Top Banner
Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida ([email protected]), Steve Plimpton, Aidan Thompson 9/5/17 Sandia National Laboratories is a multi-mission laboratory managed and operated by National Technology and Engineering Solutions of Sandia, LLC., a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525. 2014-00000
17

Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Sep 08, 2019

Download

Documents

dariahiddleston
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Spin–lattice simulations with LAMMPSLAMMPS Workshop and Symposium

Julien Tranchida ([email protected]), Steve Plimpton, Aidan Thompson

9/5/17

Sandia National Laboratories is a multi-mission laboratory managed and operated by National Technology and Engineering Solutions of Sandia, LLC., a whollyowned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.

2014-00000

Page 2: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Introduction

Objective: developing a LAMMPS package for spin–lattice simulations.

Enable study of:

Magnetostriction,Spin–lattice relaxation,Spin dynamics,Topological spinstructures,Spin liquids, ...

Simulations of bismuth oxide and fcc colbalt.

9/5/17 2

Page 3: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Introduction

Objective: developing a LAMMPS package for spin–lattice simulations.

Enable study of:

Magnetostriction,Spin–lattice relaxation,Spin dynamics,Topological spinstructures,Spin liquids, ...

Simulations of bismuth oxide and fcc colbalt.

9/5/17 2

Page 4: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

EOMs for the spin dynamicsFrom the DFT formalism, Antropov et al. derived equations forthe dynamics ot atomic spins [2]:

dsidt

=1

1 + λ2((ωi + ηi)× si + λ si × (ωi × si))

Magnetic interactions:

ωi = −1

ℏ∂HMag

∂si

Spin Hamiltonian:

HMag =N∑

i,j,i̸=jJij (rij) si · sj

+ gµBµ0

N∑i=0

si · Hext

Connection to a random bath:

⟨ηi⟩ = 0 and⟨ηα

i (t)ηβj (t

′)⟩ = 2Dδijδαβδ(t − t′)

Fluctuation–dissipationrelation:

D =2πλkBT

[2] Antropov, V. P. et al. (1996). Phys. Rev. B, 54(2), 1019.9/5/17 3

Page 5: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

EOMs for the spin dynamicsFrom the DFT formalism, Antropov et al. derived equations forthe dynamics ot atomic spins [2]:

dsidt

=1

1 + λ2((ωi + ηi)× si + λ si × (ωi × si))

Magnetic interactions:

ωi = −1

ℏ∂HMag

∂si

Spin Hamiltonian:

HMag =N∑

i,j,i̸=jJij (rij) si · sj

+ gµBµ0

N∑i=0

si · Hext

Connection to a random bath:

⟨ηi⟩ = 0 and⟨ηα

i (t)ηβj (t

′)⟩ = 2Dδijδαβδ(t − t′)

Fluctuation–dissipationrelation:

D =2πλkBT

[2] Antropov, V. P. et al. (1996). Phys. Rev. B, 54(2), 1019.9/5/17 3

Page 6: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

EOMs for the spin dynamicsFrom the DFT formalism, Antropov et al. derived equations forthe dynamics ot atomic spins [2]:

dsidt

=1

1 + λ2((ωi + ηi)× si + λ si × (ωi × si))

Magnetic interactions:

ωi = −1

ℏ∂HMag

∂si

Spin Hamiltonian:

HMag =N∑

i,j,i̸=jJij (rij) si · sj

+ gµBµ0

N∑i=0

si · Hext

Connection to a random bath:

⟨ηi⟩ = 0 and⟨ηα

i (t)ηβj (t

′)⟩ = 2Dδijδαβδ(t − t′)

Fluctuation–dissipationrelation:

D =2πλkBT

[2] Antropov, V. P. et al. (1996). Phys. Rev. B, 54(2), 1019.9/5/17 3

Page 7: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

EOMs for spin–lattice dynamics

Spin–lattice Hamiltonian [3]:

Hsl =N∑

i=1

mi|vi|2

2+

N∑i,j

V (rij) +N∑

i,j,i̸=jJij (rij) si · sj + gµBµ0

N∑i=0

si · Hext

The associated spin–lattice equations of motion are given by [3]:∂ri∂t

= vi

∂vi∂t

= Fi (rij, si,j) =N∑

j,i̸=j

[−

dV (rij)

dr+

dJ (rij)

drsi · sj

]∂si∂t

= ωi × si

[3] Beaujouan, D., et al.. (2012). Phys. Rev. B, 86(17), 174409.9/5/17 4

Page 8: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

EOMs for spin–lattice dynamics

Spin–lattice Hamiltonian [3]:

Hsl =N∑

i=1

mi|vi|2

2+

N∑i,j

V (rij) +N∑

i,j,i̸=jJij (rij) si · sj + gµBµ0

N∑i=0

si · Hext

The associated spin–lattice equations of motion are given by [3]:∂ri∂t

= vi

∂vi∂t

= Fi (rij, si,j) =N∑

j,i̸=j

[−

dV (rij)

dr+

dJ (rij)

drsi · sj

]∂si∂t

= ωi × si

[3] Beaujouan, D., et al.. (2012). Phys. Rev. B, 86(17), 174409.9/5/17 4

Page 9: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Numerical integrationAdvance operators are not commuting, a Suzuki–Trotterdecomposition has to be used [4]:

ST decomposition:

v← v + Lv.∆t/2

r← r + Lr.∆t/2

s← s + Ls.∆t

r← r + Lr.∆t/2

v← v + Lv.∆t/2

s0 ← s0 + Ls0 .∆t/2

sN−1 ← sN−1 + LsN−1 .∆t/2

sN ← sN + LsN .∆t

sN−1 ← sN−1 + LsN−1 .∆t/2

s0 ← s0 + Ls0 .∆t/2

Numerical results:

Sim. param.: λ = 0, Hext = 10T along

ez, JCo

[4] Omelyan, I. P. et al. (2001). Phys. Rev. Lett., 86(5), 898.9/5/17 5

Page 10: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Numerical integrationAdvance operators are not commuting, a Suzuki–Trotterdecomposition has to be used [4]:

ST decomposition:

v← v + Lv.∆t/2

r← r + Lr.∆t/2

s← s + Ls.∆t

r← r + Lr.∆t/2

v← v + Lv.∆t/2

s0 ← s0 + Ls0 .∆t/2

sN−1 ← sN−1 + LsN−1 .∆t/2

sN ← sN + LsN .∆t

sN−1 ← sN−1 + LsN−1 .∆t/2

s0 ← s0 + Ls0 .∆t/2

Numerical results:

Sim. param.: λ = 0, Hext = 10T along

ez, JCo

[4] Omelyan, I. P. et al. (2001). Phys. Rev. Lett., 86(5), 898.9/5/17 5

Page 11: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Numerical integrationAdvance operators are not commuting, a Suzuki–Trotterdecomposition has to be used [4]:

ST decomposition:

v← v + Lv.∆t/2

r← r + Lr.∆t/2

s← s + Ls.∆t

r← r + Lr.∆t/2

v← v + Lv.∆t/2

s0 ← s0 + Ls0 .∆t/2

sN−1 ← sN−1 + LsN−1 .∆t/2

sN ← sN + LsN .∆t

sN−1 ← sN−1 + LsN−1 .∆t/2

s0 ← s0 + Ls0 .∆t/2

Numerical results:

Sim. param.: λ = 0, Hext = 10T along

ez, JCo

[4] Omelyan, I. P. et al. (2001). Phys. Rev. Lett., 86(5), 898.9/5/17 5

Page 12: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Parallel implementation

A sectoring method, respecting the symplectic properties of thespin–lattice algorithms, was implemented [5]:

1

3

2

4

1

3

2

4

1

3

2

4

1

3

2

4

Sectoring operations for a two dimen-sional system with four processors.

A

C

B

D

Communication between four sectorsfor periodic boundary conditions.

[5] Amar, J. G. et al. (2005). Phys. Rev. B, 71(12), 125432.

9/5/17 6

Page 13: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

ResultsWeak and strong scaling results for the sectoring algorithm:

Simulation conditions: λ = 0, JCo > 0,

∆ t = 10−4 ps, Hext = 10T along ez

Simulation conditions: λ = 0, JCo > 0,

∆ t = 10−4 ps, Hext = 10T along ez

The 50 % efficiency of the algorithm is reached between 250 and300 atoms per process.

9/5/17 7

Page 14: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Conclusions

Summary:A package allowing spin lattice simulations has been developed,Mathematically rigorous integration algorithms were implemented(magnetization and energy preservation),A sectoring algorithm was implemented and tested.

Future work:Take the long range dipolar interaction into account:

HMag =N∑

i,j,i̸=jJij (rij) si · sj −

µ0µ2b

N∑i,j,i̸=j

gigjr3ij

((si · eij)(sj · eij)−

1

3(si · si)

)Find experiments to compare with.

9/5/17 8

Page 15: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Appendix 1: Magnetic interactions

Spin Hamiltonian:

HMag = HEx +HZee +HAn +HDM +HME +HDip

Magnetic anisotropy:

HAn = Ka

N∑i=0

(si · na)2

Magneto-electricinteraction:

HME =N∑

i,j,i ̸=j(E × rij) · si × sj

Zeeman interaction:

HZe = gµBµ0

N∑i=0

si · Hext

Dzyaloshinskii-moriya:

HDM =N∑

i,j,i̸=jDij · si × sj

9/5/17 9

Page 16: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Appendix 2: Poisson bracket for spin–lattice systems

Considering f (t, ri,pi, si) and g (t, ri,pi, si), one has:

{f, g} =N∑

i=1

[∂f∂ri

.∂g∂pi− ∂f

∂pi.∂g∂ri

+siℏ.

(∂f∂si× ∂g

∂si

)][] Yang, K. H. et al. (1980). Phys. Rev. A, 22(5), 1814.

9/5/17 10

Page 17: Spin–lattice simulations with LAMMPSlammps.sandia.gov/workshops/Aug17/pdf/tranchida.pdf · Spin–lattice simulations with LAMMPS LAMMPS Workshop and Symposium Julien Tranchida

Appendix 3: Parametrization of the exchange interaction

Bethe–Slater model for the parametrization:

J(ij) = 4α( rijδ

)2(1− γ

( rijδ

)2)

exp(−( rijδ

)2)Θ(rc − rij)

(1)with:

α an energy in eV,δ a characteristic distance in Å,γ an adimensional coefficient,rc a cutoff distance.

9/5/17 11