapid and slow regimes.<br />

The conditions under which fac<strong>et</strong>ons are stationary, or oscillatory,<br />

could not be d<strong>et</strong>ermined. In fact, stationary fac<strong>et</strong>ons<br />

were observed much <strong>les</strong>s frequently than, and always in coexistence<br />

with oscillating fac<strong>et</strong>ons. Moreover, some oscillating<br />

fac<strong>et</strong>ons were regular Fig. 18, but most of them were<br />

irregular Figs. 13 or 19. It is possible that the system intrinsically<br />

admits stationary, periodic, and more or <strong>les</strong>s, chaotic<br />

fac<strong>et</strong>ons. However, the following explanation is also<br />

possible.<br />

A careful inspection of Fig. 18 reveals that the transition<br />

of the oscillating fac<strong>et</strong>ons from a slow to a rapid regime<br />

corresponds to a sudden pinching off of the liquid groove,<br />

whereas the reverse transition from a rapid to a slow regime<br />

consists of a progressive deepening of the groove. If we focus<br />

on the sole groove, this behavior is strongly reminiscent<br />

of the periodic pinching off called cusp instability of the<br />

intercell grooves in nonfac<strong>et</strong>ed cellular fronts 30. This in-<br />

FIG. 17. Fac<strong>et</strong>ons appearing in response to a perturbation. Spatiotemporal<br />

diagram. V6.5 ms 1 , 0 25°, recording time: 60<br />

s. Note the opposite signs of the drift velocities of the cells and the<br />

fac<strong>et</strong>s.<br />

tainty (0.5°). Thus, the same macroscopic fac<strong>et</strong> may be in<br />

two different microscopic states, or growth regimes. One of<br />

these the ‘‘slow’’ regime is that of the stationary state, discussed<br />

in the preceding section, while the other the ‘‘rapid’’<br />

regime corresponds to a rough interface. As announced, we<br />

are thus led to assume that the crossover from vicinal to<br />

rough interfaces occurs at values of lower than 0.5° in<br />

our system. This is indeed surprising since this disorientation<br />

corresponds to a very low density of steps <strong>les</strong>s than 1 per<br />

m), but not impossible. We also note that the persistence of<br />

a macroscopic fac<strong>et</strong> while the interface is rough on a microscopic<br />

scale is explainable by the sole singularity of the <br />

plot 16.<br />

The existence of two different growth regimes of a macroscopic<br />

fac<strong>et</strong> is confirmed by the fact that fac<strong>et</strong>ons most<br />

FIG. 18. Oscillating fac<strong>et</strong>on. 0 42°, V6.5 ms 1 . a<br />

Snapshots of the front at different stages of an oscillation period. b<br />

Spatiotemporal diagram.<br />

011702-9


T. BÖRZSÖNYI, S. AKAMATSU, AND G. FAIVRE PHYSICAL REVIEW E 65 011702<br />

FIG. 19. Spatiotemporal diagram showing shallow cells, oscillating<br />

fac<strong>et</strong>ed solitary waves, and microfac<strong>et</strong>s arrow, 0 36°, V<br />

3.1 ms 1 , and G25 K cm 1 .<br />

stability, we recall, is most probably of a capillary origin<br />

Rayleigh instability 31, and very sensitive to the lattice<br />

defects that, in the nonfac<strong>et</strong>ed systems, are often attached to<br />

the groove—in fact, the grooves to which subboundaries<br />

low-angle grain boundaries are attached are not subject to<br />

the cusp instability 32. If, by analogy, we assume that the<br />

intercell groove of fac<strong>et</strong>ons, similar to that of nonfac<strong>et</strong>ed<br />

cells, is intrinsically subject to an oscillatory Rayleigh instability,<br />

we are led to the conclusion that the transition of the<br />

fac<strong>et</strong> from a slow to a rapid regime is a secondary effect due<br />

to changes occurring in the configuration of the interface<br />

near the lower edge of the fac<strong>et</strong>. The presence of lattice<br />

defects e.g., sub-boundaries emerging into the liquid at the<br />

bottom of the groove may hinder these changes, suppressing<br />

the oscillation. This would explain that fac<strong>et</strong>ons are much<br />

more often oscillatory than stationary.<br />

4. Lattice defects<br />

Some lattice defects mostly, grain boundaries may be<br />

d<strong>et</strong>ected with the optical microscope thanks to the fact that<br />

they create macroscopic depressions grooves of the growth<br />

front around the point at which they emerge into the liquid.<br />

In our system, these grooves must be partly fac<strong>et</strong>ed during<br />

solidification. We have lowered the applied thermal gradient<br />

in some experiments in order to facilitate the observation of<br />

such grooves. This allowed us to reveal that the growth front<br />

of our system is often swept by very small fac<strong>et</strong>s, called<br />

microfac<strong>et</strong>s, certainly attached to lattice defects emerging<br />

into the liquid.<br />

We observed several types of microfac<strong>et</strong>s, corresponding<br />

probably to different types of lattice defects. The microfac<strong>et</strong>s<br />

of the type shown in Fig. 19 were relatively easy to identify<br />

because they travel at a perfectly constant velocity, catching<br />

up, and running through all the other structures of the front,<br />

in particular, fac<strong>et</strong>ons. Their drift velocity has thus most<br />

probably the maximum possible value, i.e., the value corresponding<br />

to totally blocked fac<strong>et</strong>s. They must be attached to<br />

lattice defects—stacking faults, or twist subboundaries—<br />

strongly locked onto the lamella plane of the smectic. However,<br />

these microfac<strong>et</strong>s seem to have but little effect on the<br />

dynamics of the front. They indeed provoke an instantaneous<br />

slowing down of the macroscopic ‘‘rapid’’ fac<strong>et</strong>s when they<br />

collide with them see Fig. 19, but do not trigger a durable<br />

transition to the slow regime. So this observation, whatever<br />

its intrinsic interest may be, does not cast light on the question<br />

of the possible role played by lattice defects in the dynamics<br />

of the fac<strong>et</strong>ons.<br />

V. DISCUSSION<br />

We have shown that the directional solidification of a<br />

nematic–smectic-B front in the planar configuration gives<br />

rise to a wealth of interesting nonlinear phenomena, the most<br />

striking of which are the stationary or oscillatory ‘‘fac<strong>et</strong>ons’’<br />

encountered in the vicinity of the Mullins-Sekerka threshold.<br />

These observations raise numerous unsolved problems concerning<br />

the microscopic growth mechanisms of the fac<strong>et</strong>s, as<br />

well as the nonlinear dynamics of the observed macroscopic<br />

patterns. An important question is wh<strong>et</strong>her these phenomena<br />

are specific of the nematic–smectic-B fronts, or are of frequent<br />

occurrence in partly fac<strong>et</strong>ed fronts. In order to clarify<br />

this point, we are currently searching for similar phenomena<br />

in more conventional, partly fac<strong>et</strong>ed solidification fronts.<br />

Also, numerical simulations based on a phase-field m<strong>et</strong>hod<br />

are in progress in order to test the consistency of the numerous<br />

conjectures that we have been led to make in order to<br />

explain the peculiar dynamical features of the fac<strong>et</strong>ons.<br />

ACKNOWLEDGMENTS<br />

The authors wish to thank A-M. Levelut for the help in<br />

characterizing CCH4 with x-ray diffraction, T. Tóth-Katona<br />

and Á. Buka for many useful discussions, and A. Fleury and<br />

C. Picard for their technical assistance. We are also grateful<br />

to MERCK Darmstadt for kindly providing us with CCH4.<br />

T.B. would like to thank the European Commission for financial<br />

support.<br />

1 Solids Far from Equilibrium, edited by C. Godrèche Cambridge<br />

University Press, Cambridge, England, 1992.<br />

2 M. Cross and P. Hohenberg, Rev. Mod. Phys. 65, 851 1993.<br />

3 J. Langer. Rev. Mod. Phys. 52, 11980.<br />

4 S.R. Coriell and R.F. Sekerka, J. Cryst. Growth 34, 157 1976.<br />

5 P. Kopczynski, W.-J. Rappel, and A. Karma, Phys. Rev. L<strong>et</strong>t.<br />

77, 3387 1996.<br />

6 F. Melo and P. Oswald, Phys. Rev. L<strong>et</strong>t. 64, 1381 1990.<br />

7 S. Akamatsu, G. Faivre, and T. Ihle, Phys. Rev. E 51, 4751<br />

1995.<br />

011702-10


DYNAMICS OF A FACETED NEMATIC–SMECTIC-B. .. PHYSICAL REVIEW E 65 011702<br />

8 S. Akamatsu and G. Faivre, Phys. Rev. E 58, 3302 1998.<br />

9 D.K. Shangguan and J.D. Hunt, M<strong>et</strong>all. Trans. A 22A, 941<br />

1991.<br />

10 L.M. Fabi<strong>et</strong>ti and R. Trivedi, J. Cryst. Growth 182, 185 1997.<br />

11 P. Oswald, F. Melo, and C. Germain, J. Phys. France 50,<br />

3527 1989.<br />

12 F. Melo and P. Oswald, J. Phys. II 1, 353 1991.<br />

13 W.K. Burton, N. Cabrera, and F.C. Frank, Philos. Trans. R.<br />

Soc. London 243, 299 1951.<br />

14 Á. Buka, T. Tóth-Katona, and L. Kramer, Phys. Rev. E 51, 571<br />

1995.<br />

15 T. Tóth-Katona, T. Börzsönyi, Z. Váradi, J. Szabon, Á. Buka,<br />

R. González-Cinca, L. Ramirez-Piscina, J. Casademunt, and A.<br />

Hernández-Machado, Phys. Rev. E 54, 1574 1996.<br />

16 R. González-Cinca, L. Ramirez-Piscina, J. Casademunt, A.<br />

Hernández-Machado, T. Tóth-Katona, T. Börzsönyi, and Á.<br />

Buka, Physica D 99, 359 1996.<br />

17 A.A. Chernov, Contemp. Phys. 30, 251 1989.<br />

18 T. Tóth-Katona, N. Éber, and Á. Buka, Mol. Cryst. Liq. Cryst.<br />

Sci. Technol., Sect. A 328, 467 1999.<br />

19 A.M. Levelut unpublished.<br />

20 R. Brownsey and A. Leadb<strong>et</strong>ter, J. Phys. France L<strong>et</strong>t. 42, 135<br />

1981.<br />

21 P. Damman, M. Dosière, M. Brunel, and J.C. Wittmann, J. Am.<br />

Chem. Soc. 119, 4633 1997.<br />

22 T. Börzsönyi, T. Tóth-Katona, Á. Buka, and L. Gránásy, Phys.<br />

Rev. E 62, 7817 2000.<br />

23 J.C. La Combe, M.B. Koss, L.A. Tennenhouse, E.A. Winsa,<br />

and M.E. Glicksman, J. Cryst. Growth 194, 143 1998.<br />

24 C. Herring, Phys. Rev. 82, 871951.<br />

25 D.R. Ulhmann, in Advances in Nucleation and Crystallization<br />

in Glasses American Ceramic Soci<strong>et</strong>y, Columbus, OH, 1971.<br />

26 D.R. Ulhmann, in Nucleation and Crystallization in Glasses<br />

American Ceramic Soci<strong>et</strong>y, Columbus, OH, 1982.<br />

27 T. Börzsönyi, S. Akamatsu, and G. Faivre unpublished.<br />

28 J.A. Warren and J.S. Langer, Phys. Rev. E 47, 2702 1993.<br />

29 A.A. Chernov, S.R. Coriell, and B.T. Murray, J. Cryst. Growth<br />

132, 405 1993.<br />

30 P. Kurowski, S. de Cheveigné, G. Faivre, and C. Guthmann, J.<br />

Phys. France 50, 3007 1989.<br />

31 K. Brattkus, J. Phys. France 50, 2999 1989.<br />

32 S. Bottin-Rousseau, S. Akamatsu, and G. Faivre unpublished.<br />

011702-11


VOLUME 93, NUMBER 17<br />

PHYSICAL REVIEW LETTERS week ending<br />

22 OCTOBER 2004<br />

Experimental Evidence for a Zigzag Bifurcation in Bulk Lamellar Eutectic Growth<br />

Silvère Akamatsu, Sabine Bottin-Rousseau,* and Gabriel Faivre<br />

Groupe de Physique des Solides, CNRS UMR 7588, Universités <strong>Pierre</strong>-<strong>et</strong>-Marie-Curie <strong>et</strong> Denis-Diderot,<br />

Campus Boucicaut, 140 rue de Lourmel, 75015 Paris, France<br />

(Received 13 July 2004; published 20 October 2004)<br />

We present real-time observations of the directional-solidification patterns of a transparent nonfac<strong>et</strong>ed<br />

eutectic alloy (CBr 4 -C 2 Cl 6 ) in bulk samp<strong>les</strong>. The growth front of the two-phase solid is<br />

observed from the top through the liquid and the glass wall of the container with a long-distance<br />

microscope. We show that, in near-eutectic CBr 4 -C 2 Cl 6 alloys, the upper stability limit of the stationary<br />

lamellar patterns is due to a zigzag bifurcation, which occurs at an interlamellar spacing of about<br />

0:85 m , where m is the minimum-undercooling spacing. The zigzag patterns undergo a lamella<br />

breakup instability leading to the creation of new lamellae at about 1:1 m . On the other hand, the<br />

lower stability limit of the stationary patterns is due to the same instability as in thin samp<strong>les</strong>, namely,<br />

a lamella termination instability that occurs at about 0:7 m .<br />

DOI: 10.1103/PhysRevL<strong>et</strong>t.93.175701<br />

PACS numbers: 64.70.Dv, 05.70.Ln, 81.30.Fb<br />

Nonlinear pattern formation in solidification is a question<br />

of general interest in the physics of spatially extended<br />

out-of-equilibrium systems [1]. Solidification patterns or<br />

‘‘microstructures’’ are also the subject of active researches<br />

in materials sciences [2]. One of the most frequently<br />

studied examp<strong>les</strong> is that of directionally solidified<br />

(solidified at a fixed velocity V in an applied unidirectional<br />

thermal gradient), nonfac<strong>et</strong>ed binary eutectics<br />

(two-component alloys presenting a miscibility gap in<br />

the solid state). For thermodynamic reasons, the solid<br />

that grows in this case is made of two phases (called<br />

and ) of different concentrations and crystal structures.<br />

The exchange of solute b<strong>et</strong>ween the two phases during<br />

growth occurs by diffusion through the liquid. The proportion<br />

of the two phases in the solid is fixed by mass<br />

conservation, but their spatiotemporal arrangement along<br />

the front is a problem of nonlinear pattern formation<br />

[3,4]. It has been known for a long time that, within a<br />

certain range of values of the alloy concentration and V,<br />

the selected pattern is usually ‘‘lamellar’’ (periodic in<br />

one direction) and stationary (Fig. 1). This type of pattern<br />

(called ‘‘symm<strong>et</strong>rical’’ by contradistinction with the<br />

symm<strong>et</strong>ry-broken patterns [5,6]) has a wide existence<br />

range as a function of the spacing at fixed V [7].<br />

However, the nature of the predominant instability modes<br />

and the location of the instability thresholds on the<br />

scale are still largely open questions.<br />

Most previous investigations of the stability of lamellar<br />

eutectic patterns were restricted to one-dimensional<br />

(1D) eutectic fronts. Quasi-1D fronts were obtained experimentally<br />

by using ‘‘thin’’ samp<strong>les</strong>, i.e., samp<strong>les</strong> of a<br />

thickness comparable to (on the order of 10 m for V in<br />

the ms 1 range), in which the lamella plane is constrained<br />

to remain normal to the sample walls (but can<br />

undergo rotations about the normal to the sample plane).<br />

A transparent nonfac<strong>et</strong>ed eutectic alloy (CBr 4 -C 2 Cl 6 )<br />

was used, and the front was observed in real time through<br />

the wall of a glass container with a conventional optical<br />

microscope and a direction of observation normal to the<br />

wall [6]. These studies, in conjunction with numerical<br />

simulations, have led to the compl<strong>et</strong>e quantitative d<strong>et</strong>ermination<br />

of the stability diagram of 1D eutectic fronts of<br />

CBr 4 -C 2 Cl 6 as a function of , V, and the alloy concentration<br />

[8–10]. In fact, this diagram obeys the so-called<br />

V 1=2 similarity law, meaning that the actual control<br />

param<strong>et</strong>er of the instabilities is V 1=2 , or, equivalently,<br />

= m , where m is the so-called minimum-undercooling<br />

spacing, which is system dependent, and sca<strong>les</strong> with V as<br />

V 1=2 [7]. It was also shown numerically that the found<br />

qualitative features of the stability diagram were common<br />

to all nonfac<strong>et</strong>ed binary eutectics.<br />

When we turn to lamellar eutectics in bulk samp<strong>les</strong>, we<br />

are faced with a new problem. In bulk samp<strong>les</strong>, eutectic<br />

lamellae are not aligned along the normal to the sample<br />

walls—at least not quasi-instantaneously—contrary to<br />

what occurs in thin samp<strong>les</strong>. They can rotate about the<br />

growth axis, and also break up, giving rise to topological<br />

FIG. 1. (a) Sk<strong>et</strong>ch of a stationary lamellar eutectic pattern;<br />

; : eutectic crystal phases; z: axis of the thermal gradient,<br />

and normal to the growth front. (b) Sk<strong>et</strong>ch of the experimental<br />

s<strong>et</strong>up; V: pulling velocity; 0: tilt angle of the direction of<br />

observation to the horizontal.<br />

175701-1 0031-9007=04=93(17)=175701(4)$22.50 © 2004 The American Physical Soci<strong>et</strong>y 175701-1


VOLUME 93, NUMBER 17<br />

PHYSICAL REVIEW LETTERS week ending<br />

22 OCTOBER 2004<br />

defects such as lamella terminations and line defects<br />

(domain boundaries and phase jumps). Moreover, numerous<br />

experiments have shown that, in bulk samp<strong>les</strong>, lamellar<br />

patterns can disappear and be replaced by stationary<br />

patterns consisting of hexagonal arrays of inclusions of<br />

one phase embedded in the other (rodlike patterns) [7].<br />

Thus, from the viewpoint of the phenomenological theories,<br />

2D nonfac<strong>et</strong>ed eutectic fronts can be characterized<br />

by the fact that they have, at fixed V, two types (lamellar<br />

and rodlike) of stationary patterns, which exist over<br />

finite-width spacing ranges, and are degenerate in orientation.<br />

These properties are shared by other 2D patternforming<br />

systems, in particular, Rayleigh-Bénard convection<br />

[11,12]. It is thus natural to try to find out if 2D<br />

lamellar eutectics exhibit the same primary mode of<br />

instability as Rayleigh-Bénard convection rolls do,<br />

namely, the zigzag instability. This question is currently<br />

being studied numerically [13,14] but has not y<strong>et</strong> been<br />

investigated experimentally.<br />

In this L<strong>et</strong>ter, we present real-time observations of 2D<br />

lamellar eutectic fronts in bulk samp<strong>les</strong> of CBr 4 -C 2 Cl 6 .<br />

The experimental s<strong>et</strong>up is sk<strong>et</strong>ched in Fig. 1. We used<br />

300- or 350- m-thick glass samp<strong>les</strong> filled with a slightly<br />

hypereutectic, well-purified alloy under a controlled atmosphere<br />

[15]. A thermal gradient of 100 20 K cm 1<br />

was installed b<strong>et</strong>ween two isolated, temperature regulated<br />

copper blocks. The samp<strong>les</strong> were equipped with a<br />

grain selector in order to grow single, or, at least, large,<br />

weakly anisotropic eutectic grains [16,17]. These were<br />

formed by using a procedure previously used in thin<br />

samp<strong>les</strong> and described in Ref. [18]. The observations<br />

were performed in single-grain regions containing a<br />

hundred pairs of lamellae, typically. We checked that<br />

the isotherms were planar and perpendicular to z (to<br />

within 1 ) when the sample was at rest. However, a small<br />

(< 6 ) tilt of the isotherms about the x axis (thermal bias)<br />

som<strong>et</strong>imes occurred during solidification. The presence of<br />

convection flows with velocities of a few 10 ms 1 was<br />

also revealed by a lateral drift of small inert dust partic<strong>les</strong><br />

floating in the liquid close to the front, though the system<br />

is, in principle, stable against thermosolutal convection in<br />

hypereutectic CBr 4 -C 2 Cl 6 alloys—the liquid close to the<br />

front is enriched in CBr 4 , which is denser than C 2 Cl 6 .We<br />

did not d<strong>et</strong>ect any perturbation due to these flows in the<br />

regions of the samp<strong>les</strong> in which the observations were<br />

performed.<br />

The front was observed from the top through the liquid<br />

and a lateral glass plate of the container. The contrast<br />

arises from the differences of optical index b<strong>et</strong>ween the<br />

three transparent phases. A major difficulty was to separate<br />

the image of the growth front from that of the underlying<br />

two-phase solid. We solved this problem by<br />

choosing a dark-field m<strong>et</strong>hod, in which the directions of<br />

observation and lighting are both oblique, but different<br />

from each other. We used a long-distance microscope<br />

(Questar QM100) with large working distance (15 cm)<br />

and depth of field (’200 m). The light source was made<br />

of a halogen lamp and a linear bundle of optic fibers. The<br />

images were captured with a video camera and then<br />

digitized and filtered numerically for contrast enhancement.<br />

The reduction of the image in the y direction due to<br />

the oblique direction of observation was also corrected<br />

numerically. We fixed the tilt angle of the direction of<br />

observation at 50 , which corresponds approximately to<br />

the minimum value (2.85) of the image reduction factor<br />

along the y axis. [The existence of a minimum is due to<br />

the refraction of the light at the liquid-(glass)-air interface.]<br />

A corrected image of a quasistationary lamellar<br />

pattern is shown in Fig. 2. A sharp, almost uniform,<br />

optical contrast b<strong>et</strong>ween (bright) and (dark) lamellae<br />

is obtained. The shape of the front on a scale smaller than<br />

is not resolved, but this was not necessary for our<br />

purpose.<br />

We have performed solidification runs over long periods<br />

of time (several hours) for various values of the alloy<br />

concentration and V. Most generally, the pattern obtained<br />

at the end of the initial transient (about 10 min after the<br />

ons<strong>et</strong>) was quite complex and showed no preferred orientation<br />

of the lamella plane, except near the sample<br />

walls, where the no-flux condition forces the lamellae to<br />

be normal to the walls (Fig. 3). In general, a pronounced<br />

alignment normal to the sample walls was obtained after<br />

a long time through a progressive elimination of the<br />

topological defects and a slow propagation of the wall<br />

effect across the sample. The dynamics of this relaxation<br />

toward a stationary state will be reported elsewhere. In<br />

most samp<strong>les</strong>, the relaxation process led to a pattern that<br />

was essentially a symm<strong>et</strong>rical one, despite the persistence<br />

of a few undulations and lamella terminations that were<br />

probably due to imperfections of the experimental s<strong>et</strong>up,<br />

such as thermal bias, long-range concentration gradients<br />

in the liquid, convection flows, and grain boundaries<br />

(Fig. 2). In a few samp<strong>les</strong>, however, the final stationary<br />

pattern was clearly of the zigzag type [Fig. 4(a)]. This<br />

FIG. 2. Top view of a symm<strong>et</strong>ric lamellar eutectic pattern in a<br />

350- m-thick sample of a slightly hypereutectic CBr 4 -C 2 Cl 6<br />

alloy. V 0:37 ms 1 . The sample was pulled at V<br />

0:5 ms 1 for 4.5 h and then at 0:37 ms 1 for 1.5 h.<br />

Bright and dark stripes are and lamellae, respectively.<br />

Horizontal width of the image: 860 m. Average value of :<br />

0:82 m . Arrows: lamella terminations.<br />

175701-2 175701-2


VOLUME 93, NUMBER 17<br />

PHYSICAL REVIEW LETTERS week ending<br />

22 OCTOBER 2004<br />

FIG. 3. View of a pattern after 10 min at V 0:5 ms 1 ,and<br />

35 min at V 1 ms 1 in a 300- m-thick sample. Lamellar<br />

domains are separated by disordered regions containing topological<br />

defects, and rods. Horizontal width 440 m.<br />

was due to the fact that the average spacing of the system<br />

was larger than usual in these experiments (see below),<br />

but the reason of this fact itself (the larger average spacing)<br />

was unclear, except that it occurred only in samp<strong>les</strong><br />

in which the experimental imperfections happened to be<br />

particularly weak.<br />

The transition from symm<strong>et</strong>rical to zigzag patterns<br />

was studied as follows. An almost uniform zigzag pattern<br />

was obtained in a large (about 1-mm wide) eutectic grain<br />

after a 4 h pulling at V 0:5 ms 1 . It was then submitted<br />

to a sequence of four V jumps separated by 30 min<br />

pullings at constant V. We first decreased V to 0.39 and<br />

0:3 ms 1 and then reincreased it to 0.39 and<br />

0:5 ms 1 . The first three steps of the sequence are<br />

illustrated in Fig. 4. We extracted the local values of ,<br />

the amplitude A, and the wavelength L of the zigzag<br />

modulation by fitting the skel<strong>et</strong>onized image of the lamellae<br />

with a sine function over some space periods in<br />

different regions of the micrographs. We assumed that the<br />

control param<strong>et</strong>er of the instability is = m , as it is for the<br />

1D instabilities. The measured values of A and L are<br />

plotted as a function of = m in Fig. 5. Despite the<br />

dispersion of the data, it is clear that there exists a<br />

threshold c , located b<strong>et</strong>ween 0:85 m and 0:95 m , below<br />

which no zigzag pattern was observed, and above which A<br />

increased as = m increased. This is a clear sign of a<br />

bifurcation, although the character (supercritical or<br />

slightly subcritical) of this bifurcation could not be d<strong>et</strong>ermined.<br />

The existence of a region without zigzags at the<br />

lowest values of V in Fig. 4 does not necessarily mean<br />

that the bifurcation was subcritical. The zigzag pattern<br />

actually exhibited a slow global (upward) drift along the<br />

y axis, at a velocity v d of about 0:05 ms 1 for V<br />

0:5 ms 1 . Most probably, this was not an intrinsic<br />

property of the pattern, but the consequence of a thermal<br />

bias, which was comparable to tan 1 v d =V 5:7 in<br />

this experiment. This external forcing can explain the<br />

persistent absence of zigzags up to a certain distance<br />

(which increases as c decreases) from the colder<br />

wall, as observed in Figs. 4(b) and 4(c). Finally, Fig. 5(b)<br />

shows that L steeply increased when V was diminished<br />

and redecreased with some hysteresis as V was switched<br />

back to its initial value. This is a characteristic behavior<br />

of the zigzag instability, as studied, for instance, in<br />

Rayleigh-Bénard convection, which originates from its<br />

‘‘diffusive’’ character, i.e., from the fact that its amplification<br />

coefficient ! and wave vector k are related by !<br />

D k 2 at small k, where c = m is the dis-<br />

4<br />

a)<br />

A (µm)<br />

L (µm)<br />

2<br />

0<br />

80 b)<br />

60<br />

40<br />

20<br />

0.7 0.8 0.9<br />

λ/λ m<br />

1 1.1<br />

FIG. 4. Top views of zigzag patterns (a) V 0:5 ms 1 .<br />

Solidification time: 4 h; (b) V 0:39 ms 1 . Solidification<br />

time: 30 min; (c) V 0:3 ms 1 ; solidification time: 30 min.<br />

Sample thickness 300 m. Horizontal width 440 m.<br />

FIG. 5. (a) Amplitude A and (b) wavelength L of the zigzags<br />

vs the reduced spacing = m . Filled (open) symbols: data<br />

obtained by decreasing (increasing) V. The error bars represent<br />

the experimental dispersion.<br />

175701-3 175701-3


VOLUME 93, NUMBER 17<br />

PHYSICAL REVIEW LETTERS week ending<br />

22 OCTOBER 2004<br />

tance from the instability threshold, and D (phase<br />

diffusion coefficient) changes sign at 0 [12]. In conclusion,<br />

our observations prove the existence of a zigzag<br />

bifurcation, the threshold of which corresponds to the<br />

upper stability bound of the symm<strong>et</strong>ric lamellar patterns,<br />

in bulk near-eutectic CBr 4 -C 2 Cl 6 .<br />

We have also studied the lower (upper) stability bounds<br />

of the symm<strong>et</strong>rical (zigzag) patterns. The symm<strong>et</strong>rical<br />

patterns were unstable against lamella elimination below<br />

about 0:7 m —as they are in thin samp<strong>les</strong> [10]—while<br />

the zigzag patterns were unstable against a lamella<br />

breakup instability above about 1:1 m . The lamella<br />

breakups generally evolved into the creation of new lamellae,<br />

which reduced the spacing.We never observed any<br />

lamellar pattern with a value of = m lying outside the<br />

range 0.7–1.1, even in small domains. It is thus clear that<br />

the morphology diagram of nonfac<strong>et</strong>ed eutectics in bulk<br />

samp<strong>les</strong> is profoundly different from what it is in thin<br />

samp<strong>les</strong>. In thin samp<strong>les</strong>, the zigzag and the lamellar<br />

breakup are usually blocked allowing a series of<br />

symm<strong>et</strong>ry-breaking (tilt and oscillations) bifurcations<br />

to occur at spacings larger than 1:1 m [8,9]. In bulk<br />

samp<strong>les</strong>, the occurrence of the lamella breakup at relatively<br />

small spacings eliminates these bifurcations and<br />

leads to the formation of topological defects. We have<br />

observed that, sufficiently far above the threshold, the<br />

zigzag patterns always exhibited line defects corresponding<br />

to a phase jump of the pattern. Such a defect extending<br />

over about 10 can be seen at the bottom of Fig. 4(a).<br />

In this particular case, the defect disappeared when we<br />

decreased V, indicating that it was not linked to a subboundary<br />

in the crystals but was an intrinsic nonlinear<br />

property of the zigzag pattern. This may be a clue to the<br />

origin of the ‘‘line faults’’ or ‘‘mismatch surfaces’’ that<br />

always appear in large density in the cross- sections of<br />

bulk m<strong>et</strong>allic lamellar eutectics [19]. Our observations<br />

suggest that such ‘‘faulty’’ patterns might be steady at<br />

spacings much larger than the zigzag instability threshold.<br />

This would be compatible with the fact that the<br />

distributions measured in bulk m<strong>et</strong>allic eutectics generally<br />

have a larger width (>20%) [20] than that of the<br />

stability range of the symm<strong>et</strong>rical lamellar pattern<br />

(’10%).<br />

In conclusion, the experimental s<strong>et</strong>up that we have<br />

presented here is a powerful tool for the study of eutectic<br />

growth patterns in bulk samp<strong>les</strong>. It has allowed us to<br />

show that a zigzag bifurcation is the primary instability<br />

of a bulk lamellar eutectic (CBr 4 -C 2 Cl 6 ) near the eutectic<br />

concentration and that this bifurcation occurs at an unexpectedly<br />

low (0:85 m ) spacing value. A similar result<br />

was recently found numerically using a phase-field<br />

m<strong>et</strong>hod and a one-sided model (no diffusion in the solid)<br />

by Parisi and Plapp [13]. The investigation of other aspects<br />

of 2D eutectic growth, such as the dynamincs of<br />

line defects, the lamella-rod transition, and the effects of<br />

the convection flows, is in progress.<br />

We thank M. Plapp and A. Parisi for many fruitful<br />

discussions and for communicating their unpublished<br />

numerical results. We gratefully acknowledge the technical<br />

help of C. Perez and C. Picard. This research was<br />

financially supported by the Centre National d’Etudes<br />

Spatia<strong>les</strong>, France.<br />

*Electronic address: bottin@gps.jussieu.fr<br />

[1] M. C. Cross and P. C. Hohenberg, Rev. Mod. Phys. 65, 851<br />

(1993).<br />

[2] Proceedings of Solidification Processes and<br />

Microstructures: A Symposium in Honor of W. Kurz,<br />

edited by M. Rappaz, C. Beckermann, and R. Trivedi<br />

(TMS, Warrendale, 2004).<br />

[3] J. S. Langer, Phys. Rev. L<strong>et</strong>t. 44, 1023 (1980); V. Datye<br />

and J. S. Langer, Phys. Rev. B 24, 4155 (1981).<br />

[4] A. Karma, Phys. Rev. L<strong>et</strong>t. 59, 71 (1987).<br />

[5] K. Kassner and C. Misbah, Phys. Rev. A 44, 6513 (1991).<br />

[6] G. Faivre and J. Mergy, Phys. Rev. A 45, 7320 (1992).<br />

[7] K. A. Jackson and J. D. Hunt, Trans. M<strong>et</strong>all. Soc. AIME<br />

236, 1129 (1966).<br />

[8] A. Karma and A. Sarkissian, M<strong>et</strong>all. Trans. A 27, 635<br />

(1996).<br />

[9] M. Ginibre, S. Akamatsu, and G. Faivre, Phys. Rev. E 56,<br />

780 (1997).<br />

[10] S. Akamatsu, M. Plapp, G. Faivre, and A. Karma, Phys.<br />

Rev. E 66, 030501(R) (2002); M<strong>et</strong>all. Mater. Trans. A 35,<br />

1815 (2004).<br />

[11] F. H. Busse, Rep. Prog. Phys. 41, 1929 (1978)<br />

[12] P. Manneville, Dissipatives Structures and Weak<br />

Turbulence (Academic Press, Boston, 1990).<br />

[13] A. Parisi and M. Plapp (unpublished).<br />

[14] A. Karma and M. Plapp, JOM 56, No. 4, 28 (2004).<br />

[15] J. Mergy, G. Faivre, C. Guthmann, and R. Mell<strong>et</strong>, J.<br />

Cryst. Growth 134, 353 (1993).<br />

[16] B. Caroli, C. Caroli, G. Faivre, and G. Mergy, J. Cryst.<br />

Growth 118, 135 (1992).<br />

[17] G. Faivre, in Ref. [2], p. 239.<br />

[18] S. Akamatsu, S. Moulin<strong>et</strong>, and G. Faivre, M<strong>et</strong>all. Mater.<br />

Trans. A 32, 2039 (2001).<br />

[19] D. D. Double and A. Hellawell, Philos. Mag. 19, 1299<br />

(1969); J. P. Riqu<strong>et</strong> and F. Durand, Mater. Res. Bull. 10,<br />

451 (1973); H. Dean and J. E. Gruz<strong>les</strong>ki, J. Cryst. Growth<br />

21, 51 (1974).<br />

[20] R. Trivedi, J. T. Mason, J. D. Verhoeven, and W. Kurz,<br />

M<strong>et</strong>all. Trans. A 22, 252 (1991).<br />

175701-4 175701-4


ARTICLE IN PRESS<br />

Journal of Crystal Growth 299 (2007) 418–428<br />

www.elsevier.com/locate/jcrysgro<br />

Real-time study of thin and bulk eutectic growth in<br />

succinonitrile–(D)camphor alloys<br />

S. Akamatsu a, , S. Bottin-Rousseau a , M. Perrut a , G. Faivre a , V.T. Witusiewicz b , L. Sturz b<br />

a Institut des NanoSciences de Paris, CNRS UMR 7588, Université <strong>Pierre</strong>-<strong>et</strong>-Marie-Curie (Paris VI) <strong>et</strong> Université Denis-Diderot (Paris VII),<br />

Campus Boucicaut, 140 rue de Lourmel, 75015 Paris, France<br />

b ACCESS e.v., Aachen, Intzestrasse 5, D-52072 Aachen, Germany<br />

Received 27 September 2006; accepted 29 November 2006<br />

Communicated by Y. Furukawa<br />

Available online 17 December 2006<br />

Abstract<br />

We study the directional-growth patterns of the transparent nonfac<strong>et</strong>ed alloy succinonitrile–(D)camphor (SCN–DC) at eutectic<br />

concentration in thin and bulk samp<strong>les</strong> using real-time observation m<strong>et</strong>hods. We measure in situ the minimum-undercooling spacing l m<br />

and the small-spacing stability limit l c of the thin, quasi-lamellar patterns. We find l 2 m V 10:2 mm3 s 1 and l c 0:65l m . This last result<br />

is contrary to the common conjecture that l c ¼ l m in lamellar patterns, but in agreement with previous experimental and numerical<br />

studies in the CBr 4 –C 2 Cl 6 eutectic. In bulk samp<strong>les</strong>, we observe the growth of rod-like patterns in real time. We show that these patterns<br />

do not usually achieve long-range hexagonal order after long solidification times, although their average spacing reaches a constant<br />

value, which is close to l m , after a relatively short transient.<br />

r 2007 Elsevier B.V. All rights reserved.<br />

PACS: 81.30.Fb<br />

Keywords: A1. Directional solidification; A1. Eutectics<br />

1. Introduction<br />

Dynamical studies of solidification microstructures<br />

require real-time observations of the solid–liquid interface<br />

during growth. Such observations can be performed by<br />

optical m<strong>et</strong>hods using transparent nonfac<strong>et</strong>ed organic<br />

alloys [1]. Until recently, a single multiphase alloy of this<br />

type, CBr 4 –C 2 Cl 6 , was well documented [2–4]. To remedy<br />

this limitation, Witusiewicz <strong>et</strong> al. [5], have carefully studied<br />

the phase diagrams of succinonitrile–(D)camphor<br />

(SCN–DC) and several other nonfac<strong>et</strong>ed transparent<br />

eutectics [6,7]. We pursue the characterization of the<br />

SCN–DC eutectic by presenting a real-time investigation<br />

of the directional-solidification dynamics of this alloy in<br />

thin (12 mm-thick) and bulk (400 mm-thick) samp<strong>les</strong>. We<br />

focus on alloy concentrations close to the eutectic point.<br />

Corresponding author. Tel.: +33 1 44 27 63 99; fax: +33 1 43 54 28 78.<br />

E-mail address: akamatsu@insp.jussieu.fr (S. Akamatsu).<br />

The eutectic point of SCN–DC lies at T E ¼ 38:4 C and<br />

C E ¼ 13:9 mol% or 23:2 wt%DC (Fig. 1). The two eutectic<br />

solid solutions are almost pure body centered cubic SCN<br />

and hexagonal DC, respectively. The melt growth of these<br />

phases is fully nonfac<strong>et</strong>ed. Since the value of C E is low, the<br />

bulk solidification microstructures of eutectic SCN–DC are<br />

rod-like, with inclusions (rods) of DC in a continuous SCN<br />

matrix (Fig. 2). The instrument we have used for observing<br />

such patterns in situ will be described in Section 2. In thin<br />

samp<strong>les</strong>, we observed solidification microstructures consisting<br />

of a single row of DC crystals embedded in SCN<br />

(Fig. 3). Confinement effects (i.e. the absence of a diffusion<br />

flux across the walls of the container) confer a strong 2D<br />

character to the dynamics of these growth patterns, which<br />

is thus similar to that of a 2D lamellar pattern. We have<br />

studied these ‘‘quasi-lamellar’’ patterns in side view using a<br />

standard optical microscope.<br />

The presentation of the results is organized as follows.<br />

We deal first with thin quasi-lamellar patterns. We begin by<br />

0022-0248/$ - see front matter r 2007 Elsevier B.V. All rights reserved.<br />

doi:10.1016/j.jcrysgro.2006.11.271


ARTICLE IN PRESS<br />

S. Akamatsu <strong>et</strong> al. / Journal of Crystal Growth 299 (2007) 418–428 419<br />

Fig. 1. Phase diagram of the SCN–DC system [6]. Also see Table 1.<br />

Fig. 3. Quasi-lamellar eutectic growth patterns in a thin sample of a<br />

eutectic SCN–DC alloy observed in side view during directional<br />

solidification at V ¼ 0:035 (top) and 0:07 mms 1 (bottom). The SCN–liquid<br />

interfaces are hardly visible because of the small difference in optical<br />

index b<strong>et</strong>ween SCN and the liquid. Scale bars: 50 mm.<br />

Fig. 2. A rod-like eutectic growth pattern in a bulk sample of a eutectic<br />

SCN–DC alloy observed in top view during directional solidification at<br />

V ¼ 0:035 mms 1 . DC rods appear as bright disks in a dark SCN matrix.<br />

For the imaging m<strong>et</strong>hod, see Section 2. Arrow: grain boundary in the SCN<br />

matrix. Horizontal dimension: 865 mm.<br />

studying the stability of these patterns at small spacings.<br />

A longstanding conjecture predicted that 2D lamellar<br />

patterns would be unstable against a lamella elimination<br />

process when their spacing l is smaller than the value l m<br />

corresponding to the minimum of the average undercooling<br />

of the growth front [2]. We report here in situ measurements<br />

of l m , and an investigation of the response of the<br />

pattern to spatial modulations of l, which lead us to<br />

conclude that, in the explored range of growth rates, the<br />

lamella instability threshold is actually at l c 0:65l m .<br />

Next, we investigate the stability of quasi-lamellar patterns<br />

at large spacings. We show that, at slightly hypereutectic<br />

concentrations, the stability limit corresponds to a perioddoubling<br />

oscillatory instability occurring at l osc 1:8l m .<br />

These results are in good agreement with previous studies<br />

in thin CBr 4 –C 2 Cl 6 [8–10]. Then, we study 3D instabilities,<br />

which occur in thin samp<strong>les</strong> under certain conditions. This<br />

study casts light on the question of the morphology of the<br />

crystals in the transverse plane (i.e. the plane normal to the<br />

growth axis). We also give d<strong>et</strong>ails of practical interest<br />

about the initial stages of direction solidification in thin<br />

samp<strong>les</strong>. Finally, we turn to bulk rod-like patterns. Our<br />

observations point to a complicated dynamics influenced<br />

by external forcing factors, such as a slight curvature of the<br />

isotherms. We do not study this dynamics in any d<strong>et</strong>ail<br />

here, but focus on the global features (average spacing,<br />

stability limit) of the rod-like patterns. We show that they<br />

have the same order of magnitude as those of the thin<br />

quasi-lamellar patterns, and that, like the latter, they scale<br />

approximately as the<br />

1<br />

2<br />

power of V.<br />

2. Experimental m<strong>et</strong>hods<br />

We prepared alloys of nominal concentrations close to<br />

C E using zone-purified SCN ð499:99 wt%Þ [11] and<br />

sublimated DC ð499:98 wt%Þ [6]. Since DC and SCN<br />

crystals have nearly the same density, the equilibrium<br />

volume fraction of DC in the solid, Z, practically equals the<br />

DC mass fraction of the alloy. Physical param<strong>et</strong>ers of<br />

interest are given in Table 1. We used cartridges made of<br />

two 300-mm-thick glass plates separated by plastic (PET)<br />

spacers delimiting an empty space of dimensions<br />

6cm 1cm w, where w is the thickness of the PET


420<br />

ARTICLE IN PRESS<br />

S. Akamatsu <strong>et</strong> al. / Journal of Crystal Growth 299 (2007) 418–428<br />

Table 1<br />

Param<strong>et</strong>ers of the SCN–DC phase diagram<br />

T E (K) C E (mol%) m SCN (K) m DC (K)<br />

311.5 0.139 82 318<br />

T E : eutectic temperature; C E : eutectic concentration; m s ðs ¼ SCN; DCÞ:<br />

slope of the s liquidus.<br />

microscope eye-piece<br />

solidification front<br />

Light<br />

hot<br />

cold<br />

x<br />

Long-distance<br />

Microscope<br />

z<br />

y<br />

n≈1.4<br />

w<br />

w 1<br />

θ<br />

V<br />

z<br />

y<br />

Fig. 5. Left: bulk-sample directional-solidification s<strong>et</strong>up. Right: definition<br />

of the contraction factor w 1 =w.<br />

translation<br />

motor<br />

hot block<br />

coldblock<br />

Fig. 4. Thin-sample directional-solidification s<strong>et</strong>up.<br />

spacers (12 and 400 mm for thin and thick samp<strong>les</strong>,<br />

respectively). This space was filled with melted alloy under<br />

a controlled atmosphere, cooled to room temperature, at<br />

which the alloy is solid, and then sealed with a UV<br />

polymerizable glue (NOA 81).<br />

A thin-sample directional-solidification stage is shown in<br />

Fig. 4. The common axis of the thermal gradient G and the<br />

growth is denoted by z, the direction normal to the sample<br />

plane (transverse direction) by y. In this study,<br />

G ¼ 45 5Kcm 1 , and the pulling rate V pull ranges from<br />

0.0175 to 1 mms 1 . The translation of the sample is ensured<br />

by a linear, DC motor (PI) with an accuracy b<strong>et</strong>ter than<br />

1%. Residual impurities generated eutectic cells above<br />

V pull 1 mms 1 , and a nearly constant-speed recoil of<br />

the growth front at lower V pull . Un<strong>les</strong>s otherwise stated,<br />

the results are given as functions of the actual growth<br />

rate V of the eutectic solid (V pull minus the recoil speed),<br />

which was measured in situ in each experiment (typically,<br />

V ¼ 0:75V pull ). A sample could be used for several<br />

solidification runs of more than 24 h each. Observation<br />

was made using a transmission optical microscope (LEICA<br />

DMRXE) equipped with a numerical camera (Scion).<br />

The bulk-sample directional-solidification s<strong>et</strong>up was<br />

similar to the thin-sample one regarding thermal regulation<br />

and sample translation, except that the z axis was put<br />

vertically, the liquid being above the solid, in order to<br />

reduce thermosolutal convection in the liquid. We observed<br />

the growth front from the exterior of the sample with a<br />

long-distance microscope (Questar QM100) in transmission<br />

(Fig. 5). The images were processed for subtracting<br />

background. Image quality was much improved by means<br />

of a diaphragm positioned close to the entry plane of the<br />

microscope. The angle of observation y (angle b<strong>et</strong>ween the<br />

microscope axis and y) and the angle of incidence of the<br />

light y light were both adjustable. The direction of a light ray<br />

as it leaves the sample depends on the nature and<br />

orientation of the interfaces it has crossed. For a given<br />

x<br />

value of y, we selected a value of y light oy, at which DC<br />

rods appeared bright, the SCN matrix and the liquid dark.<br />

The images captured by the camera were contracted in the<br />

y direction by a factor w 1 =w given by<br />

w 1 =w ¼ cos y tan½arcsinðn 1 sin yÞŠ, (1)<br />

where n is the refractive index of the liquid. This factor has<br />

a maximum w 1M =w at an angle y M . In our case, n 1:4,<br />

y M 50 and w 1M =w 0:41. The bulk-sample micrographs<br />

shown in this paper were taken at y close to y M ,<br />

and were re-scaled numerically, i.e. str<strong>et</strong>ched by a factor of<br />

w=w 1 in the y direction. At the chosen magnification, the<br />

observation window was 440 mm wide in the x direction,<br />

and contained the whole image of the front in the transverse<br />

direction. A mosaic image of the whole growth front could<br />

be obtained in about 30 s by automatized serial shifts of the<br />

motorized solidification stage in the x direction. More<br />

d<strong>et</strong>ails about this instrument will be given elsewhere.<br />

The dynamics of solidification patterns is sensitive to<br />

deviations of the isotherms from perfect planarity and<br />

orthogonality to the growth axis. In this investigation, we<br />

were mostly concerned by the transverse bias (average tilt to<br />

the y-axis) and curvature of the isotherms. We measured the<br />

bias by direct observation of the front in side view, and s<strong>et</strong> it<br />

at a small value ðo2 Þ using the adjustable temperature<br />

difference b<strong>et</strong>ween the upper and lower copper blocks of the<br />

ovens. The isotherm curvature was evaluated by following<br />

the evolution of rod-like eutectic patterns over time (not<br />

shown), but could not be controlled. It varied from sample<br />

to sample, but never exceeded 0:2mm 1 in 400-mm thick<br />

samp<strong>les</strong>. We grew large eutectic grains using a funnel-shaped<br />

design of the spacers on the cold side and controlling the<br />

invasion process as explained in Section 3.1.5.<br />

3. Results<br />

3.1. Thin samp<strong>les</strong>. Quasi-lamellar patterns<br />

3.1.1. Minimum-undercooling spacing<br />

In this, and the next section, we shall make use of<br />

formulae, which are valid for lamellar eutectic patterns.


ARTICLE IN PRESS<br />

S. Akamatsu <strong>et</strong> al. / Journal of Crystal Growth 299 (2007) 418–428 421<br />

For the sake of consistency, we shall therefore call the DC<br />

and SCN crystals ‘‘lamellae’’, and we shall give to these<br />

lamellae thicknesses corresponding to the phase fraction Z<br />

of the alloy. It will be seen in Section 3.1.4 that, in our thin<br />

samp<strong>les</strong>, the DC crystals are, if not lamellar, at least<br />

elongated in the transverse direction. The errors that we<br />

make by assuming the observed thin patterns to be lamellar<br />

should thus be very modest compared to the other sources<br />

of experimental uncertainty.<br />

In the Jackson and Hunt (JH) approximation [2], the<br />

average undercooling DT ¼ T E T of a lamellar eutectic<br />

growth front reads as follows:<br />

DT K 1 Vl þ K 2 =l, (2)<br />

where K 1 and K 2 depend on the composition and material<br />

constants of the alloy. The minimum of DT over l gives<br />

l m l JH<br />

m ¼ðK 2=K 1 Þ 1=2 V 1=2 (3)<br />

and<br />

DT m DT JH<br />

m ¼ð4K 1K 2 Þ 1=2 V 1=2 . (4)<br />

Since many characteristic lengths of eutectic patterns scale<br />

1<br />

approximately as the<br />

2<br />

power of V, experimental and<br />

theor<strong>et</strong>ical results are often displayed as a function of the<br />

reduced spacing L ¼ l=l JH<br />

m . Hence, the importance of<br />

knowing accurately the value of l JH<br />

m or, more exactly, of the<br />

material constant l JH2<br />

m V ¼ K 2=K 1 in the alloys of interest.<br />

In a previous study [10], we d<strong>et</strong>ermined l JH<br />

m in situ in<br />

lamellar CBr 4 –C 2 Cl 6 alloys using the following m<strong>et</strong>hod:<br />

(i) we selected an area, in which l varied over a large range<br />

enclosing l JH<br />

m ; (ii) we measured the spacing distribution,<br />

lðxÞ, and the envelope of the front, zðxÞ, over this area;<br />

(iii) we eliminated x and plotted z as a function of l and<br />

(iv) fitted the data points to<br />

z ¼ G 1 DT þ z E (5)<br />

using z E (the ordinate of T E ), K 1 and K 2 as adjustable<br />

param<strong>et</strong>ers. The critical part of the procedure was the<br />

measurement of zðxÞ at stage (ii). This was performed by<br />

binarizing the optical image of the front, and then taking<br />

the average of the binarized profile over each lamella pair.<br />

We have applied this m<strong>et</strong>hod to thin SCN–DC samp<strong>les</strong>.<br />

An example is shown in Fig. 6 [stages (i)–(iii)] and the ins<strong>et</strong><br />

in Fig. 7 [stage (iv)]. We found l JH<br />

m ¼ 18:5 1:5 mm, and<br />

hence K 2 =K 1 ¼ 10:2 1:5 mm 3 s 1 . However, we m<strong>et</strong> with<br />

a difficulty: the results were sensitive to the d<strong>et</strong>ails of the<br />

binarization m<strong>et</strong>hod. This was certainly due to the poor<br />

contrast of the SCN–liquid interfaces. Most of the<br />

uncertainty originated from the regions of the front close<br />

to the junction points. To g<strong>et</strong> rid of the influence of these<br />

regions, we left out the averaging procedure, and considered<br />

only the values of the undercooling at the tips of<br />

SCN and DC lamellae denoted by DT s , where s designates<br />

SCN or DC. The DT s data from Fig. 6 are plotted in<br />

Fig. 7. Obviously, they convey more information than the<br />

DT data plotted in the ins<strong>et</strong>. We obtained equations for the<br />

DT s as functions of l from the so-called amended JH<br />

λ (µm)<br />

23<br />

21<br />

19<br />

17<br />

14<br />

0 50 100 150 200 250 300 350<br />

x (µm)<br />

Fig. 6. Top: a quasi-lamellar pattern with a smooth spacing modulation in<br />

a thin sample of eutectic SCN–DC. V ¼ 0:03 mms 1 . Bottom: binarized<br />

shape (thin line) and average envelope (thick line) of the growth front.<br />

Square symbols: spacing distribution (the line is a guide for the eye).<br />

∆T (°C)<br />

0.033<br />

0.032<br />

0.031<br />

0.030<br />

0.029<br />

0.028<br />

JH<br />

SCN<br />

DC<br />

18 20 22<br />

λ (µm)<br />

0.027<br />

14 16 18 20 22 24 26<br />

λ (µm)<br />

Fig. 7. Undercooling of the SCN and DC lamella tips as a function of<br />

spacing. Same experiment as in Fig. 6. The measurements yielded the<br />

undercoolings up to the unknown constant Gz E , which was taken as an<br />

adjustable param<strong>et</strong>er. Thin-line curves: best-fitting curves. Dashed-line<br />

curve: l JH<br />

m calculated from the best-fitting values of the material constants<br />

(see Table 2). Ins<strong>et</strong>: average undercooling and best-fitting curve according<br />

to Eq. (2).<br />

approximation, which, contrary to Eq. (2), takes the<br />

difference of average undercooling b<strong>et</strong>ween SCN and DC<br />

lamellae into account [3,12]. We fitted these equations to<br />

the DT s data using z E and the unknown material constants<br />

of the alloy as adjustable param<strong>et</strong>ers. These include the<br />

diffusion coefficient in the liquid, D, the Gibbs–Thomson<br />

coefficients, a 0:s , and the slope ang<strong>les</strong> at the junction<br />

points, y s , of the solid–liquid interfaces. The y s are related<br />

to the surface tensions of the solid–liquid and solid–solid<br />

interface (denoted by g s2L and g SCN2DC , respectively) by<br />

Young’s law. In fact, calculations showed that we could<br />

∆T (mK)<br />

20.0<br />

19.5<br />

19.0<br />

17<br />

16<br />

15<br />

z (µm)


422<br />

ARTICLE IN PRESS<br />

S. Akamatsu <strong>et</strong> al. / Journal of Crystal Growth 299 (2007) 418–428<br />

Table 2<br />

Values of material constants of eutectic SCN–DC yielded by a least-square fit of the amended JH equations [3] to the values of DT SCN and DT DC extracted<br />

from Fig. 6<br />

D ðmm 2 s 1 Þ a SCN ðmmKÞ a DC ðmmKÞ y SCN (deg.) y DC (deg.) K 1 ðKsmm 2 Þ K 2 ðK mmÞ<br />

167 0.0622 a 0.1097 50.55 65 0.0283 0.275<br />

a 0:SCN and a 0:DC are the Gibbs–Thomson coefficients of the SCN– and DC–liquid interfaces, respectively.<br />

a Value from Ref. [13].<br />

reduce the number of free param<strong>et</strong>ers without a significant<br />

loss of accuracy. We s<strong>et</strong> l JH<br />

m ¼ 18:5 mm, and kept only two<br />

free param<strong>et</strong>ers, namely, g DC2L and D. For the other<br />

material constants, we took the known values for pure<br />

SCN and DC [6,13]. The best-fitting values are given in<br />

Table 2. Of note is the substantial difference in surface<br />

tension ðg DC =g SCN 1:8Þ, and thus in slope ang<strong>les</strong> of the<br />

two solid–liquid interfaces. The best-fitting curves are<br />

plotted in Fig. 7. The fit remained satisfactory (least-square<br />

error compatible with experimental scatter) for values<br />

within about 15% of the best-fitting ones.<br />

3.1.2. Phase diffusion—Eckhaus instability<br />

It has long been known that lamellar eutectic patterns<br />

are unstable against an Eckhaus instability (i.e. a spontaneous<br />

amplification of spatial variations of l) at small<br />

spacings [2], but the exact value of the instability threshold<br />

l c is still debated. JH’s conjecture was that l c coincided<br />

with l m . In a previous work in the CBr 4 –C 2 Cl 6 alloy, it was<br />

found that l c was substantially lower than l m . This result<br />

was accounted for using the following semi-empirical<br />

formula based on numerical-simulation studies:<br />

1<br />

1<br />

L 2 c<br />

þ AG<br />

K 1 V L c ¼ 0, (6)<br />

where L c ¼ l c =l m and A 0:15 [10]. Taking for K 1 the<br />

value given in Table 2, for V and G the values<br />

corresponding to Fig. 6 and using Eq. (6), we obtain L c ¼<br />

0:77 0:15 and hence l c ¼ 13 2 mm, which is consistent<br />

with the pattern in Fig. 6 remaining stable over the<br />

duration time of the observation.<br />

We performed a more rigorous test of the validity of Eq.<br />

(6) by studying the time evolution of patterns with longrange<br />

spatial variations of l about an average value l 0 (lmodulated<br />

patterns). When l-modulations have small<br />

enough amplitudes, their spatiotemporal evolution is<br />

governed by an equation of the form<br />

q t lðx; tÞ ¼D ph ðl 0 Þq 2 xlðx; tÞ, (7)<br />

where D ph is a spacing-dependent ‘‘phase’’ diffusion<br />

coefficient. According to this equation, a sinusoid modulation<br />

of wavelength L should relax (when D ph 40), or<br />

amplify (when D ph o0), following an exponential law with<br />

a characteristic time t ¼ D ph ð2p=LÞ 2 . The Eckhaus instability<br />

threshold corresponds to the limit case in which<br />

D ph ¼ 0. A calculation by Langer [14] using Eq. (2) and the<br />

assumption that lamellae grow perpendicular to the<br />

envelope of the front gave<br />

D ph <br />

VDT<br />

JH<br />

m<br />

2G ðL 0 L0 1 Þ, (8)<br />

where L 0 ¼ l 0 =l m . According to this approximation, the<br />

instability threshold is at L 0 ¼ 1. The above-mentioned<br />

semi-empirical correction reads<br />

D ph ¼<br />

VDT<br />

JH<br />

m<br />

2G ðL 0 L 1<br />

0 ÞþAVl 0 L 0 . (9)<br />

(Note that Eq. (6) is obtained by s<strong>et</strong>ting D ph ¼ 0 in<br />

Eq. (9).) We compare this equation with experimental<br />

observations. Fig. 8 shows the time evolution of two<br />

l-modulated patterns with different values of L 0 . In one<br />

case, L 0 1, and the l-modulation relaxed over time.<br />

In the other, L 0 0:6, and the l-modulation amplified. We<br />

followed the amplitudes of long-wavelength Fourier<br />

components of the l-modulations over time, fitted the<br />

data to exponential laws, obtained values of the relaxation<br />

time t and hence of the phase diffusion coefficients.<br />

We found 0.18 and 0:01 mm 2 s 1 for D ph in the first,<br />

and the second case, respectively. From these values, we<br />

calculated K 1 and K 2 using Eq. (9). We found K 1 ¼ 3:0<br />

10 2 Ksmm 2 , K 2 ¼ 3:0 10 1 K mm, and hence K 2 =K 1 ¼<br />

10 K mm 3 s 1 and L c ¼ 0:64, in good agreement with the<br />

above results. This new validation of Eq. (6) is all the more<br />

significant since the values of K 1 in SCN–DC and<br />

CBr 4 –C 2 Cl 6 ð1:9 10 3 Ksmm 2 Þ are very different from<br />

each other.<br />

3.1.3. Large-spacing oscillatory patterns<br />

We observed two kinds of large-spacing oscillatory<br />

patterns corresponding to the 2lO (period-doubling) and<br />

1lO (period-preserving) modes, respectively. The 1lO<br />

mode (Fig. 9) was only observed in a slightly hypoeutectic<br />

sample at L42. This is in agreement with previous<br />

experimental observations in thin eutectic CBr 4 –C 2 Cl 6<br />

alloys and with numerical simulations [8,9]. However, the<br />

pattern tended to undergo a 3D deformation as it oscillated<br />

(Fig. 9b). For large enough values of L, this led to lamella<br />

branching being triggered (Fig. 9c). The spacing was then<br />

suddenly reduced and the oscillation receded.<br />

The 2lO mode was observed at large L values in<br />

slightly hypereutectic samp<strong>les</strong>. In Fig. 10, it was obtained<br />

by growing first a steady pattern (not shown) at<br />

V ¼ 0:07 mms 1 with an average spacing value corresponding<br />

to L 0 ¼ 1:5, and then imposing a velocity jump to


ARTICLE IN PRESS<br />

S. Akamatsu <strong>et</strong> al. / Journal of Crystal Growth 299 (2007) 418–428 423<br />

a<br />

b<br />

λ (μm)<br />

36<br />

32<br />

28<br />

24<br />

20<br />

A (μm)<br />

4<br />

3.5<br />

3<br />

2.5<br />

2000 6000 10000 14000<br />

t (s)<br />

λ (μm)<br />

26<br />

24<br />

22<br />

20<br />

18<br />

A (μm)<br />

2.2<br />

2<br />

1.8<br />

1.6<br />

1.4<br />

0 5 10 15 20<br />

t (10 3 s)<br />

16<br />

12<br />

t = 0<br />

t = 2800 s<br />

t = 14000 s<br />

16<br />

14<br />

t = 0<br />

t = 8000 s<br />

t = 18000 s<br />

0 50 100 150 200 250 300<br />

x (μm)<br />

0 20 40 60 80 100 120 140 160<br />

x (μm)<br />

Fig. 8. Time evolution of l-modulated quasi-lamellar patterns. The graphs show the spacing distribution lðxÞ at three successive times (see legend). The<br />

ins<strong>et</strong>s show the time evolution of a Fourier component of wavelength L (L ¼ 434 mm in (a) and 135 mm in (b)) and exponential fits to the data. (a) Stable<br />

pattern (V ¼ 0:03 mms 1 , l 0 ¼ 19 mm; l 2 0 V ¼ 10:8Kmm3 s 1 ). The micrograph was taken at t ¼ 100 s. (b) Unstable pattern (V ¼ 0:013 mms 1 ;<br />

l 0 ¼ 17 mm; l 2 0 V ¼ 3:8Kmm3 s 1 ). The micrographs were taken at t ¼ 0 and 18 s, respectively. Scale bars: 50 mm.<br />

Fig. 10. Period-doubling oscillation ð2lOÞ pattern at V ¼ 0:14 mms 1 .<br />

The slight spatial modulation of the oscillation amplitude did not<br />

correspond to any measurable spatial variation of l. Scale bar: 50 mm.<br />

Fig. 9. Period-preserving oscillation (1lO) patterns at different velocities.<br />

(a) V ¼ 0:053 mms 1 , L 1:5. (b) V ¼ 0:090 mms 1 , L 2. (c)<br />

V ¼ 0:18 mms 1 , L 2:7. Lamella branching occurs through an unidentified<br />

3D process. Scale bar: 50 mm.<br />

V ¼ 0:14 mms 1 corresponding to a jump to L 0 ¼ 2:1. The<br />

amplitude of the oscillations was observed to vary along<br />

the front according to the local value of l. The transition<br />

from steady to oscillatory lamellae occurred at L osc 1:8.<br />

The amplitude of the oscillation increased extremely<br />

rapidly as L rose above L osc , as previously observed in<br />

CBr 4 –C 2 Cl 6 samp<strong>les</strong>.<br />

Fig. 11. Period-doubling oscillation ð2lOÞ patches and travelling waves.<br />

Single eutectic grain. V ¼ 0:14 mms 1 . Horizontal dimension: 0.99 mm.<br />

The high sensitivity of the 2lO amplitude to spacing is<br />

responsible for the creation of localized patterns like that<br />

shown in Fig. 11. In this example, an oscillation patch<br />

started near a maximum of the l distribution, and then<br />

evolved in a complex, highly nonlinear way giving rise to<br />

solitary waves and intermittent oscillations.


424<br />

We also observed transitory ‘‘giant oscillations’’ (oscillations<br />

of amplitude nearly equal to l) at very large spacing<br />

values ðL43Þ (Fig. 12), which ended in triggering lamella<br />

branching.<br />

3.1.4. 3D effects<br />

We now consider the transverse morphology of the DC<br />

crystals in thin stationary patterns. In the absence of direct<br />

observations, we are reduced to conjectures. We note first<br />

that any acceptable conjecture should respect two conditions,<br />

namely, mass conservation and the no-flux condition<br />

across the glass plates. Under some simplifying assumptions<br />

about the w<strong>et</strong>ting ang<strong>les</strong> at the glass plates, the latter<br />

condition is equivalent to the following one: the rectangular<br />

arrays of DC crystals generated by mirror symm<strong>et</strong>ry<br />

about the glass plates is a stationary (but not necessarily<br />

stable) state of the infinite system. Consider now, in a thin<br />

sample, two patterns composed, the one, of a row of<br />

circular rods centered in the mid-plane of the sample, and<br />

the other, of (truncated) lamellae. The apparent widths l DC<br />

Fig. 12. Giant-oscillation pattern ðV ¼ 0:0175 mms 1 Þ. Scale bar: 50 mm.<br />

a<br />

l DC<br />

λ<br />

b<br />

DC SCN DC DC<br />

SCN<br />

Fig. 13. Hypoth<strong>et</strong>ical transverse cross-sections of DC rods at w=l ¼ 0:6<br />

and Z ¼ 2:3. (a) In a stationary pattern. (b) During the oscillation shown<br />

in Fig. 14.<br />

ARTICLE IN PRESS<br />

S. Akamatsu <strong>et</strong> al. / Journal of Crystal Growth 299 (2007) 418–428<br />

DC<br />

w<br />

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi<br />

of the DC crystals in these patterns are 4wZl=p and Zl,<br />

respectively. We introduce the aspect ratio r ¼ w=ðZlÞ, and<br />

the reduced apparent width l DC =ðZlÞ of pDC ffiffiffiffiffiffiffiffiffiffi crystals. The<br />

latter quantity is unity for lamellae and 4r=p for circular<br />

rods. Incidentally, note that circular rods cannot exist in<br />

thin samp<strong>les</strong> when ro4=p.<br />

Inpmost ffiffiffiffiffiffiffiffiffiffi of our experiments, r ranged from about 2 to 3,<br />

and 4r=p ranged from about 1.6 to 2. We measured l DC<br />

and found l DC =ðZlÞ 1:3 in a number of solidification<br />

runs. The apparent width of the DC crystals was thus<br />

intermediate b<strong>et</strong>ween those of lamellae and circular rods. A<br />

simple conjecture is that the DC crystals were rods<br />

elongated in the transverse direction (Fig. 13a). Unpublished<br />

numerical calculations by A. Parisi and M. Plapp<br />

have shown that rectangular arrays of elongated rods can<br />

be stationary in small systems. The solution outlined in<br />

Fig. 13a can thus be envisaged. Other conjectures, which it<br />

is not useful to list here, are possible. We simply wish to<br />

stress the importance of the aspect ratio r as concerns the<br />

transverse morphology of the pattern. As an illustration of<br />

this statement, we observed that thin stationary patterns<br />

were unstable to a 3D oscillation at relatively high values of<br />

r (Fig. 14). In this oscillation mode, DC crystals alternate<br />

b<strong>et</strong>ween their stationary configuration and another configuration<br />

with a much larger image width, in which they are<br />

certainly in contact with a glass plate (Fig. 13b). This last<br />

configuration is essentially unstable, and is also observed<br />

as a precursor to lamella elimination. It is worth noting<br />

that nearby DC crystals oscillate roughly in antiphase,<br />

although there is no phase correlation on a larger scale.<br />

The same oscillation was already observed in CBr 4 –C 2 Cl 6<br />

at similar values of Z and r [8], and may also be the origin<br />

of the ‘‘transverse-rod’’ microstructure shown in Fig. 10 of<br />

Ref. [5].<br />

The 3D processes through which new lamella pairs are<br />

created are generically called lamella branching. Lamella<br />

branching rarely occurs in thin samp<strong>les</strong>, except during the<br />

initial transient of directional solidification (see below). We<br />

have already mentioned the occurrence of occasional<br />

lamella branching events during large-amplitude oscillations.<br />

Fig. 14. 3D oscillations in thin samp<strong>les</strong>. (a) V ¼ 0:24 mms 1 ; r 4:5. (b) V ¼ 0:035 mms 1 ; r 4. Bars: 50 mm.


ARTICLE IN PRESS<br />

S. Akamatsu <strong>et</strong> al. / Journal of Crystal Growth 299 (2007) 418–428 425<br />

We focus here on a more spectacular mode of lamella<br />

branching, which can be triggered by a sudden increase in<br />

V applied to a stationary pattern (Fig. 15). During this<br />

process, the DC crystals split along the direction y (note<br />

that r 5 in Fig. 15), and then the branches drift in<br />

opposite directions along the axis x. The process has a<br />

relatively long lag time, which increases as l decreases<br />

indicating that it is closely connected with some longwavelength<br />

instability of the pattern. This instability might<br />

be a zigzag bifurcation [15], but this has not been<br />

established y<strong>et</strong>. The process obviously results in the local<br />

spacing being abruptly reduced by half. The resulting<br />

pattern may be unstable against lamella elimination, in<br />

which case some branches are eliminated soon after being<br />

created. Such was actually the case after the velocity jump<br />

illustrated in Fig. 15.<br />

3.1.5. Initial transient<br />

We controlled the grain structure and made checks of the<br />

concentration of the samp<strong>les</strong> during the first stages of<br />

directional solidification. Prior to solidification, the samp<strong>les</strong><br />

were directionally melted over about 4 cm, and then<br />

held at rest ðV ¼ 0Þ for some hours. The melting process<br />

yielded a collection of small crystals, which coarsened into<br />

a continuous, polycrystalline one-phased layer during the<br />

annealing at rest (Fig. 16). At the end of the process, which<br />

took about 20 h, the upper and lower boundaries of the<br />

one-phased layer were, theor<strong>et</strong>ically, at the liquidus<br />

temperature T s ðC 0 Þ for the majority phase s and T E ,<br />

respectively, so that the thickness of this layer equalled<br />

dz 0 ¼ G 1 m s ðC 0 C E Þ. In practice, however, the measured<br />

values of the layer thickness showed a large scatter,<br />

which was due to the fact that the thermal field fluctuated<br />

during the long annealing process entailing displacements<br />

of the upper, but not of the lower boundary of the onephased<br />

layer.<br />

We obtained a reliable in situ estimate of C 0 as follows.<br />

The ons<strong>et</strong> of solidification triggers a continual drift (recoil)<br />

of the interface b<strong>et</strong>ween the one-phased layer and the liquid<br />

toward cold temperatures (solute redistribution transient).<br />

A previous study showed that, during this process, crystals<br />

of the minority phase grow along some grain boundaries of<br />

the one-phased solid, and emerge at grain boundary<br />

grooves when the growth front reaches T E (Fig. 17).<br />

Fig. 15. Collective lamella branching occurring after a jump of V pull from<br />

0.28 to 0:57 mms 1 . The spacing varies from 10.6 to 11:4 mm inside the<br />

observed area.<br />

Fig. 17. Micrographs of an SCN solidification front at the ons<strong>et</strong> of pulling<br />

(right-hand side frame) and after a 1500 s pulling (same run as in Fig. 16).<br />

Ins<strong>et</strong>: enlarged view of an emerging DC crystal. The meaning of dz 0 <br />

300 mm is explained in the text.<br />

Fig. 16. Formation of the majority-phase (DC) layer at rest after partial directional melting of a hypereutectic sample ðC 0 0:25 mol%Þ; the solid layer<br />

was compl<strong>et</strong>ed after about 20 h. dz ¼ 320 mm: thickness of the DC layer.


426<br />

ARTICLE IN PRESS<br />

S. Akamatsu <strong>et</strong> al. / Journal of Crystal Growth 299 (2007) 418–428<br />

Taking for dz 0 the distance b<strong>et</strong>ween the position of the<br />

front at this moment and its initial position, we obtained<br />

values of C 0 , which were always within experimental<br />

uncertainty of the nominal values (e.g. C 0 ¼ 0:141 mol% in<br />

the case of Fig. 17). It is interesting to note that, during this<br />

process, the solid–liquid interface recoiled at a velocity<br />

practically equal to V pull , which is typical for the initial<br />

transient of alloys with an almost zero partition coefficient.<br />

A short time after crystals of the minority phase<br />

appeared, they spread along the growth front at a high<br />

velocity compared to V forming a so-called invasion<br />

tongue. Repeated lamella branching may occur, or not,<br />

near the tip of an invasion tongue, depending on the phase<br />

of which it is constituted, and the invasion rate [16]. We<br />

triggered invasions without lamella branching (one-phased<br />

invasion) in order to eliminate most of the grains present in<br />

the initial one-phased layer, and thus grow large eutectic<br />

grains. The m<strong>et</strong>hod is illustrated in Fig. 18 in the case of a<br />

slightly hypereutectic sample. A polycrystalline DC layer<br />

was covered by a monocrystalline SCN primary invasion<br />

tongue, which was in turn covered by a two-phased<br />

secondary invasion (i.e. an invasion by DC accompanied<br />

with lamella branching). Since all the lamellae of one phase<br />

originated from a single crystal, the result was a single<br />

eutectic grain. To obtain a primary one-phased invasion,<br />

we solidified until we obtained a lamellar pattern. We<br />

stopped the solidification, and waited until a DC layer reformed<br />

encapsulating all the SCN lamellae but one, which<br />

served as an invasion seed when the pulling was resumed at<br />

low speed ð0:0175 mms 1 Þ.<br />

The invasion-branching process creates a lamella pattern<br />

with a very small spacing, and is followed by a profuse<br />

lamella elimination process. When this process ends, the l<br />

distribution is entirely inside the stability range at the<br />

growth velocity used, but is modulated on a large scale<br />

compared to the average value of the spacing. This was the<br />

origin of the l-modulations we studied in Section 3.1.2.<br />

Contrary to what was observed previously in CBr 4 –C 2 Cl 6 ,<br />

we did not observe any dependence of the growth dynamics<br />

on the orientation of the crystals in SCN–DC (with one<br />

exception; see below). Such a dependence, when it exists, is<br />

due to the anisotropy of interfacial properties (surface<br />

tension and kin<strong>et</strong>ic coefficient). It manifests itself mostly<br />

through a slight grain-dependent tilt of the lamellae with<br />

respect to the growth axis [17]. The absence of such a<br />

‘‘crystallographic’’ tilt in practically all the grains of<br />

SCN–DC indicates a very small anisotropy of the SCNand<br />

DC–liquid interfaces. The exception, which we observed<br />

once, could only be due to the anisotropy of the SCN–DC<br />

interface (Fig. 19). In that particular solidification run, a<br />

lamella tilt was definitely visible within (and only within) a<br />

certain eutectic grain. We conjecture that this grain had a<br />

special orientation relationship b<strong>et</strong>ween cubic SCN and<br />

hexagonal DC, for which there existed a direction of low<br />

energy for the SCN–DC interface, which was the direction<br />

of growth of the lamellae.<br />

4. Thick samp<strong>les</strong>. Rod-like patterns<br />

Fig. 20 presents three typical rod-like eutectic patterns<br />

observed in 400 mm-thick samp<strong>les</strong> of near-eutectic<br />

SCN–DC for different values of V after several hours of<br />

solidification. It can be seen that the rod packing is<br />

essentially hexagonal, but contains a large density of<br />

topological defects. A d<strong>et</strong>ailed study of the time evolution<br />

of the patterns, which is beyond the scope of this article,<br />

showed that these defects are of a dynamical origin, except<br />

for those (rows of rods of diam<strong>et</strong>ers smaller than average)<br />

that are attached to SCN grain boundaries. We measured<br />

the average nearest-neighbor spacing l av using a Voronoi<br />

construction and gained information on the long-range<br />

order using fast Fourier transforms. At all V, we found<br />

that, after a transient, l av reached a constant value despite<br />

the fact that the pattern continued to change over time. It<br />

must therefore be understood that l av and, in fact, the<br />

whole spacing distribution are the result of a dynamical<br />

Fig. 19. The field of view contains two eutectic grains. Arrow: eutectic<br />

grain boundary. The DC lamellae are tilted rightward, and exhibit<br />

symm<strong>et</strong>ry-broken shapes in the leftmost grain, not in the rightmost one.<br />

V ¼ 0:244 mms 1 .<br />

Fig. 18. Invasion process leading to the formation of a single eutectic grain in a hypoeutectic sample. An SCN primary invasion has already taken place<br />

(the SCN–liquid interface has a vanishing contrast). A DC secondary invasion accompanied with lamella branching is running leftwards.<br />

V ¼ 0:017 mms 1 .


ARTICLE IN PRESS<br />

S. Akamatsu <strong>et</strong> al. / Journal of Crystal Growth 299 (2007) 418–428 427<br />

Fig. 20. Rod-like eutectic patterns in 400 mm-thick samp<strong>les</strong> at different solidification speeds V and solidification durations d. Top micrograph and<br />

leftmost FFT: V ¼ 0:05 mms 1 , d 22 h. Middle micrograph and FFT: V ¼ 0:15 mms 1 , d 17 h. (Note the presence of two grain boundaries.) Bottom<br />

micrograph and rightmost FFT: V ¼ 0:25 mms 1 , d 30 min. Horizontal dimension (micrographs): 0.44 mm. The scale of the FFT patterns is arbitrary.<br />

balance b<strong>et</strong>ween several different trends, among which a<br />

drift and a dilatation of the pattern induced by a tilt and<br />

a curvature of the isotherms. We have plotted l av as<br />

a function of V in Fig. 21. In the ins<strong>et</strong>, a l 2 VvsVplot<br />

1<br />

shows that l av sca<strong>les</strong> approximately as the<br />

2<br />

power of V,<br />

and is, in fact, close to the minimum-undercooling spacing<br />

for rod-like patterns (which is close to l JH<br />

m when Z ¼ 0:24<br />

[2]). This result is satisfactory, and may be considered as an<br />

overall check of the reliability of our experimental<br />

m<strong>et</strong>hods. For comparison purposes, we also plotted the<br />

minimum ðl Þ and maximum ðl þ Þ observed values of<br />

l in Fig. 21. It is interesting to note that the order of<br />

magnitude of the width of the l distribution is comparable<br />

with that found by Trivedi <strong>et</strong> al. in bulk m<strong>et</strong>allic lamellar<br />

eutectics [18].<br />

In this series of experiments, the long-range order<br />

increased when V increased, as shown by the FFT patterns<br />

displayed in Fig. 20. The same tendency has been observed<br />

previously in an Al–Al 3 Ni alloy [19]. However, we did not<br />

observe this generally. We think that the effective ordering<br />

of the rod-like pattern strongly depends on the pullingvelocity<br />

program applied during solidification. Moreover,<br />

it can be seen that, in the ordered areas of the patterns, a<br />

λ (µm)<br />

30<br />

25<br />

20<br />

15<br />

10<br />

5<br />

λ -<br />

λ av<br />

λ +<br />

λ m<br />

λ 2 V (μm 3 s -1 )<br />

25<br />

20<br />

15<br />

10<br />

5<br />

0<br />

0.02 0.04 0.06 0.08 0.10 0.12 0.14<br />

0.02 0.04 0.06 0.08 0.1 0.12 0.14<br />

V (µms -1 )<br />

V (µms -1 )<br />

Fig. 21. Minimum l , average l av and maximum l þ spacings of rod-like<br />

patterns as a function of pulling velocity. l m : calculated minimumundercooling<br />

spacing. Ins<strong>et</strong>: l 2 V values for the same data.


428<br />

ARTICLE IN PRESS<br />

S. Akamatsu <strong>et</strong> al. / Journal of Crystal Growth 299 (2007) 418–428<br />

close-packing direction was generally perpendicular to the<br />

glass plates. This illustrates the fact that the ordering<br />

process is not only history dependent, but also strongly<br />

influenced by wall effects and thermal bias. Further study<br />

will be necessary to clarify this point.<br />

5. Conclusion<br />

This study has shown that eutectic SCN–DC is capable<br />

of undergoing long-duration (several days) directionalsolidification<br />

runs without any apparent degradation, and<br />

that, despite the relatively low (hexagonal) symm<strong>et</strong>ry of<br />

DC crystals, it is not only fully nonfac<strong>et</strong>ed, but also very<br />

weakly anisotropic. At eutectic concentration, SCN–DC<br />

gives rise to rod-like patterns in bulk samp<strong>les</strong>, and quasilamellar<br />

patterns in thin samp<strong>les</strong>. We have studied in some<br />

d<strong>et</strong>ails the dynamics of thin quasi-lamellar patterns and<br />

obtained mostly two s<strong>et</strong>s of results. First, this dynamics is<br />

in entire agreement with previous experimental and<br />

theor<strong>et</strong>ical studies in other thin systems. In particular, this<br />

study confirms the recently established overstability of<br />

lamellar patterns at small spacings [10]. Second, we have<br />

d<strong>et</strong>ermined quantitatively physical properties of SCN–DC<br />

relevant to eutectic growth, in particular, the constant<br />

l 2 m V ¼ 10:2 1:5 mm3 s 1 . This value is similar to those<br />

found in many m<strong>et</strong>allic eutectics. A preliminary real-time<br />

study in bulk samp<strong>les</strong> showed that the system tends indeed<br />

to form regular hexagonal patterns, but contains many<br />

dynamical defects even after long solidification times. A<br />

real-time study of the defect dynamics in bulk eutectic<br />

SCN–DC is in progress.<br />

Acknowledgments<br />

This work was supported by Centre National d’Etudes<br />

Spatia<strong>les</strong> (CNES), France, and the Deutsches Zentrum fr<br />

Luft- und Raumfahrt e.V. DLR under Grant no.<br />

50WM0543.<br />

References<br />

[1] K.A. Jackson, J.D. Hunt, Acta M<strong>et</strong>all. 13 (1965) 1212.<br />

[2] K.A. Jackson, J.D. Hunt, Trans. M<strong>et</strong>all. Soc. AIME 236 (1966) 1129.<br />

[3] J. Mergy, G. Faivre, C. Guthmann, R. Mell<strong>et</strong>, J. Crystal Growth 134<br />

(1993) 353.<br />

[4] V. Se<strong>et</strong>haraman, R. Trivedi, M<strong>et</strong>all. Trans. 19A (1988) 2955.<br />

[5] V.T. Witusiewicz, L. Sturz, U. Hecht, S. Rex, Acta Mater. 52 (2004)<br />

4561.<br />

[6] V.T. Witusiewicz, L. Sturz, U. Hecht, S. Rex, Acta Mater. 52 (2004)<br />

5071;<br />

V.T. Witusiewicz, L. Sturz, U. Hecht, S. Rex, Acta Mater. 52 (2004)<br />

5519;<br />

V.T. Witusiewicz, L. Sturz, U. Hecht, S. Rex, Acta Mater. 53 (2005)<br />

173.<br />

[7] L. Sturz, V.T. Witusiewicz, U. Hecht, S. Rex, J. Crystal Growth 270<br />

(2004) 273.<br />

[8] M. Ginibre, S. Akamatsu, G. Faivre, Phys. Rev. E 56 (1997) 780.<br />

[9] A. Karma, A. Sarkissian, M<strong>et</strong>all. Trans. A 27 (1996) 635.<br />

[10] S. Akamatsu, M. Plapp, G. Faivre, A. Karma, Phys. Rev. E 66 (2002)<br />

030501(R);<br />

S. Akamatsu, M. Plapp, G. Faivre, A. Karma, M<strong>et</strong>all. Mater. Trans.<br />

A 35 (2004) 1815.<br />

[11] B. Kauerauf, G. Zimmermann, L. Murmann, S. Rex, J. Crystal<br />

Growth 193 (1998) 701.<br />

[12] K. Brattkus, B. Caroli, C. Caroli, B. Roul<strong>et</strong>, J. Phys. (France) 51<br />

(1990) 1847.<br />

[13] R.J. Schaefer, M.E. Glicksman, J.D. Ayers, Philos. Mag. 32 (1975)<br />

725.<br />

[14] J.S. Langer, Phys. Rev. L<strong>et</strong>t. 44 (1980) 1023.<br />

[15] S. Akamatsu, S. Bottin-Rousseau, G. Faivre, Phys. Rev. L<strong>et</strong>t. 93<br />

(2004) 175701.<br />

[16] S. Akamatsu, S. Moulin<strong>et</strong>, G. Faivre, M<strong>et</strong>all. Mater. Trans. A 32A<br />

(2001) 2039.<br />

[17] B. Caroli, C. Caroli, G. Faivre, J. Mergy, J. Crystal Growth 118<br />

(1992) 135.<br />

[18] R. Trivedi, J.T. Mason, J.D. Verhoeven, W. Kurz, M<strong>et</strong>all. Trans.<br />

22A (1991) 2523.<br />

[19] L. Ratke, J. Alkemper, Acta Mater. 48 (2000) 1939.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!