02.06.2014 Views

Siesta I - Psi-k

Siesta I - Psi-k

Siesta I - Psi-k

SHOW MORE
SHOW LESS

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

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

Linear-scaling DFT<br />

and the SIESTA method<br />

Emilio Artacho<br />

Department of Earth Sciences<br />

University of Cambridge


DFT: successful but heavy<br />

• Computationally much more expensive than<br />

empirical atomic simulations<br />

• Several thousand atoms in massively parallel<br />

supercomputers<br />

!<br />

• Computational load ~ N 3<br />

" n<br />

" m<br />

= " n<br />

*<br />

# ( r )" m<br />

( r )d 3 r = $ n,m<br />

An order of magnitude increase in computer<br />

power allows a mere doubling of system size


LINEAR SCALING<br />

CPU<br />

load<br />

~ N 3<br />

Early<br />

90’s<br />

~ N<br />

~ 100<br />

N (# atoms)<br />

G. Galli and M. Parrinello, Phys. Rev Lett. 69, 3547 (1992)


Linear-scaling KS-DFT methods<br />

• LCAO: - Gaussian based + QC machinery<br />

G. Scuseria (GAUSSIAN),<br />

M. Head-Gordon (Q-CHEM)<br />

M. Challacombe<br />

- Numerical atomic orbitals (NAO)<br />

SIESTA<br />

S. Kenny &. A Horsfield (PLATO)<br />

- Gaussian with hybrid machinery<br />

J.Vandevondele, J.Hutter, M.Parrinello (CP2K)<br />

• Bessel functions in ovelapping spheres<br />

P. Haynes & M. Payne<br />

• B-splines in 3D grid (finite-elements)<br />

D. Bowler, E. Hernandez & M. Gillan (CONQUEST)<br />

• Grid-based (plane-wave like) methods<br />

J. Bernholc (finite differences, “nearly order-N”), Fattebert<br />

A.Mostofi, C.Skylaris, P.Haynes, M.Payne (FFT-box, ONETEP)


Orbital-free linear-scaling DFT<br />

• Directly propose an approximate energy functional of<br />

the density<br />

• Difficulty: Kinetic energy functional. No ψ<br />

KS<br />

n for<br />

(and local pseudopotentials)<br />

• Linear scaling & extremely efficient<br />

(as empirical potentials)<br />

T = " h2<br />

2<br />

occ<br />

%<br />

n<br />

# 2 $ n<br />

• Simple metals<br />

!<br />

Paul Madden, Emily Carter & collaborators


TWO STEPS IN KS-DFT<br />

two different problems for linear scaling<br />

• Given a Hamiltonian, H ,<br />

obtain Ground State properties<br />

92-94 fever<br />

tight binding<br />

• Obtain H (and S) matrices<br />

(selfconsistently)<br />

- Long range interactions<br />

(electrostatics)<br />

- The rest (T,V xc , S…)<br />

Since 95<br />

DFT, HF, etc


HARTREE (electrostatics)<br />

• SIESTA (and others): real-space grids<br />

FFT, Multigrid (Numerical Recipes)<br />

• Quantum Chemistry: Fast multipoles<br />

Head-Gordon<br />

Scuseria<br />

Multigrid allows 0D, 1D, 2D & 3D bound cond<br />

and implicit solvers, Poisson-Boltzmann etc.


KEY: LOCALITY<br />

Large system<br />

x 2<br />

“Divide and Conquer”<br />

W. Yang, Phys. Rev. Lett. 66, 1438 (1992)


SIESTA method<br />

Linear-scaling DFT based on<br />

NAOs (Numerical Atomic Orbitals)<br />

P. Ordejon, E. Artacho & J. M. Soler , Phys. Rev. B 53, R10441 (1996)<br />

•Born-Oppenheimer (relaxations, mol. dynamics)<br />

•DFT<br />

(LDA, GGA: BLYP, PBE, RPBE, revPBE)<br />

•Pseudopotentials (norm conserving, factorised)<br />

•Numerical atomic orbitals as basis (finite range)<br />

•Numerical evaluation of matrix elements (3D grid)<br />

Implemented in the SIESTA program (and others)<br />

J. M. Soler, E. Artacho, J. D. Gale, A. Garcia, J. Junquera, P. Ordejon &<br />

D. Sanchez-Portal, J. Phys.: Condens. Matter 14, 2745 (2002)<br />

http://www.uam.es/siesta


Finite-support atomic orbitals as basis<br />

s<br />

p<br />

d<br />

f<br />

Strictly localised<br />

(zero beyond cut-off radius)


Hard confining potentials<br />

A single parameter<br />

Energy shift<br />

E. Artacho et al. Phys. Stat. Solidi (b) 215, 809 (1999)<br />

Fireballs<br />

O. F. Sankey & D. J. Niklewski,<br />

Phys. Rev. B 40, 3979 (1989)<br />

Convergence vs Energy shift of<br />

Bond lengths Bond energies<br />

BUT:<br />

A different cut-off radius for<br />

each orbital


Soft confining potentials<br />

1 3 5 7<br />

r (a.u.)<br />

1 3 5 7<br />

r (a.u.)<br />

• Better basis, variationally, & other results<br />

• Removes the discontinuity in the derivative<br />

J. Junquera, O. Paz, D. Sanchez-Portal & E. Artacho, Phys. Rev. B, 64, 235111 (2001)<br />

E. Anglada, J. M. Soler, J. Junquera & E. Artacho, Phys. Rev. B 66, 205101 (2002)


Multiple-zeta<br />

E. Artacho et al. , Phys. Stat. Solidi (b) 215, 809 (1999).


Polarization<br />

E. Artacho et al. , Phys. Stat. Solidi (b) 215, 809 (1999).


Linear-scaling matrix-element<br />

calculations<br />

FINITE SUPPORT BASIS FUNCTIONS<br />

S µ"<br />

,T µ"<br />

,V µ"<br />

PS,nl<br />

Two-centre integrals<br />

1D Numerical (tabulated)<br />

!<br />

!<br />

V xc µ"<br />

(#),V Hartree NA<br />

µ"<br />

($#),V µ"<br />

atoms<br />

3D integral in finite real-space grid<br />

!<br />

atoms<br />

&<br />

n<br />

"# $ # % # n<br />

atomic<br />

V NA " #{<br />

V PS,local $V Hartree (% atomic n<br />

)}<br />

n


!<br />

Grid integrations<br />

. . . . . . . . . . . . . .<br />

. . . . . . . . . . . . . .<br />

. . . .<br />

"<br />

. . . . . . . . . .<br />

. . . . . µ<br />

( r )<br />

. . . . . . . . .<br />

. . . . . . . . . . . . . .<br />

. . . . . . . . . . . . . .<br />

. . . . . . . . ". (. r r ) #<br />

. . . .<br />

. . . . . . . . . . . . . .<br />

. . . . . . . . . . . . . .<br />

Finite-range orbitals => Lists of points<br />

!<br />

Sparse matrices


Eggbox<br />

50 Ry<br />

4.4 Bohr<br />

Common to<br />

many<br />

methods!<br />

Breaking of translational invariace with 3D grid:<br />

- Grid-cell sampling (JM Soler et al JPCM 2002)<br />

- Filtering (E Anglada & JM Soler, PRB 2006)


Solving H (diagonalising)<br />

Nearsightedness principle<br />

Implies localisation:<br />

W. Kohn, Phys. Rev. Lett. 76, 3168 (1996)<br />

"( r r , r # r )<br />

"( r r # r<br />

R )<br />

' '' % 0<br />

r $ r #%&<br />

& && $ 0<br />

r # R r<br />

$%<br />

Wannier-like unitary transformations of eigenvectors:<br />

!<br />

localised basis of occupied space<br />

!


Linear-scaling solution<br />

!<br />

!<br />

• Search directly for localised quantities:<br />

-<br />

E BS<br />

"<br />

occ<br />

$ # n<br />

= E %( r<br />

n<br />

[ r , r &)]<br />

- by minimising E<br />

E BS<br />

= E<br />

[{ }]<br />

" ( r r ) i<br />

• Make them strictly local (finite support)<br />

(the approximation is here)


Functional of LWFs<br />

Where H and S are the overlap and Hamiltonian matrices in<br />

the basis of the (unknown) localised wave functions ϕ i .<br />

E. B. Stechel, A. R. Williams & P. J. Feibelman, Phys. Rev. B 49, 10088 (1994)<br />

!<br />

E BS<br />

J.-L. Fattebert & J. Bernholc, Phys. Rev. B 62 1713 (2000)<br />

The inverse of the overlap matrix can been avoided:<br />

E BS<br />

[{" }] i<br />

= 2 Tr{ S #1 H}<br />

[ ] = 2 Tr 2 # S<br />

{" } i<br />

{ }<br />

( )H<br />

P. Ordejon et al. Phys. Rev. B 48, 14646 (1993)<br />

F. Mauri & G. Galli, Phys. Rev. B 50, 4316 (1994)<br />

!<br />

The functionals are different but equal at the minimum


Localised solutions<br />

R c


Problem: local minima<br />

Excess charge<br />

on C atoms<br />

Slab of 16 layers of diamond<br />

Kim, Mauri and Galli, PRB 52, 1640 (1995)


Local minima: a way to see it<br />

Consider a square lattice with<br />

• 4 electrons per atom<br />

• 4 bonds surrounding each atom<br />

• Each bond described by a LWF<br />

• => 2 LWF per atom<br />

• Distribute: each atom “donates” 2<br />

bonds and “receives” 2


Local minima: a way to see it<br />

Consider a square lattice with<br />

• 4 electrons per atom<br />

• 4 bonds surrounding each atom<br />

• Each bond described by a LWF<br />

• => 2 LWF per atom<br />

• Distribute: each atom “donates” 2<br />

bonds and “receives” 2


Kim, Mauri & Galli’s functional<br />

Minimise the following varying φ i (LWF’s)<br />

It removes problem of local minima but<br />

requires predetermination of Fermi level<br />

Kim, Mauri and Galli, PRB 52, 1640 (1995)


KIM’s eta & ill-conditioning<br />

Effect of eta.<br />

Need to predefine it<br />

Ill-conditioning: Minimisation algorithm (CG) prop to<br />

Condition Number: Max curvature / Min curvature<br />

(problem common to many of the schemes)


Other linear-scaling solvers<br />

• Divide & Conquer (W. Yang)<br />

• Li, Nunes, Vandebilt (LNV)<br />

(and variations)<br />

[ ]<br />

E = E "( r r , r r ')<br />

BS<br />

• Goedecker’s (two) (Projecting onto Occupied space)<br />

!<br />

• Energy renormalisation group (ERG, Baer et al)<br />

METALS


Actual linear scaling<br />

Pentium III 800 MHz<br />

(single processor)<br />

1 Gb of RAM<br />

c-Si supercells, single-ζ<br />

512.000 atoms in 256 nodes<br />

(Julian Gale)


http://www.uam.es/siesta


No DC conduction in λ-DNA<br />

P. J. de Pablo et al. PRL 2000<br />

E. Artacho et al., Mol. Phys. 2003<br />

But possible polaron conduction in dry poly dG-poly dC (A-DNA)<br />

S. Alexandre, E. Artacho, J.M. Soler & H. Chacham, PRL 2003<br />

HOMO band (blue)<br />

LUMO band (red)<br />

Hole polarons:<br />

E binding = 0.3 eV<br />

Very localised<br />

Move along the<br />

HOMO band


Inhibition of cyclin-dependent kinases<br />

enzymes relevant to cell replication (and to cancer)<br />

L. Heady et al., J. Medic. Chem.2006


Geobacter pili’s pilin (geo-bio-nano!)<br />

Soil bacteria feeding from ferrous minerals by electron transfer<br />

Ambient (fluid) conditions<br />

nA, 0.1 G Ohm<br />

G. Reguera et al. Nature 435, 1098 (2005)<br />

G. Feliciano, A. J. R. da Silva, G. Reguera & EA


Candidate material for spintronics<br />

a<br />

20 ì m<br />

B,M<br />

LSMO<br />

[100]<br />

b<br />

CNT<br />

LSMO<br />

c<br />

LSMO<br />

x<br />

CNT<br />

300 nm<br />

Magnetoresistance in device made of LSMO<br />

strips bridged by CNT<br />

L. E. Hueso et al, Nature (in press; cond-mat/0511697)


Magnetoresistance measurements<br />

50<br />

40<br />

5 K<br />

25 mV<br />

a<br />

V (mV)<br />

0 100 200 300 400 500<br />

b<br />

MC (%)<br />

30<br />

20<br />

10<br />

25 mV<br />

5 K<br />

0<br />

-150 -100 -50 0 50 100 150<br />

B (mT)<br />

0 20 40 60 80 100 120<br />

T (K)<br />

Varying external B field, change relative spin orientation,<br />

important change in resistance<br />

L. E. Hueso et al, Nature 2007


Electronic structure at the LSMO/CNT interface<br />

(explicit doping; same supercell plus (6,6) CNT)


Electronic structure at the LSMO/CNT interface<br />

(explicit doping; same supercell plus (6,6) CNT)


GGA: LSMO (001) surface. x = 1/3<br />

• Supercell: Slab 2x4 (x5.5 layers + vacuum) . 224 atoms, 48 Mn. (exposing<br />

MnO 2<br />

layers on both sides)<br />

• Explicit doping: 27 La, 13 Sr (disordered, 1 realisation)


Tran<strong>Siesta</strong> (J. L. Mozos, M. Brandbyge, K. Stokbro, J. Taylor & P. Ordejon)<br />

Utility in <strong>Siesta</strong> for ballistic electron transport<br />

Non-equilibrium Green’s functions (Keldysh)<br />

Conduction through dithiol<br />

Pulling Au atoms alters conductance<br />

Transmittance vs energy<br />

R. J. C. Batista, P. Ordejon, H. Chacham & E. Artacho, PRB RC 2006<br />

Transport / energetics / geometry on the same footing<br />

But also LDA+U, Pseudo-SIC, Accelerated dynamics, STM, etc


Acknowledgments<br />

Daniel Sanchez-Portal (UPV San Sebastian)<br />

Pablo Ordejon (ICMAB Barcelona)<br />

Jose M. Soler (UAM Madrid)<br />

Julian Gale<br />

(Curtin, Australia)<br />

Alberto Garcia (UPV Bilbao)<br />

Richard Martin (U. Illinois, Urbana)<br />

Javier Junquera (U. Santander)<br />

Oscar Paz<br />

(UAM Madrid)<br />

Eduardo Anglada (UAM Madrid)<br />

SIESTA<br />

gang<br />

Miguel Pruneda, Valeria Ferrari, Nicola Spaldin, Ben Simons,<br />

Peter Littlewood<br />

Luis Hueso, Neil Mathur<br />

Ronaldo Batista, Helio Chacham<br />

Gustavo T. Feliciano, Antonio J. R. da Silva


Summary<br />

• Essentials of linear-scaling DFT around SIESTA<br />

• Exploiting locality explicitly (except for Hartree)<br />

• Still much to do (metals, generality, robustness)<br />

• Larger sizes open up to calculations<br />

• BUT: facing the complexity barrier!

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!