RECENT DEVELOPMENT IN COMPUTATIONAL SCIENCE
RECENT DEVELOPMENT IN COMPUTATIONAL SCIENCE
RECENT DEVELOPMENT IN COMPUTATIONAL SCIENCE
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>RECENT</strong> <strong>DEVELOPMENT</strong><br />
<strong>IN</strong> <strong>COMPUTATIONAL</strong> <strong>SCIENCE</strong><br />
Selected Papers<br />
from International Symposium<br />
on Computational Science<br />
International Symposium on Computational Science<br />
Kanazawa University, Japan<br />
February 2011<br />
Editors<br />
Rukman Hertadi<br />
Shinichi Miura<br />
Rinovia Simanjuntak<br />
Karel Svadlenka<br />
Volume 2
International Symposium on Computational Science 2011<br />
21-23 February 2011, Kanazawa University, Japan<br />
ISCS 2011 Organizing Committee<br />
M. A. Martoprawiro, S. Miura, H. Nagao, K. Nishikawa, S. Omata, M. Saito, R. Simanjuntak,<br />
Suprijadi, K. Svadlenka<br />
International Advisory Board<br />
E. T. Baskoro, M. A. Martoprawiro, S. Miura, H. Nagao, S. Omata, M. Saito, Suprijadi,<br />
K. Svadlenka<br />
Volume Editors<br />
� Rukman Hertadi<br />
_____________________________________________________<br />
Faculty of Mathematics and Natural Sciences, Bandung Institute of Technology, Indonesia<br />
� Shinichi Miura<br />
Institute of Science and Engineering, Kanazawa University, Japan<br />
� Rinovia Simanjuntak<br />
Faculty of Mathematics and Natural Sciences, Bandung Institute of Technology, Indonesia<br />
� Karel Svadlenka (contact person)<br />
Institute of Science and Engineering, Kanazawa University, Japan<br />
E-mail: kareru@staff.kanazawa-u.ac.jp<br />
ISSN 2223-0785<br />
Published by Kanazawa e-Publishing Co., Ltd<br />
© Organizing Committee of ISCS 2011<br />
Printed in Japan<br />
Typesetting: Camera-ready by author.<br />
Printed on acid-free paper.
Group Photo February 17, 2011 �
Preface<br />
The Fourth International Symposium on Computational Science (ISCS 2011) was held in Kanazawa,<br />
Japan, February 15-17 and continued and extended the traditions of previous conferences in the series:<br />
ISCS 2007 in Bandung, West Java, Indonesia; ISCS 2008 in Kanazawa, Japan; and ISCS 2009 in Sanur,<br />
Bali, Indonesia.<br />
The tradition of ISCS springs from the fact that it was born from the cooperation between Bandung<br />
Institute of Technology, Indonesia, and Kanazawa University, Japan, on the enhancement of graduate<br />
students’ education in the field of computational science and numerical mathematics. The aim of the<br />
symposium is to provide an occasion to present and exchange results and ideas related to these research<br />
fields not only for young researchers worldwide but also for senior professors and scientists.<br />
This year’s symposium was successful in the sense that more than 40 participants from 6 countries<br />
including several researchers from Bandung Institute of Technology and Kanazawa University joined the<br />
program. Moreover, 10 students participating in the Double-degree program between Kanazawa University<br />
and Bandung Institute of Technology also had a short talk and poster presentation about their research.<br />
Topic areas of the symposium were very rich, ranging from applied and numerical mathematics and<br />
development of new algorithms in computational science, complex system modeling and simulation,<br />
computational chemistry, up to quantum and parallel computations. This highlighted the rapid development<br />
of computational science and the flourishing international cooperation. We would like to thank all the<br />
speakers for their inspiring talks.<br />
From the submissions that ISCS 2011 received, 8 full papers were selected for this proceedings. The<br />
selected papers cover a wide variety of topics in Computational Science, such as advanced molecular<br />
dynamics computations, particle methods for fluid dynamics or simulation of crystals.<br />
We would like to thank all symposium organizers and local organizing committee, including the staff<br />
and students of Department of Computational Science at Kanazawa University for the excellent work and<br />
support of ISCS 2011. We extend our thanks also to the main organizers of the symposium, namely the<br />
Department of Computational Science, Faculty of Science, Kanazawa University and the Faculty of<br />
Mathematics and Natural Sciences, Bandung Institute of Technology for their financial help and constant<br />
support in making ISCS 2011 a success. We owe special thanks to our sponsors: Fujitsu Limited and Japan<br />
Society for the Promotion of Science (Grant-in-Aid No. 21654013 represented by prof. S. Omata) for their<br />
generous support.<br />
We invite you to visit the ISCS website (http://iscs.cc) to recount the events of the conference and to<br />
view the technical program dedicated to the fostering of Computational Science.<br />
i<br />
October 2011<br />
Editors
Compression Stress Effect on Dislocations Movement and<br />
Crack propagation in Cubic Crystal<br />
Suprijadi a, Ely Aprilia a, Meiqorry Yusfi b<br />
aFaculty of Mathematics and Natural Sciences, Institut Teknologi Bandung, Jl. Ganesha 10, Bandung 40132<br />
Indonesia. Email: supri@fi.itb.ac.id, elyaprilia@yahoo.co.id<br />
bDepartment of Physics, Andalas University, Kampus Limau Manis, Padang 25163, Indonesia.<br />
Email: meiqorry@yahoo.com<br />
Abstract. Fracture material is seriously problem in daily life, and it has connection with mechanical properties itself. The<br />
mechanical properties is belief depend on dislocation movement and crack propagation in the crystal. Information about this<br />
is very important to characterize the material. In FCC crystal structure the competition between crack propagation and<br />
dislocation wake is very interesting, in a ductile material like copper (Cu) dislocation can be seen in room temperature, but<br />
in a brittle material like Si only cracks can be seen observed. Different techniques were applied to material to study the<br />
mechanical properties, in this study we did compression test in one direction. Combination of simulation and experimental<br />
on cubic material are reported in this paper. We found that the deflection of crack direction in Si caused by vacancy of lattice,<br />
while compression stress on Cu cause the atoms displacement in one direction. Some evidence of dislocation wake in Si<br />
crystal under compression stress at high temperature will reported.<br />
Keywords: crack, dislocation, FCC, Cu, Si<br />
1 Introduction<br />
Fracture material is seriously problem in daily life, in the past and present the accident by a crack propagation<br />
growth rapidly cause of material ages. Many problem in the past was initiate by a small crack which is growth<br />
and propagate to whole body, for example in airplane crash Boeing 747-146SR which flight from Tokyo<br />
(Haneda) to Osaka (Itami) on 1985, the aircraft was crash at Mt. Osutaka by fatigue metal and crack propagation<br />
only less from 1 hour after take-off from Haneda[1], sink of 12 US Ships including SS John P. Gaines which sank<br />
on 24 November 1943 [2] and other big damage as in dam break [3]. Base on this problems, many investigation<br />
with different techniques were done to study the phenomena. In principle, almost of investigation technique<br />
using stress or shear, for example stress were use in three point bending [4], hardness test [5] and thin plate<br />
buckling experiment [6].<br />
Knowledge of material properties is useful in determining how the treatment is allowed in order not to<br />
damage the material. Loading which may result in damage to the materials is granted if the excessive load<br />
applied to them. If a material is given the excessive load will result in changes the properties of the materials.<br />
The ability of material in receiving excessive load that resulted in changes in shape or deformation makes the<br />
material can be classified into two parts, brittle and ductile [7]. In ductile material, dislocation wake is very easy<br />
to observe in room temperature, for example in steel [8] founded that dislocations more observed in unannealed<br />
steel crack propagation mechanism in Al thin film is reported [9], while in brittle material only crack were<br />
observed [10]. Increasing temperature on brittle material caused a dislocation wake, and a dislocation is believed<br />
can be occurred from a crack tip [11].<br />
To increase the understanding of material properties, a computation on failure analysis were done by many<br />
researchers using molecular dynamic methods [12], or other methods [13]. In this paper we did combination of<br />
computation and experiment to understanding those phenomena, the propagation of cracks and dislocation<br />
wake will be reported. Possibility of crack propagation and its competition with dislocation wake will be<br />
discussed.<br />
2. Experimental<br />
Experiments on real materials were done to see the crystal damage in cubic crystal. A 500µm x 4mm x 15mm<br />
thin plate of cubic crystal was pressed under continuous force with speed less than 1mm/minute as schematic in<br />
Fig.1. The sample (in this figure the sample in Copper (Cu)) will bend in the fixture as can be seen in Fig.1.b,<br />
Different material were used, Si with (111) plane is used as represent of brittle material, while Cu used as<br />
1
ISCS 2011 Selected Papers Vol.2 SUPRIJADI, E. APRILIA, M.YUSFI<br />
represent of ductile material. To observe the crystal deformation on surface, combination of etch pitch and<br />
optical microscope were applied.<br />
3. Simulation<br />
(a) Schematic (b) sample in special fixture<br />
Figure 1.Experiment apparatus<br />
In generally, the crack front will propagate if any atom loss their cohesion force, the atomic decohesion can be<br />
happen if there is an energy which bigger than critical energy. Solving for the critical energy release rate of<br />
atomic decohesion can be done completely by assuming an infinitesimal distribution of opening displacement<br />
running along the crack front develops under applied loading. The applied stress solutions are now the vertical<br />
tensile components of the stress solutions. The opening stress vs. opening displacement relationship is given by<br />
[13]<br />
where γs is the surface energy and L is the opening displacement at the peak of the stress displacement<br />
relationship. This formula must also be expressed in terms of δ, the displacement occurring at the plane of the<br />
crack front in order to solve the integral equilibrium equation expression. Varying the crack length and crack tip<br />
root radius demonstrates the effect of the pre-crack geometry parameters on the critical energy release rate of<br />
atomic decohesion.<br />
(a) Face Center Cubic (FCC) of Cu (b) (111) Si<br />
Figure 2, FCC lattice and crystal plane model<br />
2
ISCS 2011 Selected Papers Vol.2<br />
Compression effect in cubic crystal<br />
Different methods for study of material damage were applied. To study of dislocation in cubic metal<br />
the molecular dynamic based on Wei [12] algorithm was used in cooper (Cu) with FCC structure (Fig.1.a), the<br />
results was analyzed using mathematical tools freeware Scilab 5.2.2 (www.scilab.org). While for study crack<br />
propagation in brittle material was simulated using finite element in two dimensions Si with orientation in [111],<br />
the planes model can be seen in Fig.1.b. The [111] geometry choose because of simple structure and has the same<br />
length between closer atoms.<br />
Simulation on crack follows the Griffith criteria, that crack will propagate if stress in a direction higher<br />
than stress maximum, the algorithm for the crack simulation as follows:<br />
Start subprogram<br />
Initiation ;define pre-crack, F_external, MaterialLength<br />
calculate VectorPosition, CriticalForce<br />
While (F_external > CriticalForce)<br />
discritization<br />
while (VectorPosition < MaterialLength)<br />
find DOF ;degree of freedom<br />
find stress distribution each node<br />
set new VectorPosition<br />
end while<br />
end while<br />
end subprogram<br />
In simulation of crack propagation in brittle material, the Si will be use (Fig.2.b). Silicon has Young’s<br />
modulus (E)=18.5 x 10 11 dyne/cm 2 , Surface energy(γ) is 1000 erg/cm 2; Poison ratio (ν) is 0.26, and 5.43 Ǻ of<br />
crystal dimension.<br />
4. Results and discussion<br />
4.1 Brittle material<br />
Figure 3 shows the crack front of Si under stressed. In Fig.a, the experiment was done in temperature lower than<br />
transition temperature. We observed that only crack propagated along stress axis while increasing temperature<br />
just above the temperature transition [11], the dislocation start growth up, in Fig.3.b, shows TEM image of<br />
dislocation wake around the crack and crack tip as reported in [17].<br />
(a) cracks (b) dislocation wake<br />
Figure 3 Plastic deformations in Si under compression stress<br />
Simulation results that, the crack will propagate along stress direction, because of the (111) plane, we<br />
found that the crack propagated in zigzag pattern (Fig.4). To control the crack direction, at beginning, we setup<br />
3
ISCS 2011 Selected Papers Vol.2 SUPRIJADI, E. APRILIA, M.YUSFI<br />
the crack tip (Fig.2.b), different length of crack tip shows the crack easier deviated. Figure 4.a, shows the<br />
zigzagged crack by crack tip, deviated cracks more bigger in 100nm pre-crack (denote by (ii)) compare to 50nm<br />
denote by (i). All of the simulations did in perfect crystal, no vacancy as shown in shadow area in Fig.2.b.<br />
To find out effect of imperfect crystals, vacancy in some of node did in the middle of structure (Fig.2.b).<br />
The results show that increasing number of vacancy will make the crack easier to deviate. As reference, perfect<br />
crystal crack propagation (Fig. 4.a.) is used and in Fig.4.b, denoted by (i). Two atoms (ii) and four atoms (iii) was<br />
take out to give vacancy in the middle of structure (Fig.2.b). Increasing numbers of atom will bigger deflected as<br />
shown in Fig.4.b. The same problems were found in previous works as reported [7] as crack-step, the deflected<br />
crack were found in interface of Si3N4 and BN[14], the crack deflected cause of increasing energy during<br />
propagation in interface layer.<br />
(a) Crack tip effect (b) imperfect crystal effect<br />
Figure 4. Crack propagation in brittle material<br />
4.2 Ductile material<br />
Experiment on ductile material, we use the Cu, the sample was stressed in compression machine using continue<br />
stress with speed of 2 mm/minutes. The stress is stopped and release just after the sample buckle, it can be seen<br />
in stress-strain curve as below (Fig.5.a). In the center of buckling sample, we observed using optical microscope.<br />
Figure 5.b shows the image of dislocations in several places (see circles and denote by d1, d2, and d3), the<br />
dislocation occur in the same direction with stress direction (show by arrow). It is shown that the copper start<br />
deformation in room temperature under 1.938x10 8 N/m 2 (194 MPa).<br />
Figure 6, shows the result of molecular dynamic simulation in Cu under constant stress, different stress<br />
was applied start from 100 MPa to 4000 MPa. The figure shows the atoms configuration before pressed (a) and<br />
after pressed (b) the atoms were collected in the middle of cube, but it is not to easy to analyze, the atom<br />
displacement can be plot as Fig.4, it can be seen at higher stress, the atoms displacement increase rapidly. Atoms<br />
were distributed in one axis follow the compressing stress, the figure shows the dislocations start in area closed<br />
to surface and move to inner of cube. This phenomenon was reported in simulation of dislocation wake in bulk<br />
crystal [15].<br />
4
ISCS 2011 Selected Papers Vol.2<br />
Compression effect in cubic crystal<br />
(a) Stress-strain curve (b) optical microscope in Cu<br />
Figure 5 Experimental results of compressed copper<br />
(a) Before stressed (b) After stressed with 4000 MPa along z-axis<br />
Figure 6 Atoms distribution in cube metal<br />
The movement of the dislocation is made of a succession of motion of each straight segment, and the<br />
kinetics of motion of a straight segment depends upon the local effective stress and the line tension. The local<br />
effective stress is a sum of threshold stress, internal stress, external stress tensor, and image force tensor [15]. In<br />
this model, effect of external stress tensor is set to be main factor. Figure 5, shows the dislocation movement to<br />
central of the sample, it explained that the simulation give the same results as founded in other simulations<br />
method [15]. Figure 7.a, shows that if the stress applied not big enough, the dislocation only occur in small area<br />
close to surface where the stress concentrated. Different phenomenon were reported, the dislocation nucleated<br />
close to crack tip in Si at ductile brittle transition temperature (DBTT) [8] or in Cu [16], if there is a crack, the<br />
dislocation will starting occur just close to the tip, and spread in majority as stress direction applied.<br />
5
ISCS 2011 Selected Papers Vol.2 SUPRIJADI, E. APRILIA, M.YUSFI<br />
5. Conclusion<br />
(a) Stressed by 1000 MPa (b) Stressed by 4000 MPa<br />
Figure 5 Distributed of atoms displacement in Cu<br />
In brittle material, some vacancy atoms in a cubic crystal make the crack propagate in deviate direction along<br />
the compression direction. It can be shown that number of vacancies has tight relation with deviate angle of<br />
crack propagation, while the pre-crack play important role on crack propagation. In the ductile material, the<br />
dislocation wake starting from just beneath surfaces of material to middle of sample as found in [15], while the<br />
effect of compression direction has tight relation with dislocation wake.<br />
References<br />
[1] Kobayashi, H., Terada, H., (1985), Crash of Japan Airlines B-747 at Mt. Osutaka, Failure Knowledge<br />
Database, 1 - 10<br />
[2] Sawyer, L. A., Mitchell, W. H., (1970). The Liberty Ships: The history of the "emergency" type cargo ships<br />
constructed in the United States during World War II. Cambridge, Maryland: Cornell Maritime<br />
Press. ISBN 978-0-87033-152-7.<br />
[3] Biscarini,C., Di Francesco, S. ,Manciola,P., (2010), CFD modelling approach for dam break flow studies, Hydrol.<br />
Earth Syst. Sci., 14, 705–718<br />
[4] Chen, A-J, Cao J.J., (2010), Analysis of dynamic stress intensity factors of three-point bend specimen containing<br />
crack, Appl.Math.Mech.-Engl.Ed, 32 (2), 203 - 210<br />
[5] de la Fuente, O.R, Zimmerman, J.A., González, M. A., de la Figuera, J. , Hamilton, J. C., Pai, W.M, and<br />
Rojo, J. M., (2002), Dislocation Emission around Nanoindentations on a (001) fcc Metal Surface Studied by<br />
Scanning Tunneling Microscopy and Atomistic Simulations, Phys.Rev.Lett., 88, no.3,<br />
[6] Suprijadi, (2001), Transmission Electron Microscopy Study on Deformation and Fracture in Silicon Single<br />
Crystals, Nagoya University, Japan<br />
[7] Callister, Jr. William D.(2007), Material Science and Engineering . John Wiley & Sons, Inc.<br />
[8] Aguilar M.T.P., Correa, E.C.S, Monteiro, W.A., Cetlin, P.R., (2006), The effect of cyclic torsion on the<br />
dislocation structure of drawn mild steel, Materials Research, 9, 3<br />
[9] Suprijadi, Maula, R., Saka, H.,(2007), Observation of Micro Crack Propagation in a Thin Material,<br />
Indonesia Journal of Physics, 18 [2], 29 - 32<br />
[10] Saka,H, Nagaya, G., Sakuishi, T., Abe, S., (1999), Plan-view observation of crack tips in bulk materials by<br />
FIB/HVEM, Radiation Effects and Defects in Solids, 148 [1 & 4], 319 - 332<br />
[11] Suprijadi and H.Saka (1998), On the nature of a dislocation wake along a crack introduced in Si at the<br />
ductile-brittle transition temperature, Philosophy Magazine Letters., 78, 435 – 443<br />
[12] Wei,C., Bulatov, V.V., ( 2004), Mobility laws in dislocation dynamics simulations, Materials Science and<br />
Engineering A 387–389 , 277–281<br />
6
ISCS 2011 Selected Papers Vol.2<br />
Compression effect in cubic crystal<br />
[13] Fischer, L.L. and Beltz, G.E., (1999), Effect of crack geometry on dislocation nucleation and cleavage<br />
thresholds, Material Research Society Symposium. 39, 57-62<br />
[14] Desiderio Kovar,M. D. Thouless, and John W. Halloran*, (1998), Crack Deflection and Propagation in<br />
Layered Silicon Nitride/Boron Nitride Ceramics, Journal of American Ceramics Society, 81 [4] 1004–12<br />
[15] Verdieryk, M,Fivelz, M. and Gromax,I., (1998), Mesoscopic scale simulation of dislocation dynamics in fcc<br />
metals: Principles and applications, Modelling Simul. Mater. Sci. Eng. 6 , 755–770.<br />
[16] Zhu, T., Li, J., and Yip, S., (2004), Atomistic Study of Dislocation Loop Emission from a Crack Tip, Phys.<br />
Rev. Lett, 93 [2],<br />
7
Self-Siphon Simulation Using Molecular Dynamics Method<br />
SPARISOMA VIRIDI a, SUPRIJADI b, SITI NURUL KHOTIMAH a, NOVITRIAN a, FANNIA MASTERIKA c<br />
aNuclear Physics and Biophysis Research Division, Faculty of Mathematics and Natural Sciences,<br />
Institut Teknologi Bandung, Jl. Ganesha 10, Bandung 40132 Indonesia, E-mail: dudung@fi.itb.ac.id,<br />
nurul@fi.itb.ac.id, novit@fi.itb.ac.id<br />
bTheoretical High Energy Physics and Instrumentation Research Division, Faculty of Mathematics<br />
and Natural Sciences, Institut Teknologi Bandung, Jl. Ganesha 10, Bandung 40132 Indonesia, E-mail:<br />
supri@fi.itb.ac.id<br />
cMagister in Physics Teaching Program, Faculty of Mathematics and Natural Sciences, Institut<br />
Teknologi Bandung, Jl. Ganesha 10, Bandung 40132 Indonesia, E-mail: fannia_masterika@yahoo.com<br />
Abstract. A self activated siphon, which is also known as self-siphon or self-priming siphon, is simulated using<br />
molecular dynamics (MD) method in order to study its behavior, especially why it has a critical height that<br />
prevents fluid from flowing through it. The trajectory of the fluid interface with air in front of the flow or the<br />
head is also fitted the trajectory modeled by parametric equations, which is derived from geometry construction<br />
of the self-siphon. Numerical equations solved using MD method is derived from equations of motion of the<br />
head which is obtained by introducing all considered forces influencing the movement of it. Time duration<br />
needed for fluid to pass the entire tube of the self-siphon, τ, obtained from the simulation is compared<br />
quantitatively to the observation data from the previous work and it shows inverse behavior. Length of the three<br />
vertical segments are varied independently using a parameter for each segment, which are N5, N3, and N1.<br />
Room parameters of N5, N3, and N1 are constructed and the dependency of τ to these parameters are discussed.<br />
Keywords: self-siphon, critical height, duration time, molecular dynamics, room parameter<br />
1 Introduction<br />
A self-siphon is a siphon that can be primed by itself. The oldest version of this apparatus could be<br />
the one that is invented by Leonardo da Vinci [1], a cow horn-like shape siphon, as illustrated in<br />
Codex Leicester [2]. It can be used in wide range of applications, such as unique siphon with<br />
hydraulic capillary and self-venting as microfluidic tehnologies for nuleic acid extraction [3], as large<br />
siphon system to convey water over a dam or retaining wall [4], as part of wastewater treatment [5]<br />
and Hydro-Jet Screen [6], and as apparatus in stabilizing wastewater ponds [7] and treating milking<br />
parlour wastewater in cold climate [8]. Some applications do need the advantage of a self-siphon,<br />
such for subcritical supercritical flows [9]. Seeing the uses of selfsiphon, many patents based on its<br />
feature has been reported, such as a self-priming siphon filled with hydrophilic material for irrigation<br />
[10] and self-priming siphon for removing liquid from a dead storoge of a elevated storage tank [11].<br />
Many mechanisms have been used in inventing a self-siphon, which are where the flow is induced by<br />
creating a partial vacuum within priming chamber [12], where self-siphon is using one-way valve<br />
and must be inserted several times [13], where self-starting siphon has small bottle filled with air [14],<br />
or where self-siphon has several unequal length of bends [15, 16]. The last type of self-siphon will be<br />
discussed in this work. Experiments continuing previous work [17] have been conducted using<br />
segmented self-siphon and length variation for three vertical straight segments has been done [18].<br />
Simulations are performed to explain the phenomenon using molecular dynamics method, which is<br />
simple than using two-fluid model that has already implemented for flow in a siphon [19].<br />
9
ISCS 2011 Selected Papers Vol.2 S. VIRIDI, SUPRIJADI, S. N. KHOTIMAH, NOVITRIAN, F. MASTERIKA<br />
2 Simulations<br />
In order to make a model of the self-siphon, the real example in Figure 1 is mimiced also in several<br />
segments as illustrated in Figure 2. Three vertical straight segments, which are first, third, and fifth<br />
segment, have length that can be modified using several parameters: N1, N3, and N5. There are total<br />
seven segments, whose details are illustrated in Figure 2. Small segment in first, third, and fifth<br />
segment has lenght of w, which is chosen to be the same as in experiments, 2 cm. Other dimension<br />
parameters are R0 = R2 = R4 = 1 cm, L6 = 15 cm, θ0 = 0.75̟, θ2 = θ4 = ̟, and inner diameter of the tube d<br />
= 0.6 mm.<br />
Figure 1. Experimental configuration [17] of self-siphon indicated by numbers: 15 is zeroth and sixth<br />
segment, 11-14 is first segement, 10 is second segment, 6-9 is third segment, 5 is fourth segmen, and 1-<br />
4 is fifth segment. Notice that zeroth, second, and fourth segment already contribute one small<br />
segment to first, third, and fifth segment.<br />
As the self-siphon immersed into water, water will flow from inlet (fifth segment) to outlet<br />
(sixth segment). Time needed to accomplish all segments is defined as τ. Parameters N5, N3, and N1<br />
are varied and its influence on τ is observed.<br />
Equations of motion a head, the fluid interface with air in front of the flow, with mass m,<br />
cross section A, and thickness ∆h, is constructed, where forces applied on it is illustrated in Figure 3.<br />
There are two types of segment which have different equations of motion: straight segment and semicircular<br />
path. Fifth, third, first, and sixth segments are part of the first type of segment and zeroth,<br />
second, and fourth segments are part of the later type of segment.<br />
10
ISCS 2011 Selected Papers Vol.2<br />
Self-siphon simulation using molecular dynamics method<br />
Figure 2. Model of self-siphon with length variation in fifth, third, and first segment indicates by set<br />
of number, N5, N3, and N1, which in this case N5 = 2, N3 = 3, and N1 = 4.<br />
Friction of the head with the tube is defined as f, which is<br />
f = −8<br />
πη∆hv<br />
, (1)<br />
as modified from Poiseuille equation in [20]. Equation (1) holds for both type of segments. For the<br />
vertical straight segments the force equations are<br />
0 = max<br />
(2)<br />
and<br />
− mg + ρ g y − y A − f = ma . (3)<br />
( w ) y<br />
And for the semi-circular segments<br />
− mg sin θR + ρg(<br />
yw<br />
− y)<br />
AR − fR = Iα<br />
,<br />
with<br />
(4)<br />
2<br />
I = mR , (5)<br />
v<br />
ω = ,<br />
R<br />
(6)<br />
m = ∆hρA<br />
,<br />
Equation for time evolution of motion parameters are as follow<br />
(7)<br />
vi ( t + ∆t)<br />
= vi(<br />
t)<br />
+ ai∆t,<br />
i = x,<br />
y , (8)<br />
ri ( t + ∆t)<br />
= ri(<br />
t)<br />
+ vi∆t,<br />
i = x,<br />
y , (9)<br />
ω ( t + ∆t)<br />
= ω(<br />
t)<br />
+ α∆t<br />
, (10)<br />
θ ( t + ∆t)<br />
= θ(<br />
t)<br />
+ ω∆t<br />
, (11)<br />
Based on Equation (2) in the vertical segment the water does not flow in x-direction. In order to show<br />
whether the results in solving Equation (3) and (4) with MD simulation through Equation (8) – (11),<br />
there is a need to plot rx, ry, vx, vy, and t. It means that such relations<br />
0 cosθ<br />
R x x = + , (12)<br />
sinθ<br />
R y y = + , (13)<br />
0<br />
11
ISCS 2011 Selected Papers Vol.2 S. VIRIDI, SUPRIJADI, S. N. KHOTIMAH, NOVITRIAN, F. MASTERIKA<br />
are needed, for the semi-circular segments. Parameters x0 and y0 are center of the semi-circular<br />
segments.<br />
Figure 3. Diagram of considered forces applied to water element with thickness ∆h: (a) in a vertical<br />
straight segment and (b) in circular segment.<br />
The simulations are performed until the flow reaches the outlet from the inlet or when there<br />
is no flow, it must be ended when the flow changes its direction in a current segment, where is<br />
flowing.<br />
3 Results and discussion<br />
Configuration of (N5, N3, N1) = (4, 1, 4) is chosen to see whether the flow can follow the trajectory of<br />
self-siphon as it proposed in [17] using parametric equations. The result of position of the head as<br />
function of time is shown in Figure 4 and also the trajectory of self-siphon.<br />
For this 414 configuration the fifth segment is passed between 0 s and 0.000928 s, the fourth<br />
segment is passed between 0.000928 s and 0.0010867 s, the third segment is passed between 0.0010867<br />
s and 0.0014116 s, the second segment is passed between 0.0014116 s and 0.0015913 s, the first<br />
segment is passed between 0.0015913 s and 0.0021186 s, the zeroth segment is passed between<br />
0.0021186 s and 0.0022143 s, and the sixth segment is passed between 0.0022143 s and 0.002792 s. It<br />
means the for 414 configuration τ = 0.002792 s.<br />
And for the velocity evolution in time, it is illustrated in Figure 5. In the chart of vy – vx, the<br />
direction of the head can be seen for every segment, more clearer as it seen in the chart of ry – rx.<br />
These results are obtained by using parameters value: ∆h = 0.01, ∆t = 10–7, η = 10 –8. Variation of N5,<br />
N3, N1 is conducted and 125 variations are noted, each from 0 to 4 for simulations as shown in Figure<br />
6. The influece of N1, N3, N5, independently to τ is also simulated and presented in Figure 7. It can be<br />
seen that as N3 increasing the value of τ is also increasing, but as N1 increasing value of τ is<br />
decreasing, this both trend do not match previously reported result [17]. Value of τ from simulations<br />
compared to previous result in [17] is 500 times lower as it should be. It means that the MD<br />
simulations need a scaling process or the implementation of Equation (1) needs more refinement<br />
since it depends on value of ∆h which is adjustable. Precise value of ∆h has not yet been investigated.<br />
A height of inlet measured from water surface can be defined as<br />
= y − ( N − N + N )w , (13)<br />
yinlet w 1 3 5<br />
12
ISCS 2011 Selected Papers Vol.2<br />
Self-siphon simulation using molecular dynamics method<br />
and then from Figure 6 it can seen that, especially from Figure 6 (e) that the flow can occur when<br />
0. 5N<br />
1 + 1 − N3<br />
> 0,<br />
N5<br />
= 4 . (14)<br />
Current results [18] show other condition than it is suggested by Equation (14) for the flow to occur.<br />
Using Equation (14) into Equation (13) gives about<br />
yinlet < yw<br />
− 3w<br />
, (15)<br />
which is the condition that the self-siphon can flow the water. A critical height than can be defined<br />
from Equation (15) as<br />
y = yw<br />
− 3w<br />
. (16)<br />
critical<br />
Figure 4. Position of water element for configuration of (N5, N3, N1) = (4, 1, 4): r - t with symbols □<br />
and ○ indicate x and y, respectively (left) and y - x (right).<br />
Figure 5. Velocity of water element (in m/s) for configuration of (N5, N3, N1) = (4, 1, 4): v – t with<br />
symbols □ and ○ indicate vx and vy, respectively (left) and (b) vy - vx (right).<br />
Equation (16) is a requirement for the self-siphon to be able to flow the water. Height of inlet<br />
measured from water surface must be lower than this value.<br />
13
ISCS 2011 Selected Papers Vol.2 S. VIRIDI, SUPRIJADI, S. N. KHOTIMAH, NOVITRIAN, F. MASTERIKA<br />
(a) (b) (c)<br />
(d) (e)<br />
Figure 6. Variation parameter of N3 and N1 for different value of N5: (a) 0, (b) 1, (c) 2, (d) 3, and (e) 4.<br />
The symbols □ and ● indicate no flow and flow occurs, respectively, as predicted by MD<br />
simulations.<br />
14
ISCS 2011 Selected Papers Vol.2<br />
Self-siphon simulation using molecular dynamics method<br />
Figure 7. Dependency of time needed for water to flow from inlet to outlet of the self-siphon, τ (in s),<br />
to number of element in certain straight segment: N1 (upper left), N3 (upper right), and N5 (lower<br />
center).<br />
4 Conclusions<br />
An model for simulating self-siphon flow using MD has been built. From the results it can be<br />
concluded that when water or fluid can flow, τ is increasing as N5 and N1 decreasing, but it is<br />
increasing as N3 increasing. The last two results, on which τ dependent is, shows inverse behavior as<br />
previously reported in experiment. A critical height is also defined as the maximum value of inlet<br />
height, where the self-siphon still can flow the water.<br />
Acknowledgment<br />
Authors would like to thank International Conference Grant IMHERE ITB and cooperation between<br />
Kanazawa University and Institut Teknologi Bandung in 2011 fur supporting presentation of this<br />
work, cooperation between Institut Teknologi Bandung and Ministerium of Religion Affair of<br />
Republic of Indonesia in 2009-2011 for supporting experiment part cited by this work, and Institut<br />
Teknologi Bandung Alumni Association Research Grant in 2010 for supporting simulation part of this<br />
work.<br />
15
ISCS 2011 Selected Papers Vol.2 S. VIRIDI, SUPRIJADI, S. N. KHOTIMAH, NOVITRIAN, F. MASTERIKA<br />
References<br />
[1] E. Macagno (1991), Some remarkable experiments of Leonardo da Vinci, Issue La Houille<br />
Blanche 6, 463 - 471.<br />
[2] L. da Vinci (1991), Codex Leicester, Corbis, reproduced in multimedia form.<br />
[3] J. Siegrist, R. Gorkin, M. Bastien, G. Stewart, R. Peytavi, H. Kido, M. Bergeron, and M. Madou<br />
(2010), Validation of a centrifugal microfluidic sample lysis and homogenization platform for<br />
nucleic acid Extraction with clinical samples”, Lab on A Chip 10, 363-371.<br />
[4] C. W. Krusré and R. M. Lesaca (1955), Automatic siphon for the Control of Anopheles<br />
minimus var. flavirostris in the Philippines”, Am. J. Hygienise 61, 349-361 .<br />
[5] Michael G. Faram, Christopher A. Williams, and Keith G. Hutchings (2002), Storm overflow<br />
screening at wastewater treatment works, 2nd Biennial Conference on Management of Wastewaters<br />
(CIWEM/AETT), Edinburgh, Scotland, 14-17 April, 71 - 75.<br />
[6] Michael G. Faram, Robert Y. G. Andoh, and Bruce P. Smith (2001), Optimised CSO screening:<br />
a UK perspective, Novatech 2001: 4th International Conference on Innovative Technologies in Urban<br />
Drainage, Lyon, France, 25-27 June, 1031-1034.<br />
[7] Catherine Boutin, Alain Liénard, Nathalie Bilotte, and Jean-Pierre Naberac (2002), Association<br />
of wastewater stabilisation ponds and intermittent sand filters: the pilot results and the<br />
demonstration plant of Aurignac, 8th International Conference on Water Pollution Control,<br />
Arusha, Tanjanie, 16-19 September.<br />
[8] Kunihiko Kato, Tashinobu Koba, Hidehiro Ietsugu, Toshiya Saigusa, Takuhito Nozoe, Sohei<br />
Kobayashi, Katsuji Kitagawa, and Shuji Yanagiya (2006), Early performance of hybrid reed<br />
bed system to treat milking rarlour wastewater in cold climate in Japan, a poster, Lisbon,<br />
September.<br />
[9] H. A. Senturk, V. M. Basak, and T. Sahin (1978), New syphon for subcritical and supercritical<br />
flows”, J. Irrigation Drain. Div. 104, 442-446<br />
[10] Maurice Amsellem (2001), Self-priming siphon, in particular for irrigation, United States Patent<br />
6,178,984.<br />
[11] John N. Pirok an Joel H. Trammel (1962), Self-priming siphon drain, United States Patent<br />
3,019,806.<br />
[12] John H. Rice (1978), Self priming siphon, United States Patent 4,124,035.<br />
[13] Liu Songzeng (1991), Self-fill siphon pipes”, United States Patent 4,989,760.<br />
[14] R. M. Sutton (2003), Demonstration experiments in physics, American Association of Physics<br />
Teachers.<br />
[15] A. Adair (1945), Small-scale experiments for school classes, J. Chem. Edu. 22, 129.<br />
[16] M. Gadner and A. Ravielli (1981), Entertaining science experiments with everyday object, Courier<br />
Dover Publication.<br />
[17] Fannia Masterika, Novitrian, and Sparisoma Viridi (2010), Self-siphon experiments and its<br />
mathematical modeling using parametric equation, Proceeding of Conference on Mathematics and<br />
Natural Sciences 2010, 23-25 November, Bandung, Indonesia (in press).<br />
[18] Fannia Masterika, Novitrian, dan Sparisoma Viridi (2011), Eksperimen aliran fluida<br />
menggunakan self-siphon, Prosiding Seminar Nasional Inovasi Pembelajaran dan Sains 2011, 22-23<br />
Juni, Bandung, Indonesia (accepted).<br />
[19] Akihiko Minato, Takuji Nagayoshi, Kazuhide Takamori, Ichirou Harad, Masahiro Mase, and<br />
Kenji Otani, Numerical simulation of gas-fluid two-phase flow in siphon outlets of pumping<br />
plants”, a paper.<br />
[20] T. E. Faber (1995), Fluid dynamics for physicist, Cambridge University Press.<br />
16
Time benchmarks for the OpenMP and GPU parallelized calculation<br />
in the planewave pseudopotential density functional approach<br />
Junpei Gotou, a Shinya Haraguchi, a Masahito Tsujikawa, a Tatsuki Oda b<br />
a Graduate School of Natural Science and Technology, Kanazawa University, Kanazawa 920-1192,<br />
Japan, E-mail: gotou@cphys.s.kanazawa-u.ac.jp<br />
b Institute of Science and Engineering, Kanazawa University, Kanazawa 920-1192, Japan,<br />
E-mail: oda@cphys.s.kanazawa-u.ac.jp<br />
Abstract. We have investigated the computational performance of the first principles molecular<br />
dynamics code which employs planewave basis set and pseudopotential of electron-ion interaction.<br />
The efficiency of task has been measured in terms of matrix multiplication (MM), fast Fourier<br />
transformation (FFT). For the implementations of the OpenMP and GP-GPU utilities, the MM<br />
was found to mark the considerable high score, whereas the FFT a fairly good one. Analyzing the<br />
resulting whole performance, efficiency of the task other than the MM and FFT has to be improved<br />
for a higher performance.<br />
Keywords: Car-Parrinello molecular dynamics, first-principles electronic structure calculation,<br />
parallelization, OpenMP, GP-GPU<br />
1 Introduction<br />
Electronic properties in the realistic system with a nano-size scale have been becoming a target<br />
in the computational material science which explicitly includes effects on the quantum mechanics.<br />
Such a calculation needs a large amount of computational source. This is why the higher computational<br />
efficiency is required. The density functional approach has been playing an important<br />
role in material science. Indeed, the target extends to many topics associated with the increase<br />
of performance in both of hardware and software. The extension has been supported by parallel<br />
computations, which are performed in many levels of computer wares.<br />
It is important to use the parallel calculation code which is matched with its architecture.<br />
Even when we develop the computational code for the density functional calculation, some tests<br />
for new architectures are required for grasping a future efficiency in the computation. Following<br />
the message passing interface (MPI), the OpenMP has been one of the central issues for the<br />
improvement in the current computational code. Recently, the general purpose graphic processing<br />
unit (GP-GPU) is also a choice for getting a higher performance. The test of hybrid calculations<br />
could also be meaningful.<br />
In this work, we have investigated the performance about computational time in the pseudopotential<br />
planewave code for DFT calculation. The results on the matrix multiplication (MM) and<br />
fast Fourier transformation (FFT) are reported individually in detail in the implementations of<br />
the OpenMP and GP-GPU.<br />
2 Computational circumstances<br />
2.1 Properties of computer archtechture<br />
We have used three kinds of machines for the performance investigation. Their properties are<br />
summarized in Table 1. The names of machines are referred in the text.<br />
17
ISCS 2011 Selected Papers Vol.2 J. Gotou, S. Haraguchi, M. Tsujikawa, T. Oda<br />
Table 1: Properties of the computers for investigation. The machine names are used in the text.<br />
Machine Name CPU GPU CUDA<br />
M1 Quad-Core,AMD,2.3GHz no device no install<br />
M2 Intel Xeon X5590,3.33GHz Tesla C1060 CUDA Toolkit 2.3<br />
M3 Intel Xeon X5680,3.33GHz Tesla M2050 CUDA Toolkit 3.1 and 3.2<br />
The OpenMP is a parallel programming language by which the memory is shared in a node of<br />
the machine. The message communication like MPI is unnecessary in the OpenMP calculation.<br />
It is also unnecessary to make a major correction when rewriting the single processing code to a<br />
parallelized one. However, the execution may not be secure even when there is no error. When<br />
using two or more threads in the OpenMP calculation, the correspondence between a temporal<br />
thread memory and real memory may not be secured. Therefore, it is required to make attension<br />
to the synchronization between them. The speed-up factor is used for presenting parallelization<br />
performances. This factor is estimated from the inversed ratio of the time by n’s processors with<br />
respect to that by single one. Originally, GPU is a special set of many processors for graphical<br />
processes. The CUDA Toolkit developed by NVIDIA has become available to the computation for<br />
general purpose.<br />
2.2 Properties of the target code<br />
The calculation code used in the investigation is for the Car-Parrinello molecular dynamics (CPMD),<br />
which is based on the density functional theory (DFT). Details of the scheme can be found in the<br />
reference [1, 2]. The wave functions of dynamical variables are transformed by FFT between real<br />
and inverse spaces. In the ultrasoft pseudopotential scheme, the MM can be a considerable part for<br />
the time-consuming. In such a scheme of the DFT, the task of calculation is categorized to MM,<br />
FFT and the others. The code used can employ a scheme of noncollinear magnetism, affording to<br />
treat vector type spin density. Such scheme requires twice of memory for wave functions due to<br />
the two component spinor [3]. The starting code is parallelized in k points (sampling wave vectors<br />
in wave functions) with using MPI.<br />
2.3 Sample systems<br />
For investigating the computational time, the two sample systems have been considered. Both<br />
of them are slab systems; one is non-magnetic and the other magnetic with noncollinear scheme<br />
[3, 4]. Results of the former can be used for comparison with other codes and those of the latter<br />
may provide us the perspective to the application to the magnetic system with the spin-orbit<br />
interactions [5, 6]. The non-magnetic system, as shown in Fig. 1, consists of 13 monolayers (MLs)<br />
in BaTiO3 with the inplane lattice constant (a = 3.94˚A); 7 BaO MLs and 6 TiO2 MLs (32 atoms<br />
in unit cell). In both sides of the slab, the vacuum layer with dv = 5.29˚A is included. The<br />
periodic boundary condition is taken into account due to the planewave basis in which the energy<br />
cut-offs of 30Ry and 300Ry were used for the wavefunction and the density, respectively. The<br />
6 special k points are used for MPI parallel computations. The magnetic system is a thin film<br />
Fe(5MLs)/MgO(6MLs)/Fe(5MLs) like a magnetic tunnel junction (22 atoms in unit cell, a = 4.20˚A,<br />
the vacuum layer of dv = 5.29˚A, the energy cut-offs of 30Ry and 300Ry, and 4 special k points).<br />
18
ISCS 2011 Selected Papers Vol.2 Time Benchmarks for the OpenMP and GPU<br />
dv<br />
dv<br />
Figure 1: Non-magnetic and magnetic slab systems for<br />
the performance invesitigation; 13 monolayers (MLs)<br />
of BaTiO3 (left) and the magnetic tunnel junction of<br />
Fe(5MLs)/MgO(6MLs)/Fe(5MLs) (right). The dv indicates<br />
the vacuum layer.<br />
The parameters used in this work are reported in Table 2.<br />
In these systems, the main parts for time consuming are MMs and FFTs. In the magnetic<br />
system, the MM and FFT needs the 70% and 20% of the total cpu time in the single processing<br />
calculation (without using MPI, OpenMP, and GPU). In this paper, we report the result in<br />
computational performance for OpenMP parallelization and GP-GPU computation.<br />
3 Testings for the elements<br />
3.1 Matrix multiplication (MM)<br />
For the OpenMP parallelization, the MM is performed as shown in Fig. 2. Figure 3 shows<br />
parallelization efficiency for matrix multiplication. Using 8 cores, we obtain a high parallelization<br />
efficiency of 94%. It was confirmed that the MM can be parallelized with a high efficiency.<br />
The MM can be performed using the routines of xGEMM (x =S,D,C,Z) for the typical matrices<br />
in the BLAS library. The CUDA Toolkit provides the library for the GPU. Figure 4 shows the<br />
FLOPS (floating point number operations per second) in the MM for the various matrix sizes<br />
[7]. The times of data transfer from the host computer to the device (GPU) and vice versa were<br />
included for the estimation. For the architecture of Tesla C1060, the performance is not high<br />
except for the single real matrices (SGEMM and CGEMM), whereas the performance becomes to<br />
a much higher one in Tesla M2050 even for DGEMM and ZGEMM. The performance has become<br />
to a high value in archiecture of Tesla M2050. The comparison with the performance of OpenMP<br />
(see the data of ”8 thread” in Fig. 4) implies that the performance of CPMD becomes higher in<br />
the GPU of Tesla M2050 instead of the OpenMP.<br />
19<br />
dv<br />
dv
ISCS 2011 Selected Papers Vol.2 J. Gotou, S. Haraguchi, M. Tsujikawa, T. Oda<br />
Table 2: Characters and Parameters in sample systems.<br />
Sample BaTiO3 Fe/MgO/Fe<br />
lattice tetragonal lattice tetragonal lattice<br />
lattice constants a=3.94˚A, c=34.5˚A a=2.98˚A, c=37.1˚A<br />
number of k point 6 special points 4 special points<br />
number of atoms 32 (Ba:7, Ti:6, O:19) 22 (Fe:10, Mg:6, O:6)<br />
number of Kohn-Sham orbitals 148 264<br />
Energy Cutoff for wave function 25Ry 25Ry<br />
Energy Cutoff for density 300Ry 300Ry<br />
Fourier mesh for wave function 24×24×216 18×18×240<br />
Fourier mesh for density 48×48×360 32×32×400<br />
number of plane waves 7697 4731<br />
Figure 2: Parallelization schemes of matrix multiplication<br />
(MM) in the OpenMP parallelization.<br />
speedup factor<br />
8<br />
7<br />
6<br />
5<br />
4<br />
3<br />
2<br />
1<br />
0<br />
0 1 2 3 4 5 6 7 8<br />
Num.of Processors<br />
Figure 3: The parallelization efficiency (speed-up factor)<br />
of the matrix multiplication (MM) in the OpenMP<br />
of M1 for the size of 5000×5000.<br />
In the GPU calculation, the data transfer (host to device and vice versa) is time-consuming.<br />
The typical ratios are presented in Fig. 5. In the larger sizes of matrix, the ratios of data transfer<br />
become suppressed. This implies the efficiency of GPU in the application of larger systems. It<br />
is interesting to see that the transfer from the host (CPU) to the device (GPU) needs more time<br />
compared with that from the device to the host. Such feature of results is attributed to the amount<br />
of data transfer and the bandwidth for up- and down-loading between the host and device.<br />
3.2 Fast Fourier transformation (FFT)<br />
In the target code, the physical value (wave functions and densities) is efficiently calculated by using<br />
one of real and reciprocal spaces [2]. Requiring the conversion between them, three dimensional<br />
FFT (3D-FFT) is used. The parallelization scheme of OpenMP for 3D-FFT is shown in Fig. 6. In<br />
this, the 3D-FFT is divided to three times of 1D-FFTs so that the number of calculation tasks is<br />
increased and we obtain the good load balance among the threads associated with the OpenMP.<br />
This sometimes works well because in the application systems of typical size of CPMD calculation,<br />
the size of FFT is not enough large. Such sizes are only several times or several score times of the<br />
20
ISCS 2011 Selected Papers Vol.2 Time Benchmarks for the OpenMP and GPU<br />
GFLOPS<br />
GFLOPS<br />
350<br />
300<br />
250<br />
200<br />
150<br />
100<br />
50<br />
0<br />
300<br />
250<br />
200<br />
150<br />
100<br />
50<br />
0<br />
(a) real number (M2)<br />
DGEMM(8 Threads)<br />
DGEMM(Tesla C1060)<br />
SGEMM(Tesla C1060)<br />
Ratio(Single/Double)<br />
256<br />
512<br />
1024 2048<br />
Matrix size<br />
4096<br />
(c) real number (M3)<br />
Tesla C1060<br />
Tesla M2050 with cuda3.1<br />
Tesla M2050 with cuda3.2<br />
256<br />
512<br />
1024 2048<br />
Matrix size<br />
4096<br />
5120<br />
5120<br />
7<br />
6<br />
5<br />
4<br />
3<br />
2<br />
1<br />
0<br />
RATIO<br />
GFLOPS<br />
GFLOPS<br />
350<br />
300<br />
250<br />
200<br />
150<br />
100<br />
50<br />
0<br />
300<br />
250<br />
200<br />
150<br />
100<br />
50<br />
0<br />
(b) complex number (M2)<br />
ZGEMM(8 Threads)<br />
ZGEMM(Tesla C1060)<br />
CGEMM(Tesla C1060)<br />
Ratio(Single/Double)<br />
256<br />
512<br />
1024 2048<br />
Matrix size<br />
4096<br />
5120<br />
(d) complex number (M3)<br />
Tesla C1060<br />
Tesla M2050 with cuda3.1<br />
Tesla M2050 with cuda3.2<br />
256<br />
512 1024<br />
Matrix size<br />
2048<br />
Figure 4: Effective FLOPS values of the MM for various matrix sizes (256 ∼ 5120) by SGEMM, CGEMM, DGEMM<br />
and ZGEMM routines in Tesla GPUs (Tesla C1060 and Tesla M2050). The upper and lower two panels are obtained<br />
in the machines of M2 (a)(b) and M3 (c)(d) (see Table 1), respectively. The times for memory transfer are included<br />
in the estimation.<br />
Memory Transfer Rate [%]<br />
100<br />
80<br />
60<br />
40<br />
20<br />
0<br />
MM<br />
DtoH<br />
HtoD<br />
37.5%<br />
14.6%<br />
47.8%<br />
128<br />
51.8%<br />
11.3%<br />
36.8%<br />
256<br />
56.9%<br />
20.1%<br />
22.9%<br />
1024<br />
79.2%<br />
8.6%<br />
12.2%<br />
4096<br />
Figure 5: Data transfer ratio in DGEMM calculation for<br />
various matrix sizes in M3.<br />
21<br />
4096<br />
6<br />
4<br />
2<br />
0<br />
RATIO
ISCS 2011 Selected Papers Vol.2 J. Gotou, S. Haraguchi, M. Tsujikawa, T. Oda<br />
Figure 6: Parallelization scheme for three dimensional<br />
fast Fourier transformation (3D-FFT) in the OpenMP.<br />
speedup factor<br />
8<br />
7<br />
6<br />
5<br />
4<br />
3<br />
2<br />
1<br />
0<br />
0 1 2 3 4 5 6 7 8<br />
Num.of Processors<br />
Figure 7: The parallelization efficiency (speed-up factor)<br />
of 3D-FFT in the OpenMP. The size of 48 × 48 ×<br />
216 is used.<br />
number of available threads in OpenMP.<br />
The parallelization efficiency in OpenMP is presented in Fig. 7. Using 8 cores, we obtain<br />
the efficiency of 65%. The parallelization efficiency has decreased compared with the MM (see<br />
Fig. 3). This may be because in the scheme the temporal memory for 1D-FFT, as the overhead<br />
of calculation, is stored discontinuously from the original memory area of 3D-FFT. As another<br />
investigation for 3D-FFT, the automatic parallelized version of FFTW [8] in OpenMP are tested<br />
(not reported here), resulting the parallelization efficiency of 25% for using 8 cores.<br />
In this work, we have carried out the test of CUFFT for 3D-FFT. The computational times are<br />
presented in Fig. 8. The efficiency is found to depend on the size of FFT (spiky structure in the<br />
figure). There are many sizes where the efficiency is worse than that of FFTW [8]. This implies<br />
the 3D-FFT by CUFFT is not appropriate for the CPMD code in which the size of FFT depends<br />
on the input parameters.<br />
4 Results on CPMD code<br />
4.1 MPI+OpenMP (Hybrid)<br />
The hybrid version of code is organized with k point parallelization on MPI, the OpenMP calculation<br />
both in MM and FFT, and other minor parallelizations. The efficiencies are presented<br />
with those of the MPI version in Fig. 9. Using 16 cores (2 MPI+ 8 OpenMP), the performance<br />
amounts to 37%. This value is still fearly well, but the additional improvement must be required<br />
for an more efficient calculation.<br />
4.2 MPI+GPU<br />
The MPI+GPU version of CPMD code has been checked mainly in M2, which has 8 cores and 3<br />
GPUs. The BaTiO3 system is used for the computation. The time report is presented in Figs. 10<br />
and 11. In the latter, the times for MM, FFT are included. In these figures, the labels of BLAS<br />
and FFTW indicate the single processing calculation for the MM and the FFT, respectively. The<br />
22
ISCS 2011 Selected Papers Vol.2 Time Benchmarks for the OpenMP and GPU<br />
Elapsed time (sec)<br />
5<br />
4<br />
3<br />
2<br />
1<br />
FFTW<br />
CUFFT(DOUBLE)<br />
CUFFT(S<strong>IN</strong>GLE)<br />
0.02<br />
0.01<br />
0<br />
0 50 100<br />
0<br />
0 100 200 300 400<br />
Transform Size (N^3)<br />
Figure 8: The computational times with respect to the<br />
various sizes for the CUFFT routine of 3D-FFT (GPU)<br />
in M2.<br />
Elapsed time (sec)<br />
70<br />
60<br />
50<br />
40<br />
30<br />
20<br />
10<br />
MPI + GPU<br />
MPI(BLAS+FFTW)<br />
MPI(CUBLAS+CUFFT)<br />
MPI(CUBLAS+FFTW)<br />
0<br />
0 1 2 3 4 5 6<br />
Num.of Processors<br />
Figure 10: Conputational times (single CPMD iteration<br />
in second) with respect to the number of MPI<br />
threads for MPI+GPU (Tesla C1060) version. The<br />
single MPI thread can access to a single GPU in M2.<br />
Speed Up Factor<br />
16<br />
14<br />
12<br />
10<br />
8<br />
6<br />
4<br />
2<br />
6<br />
4<br />
2<br />
MPI + OpenMP<br />
MPI<br />
0<br />
0 2 4 6<br />
0<br />
0 2 4 6 8 10 12 14 16<br />
Num.of Processors<br />
Figure 9: Parallelization efficiency for the MPI and<br />
Hybrid versions for BaTiO3 system.<br />
Elapsed time (sec)<br />
35<br />
30<br />
25<br />
20<br />
15<br />
10<br />
5<br />
0<br />
6.30<br />
5.09<br />
20.0<br />
33.9<br />
2.35<br />
16.6<br />
2.08<br />
21.0<br />
Other<br />
FFT<br />
Matrix<br />
8.70<br />
2.02<br />
4.45<br />
2.23<br />
FFTW+BLAS CUFFT+CUBLAS FFTW+CUBLAS<br />
Figure 11: Conputational times (single CPMD iteration<br />
in second) of MM, FFT, and others for BaTiO3<br />
by MPI+ GPU (Tesla C1060) version (3MPI+3GPU).<br />
GPU calculation for the MM (CUBLAS) is improved considerably, whereas the FFT by CUFFT<br />
is more time-consuming than the FFTW, as expected from the result in Fig. 8.<br />
4.3 MPI+OpenMP+GPU<br />
In the experience which is encountered in the previous sections, the FFT by GPU cannot be carried<br />
out so high performance. Taking into account this, the MPI+OpenMP+GPU (hybrid+GPU)<br />
version is organized with the MPI (k point parallelization), the MM by GPU, the FFT by OpenMP<br />
(6 threads in OpenMP), and other minor parallelization. This is tested in the magnetic system<br />
(Fe/MgO/Fe) with M3 (4 threads in the MPI and 2 GPUs). The result with the CUDA Toolkit<br />
23
ISCS 2011 Selected Papers Vol.2 J. Gotou, S. Haraguchi, M. Tsujikawa, T. Oda<br />
Elapsed time (sec)<br />
25<br />
20<br />
15<br />
10<br />
5<br />
0<br />
4.98<br />
4.44<br />
13.9<br />
23.3<br />
MPI<br />
4.87<br />
1.41<br />
2.75<br />
9.03<br />
Hybrid<br />
5.87<br />
4.09<br />
1.79<br />
Other<br />
FFT<br />
Matrix<br />
12.08<br />
4.71<br />
1.63<br />
1.93<br />
7.97<br />
MPI+GPU Hybrid+GPU<br />
Figure 12: Computation times (single CPMD iteration in<br />
second) for the slab of Fe(5MLs)/MgO(6MLs)/Fe(5MLs)<br />
in M3.<br />
3.1 is represented in Fig. 12. In the MPI+GPU calculation, the FFTW is used in the 3D-FFT for<br />
each k point sampling. The hybrid+GPU calculation marks the highest score among the versions<br />
investigated. By introducing OpenMP and GPU, the time ratio of MM has become to 24% from<br />
70% in the single processing calculation and from 60% in the calculation. If preparing one GPU per<br />
MPI thread, instead of one GPU per two MPI threads, the MM can be accelarated. Introducing<br />
the MPI, OpenMP, and GPU to the CPMD code, the computation other than the MM and FFT<br />
becomes to the major part of time-consuming. It should be improved in future.<br />
5 Conclusions<br />
We have investigated computational performance of the CPMD code for the OpenMP parallelization<br />
and the GPU architecture in addition to the MPI parallelization. Our test has succeeded to<br />
reduce the computing time considerably, compared with the MPI version for the magnetic slab<br />
system which is an accesible model for typical magnetic tunnel junction. Based on such development<br />
of the code which accords with the new architectures, we may approach electronic structures<br />
and molecular dynamics for larger realistic systems with the high performance computation.<br />
Acknowledgment<br />
One of authors (T.O.) would like to thank the Japan Society for the Promotion of Science (JSPS) for<br />
financial supports (Grant 22104012, 22360014, and 22340106). One of authors (M.T.) acknowledges<br />
the JSPS Research Fellowships (Grant No.20-6647) for Young Scientists.<br />
References<br />
[1] D. Vanderbilt (1990). Soft self-consistent pseudopotentials in a generalized eigenvalue formalism,<br />
Phys. Rev. B, 41, 7892-7895.<br />
24
ISCS 2011 Selected Papers Vol.2 Time Benchmarks for the OpenMP and GPU<br />
[2] K. Laasonen, A. Pasquarello, R. Car, C. Lee, and D. Vanderbilt (1990). Car-Parrinello molecular<br />
dynamics with Vanderbilt ultrasoft pseudopotentials, Phys. Rev. B, 47, 10142-10153.<br />
[3] T. Oda, A. Pasquarello, and R. Car (1998). Fully Unconstrained Approach to Noncollinear<br />
Magnetism: Application to Small Fe Clusters, Phys. Rev. Lett., 80, 3622-3625.<br />
[4] T. Oda and A. Hosokawa (2005), Fully relativistic two-component-spinor approach in the<br />
ultrasoft-pseudopotential plane-wave method, Phys. Rev. B, 72, 224428(1-4).<br />
[5] K. Sakamoto, T. Oda, A. Kimura, K. Miyamoto, M. Tsujikawa, A. Imai, N. Ueno, H. Namatame,<br />
M. Taniguchi, P. E. J. Eriksson, R. I. G. Uhrberg (1997). Abrupt Rotation of the<br />
Rashba Spin to the Direction Perpendicular to the Surface, Phys. Rev. Lett., 102, 096805(1-4).<br />
[6] M. Tsujikawa and T. Oda (2009). Finite Electric Field Effects in the Large Perpendicular<br />
Magnetic Anisotropy Surface Pt/Fe/Pt(001): A First-Principles Study, Phys. Rev. Lett., 102,<br />
247203(1-4).<br />
[7] http://www.softek.co.jp/SPG/Pgi/TIPS/public/accel/cublas-matmul.html<br />
[8] Matteo Frigo and Steven G. Johnson, ”manual for FFTW 3.2.2” July 2009,<br />
http://www.fftw.org/.<br />
25
Simulation Of Fluid-Solid Interaction<br />
Using Moving Particle Semi-Implicit<br />
And Mass Spring System<br />
Mourice Woran a,b , Seiro Omata b<br />
a Faculty of Mathematics and Natural Sciences, Institut Teknologi Bandung, Jl. Ganesha 10,<br />
Bandung 40132 Indonesia.<br />
b Institute of Science and Engineering, Kanazawa University, Kakuma, Kanazawa 920-1192 Japan,<br />
E-mail: moris 3gun@polaris.s.kanazawa-u.ac.jp<br />
Abstract. Interactions between fluids and deformable solid objects are very important in practical<br />
applications and it is an essential task to be able to simulate them correctly. A model for this kind<br />
of simulation is presented. The fluid is handled by use of a particle method, the so-called moving<br />
particle semi-implicit (MPS), whereas the solid is modeled by a spring-mass system which maintains<br />
its volume throughout the simulation. The interaction relies on the coupling force between<br />
the solid and the fluid, through which the movements of both parts influence each other.<br />
Keywords: Fluid-solid interaction, particle method, MPS, mass spring system, volume preservation,<br />
weak coupling.<br />
1 Introduction<br />
Fluid-deformable solid interaction is the interaction of some movable or deformable solid with an<br />
internal or surrounding fluid flow [1]. The consideration of fluid-deformable solid interactions is<br />
crucial not only in the design of many engineering systems but also in the medical science, where<br />
as examples we can name the treatment of aneurysms in large arteries or artificial heart valves.<br />
Animation of fluids is a particularly difficult task due to the underlying laws that govern fluid’s<br />
motion. These laws, known as the Navier-Stokes equations, are highly unstable partial differential<br />
equations that are quite difficult to solve in an efficient way. Recently, particle methods have<br />
been developed to handle fluid motion. Particle methods draw attention for their advantages<br />
of Lagrangian and meshless formulation. Large motion of free surfaces can be tracked without<br />
numerical diffusion while fluid disintegration and merging can be analyzed without mesh tangling.<br />
Moreover particle methods are also fitted to large deformation and fracture of solid materials [10].<br />
Moving Particle Semi-implicit (MPS) is one of the particle methods that was developed by<br />
Koshizuka and coworkers for incompressible flow [7] [8] [9]. Incompressibility is achieved by keeping<br />
the density of particles constant during the computation. MPS is based on Taylor series expansion<br />
that employs an original form of the weight function to approximate spatial derivatives of the<br />
Lagrangian Navier-Stokes system by a deterministic particle interaction model [6].<br />
Although fluid is simulated through a rather complex process, solids are much easier to handle.<br />
While solids can be represented as rigid bodies delineated by a surface made of triangles, they can<br />
also be represented as a set of linked point masses [12]. This model, for which a simple example is<br />
given in Figure 3, is especially well suited to animation. It is used in the presented method because<br />
of its flexibility to handle both rigid and non-rigid bodies, its easy manipulation, and the way it<br />
fits in the interaction model.<br />
27
ISCS 2011 Selected Papers Vol.2 M. Woran, S. Omata<br />
The purpose of this paper is to make a simple fluid-solid interaction with volume preservation<br />
for the solid, by using MPS and mass spring model. Several models have been proposed to handle<br />
this kind of interaction [5] [3], including those using MPS and mass-spring system [13]. However,<br />
our model is different in the way we handle the mass-spring system. Through this model we<br />
simulate the motion of a membrane with external force applied from above (see Figure 1a) and<br />
the deformation and motion of a box inside parallel flow (Figure 1b).<br />
(a) Test model 1 (b) Test model 2<br />
Figure 1: Test models.<br />
2 MPS in Computational Fluid Dynamics (CFD)<br />
Finite difference method (FDM), finite volume method (FVM) and finite element method FEM are<br />
very powerful, popular and widely used methods in CFD. In these methods, continuum domain is<br />
discretized into a fixed discrete grid or mesh. The strategy with fixed grids assures both robustness<br />
and accuracy when obtaining the solution of a differential equation using the discrete system. However,<br />
this strategy also gives several limitations in analyzing complex geometries and multiphysics<br />
problems, which consequently limits the application of these methods to practical problems. The<br />
limitations of the fixed grid and mesh-based methods have triggered the development of a mesh<br />
free method which uses only a set of distributed nodes to express the mechanical system in a<br />
discretized form. The particle method developed for incompressible flow is called moving particle<br />
semi-implicit (MPS). In the MPS method, fluids are represented by a set of moving particles and<br />
governing equations are discretized by particle interaction models, hence grids are not necessary.<br />
2.1 Governing Equations<br />
Governing equations express the conservation laws of mass and momentum,<br />
1 ∂ρ<br />
+ ∇ · u = 0, (1)<br />
ρ ∂t<br />
Du 1<br />
= −<br />
Dt ρ ∇P + v∇2u + f, (2)<br />
while incompressible fluids must satisfy<br />
∂ρ<br />
= 0.<br />
∂t<br />
28
ISCS 2011 Selected Papers Vol.2 Simulation of fluid-solid interaction by MPS and mass-spring system<br />
Here ρ denotes density, P pressure, v viscosity, u velocity and f external force. The left-hand<br />
side of Eq.(2) is the Lagrangian time differentiation involving advection terms. In MPS methods<br />
advection terms are directly incorporated into the calculation of moving particles. In this study,<br />
we solve two-dimensional problems and the only outer force considered is the gravity [9]. The<br />
Navier-Stokes solution process by MPS is a semi-implicit prediction-correction process. Namely,<br />
the basic Eq.(2) is divided into two parts,<br />
and<br />
( ) explicit<br />
Du<br />
= υ∇<br />
Dt<br />
2 u + f (3)<br />
( ) implicit<br />
Du<br />
Dt<br />
= − 1<br />
∇P. (4)<br />
ρ<br />
We solve Eq.(3) first to predict the velocity and position of fluid particles. Therefore, in the first<br />
part the accelerations of the particles taking into account all the terms except for pressure are<br />
evaluated, and the temporal velocities and positions are calculated by<br />
u ∗ = u k + ∆t<br />
r ∗ = r k + ∆tu ∗ .<br />
( ) explicit<br />
Du<br />
Dt<br />
The pressure term Eq.(4) is included by solving Poisson equation for pressure in the second part<br />
of the algorithm. The velocities and positions of each particle are modified as<br />
u k+1 = u ∗ + ∆t<br />
r k+1 = r ∗ + ∆t 2<br />
2.2 Particle Interaction Model<br />
( ) implicit<br />
Du<br />
Dt<br />
( ) implicit<br />
Du<br />
.<br />
Dt<br />
A particle interacts with its neighboring particles within the support of weight function w(r), where<br />
r is a distance between two particles. The weight function employed in this study has the following<br />
form<br />
{ re<br />
w(r) = r − 1 0 ≤ r < re<br />
0 re ≤ r.<br />
Hence, a particle interacts with only a finite number of particles located within the distance of<br />
re. The cutoff radius re limits the number of the neighboring particles and allows to reduce the<br />
computational cost. The accuracy may be deteriorated with few neighbor particles when we adopt<br />
an excessively small radius. On the other hand, a larger radius will make the spatial resolution<br />
lower [6].<br />
To calculate the weighted average, a normalization factor called particle number density between<br />
particle i and its neighbors j located at positions ri and rj is used:<br />
〈n〉i = ∑<br />
w(|rj − ri|). (5)<br />
j�=1<br />
29
ISCS 2011 Selected Papers Vol.2 M. Woran, S. Omata<br />
Figure 2: Particle interaction given by weight function.<br />
This value is assumed to be proportional to the density. As the density is almost constant in<br />
incompressibility calculation, we use the constant n 0 instead of 〈n〉i in the formulation of weighted<br />
average.<br />
The gradient operator is modeled using the weight function. A gradient vector between two<br />
neighboring particles i and j is evaluated as (φj - φi)(rj − ri) / |rj − ri| 2 , where φ is a physical<br />
quantity. The gradient vector at ri is then a weighted average of these vectors :<br />
〈∇φ〉 i = d<br />
n0 ∑ (φj − φ<br />
j�=i<br />
′<br />
i )<br />
|rj − ri| 2 (rj − ri)w(|rj − ri|), (6)<br />
where d is the number of space dimensions and n 0 is the reference value of the particle number<br />
density. In Eq.(6) a different value φ ′<br />
i is used in place of φi because of stability reasons. If the<br />
configuration of neighboring particles is isotropic, Eq.(6) is not sensitive to absolute value and any<br />
value can be used for φ ′<br />
i . However, the configuration is not isotropic in general. In that case, the<br />
in this study is calculated by<br />
value of φ ′<br />
i<br />
φ ′<br />
i = min φj for {j | w(|rj − ri|) �= 0}.<br />
This means that the minimum value is selected among the neighboring particles within the distance<br />
of re. Using Eq.(6), forces between particles are always repulsive because φj − φ ′<br />
i is positive. As<br />
noted in [6], this contributes to numerical stability.<br />
Laplacian operator is modeled by the following identity:<br />
∑<br />
(φj − φi)w(|rj − ri|).<br />
〈∇ 2 φ〉 i = 2d<br />
λn 0<br />
j�=i<br />
Here, the parameter λ is introduced so that the variance increase is equal to the analytical solution,<br />
which yields<br />
∑<br />
j�=i<br />
λ =<br />
|rj − ri| 2w(|rj − ri|)<br />
∑<br />
j�=i w(|rj<br />
.<br />
− ri|)<br />
30
ISCS 2011 Selected Papers Vol.2 Simulation of fluid-solid interaction by MPS and mass-spring system<br />
3 Mass Spring Model<br />
The mass-spring model is an array of masses linked by springs. Each mass connects to its adjacent<br />
masses as in the figure below. The movement of each mass is determined by the spring force and<br />
damp force between the masses.<br />
Figure 3: Basic mass-spring model.<br />
The spring force complies with the Hooke’s law. Supposed there are springs connecting the mass<br />
i with its neighbors denoted by Ω. According to Hooke’s law, the spring force is written as<br />
and the damp force has the form<br />
Fint = − ∑<br />
j∈Ω<br />
ki,j(|li,j| − l 0 i,j) li,j<br />
|li,j| ,<br />
Fdamp = − ∑<br />
bi,j(vi − vj),<br />
j∈Ω<br />
where kij is the stiffness of the spring, l 0 ij is the rest length of the spring, bij is the coefficient of<br />
elasticity of the spring between the masses i and j , li,j is the vector connecting neighboring masses<br />
and vi is the speed of the mass i.<br />
According to Newton’s second law of motion, the total force on each node (mass) is<br />
Fi = Fint + Fdamp + Fext,<br />
where Fext is the external force that also influences the movement of nodes.<br />
To solve this equation the Verlet method is used. Beginning at a time step n and given position<br />
rn, velocity vn and force Fn acting on each node, the following equations are used to obtain values<br />
for the next step [11].<br />
v n+ 1<br />
2 = vn + ∆t<br />
2 m−1 Fn<br />
rn+1 = rn + v n+ 1<br />
2 ∆t<br />
Fn+1 = F (rn+1) + Fdamp + Fext<br />
∆t<br />
vn+1 = v 1 n+ +<br />
2 2 m−1Fn+1. 31
ISCS 2011 Selected Papers Vol.2 M. Woran, S. Omata<br />
4 Volume Preservation<br />
During deformable solid object simulation we need to keep its volume constant to maintain reliability<br />
of our model. Here we adopt the method of [4]. The idea is to use the divergence theorem<br />
to transform the expression for the volume of the solid body to an integral over the surface of the<br />
body. This integral is then evaluated using a triangular mesh for the surface and there is no need<br />
to analyze the structure of the whole object.<br />
The condition of total volume preservation is cast into a constrained dynamic system. The<br />
constrained formulation using Lagrange multipliers results in a mixed system of ordinary differential<br />
equations and algebraic expressions. The system of equations is represented using dn generalized<br />
coordinates q, where n is the number of discrete masses. The generalized coordinates which are<br />
simply the Cartesian coordinates of the discrete masses are defined in our 2D setting as<br />
q = [x1, y1, x2, y2, . . . , xn, yn] T .<br />
The difference between V0 (original volume) and V (current volume) should be 0 for each<br />
iteration process, which is written using the constraint function Φ(q, t) as follows:<br />
Φ(q, t) = V − V0 = 0.<br />
To maintain the volume at a constant value, implicit method is applied. The equation of motion<br />
and kinematic relationship between q and ˙q are discretized in the following way:<br />
˙q(t + ∆t) = ˙q(t) − ∆tM −1 ∇Φ(q, t)λ + ∆tM −1 F A (q, t), (7)<br />
q(t + ∆t) = q(t) + ∆t ˙q(t + ∆t). (8)<br />
Here F A are spring forces acting on the discrete masses, M is a dn×dn diagonal matrix containing<br />
discrete nodal masses, λ is a Lagrange multiplier, and ∇Φ = ∇qV . We treat the constrained<br />
equations at new time implicitly,<br />
Φ(q(t + ∆t), t + ∆t) = 0. (9)<br />
Eq. (9) is now approximated using a truncated first-order Taylor series:<br />
Φ(q, t) + ∇Φ T (q, t)(q(t + ∆t) − q(t)) + Φt(q, t)∆t = 0. (10)<br />
Substituting ˙q(t + ∆t) from Eq. (7) into Eq. (8) we obtain<br />
q(t + ∆t) = q(t) + ∆t ˙q(t) + (∆t) 2 {M −1 F A (q, t) − M −1 ∇Φ(q, t)λ}. (11)<br />
Substituting this result into Eq. (10) results in the following identity, which yields the Lagrange<br />
multiplier by a simple division:<br />
∇Φ T (q, t)M −1 ∇Φ(q, t)λ = 1<br />
1<br />
2 Φ(q, t) +<br />
∆t ∆t Φt(q, t) + ∇Φ T (q, t)( 1<br />
∆t ˙q(t) + M −1 F A (q, t)). (12)<br />
5 Weak Coupling of Fluid and Solid<br />
In this simulation the interaction between the fluid and the solid is carried out in the manner of<br />
weak coupling, where the effect of the fluid on the solid is calculated after one iteration of the fluid<br />
[13]. Steps of the interaction between the fluid and the solid are described as follows.<br />
32
ISCS 2011 Selected Papers Vol.2 Simulation of fluid-solid interaction by MPS and mass-spring system<br />
1. Determine the motion of fluid particles for one calculation step using the MPS method with<br />
time step ∆t fluid .<br />
2. Calculate the fluid force F fluid<br />
i , consisting of pressure and viscous force acting on each solid<br />
particle i that interacts with fluid:<br />
where F viscosity<br />
i<br />
mi( Du<br />
Dt )implicit .<br />
F fluid<br />
i<br />
= F pressure<br />
i<br />
+ F viscosity<br />
i ,<br />
= mi( Du<br />
Dt )explicit (without considering external force f), and F pressure<br />
i<br />
3. Determine the motion of the solid particles using the spring network model for time ∆t fluid .<br />
This amounts to solving the set of equations of motion with respect to the solid particles,<br />
mi¨ri + bi,j( ˙ri − ˙rj) = Fi + F fluid<br />
i<br />
with the Verlet method. Here, the dot denotes the time derivative, j is a neighbor of the<br />
solid particle i and Fi includes the spring forces and volume constraint. Since the time step<br />
∆t solid used in mass spring calculation is different from the time step ∆t fluid , the motion of<br />
solid was calculated for ∆t fluid /∆t solid iterative steps.<br />
4. Return to step 1 and repeat procedures 1. - 4. until the final time is reached.<br />
6 Results and Discussion<br />
First, we tested the mass spring and volume preservation model using a 7 × 7 mass spring system<br />
with the configuration of springs as in Figure 3. The response of the object was compared for the<br />
cases with and without volume constraint after applying an external force from above for a period<br />
of time and then releasing it. The difference is shown in Figure 4 and Figure 5 shows the evolution<br />
of the relative volume error computed by the formula<br />
V − V0<br />
ε = .<br />
Next, we tested the fluid-solid interaction model using two problems (Figure 6). In the first<br />
computation, the upper part of a two-dimensional tube corresponds to a deformable solid wall and<br />
acts like a membrane, while the the bottom part is considered as a rigid wall. External force is<br />
applied on the upper wall from above for a period of time. Solid deformations due to the external<br />
force cause oscillations of the membrane. This vibration triggers an interaction between fluid and<br />
solid and a flow is observed inside the tube. In order to preserve tube’s volume, we connected the<br />
bottom and upper part with a spring. In this way, the whole domain is included in the mass-spring<br />
system and the volume is thus preserved. For the upper membrane, 3 × 39 nodes are used.<br />
In the second test problem, both upper and lower part of the tube are rigid walls, while we<br />
insert a deformable solid object inside the tube. The boundary condition on the inlet is selected as<br />
uniform velocity U0 in the negative x-direction Here we use 9 × 5 nodes to approximate the box,<br />
which deforms into a parachute-like shape, while being transported by the fluid flow.<br />
33<br />
V0<br />
=<br />
(13)
ISCS 2011 Selected Papers Vol.2 M. Woran, S. Omata<br />
7 Conclusion<br />
Figure 4: Mass spring model without and with volume preservation.<br />
Figure 5: Relative error of volume during the simulation.<br />
We have succeeded in simulating a simple fluid-solid interaction by use of the Moving Particle<br />
Semi-implicit method for fluid flow and mass-spring system for solid deformation keeping the<br />
volume constant throughout the simulation process. The key of the model is a weak coupling<br />
34
ISCS 2011 Selected Papers Vol.2 Simulation of fluid-solid interaction by MPS and mass-spring system<br />
Figure 6: Results of test problems.<br />
interaction between the fluid and the solid, where the effect of viscosity and pressure count as the<br />
external forces for mass-spring system after one fluid iteration. We use a smaller time steps for<br />
solid calculations then for fluid calculations. To be able to simulate a real part of a human body<br />
like in [13] or [5], parameter tuning and increase in resolution are indispensable.<br />
References<br />
[1] H.-J. Bungartz, M. Schäfer (2006). Fluid-structure Interaction: Modelling, Simulation, Optimization,<br />
Springer-Verlag.<br />
[2] D. Bourguignon, M.-P. Cani. Controlling anisotropy in mass-spring systems, Eurographics<br />
Workshop on Computer Animation and Simulation (EGCAS), 113–123.<br />
[3] O. Genevaux, A. Habibi, J.H. Dischler. Simulating fluid-solid interacion, LSIIT UMR CNRS-<br />
ULP 7005.<br />
[4] M. Hong, S. Jung, M.H. Choi, S.W.J. Welch (2006). Fast Volume Preservation for a Mass-<br />
Spring System, IEEE Computer Graphics and Applications 26(5), 83–91.<br />
[5] M. Kazama, S. Omata (2009). Modeling and computation of fluid-membrane interaction, Nonlinear<br />
Analysis 71, e1553-e1559.<br />
35
ISCS 2011 Selected Papers Vol.2 M. Woran, S. Omata<br />
[6] M. Kondo, S. Koshizuka (2011). Improvement of stability in moving particle semi-implicit<br />
method, Int. J. Numer. Meth. Fluids 65, 638?-654.<br />
[7] S. Koshizuka, Y. Oka (1995). A particle method for incompressible viscous flow with fluid<br />
fragmentation, Comput. Fluid Dyn. J. 4 (1), 29-46.<br />
[8] S. Koshizuka, Y. Oka (1996). Moving particle semi-implicit method for fragmentation of incompressible<br />
fluid, Nucl. Sci. Eng. 123 (3), 421-434.<br />
[9] S. Koshizuka, A. Nobe , Y. Oka (1998). Numerical Analysis of Breaking Waves Using The<br />
Moving Particle Semi-Implicit Method, International Journal for Numerical Methods in Fluids<br />
26, 751-769.<br />
[10] S. Koshizuka (2005). Moving Particle Semi-Implicit (MPS) Method : A Particle Method for<br />
Fluid and Solid Dynamics, in IACM expressions. Bulletin for The International Association<br />
for Computational Mechanics, 4–9.<br />
[11] J. Li, D. Zhang, G. Lu, Y. Peng, X. Wen, Y. Sakaguti (2005). Flattening Triangulated Surfaces<br />
Using a Mass-Spring Model, Int. J. Adv. Manuf. Technol. 25, 108–117.<br />
[12] G.S.P.Miller (1988). The motion dynamics of snakes and worms, SIGGRAPH 88 Confrence<br />
Proceedings, 198 – 178.<br />
[13] K. Tsubota, S. Wada, T. Yamaguchi (2006). Particle method for computer simulation of red<br />
blood cell motion in blood flow, Computer Methods and Programs in Biomedicine 83, 139–146.<br />
36
High-Pressure Crystal Structure Prediction<br />
Using Evolutionary Algorithm Simulation<br />
Athiya M. Hanna 1,2 , Toru Yoshizaki 1 , Muhamad A. Martoprawiro 2 , Tatsuki Oda 1<br />
1 Institute of Science and Engineering, Kanazawa University, Kakuma, Kanazawa 920-1192 Japan,<br />
E-mail: oda@cphys.s.kanazawa-u.ac.jp<br />
2 Faculty of Mathematics and Natural Sciences, Institut Teknologi Bandung, Jl. Ganesha 10,<br />
Bandung 40132 Indonesia, E-mail: athiya.mh@students.itb.ac.id, muhamad@chem.itb.ac.id<br />
Abstract. We implemented a high-pressure crystal structure prediction using evolutionary algorithm<br />
method, one of successful method to deal with this kind of problems. This method employs<br />
three evolution operators to generate a new offspring from its parents; heredity operator, permutation<br />
operator, and mutation operator. We run two simulation tests to this method and found<br />
results having a good agreement with experimental results. We also found some metastable structures<br />
produced by this method.<br />
Keywords: high-pressure, evolutionary algorithm, ab-initio, crystal structure prediction.<br />
1 Introduction<br />
Crystal structure prediction at particular conditions are of the important challenge declared by<br />
John Maddox more than twenty years ago in his article [1]. The underlying problem of this challenge<br />
is searching the global minimum among the very high diversity of crystal structures even for the<br />
simple chemical compositions. Trimarchi et al. classify the optimization problem into three type<br />
of problems [2]. The type-I problems are structural relaxation problem of given crystal structures,<br />
which are the search of visible local minimum of given structure. The type-II problems are the<br />
search of correct chemical arrangement among the possible ones of given chemical composition and<br />
lattice type crystal. The type-III problems are to find the stablest structure where the lattice type<br />
and atomic arrangement are unknown. Maddox’s challenges are all about the type-III problem<br />
and the crystal structures at high-pressure now become one of his challenges to be solved.<br />
There are also other reasons why crystal structure prediction at certain condition, especially<br />
high-pressure, is important. The article written by McMillan [3] summarizes the current progress<br />
on high-pressure materials and there are some of these reasons in it. High-pressure condition is<br />
leading us to new physical behavior of materials; superhard materials synthesized in high-pressure<br />
condition to replace diamond, new optoelectronics properties of materials, and expansion region<br />
of superconductivity phenomena over all elements and increasing Tc of superconductivity. The<br />
crystal structures at high-pressure now become a challenge to be solved.<br />
To overcome this problem, many scientists have developed several powerful method to predict<br />
crystal structure at particular condition. Among these methods, one of the successful ones was<br />
evolutionary algorithm method [4]. From biological inspiration, the evolutionary algorithm method<br />
can be derived into two schemes; Lamarckian and Darwinian ones. Lamarckian scheme within<br />
evolutionary algorithm incorporates structure optimization procedure while evaluating fitness value<br />
and takes the fitness value of the optimized structure. While the Darwinian scheme skips this<br />
optimization procedure and goes directly to evaluating the fitness value. Woodley et al. proves<br />
that the Lamarckian scheme is more efficient and successful rather than the Darwinian one [5].<br />
In this research, we used the Lamarckian scheme of evolutionary algorithm method adopted<br />
from the one developed by Oganov and his research group [4, 6, 7].<br />
37
ISCS 2011 Selected Papers Vol.2 A. M. Hanna, T. Yoshizaki, M. A. Martoprawiro, T. Oda<br />
2 Evolutionary Algorithm<br />
In this method, we treat lattice parameters and atomic coordinates of a crystal structure as a set of<br />
variables. For each crystal structure, there must be an evaluation parameter which is a function of<br />
the set of variables, namely, fitness value. This fitness value is taken from thermodynamics potential<br />
calculated by the optimizer code. In this research, we use enthalpy as the fitness value. As the<br />
enthalphy becomes lower, the fitness value represents the better structure. A set of crystal structure<br />
and its fitness value is an entity called individual. Some individuals produced in a generation cycle<br />
are called population or generation depending on the context. The above concepts are illustrated<br />
in Figure 1 (a) and (b).<br />
(a) (b)<br />
Figure 1: (a) A set of crystal structure and its fitness value is fused in an individual, (b) some individuals are<br />
grouped to be a population or generation.<br />
START<br />
Previous<br />
Population<br />
Structure<br />
Generator<br />
No<br />
Volume<br />
Rescalation<br />
Initial<br />
Selection<br />
Stage, pass?<br />
Yes<br />
Ab-Initio<br />
Computation<br />
Figure 2: Flowchart of single generation in the method.<br />
Elite<br />
Selection<br />
Flowchart of single generation of our evolutionary algorithm method is described in Figure<br />
2. This flowchart can be partitioned into two parts; structure generation part and selection and<br />
adaptation part.<br />
2.1 Structure Generation<br />
Structure generation plays a keyrole in the evolutionary algorithm method. Structure generation<br />
generate a new crystal structure randomly if the cycle is initial generation or from previous generation<br />
if it is not initial generation. There are three evolution operators used in this method:<br />
heredity operator, permutation operator, and mutation operator.<br />
38<br />
Stage<br />
END
ISCS 2011 Selected Papers Vol.2 Crystal Structure Prediction Using Evolutionary Algorithm<br />
Heredity Operator<br />
Heredity operator needs two parents to generate an offspring. As described in the previous work<br />
[2, 4, 6, 7] and illustrated in Figure 3, first all the parents are transformed arbitrarily. Then the<br />
heredity operator chooses a random number between 0.0 to 1.0 to slice one of lattice vectors of<br />
the two parents randomly. The plane of this cut is parallel to two other lattice vectors. Then the<br />
offspring is constructed by combining atomic position of the two parents delimited by the cut on<br />
crystal slab. Lattice parameters of the offspring are weighted average of the two parent’s whose<br />
weight is also determined randomly. If the number of atoms in a unit cell is not equal to the<br />
parents, the corresponding atom is added or drawn stochastically, so that the number of atoms of<br />
the offspring is equal to its parents.<br />
Permutation Operator<br />
Figure 3: Heredity operator.<br />
Permutation operator is a one-parent operator. It needs only one parent to generate an offspring.<br />
As illustrated in Figure 4, it chooses two lattice points of different atomic species and then swap<br />
their atomic identity. This procedure is repeated as many as parameter given by the user.<br />
Figure 4: Permutation operator.<br />
The permutation operator explores correct chemical arrangement of the crystal structure in<br />
the configurational space [7]. It works only for crystal structure which consists of more than one<br />
atomic species. It is completely useless in single species crystals.<br />
Mutation Operator<br />
Mutation operator is also a one-parent operator. It distorts lattice vectors of the parent (Figure<br />
5), while atomic coordinates remain unchanged relative to the lattice vectors. This transformation<br />
is done using random symmetric matrix [2, 4]. A new lattice vector of the offspring is produced<br />
by this following formula:<br />
a ′ = (I + ɛ)a , (1)<br />
39
ISCS 2011 Selected Papers Vol.2 A. M. Hanna, T. Yoshizaki, M. A. Martoprawiro, T. Oda<br />
where a ′ , a are lattice vectors before and after mutation operation respectively. The (I + ɛ) is the<br />
symmetric matrix defined as<br />
⎛<br />
⎞<br />
ɛ12 ɛ13<br />
1 + ɛ11 2<br />
2<br />
(I + ɛ) = ⎝ ɛ12<br />
ɛ23<br />
2 1 + ɛ22 ⎠<br />
2 , (2)<br />
1 + ɛ33<br />
ɛ13<br />
2<br />
where ɛij are generated randomly by 0-mean gaussian random distribution between 0.0 and 1.0.<br />
2.2 Selection and Adaptation<br />
ɛ23<br />
2<br />
Figure 5: Mutation operator.<br />
Each new structure produced by one of the three operators raised above is set to certain volume<br />
which is of the best crystal structure of previous generation [7]. This rescalation compensates the<br />
pre-condition of the crystal structure generated by the evolution operator which may not be a<br />
good initial structure before the crystal structure is relaxed to enhance the structure relaxation in<br />
ab-initio computation stage. The volume of initial generation is approximated by atomic radii.<br />
To enhance the structure relaxation more effectively before it is performed, each generated<br />
structure must be passed through an initial selection stage. There are three criteria in this selection<br />
[2, 4, 6, 7]; minimum interatomic distance, minimum lattice vector length, and minimum and<br />
maximum angle of lattice parameters. The structure which violates these criteria is rejected and a<br />
new structure is generated again by structure generation using the same operator. This procedure<br />
is repeated until it matches to the criteria.<br />
Fitness value of each individual as a function of crystal structure set of varibles is computed<br />
in ab-initio computation using first-principle method. Details of the ab-initio method will be<br />
described in Section 3.<br />
Elite selection stage plays a role as nature selection of the Lamarckian evolution theory. In this<br />
stage, a few worst individuals are discarded and other remaining elite structures are passed to the<br />
next cycle of generation and also ranked based on their fitness values. The ranking by the fitness<br />
value is useful for procreation or parent selection to generate a new population.<br />
3 Ab-Initio and Simulation Methods<br />
We used density functional theory (DFT) scheme of ab-initio computation [8] . Within the scheme,<br />
we employed planewave basis function and ultrasoft pseudopotential [9], and used generalized<br />
gradient approximation (GGA) for exchange-correlation term [10]. We also used two kinds of kpoint<br />
meshes used for ab-initio calculation: low k-point mesh for generating all possible individuals<br />
throughout the generations, and high k-point mesh for reoptimization of the last stablest structure<br />
of the generations. Our simulation test was carried out on two cases: single phase simulation and<br />
phase transition simulation.<br />
40
ISCS 2011 Selected Papers Vol.2 Crystal Structure Prediction Using Evolutionary Algorithm<br />
3.1 Single Phase Simulation<br />
Single phase simulation was run on silicon (Si) and phosphorus (P). Some details of ab-initio and<br />
evolutionary parameters are presented in Table 1.<br />
Table 1: Ab-initio and evolutionary parameters used in single phase simulation.<br />
Parameter Silicon (Si) Phosphorus (P)<br />
Pressure 20 GPa 190 GPa<br />
Energy cut-off wave-function 15 Ry 20 Ry<br />
Energy cut-off density 150 Ry 240 Ry<br />
k-point mesh 4 x 4 x 4 8 x 8 x 8<br />
Number of indiv. per generation 10 10<br />
Probability heredity 0.6 0.7<br />
Probability mutation 0.4 0.3<br />
3.2 Phase Transition Simulation<br />
Phase transition simulation was run on phosphorus (P) to see phase transition behavior in highpressure<br />
range using this method. We used several pressure, 95, 100, 110, 120, 130, 140, 150,<br />
160, 170, and 190 GPa. Cut-off energy for wave function and electron density used in this case<br />
are 20 Ry and 240 Ry respectively. For generating part of simulation, we used 4 x 4 x 4 k-point<br />
mesh. The evolutionary parameters used in this simulation are showed in Table 2. At the end of<br />
Table 2: Evolutionary parameters used in phase transition simulation on phosphorus.<br />
Pressure Number of indiv. per gen. Prob. heredity Prob. mutation<br />
95 GPa 10 0.6 0.4<br />
100 GPa 10 0.6 0.4<br />
110 GPa 10 0.6 0.4<br />
120 GPa 8 0.625 0.375<br />
130 GPa 8 0.625 0.375<br />
140 GPa 8 0.625 0.375<br />
150 GPa 10 0.6 0.4<br />
160 GPa 8 0.625 0.375<br />
170 GPa 8 0.625 0.375<br />
180 GPa 8 0.625 0.375<br />
190 GPa 10 0.7 0.3<br />
simulation, for each different pressure, we then reoptimized the last stablest individual using 16 x<br />
16 x 16 k-point mesh to enhance the accuracy of our calculations.<br />
41
ISCS 2011 Selected Papers Vol.2 A. M. Hanna, T. Yoshizaki, M. A. Martoprawiro, T. Oda<br />
4 Result and Discussion<br />
4.1 Single Phase Simulation<br />
Profiles of silicon and phosphorus in the series of generation are shown in Figure 6. In the profile of<br />
silicon, there is a divergent behavior along the fitness value (enthalpy), whereas in phosphorus the<br />
enthalpies are converged in the narrow range. Such convergence in the profile may be controlled by<br />
the probability of evolution operators. Indeed, in the comparison between two figures, the higher<br />
probability of heredity results in the convergence profile on generation series. This convergence is<br />
an advantage if one can prevent the same structure to be generated again. However, if one cannot<br />
prevent such kind of procedure, this convergence become a disadvantage. It may be effective to<br />
prevent redundancy of same structures as Oganov et al. did [6, 11].<br />
(a) (b)<br />
Figure 6: (a) Flow of generation series of silicon at 20 GPa and (b) phosphorus at 190 GPa.<br />
The reoptimized results of the last stablest individual shown in Table 3 give a good agreement<br />
with experimental result of the similar materials [12, 13]. Although there is some redundancy of<br />
same structures, these resutls show the prediction power of evolutionary algorithm method.<br />
Table 3: Comparison of our result with experimental data.<br />
Lattice Parameters a (˚A) b (˚A) c (˚A) α β γ Vol. (˚A 3 )<br />
Si last generation (20 GPa) 2.471 2.471 2.362 89.978 89.932 115.014 13.071<br />
Si re-optimized (20 GPa) 2.512 2.512 2.380 89.998 89.997 118.293 13.229<br />
Si Experiment [12] (∼ 16 GPa) 2.551 2.551 2.387 90 90 120 13.45<br />
P last generation (190 GPa) 2.119 2.132 2.047 89.788 90.660 118.955 8.088<br />
P re-optimized (190 GPa) 2.122 2.126 2.045 89.848 90.463 118.986 8.066<br />
P Experiment [13] (151 GPa) 2.175 2.175 2.0628 90 90 120 8.452<br />
42
ISCS 2011 Selected Papers Vol.2 Crystal Structure Prediction Using Evolutionary Algorithm<br />
4.2 Phase Transition Simulation<br />
Result of generation series for each pressure is similar to single phase simulation. The reoptimized<br />
last stablest generation for all pressures are summarized in Table 4, and we also found metastable<br />
structures for some pressures.<br />
Table 4: Summary of the reoptimized last generation of all pressures in phase transition simulation.<br />
Press.(GPa) Struct.[S] Vol.(˚A 3 )[S] H(eV)[S] Struct.[M] Vol.(˚A 3 )[M] H(eV)[M]<br />
95 SC 10.1469 -237.29666<br />
100 SC 10.0380 -236.98075<br />
110 SC 9.8366 -236.35717 SH 9.2032 -236.28198<br />
120 SC 9.6456 -235.74231 SH 9.0181 -235.71340<br />
130 SC 9.4509 -235.15132 SH 8.8409 -235.15565<br />
140 SC 9.2821 -234.56238 d-SC 8.9916 -234.55954<br />
150 SH 8.5704 -234.04512 SC 9.1134 -233.99563<br />
160 SH 8.4116 -233.53843 SC 8.9842 -233.42351<br />
170 SH 8.2897 -233.01649 SC 8.8512 -232.86837<br />
180 SH 8.1739 -232.50200 SC 8.7302 -232.31919<br />
190 SH 8.0660 -232.00137<br />
SC = Simple Cubic; SH = simple Hexagonal; d-SC = Distorted Simple Cubic<br />
(a) (b)<br />
Figure 7: Profiles of (a) volume versus pressure and (b) enthalpy versus pressure.<br />
From Table 4, we ploted a graph of volume versus pressure and enthalpy versus pressure, as<br />
depicted in Figure 7, with the curve of Murnaghan equation of state in which its parameters are<br />
taken from the works of Akahama et al. [13, 14]. It is shown in Figure 7 that the transition<br />
pressure obtained from our simulation occurs at around 140 GPa. It is overestimated, compared<br />
to the experimental result (137 GPa) of Akahama et al. [13]. Our prediction volumes at the<br />
pressures above 150 GPa are also lower than the one predicted by the experiment (Murnaghan<br />
state of equation). This underestimation is probably due to the k-point mesh which should be finer<br />
43
ISCS 2011 Selected Papers Vol.2 A. M. Hanna, T. Yoshizaki, M. A. Martoprawiro, T. Oda<br />
at high-pressure condition or the typical accuracy of GGA approach. The pressure dependences of<br />
the volume and enthalpy are consistent with a typical behavior at the first-order phase transition.<br />
Around the transition point, the lattice type is switched over between simple cubic and simple<br />
hexagonal structures for both of the ground state structure and the metastable one.<br />
5 Conclusion<br />
We have performed simulations of crystal structure prediction using the evolutionary algorithm<br />
method in combination with ab-initio calculation. Results of our simulation have a good agreement<br />
with the previous experimental works. We found that this implementation of evolutionary<br />
algorithm method on crystal structure prediction formed a good new path to answer Maddox’s<br />
challenge of the crystal structure prediction. We also found the metastable structure was switched<br />
in accordance with the transition pressure. The evolutionary algorithm simulation experienced in<br />
this work strongly supports us to the research of crystal structure predictions.<br />
References<br />
[1] J. Maddox. (1988). Crystal from first principles. Nature, 335(6187), 210.<br />
[2] G. Trimarchi and A. Zunger. (2007). Global space-group optimization problem: Finding the<br />
stablest crystal structure without constraints. Phys. Rev. B, 75, 104113-1–104113-8.<br />
[3] P.F. McMillan. (2002). New materials from high-pressure experiments. Nat. Mater., 1(1),<br />
19-25.<br />
[4] A. R. Oganov and C. W. Glass. (2006). Crystal structure prediction using ab initio evolutionary<br />
techniques J. Chem. Phys., 124(24), 244704.<br />
[5] S.M. Woodley, and C.R.A. Catlow. (2009). Structure prediction of titania phases. Computational<br />
Materials Science, 45(1), 84–95.<br />
[6] A. R. Oganov, Y. Ma, A. O. Lyakhov, M. Valle, and C. Gatti. (2010). Evolutionary crystal<br />
structure prediction as a method for the discovery of minerals and materials Reviews in<br />
Mineralogy and Geochemistry, 71(1), 271–298.<br />
[7] C. W. Glass, A. R. Oganov and N. Hansen. (2006). USPEX-evolutionary crystal structure<br />
prediction Computer Physics Communications, 175(11-12), 713–720.<br />
[8] P. Hohenberg and W. Kohn. (1964). Inhomogeneous electron gas Phys. Rev. 136, B864-B871.;<br />
W. Kohn and L. J. Sham. (1965). Self-consistent equations including exchange and correlation<br />
effects Phys. Rev. 140, A1133–A1138.<br />
[9] D. Vanderbilt (1990). Soft self-consistent pseudopotentials in a generalized eigenvalue formalism<br />
Phys. Rev. B, 41, 7892-7895.<br />
[10] J. P. Perdew, J. A. Chevary, M. R. Pederson, D. J. Singh, and C. Fiolhais. (1992). Atoms,<br />
molecules, solids, and surfaces: Applications of the generalized gradient approximation for<br />
exchange and correlation, Phys. Rev. B, 46, 6671-6687.<br />
[11] A.R. Oganov and M. Valle. (2009). How to quantify energy landscapes of solids. J. Chem.<br />
Phys., 130(10), 104504.<br />
44
ISCS 2011 Selected Papers Vol.2 Crystal Structure Prediction Using Evolutionary Algorithm<br />
[12] J. Z. Hu, L. D. Merkle, C. S. Menoni, and I. L. Spain. (1986). Crystal data for high-pressure<br />
phases of silicon Phys. Rev. B, 34(7), 4679.<br />
[13] Y. Akahama, M. Kobayashi, and H. Kawamura. (1999). Simple-cubic-simple-hexagonal transition<br />
in phosphorus under pressure. Phys. Rev. B, 59(13), 8520–8525.<br />
[14] Y. Akahama, H. Kawamura, S. Carlson, T. L. Bihan, and D. Häusermann. (2000). Structural<br />
stability and equation of state of simple-hexagonal phosphorus to 280 GPa: Phase transition<br />
at 262 GPa. Phys. Rev. B, 61(5), 3139–3142.<br />
45
Quantum control of high harmonic generation in anharmonic potential<br />
Muhamad Koyimatu a,b , Kiyoshi Nishikawa b<br />
a Faculty of Mathematics and Natural Sciences, Institut Teknologi Bandung, Jl. Ganesha 10,<br />
Bandung 40132 Indonesia, E-mail: s105koyimatu@mail.chem.itb.ac.id<br />
b Institute of Science and Engineering, Kanazawa University, Kakuma, Kanazawa 920-1192 Japan,<br />
E-mail: kiyoshi@wriron1.s.kanazawa-u.ac.jp<br />
Abstract.<br />
In this work, we first aim to realize the high harmonic generation(HHG) spectrum in a classical<br />
and quantum treatment, and then to achieve the quantum control of the HHG spectrum. Here, we<br />
consider the HHG from the Duffing oscillator well-known as a typical anharmonic system. The<br />
HHG induced by the intense laser is a nonlinear optical process, where a molecular system absorbs<br />
n photons with energy �ω and emits one high energy photon with energy n�ω. The light generated<br />
by HHG is a promising light source with an attosecond ultrashort pulse, which could be used to<br />
analyze the electronic structure and motion in molecule.<br />
First, we perform a classical simulation to explain the structure of the HHG spectrum. Then, we<br />
investigate quantum-mechanically the mechanism of HHG, and compare the classical and quantum<br />
result. In quantum treatment, we apply the Fourier Grid Method (FGM) to obtain the eigenvalues<br />
and wavefuntion of the Duffing oscillator and use the Split Operator Method (SOM) to solve the<br />
time-dependent Schrödinger equation. Here we study the laser intensity dependence of the HHG<br />
with a single laser pulse, and then from the interference point of view, we focus on the role of<br />
the relative phase of two laser pulses and simulate to find out the phase dependence of the HHG.<br />
Finally we investigate the laser control of the HHG by changing the initial quantum state, which<br />
is a kind of a wave packet due to the superimposition of few eigenstates.<br />
Keywords: HHG, SOM, anharmonic oscillator, quantum control<br />
1 Introduction<br />
Nonlinear optics process (NLOP) describes the response of the molecular system to the strong<br />
electromagnetic wave, and is usually induced only by a strong light such as a laser with high<br />
intensity. If the induced dipole moment of the material respond instantaneously to an applied<br />
electric field E(t), the induced polarization P (t) can be expressed in terms of the power of the<br />
electric field E(t),<br />
P (t) = χ (1) E(t) + χ (2) E 2 (t) + χ (3) E 3 (t) + · · · . (1)<br />
Above perturbation treatment of the nonlinear optics could describe accurately the characteristics<br />
of the ordinary nonlinear optical phenomenon. With sufficiently intense laser field, the higher order<br />
terms play an very important role, and give rise to Fourier components of the polarization with high<br />
harmonics of the laser frequency, emitting the electromagnetic wave with the same frequency as the<br />
polarization. The harmonic generation is an interesting phenomenon, which describe a nonlinear<br />
process of light matter interaction, and a typical example of the process is a second-order harmonic<br />
generation.<br />
47
ISCS 2011 Selected Papers Vol.2 M. Koyimatu, K. Nishikawa<br />
Due to the great progress of the laser technique, an intense laser with the intensity 10 8 <<br />
I < 10 12 W/cm 2 have become available, and we have discovered the noble interesting multiphoton<br />
processes in molecule [1], such as multi-photon absorption (MPA), multi-photon ionization<br />
(ATI), multi-photon dissociation (MPD), above threshold dissociation (ATD), the high harmonic<br />
generation (HHG), and so on, which are nonlinear optical processes arising from the higher-order<br />
effect of the external laser field.<br />
Here we focus on the HHG, where a molecule absorbs n photons (with energy �ω ) and emits<br />
one high photon with energy n�ω. HHG is experimentally observed by irradiating a noble gas<br />
with a near-IR laser pulse with a ultra high intensity 10 14 W/cm 2 , and the energy of the radiating<br />
electromagnetic wave is in the region of the extreme-ultraviolet (XUV) [2]. It have been known<br />
that the emitted photon energy is found to be the odd harmonics of the laser frequency (2m+1)�ω,<br />
and in some cases, the harmonic order n = 2m + 1 becomes larger than 100. The HHG spectrum<br />
has a characteristic structure, namely in a low energy region of HHG spectrum, the intensity of<br />
HHG rapidly decreases as the harmonic order increases, and then the typical harmonic shows a<br />
plateau, followed by a sharp cut-off. Thus the HHG provides an attractive light source of ultrashort<br />
coherent radiation in the XUV and soft-X-ray range, and it is expected that using the attosecond<br />
laser pulses by HHG, we could observe the electronic structure in molecule and control directly<br />
the motion of electron in molecule.<br />
On the other hand, several theoretical methods have been developed to understand the basic<br />
mechanism of the HHG. Most familiar model is the sp-called three-step model by P.B.Corkum<br />
to explain semiclassically the basic generation mechanism of higher-order harmonics. Usually the<br />
potential induced by the ultrahigh intensity of the external oscillating field is comparable with the<br />
Coulomb potential due to the nucleus, so that electron moves in the deformed oscillating Coulomb<br />
potential, being a superposition of the Coloumb potential and the time dependent potential due<br />
to the laser field. When the intensity of the external field becomes close to the maximum, the<br />
electron go through the potential barrier by the tunneling ionization. Then the ionized electron is<br />
accelerated by the strong laser electric field, and acquire a lot of energy form the external field.<br />
Finally this accelerated ionized electron return back to the nucleus to recombine with it, emitting<br />
the high energy photon. Namely an electron emits the electromagnetic wave by three steps, i.e.,<br />
(1) field induced tunnel ionization, (2) acceleration in the laser field, and (3) recombination and<br />
photoemission. Thus the energy of the emitted photon depends on the ionization potential of the<br />
atom and on the kinetic energy of the electron upon its return to its parent ion.<br />
The paper is organized in the following way. In section 2,we develop the theoretical treatment,<br />
where we introduce the split operator method (SOM) to solve the time-dependent Scrödinger equation,<br />
and the Fourier grid Hamiltonian (FGH) method to solve the time-independent Scrödinger<br />
equation. In SOM, the time development operator U(t) to generate the wavefunction |Ψ(t)〉 from<br />
the initial state |Ψ(0〉) is an unitary transformation so that the norm of the wavefunction is always<br />
kept constant. Furthermore it uses the fast Fourier transformation (FFT), so that it solves the<br />
dynamics of the system in a very efficient way. On the other hand, FGH method is a effective<br />
way to solve the eigenvalue problem of the Hamiltonian matrix, and yields the eigenvalues and<br />
eigenvectors, which is required to construct the initial wavepacket for the quantum control of HHG.<br />
In section 3, we describe the details of the target Duffing oscillator, which is a typical anharmonic<br />
oscillator. The anharmonic term plays an fundamental role to generate the HHG. In section 4,<br />
we perform the classical simulation of the HHG, From the numerical solution of the Newton’s<br />
equation, we estimate the dependence of the HHG spectra on the field intensity. In section 5,<br />
we develop the quantum mechanical treatment of the HHG, and calculate the HHG spectrum by<br />
changing the field intensity, and moreover investigate the relative phase dependence in the case of<br />
48
ISCS 2011 Selected Papers Vol.2 Quantum control of high harmonic generation in anharmonic potential<br />
the two color lase pulse. Finally we study the initial state dependence on HHG spectrum, where<br />
we adopt some initial wavepackets as a initial state. Last section treat the summary of this work.<br />
2 Theoretical Treatments and Numerical Methods<br />
2.1 Time-dependent Schrödinger Equation and Split Operator Method<br />
Now we develop the semiclassical theory of the light-molecule interaction, where the molecular<br />
system is treated quantum-mechanically, but the electromagnetic filed is described by the timedependent<br />
external field. Then the time-dependent Schrödinger equation is given by<br />
i� d<br />
dt |Ψ(t)〉 = [H0 + V (t)]|Ψ(t)〉, (2)<br />
where H0 and V (t)is the unperturbed molecular Hamiltonian and the perturbation Hamiltonian,<br />
respectively. In our simulation of the light-molecule interaction, the interaction Hamiltonian between<br />
the classical electronic field E(t) and the dipole moment operator µ(r) is given by<br />
V (t) = −µ(r) · E(t) (3)<br />
where the applied electric field E(t) is assumed as E(t) = E0g(t) sin(ωt), where E0 and ω mean<br />
the amplitude vector and the frequency of the external electric field, and the shape function g(t)<br />
of the laser pulse is assumed to be Gaussian function g(t) = e−(t−T0)2 /σ 2<br />
, where T0 and σ are the<br />
temporal center and the pulse width of the Gaussian pulse, respectively.<br />
The time-dependent Schrödinger equation is generally solved by expanding the state ket |Φ(t)〉<br />
in terms of the eigenstate |n〉 of the molecular Hamiltonian H0 for the set {|n〉|n = 0, 1, 2, · · ·} is<br />
a complete orthonormal set, namely |Φ(t)〉 = �<br />
n Cn(t)|n〉.<br />
However, we here introduce the time-development operator U(∆t) for the infinitesimal time<br />
increment ∆t in order to analyze the dynamics of the time-dependent system. Then the formal<br />
solution of time dependent Schrödinger equation is given by<br />
|Ψ(∆t)〉 = U(∆t)|Ψ(0)〉, (4)<br />
where the infinitesimal time-development operator is given in terms of the total Hamiltonian H<br />
by U(∆t) = e −iH∆t/� .<br />
Here we divide the Hamiltonian operator into two parts i.e., H ≡ H0 + V (t) = T + W (t). Note<br />
that the kinetic operator T = p 2 /2m and the potential term W (t) = V0 + V (t) does not commute.<br />
Now we employ the split operator method [3, 4] to solve the TDSE of the system numerically. The<br />
infinitesimal time-development operator is approximated by<br />
−i(T +W )∆t/�<br />
U(∆t) ≡ e<br />
≈ e −iW ∆t/2� e −iT ∆t/� e −iW ∆t/2� + O(∆t 3 ). (6)<br />
This approximation is correct up to the order ∆t 3 . Moreover, this approximated operator is unitary,<br />
so that the norm of the wavefunction obtained by applying this operator is kept constant while<br />
the numerical simulation. In order to get a state vector |Ψ(t)〉 at finite time t, we just apply this<br />
infinitesimal time-development operator to the initial state n-times, namely |Ψ(t)〉 = U(∆t) n |Ψ(0)〉,<br />
where n is the number of the simulation step, so the arbitrary time is given by t = n∆t.<br />
In general, the wavefunction and the potential terms are diagonal in the coordinate space,<br />
and the kinetic operator is diagonal in the momentum space, Therefore we adopt the fast Fourier<br />
49<br />
(5)
ISCS 2011 Selected Papers Vol.2 M. Koyimatu, K. Nishikawa<br />
transformation(FFT) in our numerical simulation, then we could effectively calculate the timedependent<br />
wavefunction as follows;<br />
Ψ(r, t0 + ∆t) ≡ 〈r|e −iW ∆t/2� e −iT ∆t/� e −iW ∆t/2� |Ψ(t0)〉 (7)<br />
= e −iW (r,t0)∆t/2� FFT −1 p2<br />
−i<br />
[e 2m ∆t/� FFT[e −iW (r,t0)∆t/2� Ψ(r, t0)]], (8)<br />
where the last line shows schematically the procedure of the programing, and FFT stand for the<br />
fast Fourier transform from the coordinate space to the momentum space, while FFT −1 represents<br />
the inverse FFT from the momentum space to the coordinate space.<br />
2.2 Time-independent Schrödinger Equation and Fourier Grid Hamiltonian<br />
Method<br />
In quantum simulation of the HHG, we need the wavefunction and the corresponding energy of<br />
the molecular system. In order to calculate correctly the eigenfunction and its energy in one and<br />
two dimensional systems, we here develop the Fourier Grid Hamiltonian(FGH) Method.<br />
Now let us consider the general Hamiltonian H with a kinetic energy T and a local potential<br />
V (x). In actual numerical simulation, we adopt the discrete variable representation, namely, we<br />
discretize the coordinate variable to construct the Hamiltonian matrix, where every spatial point<br />
is described by the set {xi = x0 + (i − 1)∆x | i = 1, 2, 3, · · ·, n} and ∆x is an spatial increment. We<br />
apply the second order differencing(SOD) to the 2nd-order spatial differenciation, then the explicit<br />
form of the matrix element for the kinetic and potential operator is given by, in the discretized<br />
coordination representation 〈〈x〉〉,<br />
Ti,j ≡ 〈xi| p2<br />
2m |xj〉 =<br />
�2<br />
2m∆x2 {2δi,j − δi,j+1 − δi,j+1a}, (9)<br />
Vi,j ≡ 〈xi|V |xj〉 = V (xi)δi,j. (10)<br />
By means of the diagonalization of the Hamiltonian matirix Hi,j, we could obtain the eigenfunctions<br />
and eigenvalues of the system described by the Hamiltonian operator H. However, the<br />
accuracy of this method is mainly due to the evaluation of the kinetic operator, so we need a larger<br />
basis set to get a accurate results.<br />
Now we develop the more efficient and accurate method called the Fourier Grid Hamiltonian(FGH)<br />
method. First we note that the kinetic operator is diagonal in the 〈〈p〉〉 representation.<br />
Using the wavefunction of the momentum eigenstate 〈x|p〉 = eipx / √ 2π, then the matrix element<br />
of the kinetic operator T is given by, in 〈〈x〉〉 representation,<br />
�<br />
〈x|T |y〉 =<br />
dp 〈x|T |p〉〈p|y〉 = 1<br />
2π<br />
�<br />
dp e −ip(x−y) �2 p 2<br />
. (11)<br />
2m<br />
This final expression means that this matrix element is given in terms of the Fourier transformation<br />
of the kinetic energy (�p) 2 /2m. Speaking exactly, p and q in above equation mean the wave vector,<br />
and we use this notation in the following.<br />
Next we turn to the discrete variable representation, namely we introduce the discretized<br />
momentum basis set {|pi〉 | i = 1, 2, 3, · · ·, N} in addition to the discretize coordinate basis set<br />
{|xi〉 | i = 1, 2, 3, · · ·, N}, where N means the dimension of the basis set, and usually is taken<br />
to be odd integer N = 2n + 1. Let the interval L of the coordinate under consideration be<br />
50
ISCS 2011 Selected Papers Vol.2 Quantum control of high harmonic generation in anharmonic potential<br />
xmin ≤ x ≤ xmax, so that x1 = xmin and xN = xmax. Now we set L = xmax − xmin, then the<br />
spatial increment is ∆x = L/(N − 1).<br />
On the other hand, in the discretized momentum space, we could take the domain of the<br />
momentum as 0 ≤ p ≤ pmax, where p1 = 0 and pN = pmax. Taking into account the periodicity in<br />
both spaces i.e., x1 = xN+1 and p1 = pN+1 in the discrete Fourier transformation, the momentum<br />
increment ∆p and maximum momentum pmax are given by in terms of ∆x and N as ∆p =<br />
2π/(N∆x). Each basis point of two spaces is given by pi = (i − 1)∆p and xi = x0 + (i − 1)∆x,<br />
however we could shift the domain of the momentum space due to the periodicity, namely we<br />
change the domain of the momentum point such as −pmax/2 ≤ pi ≤ pmax/2.<br />
Now we calculate the matrix element of kinetic operator in the discretized basis set,<br />
〈xi|T |yj〉 = 1<br />
2π<br />
=<br />
1<br />
∆x<br />
n�<br />
∆p e −ipl(xi−xj)<br />
Tl, (12)<br />
l=−n<br />
n�<br />
l=1<br />
2Cos(2πl(i − j)/N)<br />
N<br />
Tl, (13)<br />
where, for integer l, we define Tl = (�∆p) 2 /(2m) × l 2 . Thus we obtain the explicit form of the<br />
Hamiltonian matrix as<br />
Hi,j = 1 2<br />
{<br />
∆x N<br />
n�<br />
Cos(2πl(i − j)/N) Tl + V (xi)δi,j}. (14)<br />
l=1<br />
In this case, the Hamiltonian matrix Hi,j∆x include more element than in the second-order<br />
differencing method, but a diagonalization of this Hamiltonian yield numerically more accurate<br />
result of the eigenvalues and eigenvectors.<br />
3 Duffing Oscillator<br />
The purpose of the paper is to present a detailed analysis of HHG by an anharmonic oscillator.<br />
Since HHG by noble gas atoms irradiated with intense laser pulses has been observed, the theory<br />
of the highly nonlinear optical processes has been developed to investigate the characteristic<br />
features of HHG spectrum, which consists of a first decrease, a long plateau, and a sharp cutoff.<br />
Many model atoms are studied to explain the characteristic response. Here we adopt the typical<br />
anharmonic oscillator with a quartic anharmonicity, so-called Duffing oscillator. The anharmonic<br />
oscillator generally presents several appealing properties to investigate the nonlinear dynamics of<br />
one dimensional systems, especially this allow us to make a comparison between the classical and<br />
quantum-mechanical treatments [5]. Thus we will show that we can be able to understand the<br />
some important aspects of HHG just by this very simple model.<br />
The Duffing potential is written as<br />
V (x) = 1<br />
2 ω2 0 x 2 + 1<br />
4 v x4 , (15)<br />
where the first term represents the ordinary harmonic oscillator with the fundamental frequency<br />
ω0, and the second one corresponds to the quartic anharmonicity with the anharmonic coefficient<br />
v. In this work, we restrict the motion of the electron to the bound state, namely we consider only<br />
the confining case v > 0.<br />
51
ISCS 2011 Selected Papers Vol.2 M. Koyimatu, K. Nishikawa<br />
Next, we turn to the quantum mechanical treatment of the Duffing oscillator. We calculate<br />
the wavefunction and the corresponding energy of the Duffing oscillator using the FGH method<br />
mentioned in the previous section. Here we used the parameter set {ω0 = 1, v = 5}. The resultant<br />
wavefunctions are used to construct the initial wavepacket of the HHG simulation in order to study<br />
the initial state dependence on the HHG spectrum, namely to investigate the quantum control of<br />
the HHG. Figure 1 shows the Duffing potential and five lowest wavefunctions.<br />
Figure 1: Plot of Duffing potential and wavefunction<br />
In tis figure, the narrow quadratic curve means the Duffing potential, and the other one is<br />
the harmonic potential. The Duffing potential confines the electronic motion in the more narrow<br />
region. The behavior of the wavefunction of the Duffing oscillator is similar to the harmonic<br />
oscillator.<br />
4 Classical Simulation<br />
Here we develop the classical treatment of the HHG by the Duffing potential, so we investigate the<br />
classical motion of the electron in the quartic confining anharmonic oscillator under the intense<br />
electric field of the laser, E(t) = E0 sin(ω1t), where E0 and ω1 are the intensity and the frequency<br />
of the applied electric field, respectively. Then the total time-dependent potential of this system<br />
is given by<br />
V (x, t) = 1<br />
2 ω2 0 x 2 + 1<br />
4 v x4 − ex E0 sin(ω1t), (16)<br />
where the last term means the interaction of the electric field E(t) and the electric dipole moment<br />
ex, and e is the charge of the electron. The harmonic frequency ω0 is greater than the one of the<br />
external field ω1, namely ω0 > ω1. This physically means that the electron doing fast motion due<br />
to the Duffing potential moves under the slowly oscillating potential.<br />
In this model, we study not only the details of the HHG spectrum, but also the damping effect<br />
to the HHG spectrum. The quantum theory does not treat this damping effect, because there is<br />
no Hamiltonian to describe correctly the damping effect. Namely, the system with the damping<br />
term is dissipative and do not conserve the energy, but practically this effect is very important.<br />
Now consider the Newton’s equation of motion with damping effect given by<br />
¨x + Γ ˙x + ω 2 0 x + v x 3 = E0 sin(ω1t), (17)<br />
52
ISCS 2011 Selected Papers Vol.2 Quantum control of high harmonic generation in anharmonic potential<br />
where Γ is a damping coefficient. In order to analyze the oscillator dynamics given by the potential<br />
of Eq.16, we numerically integrate Eq.17. We assumed that the electron is rest at the origin at an<br />
initial time, namely we set the initial condition of the trajectory x(t) as {x(0) = 0 and ˙x(0) = 0}<br />
to solve the second order differential equation.<br />
As mentioned in the previous section, the power spectrum I(ω) is proportional to the absolute<br />
square of the Fourier transform of the acceleration α(t), i.e., I(ω) ∝ |α(ω)| 2 . Thus in the following,<br />
we calculate α(ω) to discuss the HHG spectrum. In the following, we will show the results of the<br />
HHG simulation in order to investigate the damping effect, the field intensity dependence, and the<br />
anharmonicity dependence to the HHG.<br />
1. The damping effect on the HHG spectrum:<br />
Here we show the typical example of the trajectory x(t) and the related HHG spectrum, where<br />
the x-axis is the harmonic order n = ω/ω1 In the following simulation we set the harmonic frequency<br />
ω0 = 1, namely we adopt this scaled unit. Here we used the parameters {v = 1, E0 = 5, and ω1 =<br />
0.1}. In figure 2, two trajctories x(t) corresponding to the weak and strong damping cases and the<br />
harmonic oscillator without the damping (v = 0 and Γ = 0) are shown, and the corresponding HHG<br />
spectrum are shown in figure 3 The spectrum of the harmonic oscillator (without anharmonicity<br />
Figure 2: Time variation of trajectory x(t) with respect<br />
to { Γ = 10(Green), 0.01(Blue), and harmonic oscillator<br />
(Red) }<br />
Figure 3: The corresponding HHG spectrum<br />
and damping) shown by red peak in Fig.3 has only two peaks corresponding to the external<br />
frequency ω1 = 0.1 and the harmonic frequency ω0 = 1. The harmonic oscillator does not show<br />
any interesting optical response, while Duffing oscillator with the interaction of the electron and<br />
the field shows several peaks of the HHG, but for large damping suppress the HHG spectrum, i.e.,<br />
there are small peaks (green) for Γ = 10. Therefore we are sure that the anharmonicity plays an<br />
very important role in the nonlinear optical response.<br />
The amplitude of all trajectories if Fig.?? have two frequency components of the periodic motion<br />
corresponding to ω0 and ω1. As the damping becomes large, the high frequency component<br />
of the x(t) decrease, so that the largest damping Γ = 10 gives a smooth curve (green) and this<br />
behavior of the trajectory makes the HHG spectra disappear drastically. On the other hand,for<br />
small damping, several sharp spectrum peaks of the high harmonic order with 13 ≤ n ≤ 31 appear.<br />
2. Field intensity dependence:<br />
53
ISCS 2011 Selected Papers Vol.2 M. Koyimatu, K. Nishikawa<br />
In order study the intensity dependence, we set parameters {v = 5, Γ = 0.01, and ω1 = 0.1},<br />
and we perform the simulation by changing the field intensity such as E0 = 1, 5, and 10. As shown<br />
from Fig.4, as the field intensity increase, the amplitude of the trajectory also become large. For<br />
large field intensity, there appear the large ripple with the large amplitude corresponding to the<br />
high frequency component shown by blue curve in Fig.4. From the HHG spectrum shown in Fig.5,<br />
Figure 4: Time variation of trajectory x(t) with respect<br />
to {E0 = 1(red), 5(green), and 10(Blue)}<br />
Figure 5: The corresponding HHG spectrum<br />
we easily find out that as the intensity become large, both the intensity and the harmonic order<br />
of the HHG spectrum also become large. For strong field E0 = 10 case, two peak groups appear<br />
around the harmonic order 15 < n < 40 and 50 < n < 70.<br />
3. Anharmonicity dependence:<br />
As we mentioned before, anharmonicity is very important to induce the HHG. In order to<br />
study the anharmonicity dependence, we set parameters {E0 = 5, Γ = 0.01, and ω1 = 0.1}, and we<br />
perform the simulation by changing the anharmonicity coefficient such as v = 0.1, 1, and 10. Here<br />
we use the time variation of the acceleration α(t) instead of the trajectory x(t). From the Fig.6, the<br />
behavior of the acceleration apparently looks like one of the trajectory, and as the anharmonicity<br />
becomes large, the high frequency component corresponding to the HHG shows the oscillation with<br />
more high frequency. While the low frequency component shows quite different behavior, namely<br />
the envelop of the large anharmonicity oscillate alternately with large and small amplitude. Fig.7<br />
clearly shows that the global pattern of the HHG spectrum is similar for three cases, but its peaks<br />
shift to the high frequency (i.e., high harmonic order).<br />
5 Quantum Simulation<br />
Here, we develop the quantum mechanical treatment of the HHG. The time-dependent Schr—”odinger<br />
equation determines how a given initial wavefunction will change with time. It differs from Newton’s<br />
law in two important ways: it involves a first time derivative instead of a second, and the<br />
square root of -1 appears explicitly in it. A wavefunction in an infinite square well is similar in<br />
some respects to a classical string clamped at both ends. In this section, we especially focus on<br />
the initial state dependence, the relative phase dependence in two color external field, and the<br />
interference effect on the HHG spectrum.<br />
54
ISCS 2011 Selected Papers Vol.2 Quantum control of high harmonic generation in anharmonic potential<br />
Figure 6: Time variation of acceleration α(t) with respect<br />
to {v = 0.1(red), 1(green), and 10(Blue)}<br />
1. Initial state dependence:<br />
Figure 7: The corresponding HHG spectrum<br />
Initial wave function is calculated numerically by Fourier Grid Hamiltonian method, and we<br />
consider the lowest three states of the Duffing system, {Ψ0, Ψ1 and Ψ2}. By changing these initial<br />
wave functions, we make some quantum simulation and get the information about initial state<br />
dependence to HHG spectrum. Figure 8 show that we could see clearly some groups in the HHG<br />
Figure 8: Initial state dependence on the HHG spectrum, where Ψ0 (red), Ψ1(green) and Ψ2(blue)<br />
spectrum peak, and in generally as the energy of the initial wave function becomes higher, the<br />
intensity of the give more intense HHG spectrum. But in the lowest region of the harmonic order,<br />
spectrum peaks from Ψ0 and Ψ1 is more intense, and the higher harmonic order region, the HHG<br />
pek groups from Ψ2 look like more intense than Ψ0 and Ψ1.<br />
2. Relative phase dependence of two-color laser pulses:<br />
55
ISCS 2011 Selected Papers Vol.2 M. Koyimatu, K. Nishikawa<br />
Next we performed the HHG simulation with a two-color laser, and examined the effect of the<br />
relative phase difference between two laser pulses. We here consider the following external field<br />
with two-color {ω1, 3ω1},<br />
E(t) = E0g(t) (sin(ω1t) + sin(3ω1t + δ)) , (18)<br />
where, δ is a relative phase of two laser pulses. The reason assuming the frequency of second<br />
pulse with three times the original pulse is that we expected the interference effect of two pathway<br />
by the two pulses in the HHG spectrum may occur. We observed the small difference in some<br />
peaks of the HHG spectrum, however we did not find out the drastic change of the some ports of<br />
the HHG spectrum due to the change of the relative phase of two pulses.<br />
3. Analysis of interference effect:<br />
Finally we have tried to control the HHG spectrum by using the wavepacket, which is a superposition<br />
of the three lowest states, {Ψ0, Ψ1 and Ψ2}. where we In this simulation we adopted<br />
the three states, {ψ0, (ψ0 + ψ1)/ √ 2 and (ψ0 + ψ1 + ψ2)/ √ 3}, as the initial state. If we consider<br />
the wavepacket made from three stationary states, then we could consider three pathway from<br />
three initial states to induce the HHG spectrum, and the HHG spectrum induced by three initial<br />
states could have a interference effect. Thus we have expected that the HHG spectrum from each<br />
component wavefunction constituting the wavepacket could make a interference, so that the HHG<br />
spectrum have a different structure from one of the each component wavefunction. The resultant<br />
HHG spectrum is shown Fig.9. The superposed initial state with higher energy gives stronger<br />
Figure 9: HHG spectrum from the superposed initial state, where Ψ0(red),(Ψ0 + Ψ1)/ √ 2(green) and (Ψ0 + Ψ1 +<br />
Ψ2)/ √ 3(blue)<br />
HHG spectrum in almost all region except the lowest high harmonic order region.<br />
Next we investigate the details of the difference due to the interference effect. In Figs.10 and<br />
11, we show the averaged HHG spectrum (red curve) from two independent states and the HHG<br />
spectrum (blue curve) form the superposed state with interference effect. <strong>IN</strong> Fig.10, the red curve<br />
56
ISCS 2011 Selected Papers Vol.2 Quantum control of high harmonic generation in anharmonic potential<br />
show the averaged HHG spectrum with the initial state Ψ0 and Ψ1, and the blue curve is the<br />
spectrum of the superposed state (Ψ0 + Ψ1)/ √ 2, while Fig.11 show the same spectrum related to<br />
the three initial states, Ψ0, Ψ1, and Ψ2. We find clearly out that, in both cases, the intensity of<br />
the HHG spectrum from the wavepacket is larger than one of the averaged spectrum. However<br />
we could not realize the increase or decrease of spectrum peaks only with respect to the specified<br />
region. Namely it seems difficult to control only the specified peaks of the HHG spectrum.<br />
Figure 10: The averaged HHG spectrum from two states Figure 11: The interference effect for three-state case : the<br />
(res) and the HHG spectrum from the superposed state averaged HHG spectrum (red) and the HHG spectrum of<br />
(blue)<br />
the superposed state (blue)<br />
6 Summary<br />
The purpose of this work is to investigate the mechanism and spectrum of the high harmonic<br />
generation(HHG) in a classical and quantum-mechanical point of view, and finally we have made<br />
an simulation to achieve the quantum control of the HHG spectrum. Here, we considered the<br />
HHG from the Duffing oscillator well-known as a typical anharmonic system. HHG spectrum is<br />
calculated in terms of the Fourier transform of the acceleration (α(ω)).<br />
In classical simulation, we have studied the damping effect, the intensity dependence, and<br />
the anharmonicity dependence on the HHG spectrum. In first simulation, we have found that<br />
the harmonic oscillator does not give rise to the nonlinear optical response, and then the large<br />
damping generates the weak HHG spectrum. Thus we understand that the anharmonicity is the<br />
fundamental quantity to induce the nonlinear optical response including HHG. Next by changing<br />
the field intensity and analyzing the time variation of the trajectory, we have clearly found that<br />
there are two oscillating components, corresponding to the harmonic (high) frequency ω0 and the<br />
(low) frequency of external field ω1. The HHG comes from the high frequency component, and<br />
the intense field yields the HHG with higher harmonic order. Then we have investigated the role<br />
of the anharmonicity. We have reconfirmed that as the anharmonic coefficient becomes large, the<br />
HHG spectrum with high harmonic order appears and the cut off energy increases.<br />
In quantum simulation, we have studied the initial state dependence, the relative phase dependence<br />
in two color external field, and the interference effect on the HHG spectrum. First we<br />
have took the lowest three states of Duffing oscillator {|Ψ0〉, |Ψ1〉, and |Ψ2〉} as an initial state<br />
57
ISCS 2011 Selected Papers Vol.2 M. Koyimatu, K. Nishikawa<br />
|Ψ(0)〉, where we calculated the eigenfunctions and eigenvalues of Duffing oscillator by means of<br />
the FGH method. The initial state with more higher energy gives rise to the HHG spectrum<br />
with the more higher harmonic order. Next we performed the HHG simulation with a two-color<br />
laser, and examined the effect of the relative phase difference. However the change of the relative<br />
phase of two pulses does not show the clear difference on the HHG spectrum. Finally we have<br />
tried to control the HHG spectrum by using the wavepacket, where we adopted the three states<br />
{|Ψ0〉, (|Ψ0〉 + |Ψ1〉)/ √ 2, and (|Ψ0〉 + |Ψ1〉 + |Ψ2〉)/ √ 3} as the initial state |Ψ(0)〉. It is expected<br />
that the superposed state has many channel corresponding to each state to generate the HHG, and<br />
the HHG from some channel could give rise to the interference among some channels. From the<br />
resultant HHG spectrum, in almost all region, the superposed initial state with the higher energy<br />
state shows the stronger HHG spectrum, however, we could not find out clearly the interference<br />
effect. Namely three spectrum has almost the same shape, but we just found some peaks have a<br />
little difference in some region of the HHG spectrum.<br />
Last I would like to comment on the mechanism of the HHG. In order to explain the HHG from<br />
the rare gas by the intense laser, P.B. Corkum proposed the three-step model, where the ionization<br />
and tunneling of the electron due to the deformation of the Coulomb attractive potential by the<br />
strong electric field of laser pulse plays an very fundamental role. However in the Duffing model<br />
only with the bound states, we could observe the HHG spectrum. From these simulation, it seems<br />
that the anharmonicity is the fundamental quantity to induce the nonlinear optical response.<br />
References<br />
[1] P. B. Corkum (1993). Plasma perpective on strong field multiphoton ionization. J. Phys. Rev.<br />
A, 71, 1994 – 1997<br />
[2] C. Winderfeld, Ch. Spielmann, and G. Gerber (2008). Optimal control of high-harmonic generation.<br />
Rev. Mod. Phys., 80, 117 – 140<br />
[3] M.D. Feit, J.A. Fleck Jr. and A. Steiger (1982). Solution of the Schrödinger equation by a<br />
spectral method. J. Comp. Phys., 47, 412 – 433<br />
[4] M. D. Feit and J. A. Fleck (1983). Solution of the Schrödinger equation by a spectral method<br />
II: Vibrational energy levels of triatomic molecules a) . J. CHem. Phys., 78, 301 – 308<br />
[5] Ph. Balcou, Anne L’Huillier, and D. Escande (1996). High-order harmonic generation processes<br />
in classical and quantum anharmonic oscillators. J. Phys. Rev. A , 53, 3456 – 3468<br />
58
Molecular Dynamics Studies on Structure and Dynamics of Spherical Micelles<br />
Micke Rusmerryani a,b , Gia Septiana Wulandari a,b , Shuhei Kawamoto b , Hiroaki<br />
Saito b , Kiyoshi Nishikawa b ,Hidemi Nagao b<br />
a Faculty of Mathematics and Natural Sciences, Institut Teknologi Bandung, Jl. Ganesha 10,<br />
Bandung 40132 Indonesia<br />
b Institute of Science and Engineering, Kanazawa University, Kakuma, Kanazawa 920-1192 Japan<br />
E-mail: micke@wriron1.s.kanazawa-u.ac.jp<br />
Abstract. Due to the soft and flexible structure, lipid molecule forms various shapes such as<br />
micelle, bilayer and vesicle. In these systems, since the spherical micelle can be applied to the<br />
application of drag delivery system, the analysis of detailed structure and dynamics of the micelle<br />
is important. In this study, we carried out molecular dynamics (MD) simulations of the spherical<br />
micelles dimer in water solvent to investigate the dynamical structure and correlated motion of<br />
the micelle dimer. The MD simulations were run under the constant NPT and NVT conditions<br />
with periodic boundary. We adopted the ASIC analysis, which is based on the aperture, symmetry,<br />
isotropy, and compactness of the micelle structure to analyze the shape fluctuations for each<br />
micelle. From this analysis, we show the stability and correlated motion of spherical micelle and<br />
investigate the patterns of synchronization motions between micelle dimer. The mutual fluctuations<br />
were periodically shown in the constant NVT simulation, implying that the existence of synchronization<br />
phenomena between micelle dimer.<br />
Keywords: spherical micelles, molecular dynamics, dynamics, synchronization<br />
1 Introduction<br />
Biological phospholipids show a self-assembly processes to form a certain aggregate such as micelles,<br />
vesicles, and membranes. In these systems, the micelle is an aggregate of surfactant molecules in<br />
aqueous solution. The micelle is formed by the competition between the hydrophobic and the<br />
electrostatic interactions of lipid molecule. The shape of micelle also depends on the molar density<br />
of lipid in aqueous solution. In the high lipid density condition, the lipid molecule aggregates so<br />
as to direct the head group of lipid to each other and forms the inverse micelle.<br />
Because of the soft and flexible structure, the structure (shape and size) of micelle fluctuates<br />
in aqueous solution and depends on both the component of the surfactant molecule and solution<br />
conditions such as the temperature and presence of impurities [1]. Also, it forms various structures<br />
such as the spherical, cylindrical, and rod-like structure. In these structures, because the spherical<br />
micelle structure is relatively stable in aqueous solution, the spherical micelle is expected to apply<br />
to drag delivery system.<br />
In the previous research [2], four structural parameters, aperture A, symmetry S, isotropy I,<br />
and compactness C (ASIC), were introduced to investigate the shape uctuation of micelle system.<br />
From the analysis of ASIC parameters at each time step, the structural fluctuation and correlated<br />
motion of the micelle were shown in detail. This analysis clearly showed the correlation between<br />
the isotropy I and compactness C. This technique could be expanded for other cases in biological<br />
dynamics. Other valuable informations can be shown by combining with the other parameters. In<br />
other research [3], the synchronization motion in mutual micelle clusters were implied in aqueous<br />
solution. Based on the previous research, it should be interesting to investigate the dynamical<br />
59
ISCS 2011 Selected Papers Vol.2 Micke Rusmerryani et al.<br />
(a) Chemical Structure of POPC (b) POPC Spherical Micelle<br />
(c) Spherical Micelles in Boxwater<br />
Figure 1: Initial Condition<br />
structure and the synchronization motion in mutual clusters such as two spherical micelles. Thermodynamic<br />
conditions will also be studied to figure out the effects on lipids dynamics. Our goal<br />
is to find out whether the dynamics of both micelle have any correlation. In this study, we thus<br />
carried out molecular dynamics (MD) simulations of the spherical micelle dimer in water solvent<br />
to reveal the dynamical structure and correlated motion of the micelle dimer system. We analyze<br />
the fluctuation of the spherical micelle by the ASIC analysis and investigate the synchronization<br />
motion between micelle dimer by the time correlation analysis.<br />
2 Computational Methods<br />
2.1 Initial Structure<br />
In this study, we used 8 POPC (1-palmitoyl-2-oleoyl-phosphatidycholine) lipids for each micelle.<br />
Figure 1 (a) shows the structure of POPC lipid. The POPC lipid has two hydrocarbon chains<br />
and one phosphatidylcholine (PC) head group [4]. The initial coordinates of the micelle dimer and<br />
water molecule were placed in the MD box by Packmol program [5]. The 11,326 TIP3P water were<br />
filled in the MD box (8.2 × 15.8 × 8.2 nm). Figure 1(c) shows the snapshot of initial structure of<br />
micelle dimer in the MD box.<br />
2.2 Molecular Dynamics Simulation<br />
In this study, two molecular dynamics simulations under the constant NPT and NVT conditions<br />
were carried out by NAMD 2.7b3 program package. We used the CHARMm36 force field [6] and<br />
TIP3P model for the POPC and water molecule, respectively. The periodic boundary condition<br />
60
ISCS 2011 Selected Papers Vol.2 Molecular Dynamics Studies on Spherical Micelles<br />
(a) After 10 ns NPT simulation<br />
(b) After 10 ns NVT simulation<br />
Figure 2: Final Snapshot<br />
was applied to the MD box, and the particle mesh Ewald (PME) method was adopted for the<br />
electrostatic interaction. The cutoff length for nonbond interaction in real space was 12 ˚A [7].<br />
The constant NPT simulation is used for the system equilibrium. The MD box decreased until<br />
5.85 × 11.11 × 5.84 nm after 10 ns. The length of MD box fluctuated 0.36% during the simulation.<br />
We conformed the micelle dimer sufficiently equilibrated in 10 ns. After 10 ns, the MD simulation<br />
was continued to the constant NVT simulation for 10 ns. The effects due to the difference of the<br />
MD conditions on structural and dynamical behavior of micelle dimer are also investigated in this<br />
study. Both MD simulations were run under the constant temperature (T=300 K). Figure 2 shows<br />
the last snapshots of micelle dimer for each simulation at 10 ns.<br />
3 Analysis<br />
3.1 ASIC analysis<br />
In this study, the ASIC analysis was adopted to investigate the dynamical structure and correlated<br />
motion of the POPC spherical micelle dimer [2]. The ASIC analysis gives four structural<br />
parameters, aperture A, symmetry S, isotropy I, and compactness C of the micelle system. Each<br />
parameter can be calculated by using the defined vectors �r1(t), �r2(t) and � R(t) in the POPC lipid.<br />
The vector �r1(t) is applied for the unsaturated acyl chain, the vector �r2(t) is applied for the saturated<br />
acyl chain, and the vector � R(t) is given by averaging the vectors �r1(t) and �r2(t). The<br />
61
ISCS 2011 Selected Papers Vol.2 Micke Rusmerryani et al.<br />
time-dependent ASIC parameters are expressed as following equations;<br />
A(t) = 1<br />
�<br />
�<br />
�<br />
�<br />
N<br />
N �<br />
|�r1i(t) × �r2i(t)|, (1)<br />
i=1<br />
S(t) = 1<br />
N |<br />
N�<br />
�Ri(t)|, (2)<br />
i=1<br />
I(t) = 1<br />
�<br />
�<br />
�<br />
N� N�<br />
�<br />
|<br />
N<br />
� Rj(t) × � Rk(t)|, (3)<br />
j=1 k=2,(k>j)<br />
C(t) = 1<br />
N<br />
N�<br />
| � Ri(t)|, (4)<br />
where, N is the total number of lipids in a micelles. We analyze the flexibility of micelle by<br />
assessing these ASIC parameters at each MD time steps. The detailed dynamics information<br />
about the dynamics of spherical micelles can be obtained by these four parameters.<br />
3.2 Time correlation and delayed time analysis<br />
The time correlation analysis was performed to investigate the correlated motion between micelle<br />
dimer in water solvent. The relaxation properties of micelle structure were evaluated by using<br />
time correlation function (TCF). The TCF is one of the most common tools to analyze the timedependent<br />
properties of physical variable and is useful to investigate the dynamics of micelle<br />
structure. The relaxation of structural parameter is estimated by the sufficient statistical average<br />
of the time series of structural parameter. The time-correlation function (TCF) of variable A(t) is<br />
defined as following equation;<br />
CAA(dτ) = 1<br />
T − τ<br />
i=1<br />
� T −τ<br />
0<br />
A(t) · A(t + τ)dt, (5)<br />
where τ is lag time[8]. This equation is the same as the form of auto-correlation function, when<br />
the A(t) and A(t + τ) are the same variable, whereas the cross-correlation function correlation is<br />
defined by the two different variables. For example, the time correlation of dynamical properties<br />
A(t) and B(t) is expressed as<br />
CAB(dτ) = 1<br />
T − τ<br />
� T −τ<br />
0<br />
A(t) · B(t + τ)dt. (6)<br />
Delayed time analysis was evaluated by calculating the direction correlation between the micelle<br />
dimer [3]. From the Eq.2, we dene symmetry vector � S(t), by using the vector � R(t). The direction<br />
correlation (DC) is expressed by following equation;<br />
Dc ij (t, t ′ ) = � Si(t) · � Sj(t ′ )<br />
| � Si(t)|| � Sj(t ′ . (7)<br />
)|<br />
This analysis gives the degree of parallel between the i th and j th micelle. When the DC equals<br />
to 1, the micelle dimer should have parallel orientation. Conversely, when the DC equals to -1,<br />
the micelle dimer are in antiparallel condition. This analysis can be applied to investigate the<br />
synchronization motion between micelle dimer.<br />
62
ISCS 2011 Selected Papers Vol.2 Molecular Dynamics Studies on Spherical Micelles<br />
4 Results and Discussion<br />
Figure 3: Time (ns) vs RMSD (nm 2 )<br />
Figure 4: Time (ns) vs Distance (nm)<br />
From each MD simulation, we obtained trajectories of micelle dimer in water solvent. Here, we<br />
simply call those micelles as the micelle 1 and micelle 2. We first calculated the root mean square<br />
displacement (RMSD) of each micelle to assess the equilibrium of the system during the simulations.<br />
The observed RMSD in the constant NPT and NVT simulation is shown in Figure 3. The results<br />
showed that the structures of both micelles in the constant NPT condition were equilibrated after<br />
5 ns. However in NVT simulation, the RMSD of micelle 1 increased after 5 ns, showing the large<br />
displacement of micelle 1 in the system.<br />
Figure 4 shows the distance between the center of mass of each micelle. We found that the<br />
micelles were getting closer to each other, and then went back to the initial distance in the constant<br />
NPT MD simulation. In NVT simulation, the observed micelle distance was found to be flctuated<br />
around the 5.5 nm. This difference could be cased by the difference of simulation conditions; the<br />
fluctuation of MD box possibly affects the dynamics of micelle. The distance between micelle dimer<br />
in NVT simulation was found to be larger than that in NPT simulation.<br />
63
ISCS 2011 Selected Papers Vol.2 Micke Rusmerryani et al.<br />
Figure 5: Time (ns) vs ASIC-NPT plotted every 20 ps<br />
Figure 5 and 6 shows the observed ASIC parameters as a function of MD time steps. We found<br />
that the structural parameter of micelle A, I and C were high values whereas S showed low value<br />
in both constant NPT and NVT condition. In the constant NVT condition, the ASIC parameters<br />
were found to fluctuate around average values. In NPT simulation, the motion of lipid chain tail<br />
are more exible than in NVT simulation, because it has wider range of A-uctuation. However, the<br />
constant NVT simulation shows a higher A value. In the symmetry parameter S, the micelle 2<br />
was more symmetric than micelle 1 in NPT simulation, whereas the opposite symmetry character<br />
was shown in NVT simulation. We also found that the most symmetric structure was micelle 2 in<br />
NPT ensemble. Figure 5 shows that the micelle 2 reached nearly zero at some points, indicating<br />
that the micelle 2 has sufficient symmetry at these time steps.<br />
Some interesting results were shown in the parameters of isotropy I and compactness C. In<br />
NPT simulation, the highest (or the lowest) I was exactly shown at the same time when the<br />
parameter C showed the highest (or the lowest) value. This implies that there is a correlation<br />
between the isotropy I and compactness C. We also found that both micelles have quite similar<br />
distribution of C fluctuation in NPT simulation.<br />
To reveal the correlation between micelle dimer and the correlation in themself, the time correlation<br />
were calculated by using Eq. (5) and (6). The observed TCF are shown in Figure 7. The<br />
observed TCF of micelle 1 in the constant NPT simulation monotonically decrease value in first 3<br />
64
ISCS 2011 Selected Papers Vol.2 Molecular Dynamics Studies on Spherical Micelles<br />
Figure 6: Time (ns) vs ASIC-NVT plotted every 20 ps<br />
ns, implying the relaxation of TCF of each micelle in this time. However, further statistic average<br />
should be taken for the clear this relaxation property. On the other hand, in the constant NVT<br />
simulation, we did not see the usual relaxation in each micelle in this time range. This implies the<br />
there is no correlation in each micelle. The larger correlation time or sufficient statistic average<br />
should be used for the observation of relaxation of the TCF.<br />
The direction correlation was estimated by Eq. 7. The calculated DC in the constant NPT<br />
and NVT condition are shown in Figure 8. We found that the observed correlation values sometimes<br />
show parallel direction. The probability distribution of DC in the constant NPT simulation<br />
shows that the peak of distribution appears around zero. However, in NVT simulation, mutual<br />
fluctuations were shown at many times, suggesting that two micelles have the similar orientation<br />
during the simulation.<br />
Figure 9 shows the calculated direction correlation with delayed time of 1 ns. In this figure,<br />
the direction correlation was plotted with gray scaled color; DC = 1 was represented as white<br />
color, and DC = 1 was represented as black color. Mutual fluctuations periodically occur in NVT<br />
simulation. On the other hand, there are unsmooth patterns at several times. This result shows<br />
that the dynamics of each micelle are correctively oscillated in this period of time, indicating the<br />
synchronization motion of micelle dimer.<br />
65
ISCS 2011 Selected Papers Vol.2 Micke Rusmerryani et al.<br />
5 Summary<br />
Figure 7: Autocorrelation of Symmetry<br />
Figure 8: Direction correlation at the same time<br />
We have carried out the molecular dynamics (MD) simulation of the micelle dimer in water solvent<br />
to investigate the dynamical structure and correlation motion of the micelle system. We found<br />
that the structure of micelle was randomly fluctuated during MD time steps. The simulation also<br />
showed the sufficient correlation between the isotropy and compactness of micelle structure. We<br />
analyzed the time correlation function of each micelle in both constant NPT and NVT simulations<br />
and showed the difference of relaxation of symmetry fluctuation of micelle. We however found that<br />
further statistic average should be taken for the clear this relaxation property. In the analysis of<br />
direction correlation between micelle dimer, the dynamics of each micelle were found to be oscillated<br />
correctively in the period of time, indicating the synchronization motion of micelle dimer. In<br />
the future, it is interesting to investigate another method to reveal the existence of synchronization.<br />
Furthermore, the concepts of limit cycle and synchronization might help us to analyze the<br />
phenomena of synchronization.<br />
66
ISCS 2011 Selected Papers Vol.2 Molecular Dynamics Studies on Spherical Micelles<br />
References<br />
Figure 9: Color-mapped of DC with delayed time 1 ns<br />
[1] P. S. Goyal, V. K. Aswal (2001). Micellar structure and inter-micelle interactions in micellar<br />
solutions: Results of small angle neutron scattering studies. Current Science.<br />
[2] A. Purqon, A. Sugiyama, H. Nagao, K. Nishikawa (2007). Aperture, symmetry, isotropy, and<br />
compactness analysis and their correlation in spaghetti-like nanostructure dynamics. Chem.<br />
Phys. Lett., 443, 356 - 363.<br />
[3] A. Purqon, H. Nagao, K. Nishikawa (2008). Synchronization Patterns in Spaghetti-Like Nanoclusters.<br />
Int. Journal of Quantum Chem., 108, 2870 -2880.<br />
[4] Hanahan, Donald J. (1997). A Guide to Phospholipid Chemistry, Oxford University Press,<br />
New York.<br />
[5] L. Martinez, R. Andrade, E. G. Birgin, J. M. Martinez (2008). Packmol: A Package for<br />
Building Initial Configurations for Molecular Dynamics Simulations. Journal of Comp. Chem.,<br />
30, 2157 - 2164.<br />
[6] J. B. Klauda, et all (2010). Update of the CHARMM All-Atom Additive Force Field for Lipids:<br />
Validation on Six Lipid Types. J. Phys. Chem., 114, 7830 - 7843.<br />
67
ISCS 2011 Selected Papers Vol.2 Micke Rusmerryani et al.<br />
[7] D. Frenkel, B. Smith (2002). Understanding Molecular Simulation: From Algorithms to Applications,<br />
Academic Press, USA.<br />
[8] J. Ramirez, S. K. Sukumaran, B. Vorselaars, A. E. Likhtman (2010). Efficient on the fly<br />
calculation of time correlation functions in computer simulations. The Journal of Chem. Phys.,<br />
133, 154103.<br />
68
Temperature Effects on Dynamics of Spherical Micelles :<br />
A Molecular Dynamics Study<br />
Gia Septiana Wulandari 1,2 , Micke Rusmerryani 1,2 , Shuhei Kawamoto 1 ,<br />
Hiroaki Saito 1 , Kiyoshi Nishikawa 1 , Hidemi Nagao 1<br />
1 Computational Science, Graduate School of Natural Science and Technology, Kanazawa<br />
University, Japan<br />
2 Computational Science, Graduate School of Mathematics and Natural Science, Bandung<br />
Institute of Technology, Indonesia<br />
E-mail: gia@wriron1.s.kanazawa-u.ac.jp<br />
Abstract. We studied the temperature effect on the dynamics of spherical micelle dimer in water<br />
solvent by using molecular dynamics (MD) simulation. For this purpose, we carried out the<br />
MD simulations of 8-8 spherical POPC micelle dimer in 11,326 TIP3P water molecules under<br />
the constant NPT condition for 10 ns and continued the MD simulation under the constant NVT<br />
condition for 10 ns. We ran the MD simulations of this system at two different temperatures, 340<br />
K and 370 K. The dynamical behavior of the micelle was analyzed by calculating ASIC parameters.<br />
We found that the effects of temperature on stuctures and dynamics of micelle were different for<br />
the constant NPT and NVT condition.<br />
Keywords: Molecular Dynamics, lipid, POPC, temperature effects<br />
1 Introduction<br />
Biological phospholipids show self-assembly processes to form certain clusters such as micelle,<br />
vesicle, and membrane[1, 6]. Micelle is an aggregate of surfactant molecules in aqueous solutions[9].<br />
The self-assembly is a term used to describe processes in which a disordered system of pre-existing<br />
components forms an organized structure or pattern as a consequence of specific local interactions<br />
among the components themselves without external direction.<br />
In the computational study of micelle systems, several groups have studied the temperature<br />
effects on dynamics of micelle so far. Acep Purqon in his doctoral thesis gave an interesting<br />
inspiration to this study. He studied seven issues on bionanocluster fluctuations[1]. One of the<br />
issues is to identify solvent effects on the micelle. He analyzed the effects of salty water and<br />
temperature on the phospholipids of micelle. In this study, four parameters, aperture A, symmetry<br />
S, isotropy I, and compactness C (ASIC) of the micelle system were introduced to investigate the<br />
structural character of the micelle in water solvent. From these structural parameter analysis, he<br />
found that the symmetry S and aperture A parameters increase as the temperature rises, showing<br />
the irregular structure and rapid tail fluctuation of the micelle system. The effects of salt and<br />
temperature consequently contribute to shape fluctuations as well. In contrast, the micelle system<br />
shows wider fluctuation in pure water.<br />
However, the temperature effects on the self-aggregation of micelle and structural stability of<br />
the lipid of the micelle in water solvent are still not clear. In this study, we therefore analyze the<br />
structures and dynamics of small spherical micelle dimer consisting of phospholipids molecules by<br />
molecular dynamics (MD) simulations. We also study the stability of spherical micelles at two<br />
temperatures and describe the temperature effects on the dynamics of the spherical micelles.<br />
69
ISCS 2011 Selected Papers Vol.2 Gia et al.<br />
2 Computational Details<br />
In this study, we used 16 palmitoyloleoyl-phosphatidylcholine (POPC) lipids, which is a diacylglycerol<br />
and phospholipid found in human or animal, for the micelle dimer simulation. To construct<br />
the initial condition for the system, we used Packmol program [4]. We divided those lipids into<br />
two spherical initial conditions, 8 lipids for each spherical system as in Figure 1(a). Then we put<br />
those lipids into the MD box (82 × 158 × 82 ˚ A) filed with 11,326 TIP3P water molecules (Figure<br />
1(b)).<br />
2.1 Molecular Dynamics Simulation<br />
(a) lipids (b) initial condition (c) POPC<br />
Figure 1: POPC. (a) shows 8-8 spherical POPC lipid we used as initial condition for this system. (b) shows initial<br />
condition for this system. (c) shows how did we define two vectors to calculate ASIC parameter<br />
We carried out the MD simulations by using NAMD2.7.b3 [7] program with CHARMM36 force<br />
field [10, 5] with 12 ˚ A cutoff radii for nonbond interactions. We ran the MD simulations of the<br />
system at 340 K and 370 K under the constant pressure condition (NPT ensemble) for 10 ns then<br />
continued the simulation under the constant volume condition (NVT ensemble) for 10 ns. By using<br />
NPT ensemble, the water and MD box size should be equilibrated in a few ns. The constant NPT<br />
simulation was adopted to prevent the effect of the fluctuation of MD box on the dynamics of<br />
micelle dimer. We then continued the MD simulation under the constant NVT conditions, without<br />
change the volume of MD box. We used the last coordinate of micelle dimer in the constant<br />
NPT simulation as the initial condition for the constant NVT simulation. The effects due to<br />
the difference of the MD conditions on structural and dynamical behavior of micelle dimer were<br />
investigated in this study. We carried out the MD simulations of micelle dimer system at 340 K<br />
and 370 K to investigate the effect of temperature difference on this system.<br />
2.2 Analysis Method<br />
After the MD simulations, we analyzed the structural and dynamical behavior of the micelle dimer.<br />
First, we calculated the root mean square displacement (RMSD) to assess the system equilibrium.<br />
The RMSD is frequently used to measure the displacement of the system at time t form the initial<br />
70
ISCS 2011 Selected Papers Vol.2 Temperature Effects on Dynamics of Spherical Micelle<br />
coordinate. The RMSD ise calculated by following equation;<br />
�<br />
�<br />
�<br />
RMSD(t) = � 1<br />
N�<br />
(R(i, t) − R(i, 0))<br />
N<br />
2<br />
i=1<br />
where N is the number of lipid molecule, t is the MD time step, and R(i, t) is the coordintes of lipid<br />
molecules pf micelle. We also calculated symmetry parameter S, which is one of four structural<br />
parameters given by Acep et al. [1]. Although there are four parameters, aperture A, symmetry<br />
S, isotropy I, and compactness C for analysis of micelle system, we only investigate the symmetry<br />
parameter S of micelle in this study. To calculate symmetry S, we defined three vectors −→ R1(t),<br />
−→<br />
R2(t), and −→ r (t) in each POPC lipid as shown in Figure 1(c). These vectors are defined by the two<br />
atoms, which are a phosphate atom in the head group and the hydrocarbon atom of lipid chain<br />
tail. The vector −→ R1(t) is applied for the unsaturated acyl chain, the vector −→ R2(t) is applied for the<br />
saturated acyl chain, and the vector −→ r (t) is given by averaging the vectors −→ R1(t) and −→ R2(t). The<br />
symmetry parameter S(t) at each time steps is estimated by following equation;<br />
S(t) = 1<br />
N<br />
� �<br />
� N� �<br />
�<br />
�<br />
−→ �<br />
ri (t) �<br />
� � .<br />
By this calculation, we can assess a symmetry property of the micelle. The low S value means<br />
that the system has high symmetry. From the time series of the calculated symmetry value S(t),<br />
we analyze the dynamics of spherical micelle. We also calculated time correlation function (TCF)<br />
and direction correlation (DC) to investigate the correlation between micelle dimer in system. The<br />
time correlation function is defined as<br />
C(t) = 1<br />
T<br />
�T<br />
0<br />
i=1<br />
S1(t ′ )S2(t ′ + t)dt ′ ,<br />
where Si is a symmetry value of micelle i. The time correlation between S1(t) and S2(t ′ +t) should<br />
be evaluated by ensemble average of these values. We can also estimate autocorrelation function<br />
of symmetry parameter of each micelle when we take Si of the same micelle in this equation. We<br />
compare the calculated TCF between S1 and S2, and the autocorrelation function of symmetry Si<br />
of each micelle.<br />
We also calculated direction correlation (DC) to see the correlation between micelle dimer in<br />
direction. This calculation shows the degree parallel of micelle dimer. We evaluated this correlation<br />
by following equation;<br />
D ij<br />
C (t, t′ −→<br />
Si(t)<br />
) =<br />
−→ Sj(t ′ )<br />
| −→ Si(t)|| −→ Sj(t ′ )| ,<br />
where −→ S (t) means the average of lipid vector in the micelle, −→ S (t) = 1<br />
N<br />
DC value equals to zero, they have anti-parallel direction.<br />
3 Results and Discussion<br />
N� −→<br />
ri (t). If the calculated<br />
From each MD simulation, we obtained trajectories of two micelles in water solvent. Here, we<br />
simply call those micelles as the micelle 1 and micelle 2. Figure 2 shows the distance between the<br />
71<br />
i=1
ISCS 2011 Selected Papers Vol.2 Gia et al.<br />
(a) NPT at 340 K (b) NPT at 370 K<br />
(c) NVT at 340 K (d) NVT at 370 K<br />
Figure 2: Distance between the center of micelle 1 and the center of micelle 2. At higher temperature di distance<br />
is getting closer.<br />
center of mass of micelle 1 and the center of micelle 2. We found that the micelle 1 and micelle<br />
2 were getting closer at higher temperature (T = 370K) as the MD time step increases in the<br />
constant NVT condition. In order to assess the system equilibrium, we calculated RMSD of the<br />
micelles.<br />
Figure 3 shows the calculated RMSD of each micelle in several conditions. We found that the all<br />
RMSD were roughly equilibrated in 10 ns MD time. We then analyzed the fluctuation of symmetry<br />
parameter S(t) as a function of time. The results in both constant NVT and NPT simulations at<br />
two temperatures are shown in Figure 4. The results show that the system is randomly fluctuate<br />
during the simulation time.<br />
Table 1 shows standard deviation of symmetry parameter S. The small standard deviation indicates<br />
that the data points tend to be very close to the average value, whereas the large standard<br />
deviation indicates that the data spread out in wide range. We found that the standard deviation<br />
obtained from the constant NPT simulation shows a larger value than that in the constant<br />
NVT simulation, indicating that the structure of micelle in the NPT simulation should be largely<br />
fluctuated.<br />
For further analysis, we calculated the TCF of micelle in each MD condition. The calculated<br />
results are shown in Figure 5. The dotted line shows the time correlation function as a function<br />
72
ISCS 2011 Selected Papers Vol.2 Temperature Effects on Dynamics of Spherical Micelle<br />
(a) NPT at 340 K (b) NPT at 370 K<br />
(c) NVT at 340 K (d) NVT at 370 K<br />
Figure 3: RMSD calculation. At lower temperature, system seems more equilibrium than at higher temperature.<br />
Ensemble T Micelle SD Symmetry<br />
NPT<br />
340 K<br />
370 K<br />
1<br />
2<br />
1<br />
2<br />
2.62<br />
1.99<br />
6.48<br />
3.13<br />
NVT<br />
340 K<br />
370 K<br />
1<br />
2<br />
1<br />
2<br />
4.58<br />
4.37<br />
2.16<br />
4.32<br />
Table 1: Standard Deviation of Symmetry<br />
of time steps between the micelle 1 and micelle 2. We did not see the usual relaxation between<br />
the symmetry of micelle dimer in this time range. This imply the there is no correlation between<br />
micelle 1 and 2. The larger correlation time or sufficient statistic average should be used for the<br />
observation of relaxation of the TCF. The other lines, which indicate the autocorrelation function<br />
of each micelle, show different result with the dotted line. The observed TCF monotonically<br />
decrease value, implying the relaxation of TCF of each micelle in this time. However, further<br />
statistic average should be taken for the clear this relaxation property.<br />
73
ISCS 2011 Selected Papers Vol.2 Gia et al.<br />
(a) NPT at 340 K<br />
(b) NPT at 370 K<br />
(c) NVT at 340 K<br />
(d) NVT at 370 K<br />
Figure 4: Symmetry. It shows that system are fluctuative.<br />
Besides analyzing correlation in time, we investigate the correlation of micelles in direction. As<br />
shown in Figure 6, the direction of micelle 1 and micelle 2 were shown to be randomly changed<br />
during the MD time steps in all simulation conditions. Figure 7 shows that, the micelle 1 and micelle<br />
2 are more anti-parallel under the constant NPT condition at higher temperature (T = 370K),<br />
while they are more parallel under constant NVT condition at the same temperature.<br />
4 Conclusion<br />
We have carried out the molecular dynamics simulations of spherical micelle dimer in water solvent<br />
at two different temperatures under the constant NPT and NVT condition to investigate the<br />
temperature effects on this system. We evaluated the symmetry parameter S of the micelle dimer<br />
system at each MD time steps and showed that the micelle largely uctuates at higher temperature<br />
under the constant NPT condition, while the large fluctuation was shown at lower temperature<br />
in the constant NVT condition. In the calculation of the time correlation function of micelle<br />
74
ISCS 2011 Selected Papers Vol.2 Temperature Effects on Dynamics of Spherical Micelle<br />
(a) NPT at 340 K (b) NPT at 370 K<br />
(c) NVT at 340 K (d) NVT at 370 K<br />
Figure 5: Time Correlation Function<br />
(a) NPT at 340 K (b) NPT at 370 K<br />
(c) NVT at 340 K (d) NVT at 370 K<br />
Figure 6: Direction Correlation<br />
75
ISCS 2011 Selected Papers Vol.2 Gia et al.<br />
(a) NPT at 340 K (b) NPT at 370 K<br />
(c) NVT at 340 K (d) NVT at 370 K<br />
Figure 7: Distribution of Direction Correlation<br />
system, we found that there is not any correlation between micelle 1 and micelle 2, whereas the<br />
micelles have sufficient correlation with themselves. Also at constant NPT condition, the micelle 1<br />
and micelle 2 are more anti-parallel at higher temperature, while they are more parallel at higher<br />
temperature in the constant NVT condition.<br />
References<br />
[1] A. Purqon (2008). Shape Fluctuation Modes and Synchronization Patterns in Self-Assembly<br />
Aggregate Bionanocluster, Kanazawa University, Japan.<br />
[2] C. J. Högberg; A. M. Nikitin; A. P. Lyubartsev. (2008). Modification of the CHARMM Force<br />
Field for DMPC Lipid Bilayer. J. Comput. Chem., 29, 2359-2369.<br />
[3] C. Moitzi; I. Portnaya; O. Glatter; O. Ramon; D. Danino. (2008). Effect of Temperature<br />
on Self-Assembly of Bovine -Casein above and below Isoelectric pH. Structural Analysis by<br />
Cryogenic-Transmission Electron Microscopy and Small-Angle X-ray Scattering. Langmuir,<br />
24, 3020-3029.<br />
[4] L. Martinez, R. Andrade, E. G. Birgin, J. M. Martinez (2008). Packmol: A Package for<br />
Building Initial Configurations for Molecular Dynamics Simulations. Journal of Comp. Chem.,<br />
30, 2157 - 2164.<br />
76
ISCS 2011 Selected Papers Vol.2 Temperature Effects on Dynamics of Spherical Micelle<br />
[5] J. B. Klauda, et al. (2010). Update of the CHARMM All-Atom Additive Force Field for Lipids:<br />
Validation on Six Lipid Types. J. Phys. Chem., 114, 7830 - 7843.<br />
[6] John, K. and Bar, M. (2005). Alternative Mechanisms of Structuring Biomembranes: Self-<br />
Assembly versus Self-Organization. Phys. Rev. Lett., 95, 198101.<br />
[7] M T Nelson, et al. (1996). NAMD: a Parallel, Object-Oriented Molecular Dynamics Program.<br />
International Journal of High Performance Computing Applications, 10, 251-268.<br />
[8] O. Domenech; S. M. Montero; J. H. Borell. (2006). Colloids and Surface, 47, 102.<br />
[9] S. Pal; S. Balasubramanian; B. Bagchi. (2002). Temperature Dependece of Water Dynamics<br />
at an Aquoeous Micellar Surface : Atomistic Molecular Dynamics Simulation Studies of a<br />
Complex System. J. Chem. Phys., 117, 2852.<br />
[10] X. Toxvaerd. (1993). Molecular Dynamics at Constant Temperature and Pressure. Phys. Rev.<br />
E, 47, 343-350.<br />
77
International Symposium on Computational Science 2011<br />
Date: February 15 – 17, 2011<br />
Venue: Kanazawa University, Kakuma Campus, Kanazawa, Japan<br />
Organized by:<br />
Building of Natural Science and Technology,<br />
Lecture rooms n.103 and n.104 (Feb. 15-16),<br />
Lecture room n.301 (Feb. 17)<br />
Organizing Committee:<br />
Department of Computational Science,<br />
Faculty of Science, Kanazawa University<br />
Faculty of Mathematics and Natural Sciences,<br />
Bandung Institute of Technology<br />
K. Svadlenka (Kanazawa University)<br />
M. Saito (Kanazawa University)<br />
K. Nishikawa (Kanazawa University)<br />
S. Omata (Kanazawa University)<br />
H. Nagao (Kanazawa University)<br />
S. Miura (Kanazawa University)<br />
M. A. Martoprawiro (Bandung Institute of Technology)<br />
S. Haryono (Bandung Institute of Technology)<br />
R. Simanjuntak (Bandung Institute of Technology)<br />
R. Hertadi (Bandung Institute of Technology)<br />
Sponsored by:<br />
Fujitsu Limited<br />
JSPS Grant-in-Aid (No. 21654013, S. Omata)<br />
v
February 15 (Tuesday): 1 st day<br />
10:00-10:30 Registration<br />
10:30-10:40 Opening address<br />
Symposium Program<br />
Morning session (room 103, chairperson: Mineo Saito)<br />
10:40-11:20 Laksana Tri Handoko (Indonesian Institute of Sciences)<br />
Modeling nano particle dynamics using statistical mechanics approach<br />
11:20-12:00 Suprijadi Haryono (Bandung Institute of Technology)<br />
Uniaxial stress effect on crystal dislocations movement and crack propagation in cubic crystal<br />
Lunch break<br />
Afternoon session 1 (room 103, chairperson: Kiyoshi Nishikawa)<br />
14:00-14:40 Harno Pranowo (Universitas Gadjah Mada)<br />
Microsolvation conformation of crown ether-transition metal cation complexes by ab initio<br />
method<br />
14:40-15:20 Muhamad A. Martoprawiro (Bandung Institute of Technology)<br />
Computational study of iron(II)-based spin transition complex<br />
Afternoon session 2 (room 104, chairperson: Karel Svadlenka)<br />
14:00-14:40 Shinya Okabe (Iwate University)<br />
The existence of shortening-straightening flow for non-closed planar curves with infinite length<br />
14:40-15:20 Hideki Murakawa (University of Toyama)<br />
Numerical simulations of nonlinear cross-diffusion systems using a linear scheme<br />
Coffee & Poster break<br />
Evening session (room 103, chairperson: Shinichi Miura)<br />
16:00-16:40 Makoto Iima (Hokkaido University)<br />
Numerical simulation and mathematical analysis of flapping flight problem<br />
16:40-16:55 Muhamad Koyimatu (Kanazawa University & Bandung Institute of Technology)<br />
Quantum control of high harmonic generation in anharmonic potential<br />
16:55-17:10 Athiya M. Hanna (Kanazawa University & Bandung Institute of Technology)<br />
High-pressure crystal structure prediction using evolutionary algorithm simulation<br />
17:10-17:25 Nyayu Siti Nurainun (Kanazawa University & Bandung Institute of Technology)<br />
Hydrogen impurities in graphene: first principles study<br />
17:25-17:40 Micke Rusmerryani (Kanazawa University & Bandung Institute of Technology)<br />
Molecular dynamics studies on structure and dynamics of spherical micelles<br />
17:40-17:55 Gia Septiana Wulandari (Kanazawa University & Bandung<br />
Institute of Technology):<br />
Temperature effects on dynamics of spherical micelles: a molecular dynamics study<br />
vi
February 16 (Wednesday): 2 nd day<br />
Morning session (room 103, chairperson: Shinichi Miura)<br />
10:00-10:40 Hitoshi Imai (University of Tokushima)<br />
Numerical simulation on non-existence and non-uniqueness of solutions for the Tricomi<br />
equation<br />
10:40-11:20 Sparisoma Viridi (Bandung Institute of Technology)<br />
Self-siphon simulation using molecular dynamics method: a preliminary study<br />
11:20-12:00 Zaki Su’ud (Bandung Institute of Technology)<br />
Unprotected loss of flow simulation in fast reactor safety analysis<br />
Lunch break<br />
Afternoon session 1 (room 103, chairperson: Fumiyuki Ishii)<br />
14:00-14:25 Hiroaki Saito (Kanazawa University)<br />
Hydration property of globular proteins: a molecular dynamics study<br />
14:25-14:50 Kazutomo Kawaguchi (Kanazawa University)<br />
The molecular dynamics simulation of protein complexes<br />
14:50-15:15 Tatsuki Oda (Kanazawa University)<br />
Toward a computer modeling in magnetic anisotropy and its electric-field-control for<br />
nano-structures<br />
15:15-15:40 Shinichi Miura (Kanazawa University)<br />
Variational path integral molecular dynamics method applied to molecular vibrational<br />
fluctuations<br />
Afternoon session 2 (room 104, chairperson: Hideki Murakawa)<br />
14:00-14:40 Katsuyuki Ishii (Kobe University)<br />
Mathematical analysis for an approximation scheme to mean curvature flow<br />
14:40-15:20 Takeshi Fukao (Kyoto University of Education)<br />
Variational inequality for the Navier-Stokes equations with time dependent constraint<br />
15:20-15:40 Elliott Ginder (Kanazawa University)<br />
A variational method for volume-controlled membrane motions<br />
Coffee & Poster break<br />
Evening session 1 (room 103, chairperson: Tatsuki Oda)<br />
16:20-16:50 Phung Thi Viet Bac (AIST)<br />
Theoretical investigations of hydrogen diffusion into metallic nanoparticles<br />
16:50-17:15 Fumiyuki Ishii (Kanazawa University)<br />
First-principles study of ferroelectricity in hydrogen-bonded molecular systems<br />
17:15-17:40 Mineo Saito (Kanazawa University)<br />
First-principles calculations of defects in graphenes<br />
17:40-17:55 Shuhei Kawamoto (Kanazawa University)<br />
Free energy of peptide to permeate lipid bilayer membrane -- coarse-grained simulation<br />
17:55-18:10 Muhammad Ilyas (Kanazawa University & Bandung Institute of Technology)<br />
Generalized extended Hamming codes over Galois ring of characteristic 2 n<br />
vii
Evening session 2 (room 104, chairperson: Takeshi Fukao)<br />
16:20-17:00 Naoto Nakano (Hokkaido University)<br />
On steady simple shear flows of a continuum model with density gradient-dependent stress<br />
17:00-17:25 Hiroshi Iwasaki (Kanazawa University)<br />
Numerical simulation of Biot's consolidation problem<br />
17:25-17:40 Triati Dewi Kencana Wungu (Osaka University)<br />
Study of lithium montmorillonite by ab initio calculation<br />
February 17 (Thursday): 3 rd day<br />
Morning session (room 301, chairperson: Hiroaki Saito)<br />
9:20-9:35 Putu Harry Gunawan (Kanazawa University & Bandung Institute of Technology)<br />
Simulation of surface detection and surface tension with smoothed particle hydrodynamics<br />
9:35-9:50 Mourice C. K. Woran (Kanazawa University & Bandung Institute of Technology)<br />
Simulation of fluid-solid interaction using moving particle semi-implicit and spring-mass<br />
system<br />
9:50-10:05 Ruddy Kurnia (Kanazawa University & Bandung Institute of Technology)<br />
A novel scheme smoothed particle hydrodynamics to overcome energy loss<br />
10:05-10:20 Christian Fredy Naa (Kanazawa University & Bandung Institute of Technology)<br />
Explicit step improvement on moving particle semi-implicit and its analysis by using surface<br />
detection algorithm<br />
Coffee & Poster break<br />
10:50-11:30 Akhmaloka (Bandung Institute of Technology)<br />
Thermophilic microorganisms and thermostable enzyme from Indonesia hot springs<br />
11:30-12:00 Hidemi Nagao (Kanazawa University)<br />
Reduction potential of blue copper proteins<br />
viii
List of Participants<br />
Akhmaloka Rector of Bandung Institute of Technology, Indonesia<br />
Phung Thi Viet Bac<br />
Takeshi Fukao<br />
Elliott Ginder<br />
Putu Harry Gunawan<br />
Laksana Tri<br />
Handoko<br />
Athiya Mahmud<br />
Hanna<br />
Suprijadi Haryono<br />
Nanosystem Research Institute, National Institute of Advanced Industrial Science<br />
and Technology (AIST), Japan, E-mail: phung.bac@aist.go.jp<br />
Department of Mathematics, Faculty of Education, Kyoto University of Education,<br />
Japan, E-mail: fukao@kyokyo-u.ac.jp<br />
Graduate School of Natural Science and Technology, Kanazawa University, Japan,<br />
E-mail: eginder@polaris.s.kanazawa-u.ac.jp (PhD student)<br />
KU-ITB Double-Degree Program* Master Student,<br />
E-mail: erik_mathboy@yahoo.com<br />
Group for Theoretical and Computational Physics Research Center for Physics,<br />
Indonesian Institute of Sciences, Indonesia, E-mail: handoko@teori.fisika.lipi.go.id<br />
KU-ITB Double-Degree Program* Master Student,<br />
E-mail: athiya_mh@yahoo.com<br />
Faculty of Mathematics and Natural Sciences, Bandung Institute of Technology,<br />
Indonesia, E-mail: supri@fi.itb.ac.id<br />
Yasuaki Hiwatari Toyota Physical and Chemical Research Institute, Japan<br />
Makoto Iima<br />
Department of Mathematics, Hokkaido University, Japan,<br />
E-mail: makoto@nsc.es.hokudai.ac.jp<br />
Muhammad Ilyas KU-ITB Double-Degree Program* Master Student, E-mail: reiken7@gmail.com<br />
Hitoshi Imai<br />
Fumiyuki Ishii<br />
Katsuyuki Ishii<br />
Hiroshi Iwasaki<br />
Kazutomo<br />
Kawaguchi<br />
Shuhei Kawamoto<br />
Masaki Kazama<br />
Muhamad Koyimatu<br />
Ruddy Kurnia<br />
Wicharn<br />
Lewkeeratiyutkul<br />
Department of Applied Physics and Mathematics, Faculty of Engineering,<br />
University of Tokushima, Japan, E-mail: imai@pm.tokushima-u.ac.jp<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: fumiyuki@kanazawa-u.ac.jp<br />
Faculty of Maritime Sciences, Kobe University, Japan,<br />
E-mail: ishii@maritime.kobe-u.ac.jp<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: iwasaki@cs.s.kanazawa-u.ac.jp<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: kkawa@wriron1.s.kanazawa-u.ac.jp<br />
Graduate School of Natural Science and Technology, Kanazawa University, Japan,<br />
E-mail: kawamoto@wriron1.s.kanazawa-u.ac.jp (PhD student)<br />
Fujitsu Limited Next Generation Technical Computing Unit,<br />
E-mail: kazama.masaki@jp.fujitsu.com<br />
KU-ITB Double-Degree Program* Master Student,<br />
E-mail: koyimatu@wriron1.s.kanazawa-u.ac.jp<br />
KU-ITB Double-Degree Program* Master Student,<br />
E-mail: ruddy.kurnia@gmail.com<br />
Department of Mathematics, Faculty of Science, Chulalongkorn University,<br />
E-mail: Wicharn.L@chula.ac.th<br />
ix
Muhamad<br />
Abdulkadir<br />
Martoprawiro<br />
Khamron Mekchay<br />
Shinichi Miura<br />
Hideki Murakawa<br />
Christian Fredy Naa<br />
Hidemi Nagao<br />
Faculty of Mathematics and Natural Sciences, Bandung Institute of Technology,<br />
Indonesia, E-mail: muhamad@chem.itb.ac.id<br />
Department of Mathematics, Faculty of Science, Chulalongkorn University,<br />
Email: k.mekchay@gmail.com<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: smiura@mail.kanazawa-u.ac.jp<br />
Graduate School of Science and Engineering for Research, University of Toyama,<br />
Japan, E-mail: murakawa@sci.u-toyama.ac.jp<br />
KU-ITB Double-Degree Program* Master Student,<br />
E-mail: chris_mail@yahoo.com<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: nagao@wriron1.s.kanazawa-u.ac.jp<br />
Naoto Nakano Hokkaido University, Japan, E-mail: nakano.naoto@gmail.com<br />
Kiyoshi Nishikawa<br />
Nyayu Siti Nurainun<br />
Tatsuki Oda<br />
Shinya Okabe<br />
Seiro Omata<br />
Harno Dwi Pranowo<br />
Micke Rusmerryani<br />
Hiroaki Saito<br />
Mineo Saito<br />
Zaki Su'ud<br />
Karel Svadlenka<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: kiyoshi@wriron1.s.kanazawa-u.ac.jp<br />
KU-ITB Double-Degree Program* Master Student,<br />
E-mail: nyayu@cphys.s.kanazawa-u.ac.jp<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: oda@cphys.s.kanazawa-u.ac.jp<br />
Faculty of Humanities and Social Sciences, Environmental Sciences, Iwate<br />
University, Japan, E-mail: okabes@iwate-u.ac.jp<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: omata@kenroku.kanazawa-u.ac.jp<br />
Austrian-Indonesian Center for Computational Chemistry, Chemistry Department,<br />
Faculty of Mathematics and Natural Sciences, Universitas Gadjah Mada,<br />
Yogyakarta, Indonesia, E-mail: harnodp@ugm.ac.id<br />
KU-ITB Double-Degree Program* Master Student,<br />
E-mail: micke@wriron1.s.kanazawa-u.ac.jp<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: saito@wriron1.s.kanazawa-u.ac.jp<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: m-saito@cphys.s.kanazawa-u.ac.jp<br />
Department of Physics, Bandung Institute of Technology, Indonesia,<br />
E-mail: szaki@fi.itb.ac.id<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: kareru@staff.kanazawa-u.ac.jp<br />
Masako Takasu Tokyo University of Pharmacy and Life Sciences, E-mail: takasu@toyaku.ac.jp,<br />
Nuclear Physics and Biophysics Research Division, Faculty of Mathematics and<br />
Sparisoma Viridi Natural Sciences, Bandung Institute of Technology, Indonesia,<br />
E-mail: dudung@fi.itb.ac.id<br />
Mourice Ceisper<br />
Kevin Woran<br />
KU-ITB Double-Degree Program* Master Student,<br />
E-mail: moris_3gun@polaris.s.kanazawa-u.ac.jp<br />
x
Gia Septiana<br />
Wulandari<br />
Triati Dewi<br />
Kencana Wungu<br />
Mieko Yamada<br />
KU-ITB Double-Degree Program* Master Student,<br />
E-mail: gia@wriron1.s.kanazawa-u.ac.jp<br />
Department of Precision Science & Technology and Applied Physics, Graduate<br />
School of Engineering, Osaka University, Japan,<br />
E-mail: triati@dyn.ap.eng.osaka-u.ac.jp<br />
Institute of Science and Engineering, Kanazawa University, Japan,<br />
E-mail: myamada@kenroku.kanazawa-u.ac.jp<br />
* This Double-Degree Program is an educational program based on the agreement of cooperation<br />
between Kanazawa University and Bandung Institute of Technology. Therefore, the affiliation of<br />
students participating in this program is as follows:<br />
Graduate School of Natural Science and Technology, Kanazawa University, Japan &<br />
Faculty of Mathematics and Natural Sciences, Bandung Institute of Technology, Indonesia<br />
xi