gipaw&–&linear&response&& inthe& presenceof magnec &fields& ·...
TRANSCRIPT
GIPAW – Linear response in the presence of magne7c fields
Uwe Gerstmann University of Paderborn, Germany
Trieste, January 18 – 21, 2016
QE developers’ mee7ng Linear Response
Davide Ceresoli Emine Kucukbenli Ari Parvo Seitsonen
Originally aimed to calculate NMR chemical shiBs, later also the EPR electronic g-‐tensor, ...
Standard version: qe-gipaw-5.3.tar @ qe-‐forge
Uwe Gerstmann University of Paderborn, Germany
Trieste, January 18 – 21, 2016
QE developers’ mee7ng Linear Response
Davide Ceresoli Emine Kucukbenli Ari Parvo Seitsonen
Originally aimed to calculate NMR chemical shiBs, later also the EPR electronic g-‐tensor, ...
Standard version: qe-gipaw-5.3.tar @ qe-‐forge
GIPAW – Linear Magne7c Response
Trieste, January 18 – 21, 2016
QE developers’ mee7ng Linear Response
Originally aimed to calculate NMR chemical shiBs, later also the EPR electronic g-‐tensor,...
GIPAW rouOnes/pseudopoten7als are also used for
-‐ X-‐ray spectra (Xspectra): XANES, (XMCD)
-‐ orbital magneOzaOon, converse NMR-‐approach
-‐ SOC including two component noncolinear scheme using scalar-‐relaOvisOc GIPAW pseudos
GIPAW – Linear Magne7c Response
QE Developers’ MeeOng, Trieste, January 18 – 21, 2016
EssenOal: B-‐field brings the phase of the wfc into play (to insure translaOonal invariance within PBC) The PAW augmentaOon scheme has to be extended in a Gauge Including way: à GIPAW
GIPAW-‐ Linear Magne7c Response
Basic quan7ty for NMR (EPR): (spin) currents induced by the B-‐field
à Green‘s funcOon required
spin-‐currents
EPR:
GIPAW-‐ Linear Magne7c Response QE Developers’ MeeOng, Trieste, January 18 – 21, 2016
Basic quan7ty for NMR (EPR): (spin) currents induced by the B-‐field
à Green‘s funcOon required
spin-‐currents
EPR:
NMR: Bind(r‘) = ... ... |r‘-‐ r|3
GIPAW-‐ Linear Magne7c Response QE Developers’ MeeOng, Trieste, January 18 – 21, 2016
r‘-‐ r
Central rou7ne: greenfunction.f90
GIPAW-‐ Linear Magne7c Response QE Developers’ MeeOng, Trieste, January 18 – 21, 2016
Green‘s funcOon@k+q to be applied on a modified version of the unperturbed wfc: Gk+q |psi > = Gk+q { Vk+q,k|evc>}; q = {0, ± Δq ei with i=x,y,z} Similar to solve_linter.f90 in PH, it calls cgsolve_all.f90 iteraOve solver of linear systems cg_psi.f90 for (simple) precondiOoning (via diagonals of H), ch_psi_all.f90 for applicaOon of H-‐eS+P_cv orthogonalize.f90 for compuOng P_cv
( h_psi_q.f90 for BAND parallelzaOon only; otherwize: h_psi, calbec, s_psi )
Central rou7ne: greenfunction.f90
GIPAW-‐ Linear Magne7c Response QE Developers’ MeeOng, Trieste, January 18 – 21, 2016
Green‘s funcOon@k+q to be applied on a modified version of the unperturbed wfc: Gk+q |psi > = Gk+q { Vk+q,k|evc>}; q = {0, ± Δq ei with i=x,y,z} Similar to solve_linter.f90 in PH, it calls cgsolve_all.f90 iteraOve solver of linear systems cg_psi.f90 for (simple) precondiOoning (via diagonals of H), ch_psi_all.f90 for applicaOon of H-‐eS+P_cv orthogonalize.f90 for compuOng P_cv
( h_psi_q.f90 for BAND parallelzaOon only; instead: h_psi, calbec, s_psi )
Central rou7ne: greenfunction.f90
GIPAW-‐ Linear Magne7c Response QE Developers’ MeeOng, Trieste, January 18 – 21, 2016
Gk+q |psi > = Gk+q { Vk+q,k|evc>}; q = {0, ± Δq ei with i=x,y,z} in addiOon: apply_operators.f90 , e.g. apply Vk+q,k to rhs compute_u_kq.f90 prepares unperturbed wfc@k+q
NEW: Speed-‐up for small Δq: resuse Gk+q |psi > from previous q, ch_psi.f90 only for q=0. Includes all things for precondiOoning, e.g. also small parts from phq_init.f90: eprec = 1.35*zdotc(evq,work)
Central rou7ne: greenfunction.f90
GIPAW-‐ Linear Magne7c Response QE Developers’ MeeOng, Trieste, January 18 – 21, 2016
Gk+q |psi > = Gk+q { Vk+q,k|evc>}; q = {0, ± Δq ei with i=x,y,z}
in addiOon: apply_operators.f90 , e.g. apply Vk+q,k to rhs
compute_u_kq.f90 prepares unperturbed wfc@k+q
Specific topics:
-‐ quanOOes at k and k+q must have the same G-‐vector ordering; we call gk_sort only for k, not for k+q.
-‐ symmetry operaOons that do not map cartesian axes might be – in principle – removed (as done in 5.3.0), but: NEW: it works also with full symmetry, if symmetrizaOon is applied at the very end (onto the full tensors) excusively.
LR related rou7nes (as a general observa6on):
LR in GIPAW has very few dependencies, can be easily decoupled from the code, and can be „librarized“ away (into LR_Modules). As my present personal opinion: greenfunction.f90 (or something like that) should be either kept GIPAW-‐specific or solve_linter.f90 should be reorganized, split-‐up into less PH-specific logical subrouOnes, may be all of them kept within the same file.
interface between GIPAW/LR_Modules can be changed accordingly.
GIPAW-‐ Linear Magne7c Response QE Developers’ MeeOng, Trieste, January 18 – 21, 2016
Open Ques7on (loosely related to LR):
GIPAW-‐ Linear Magne7c Response QE Developers’ MeeOng, Trieste, January 18 – 21, 2016
-‐ Zero-‐Field Spliqng (ZFS) of EPR, both DS-‐S and DSO! -‐ NMR/EPR for hybrid funcOonals -‐ NMR with spin-‐orbit coupling, colinear & non-‐colinear based on relaOvisOc & scalar relaOvisOc (GI)PAW pseudos -‐ Circular Dichroism of X-‐ray AdsorpOon (XMCD) (Mateo Calandra, UG)
Work in progress, perspec7ves:
-‐ Where to put the Berry-‐phase rouOnes applicable onto NMR/EPR, GIPAW-‐tree or (parOally) PW-‐core? -‐ orbital magneOzaOon as a more general quanOty (MTM analogue to MTM, suggesOng PW), but to compute accurate du/dk: greenfunction.f90
Proposal: keep it close to “NMR/EPR“, should be found in GIPAW, e.g.:
Thanks for your aZen7on!
QE Developers’ MeeOng, Trieste, January 18 – 21, 2016
GIPAW-‐ Linear Magne7c Response
2nd method: beyond perturba7on theory Orbita
l magne
7za7
on
EPR
htp://www.quantum-‐espresso.org
2nd method: beyond perturba7on theory Orbita
l magne
7za7
on
EPR
htp://www.quantum-‐espresso.org
Both methods are now applicable also on metallic systems, e.g. electrons trapped by the conducOon band minimum in Si bulk:
Strongly delocalized states: conducOon band electrons in Si: gexp = 1.9995 cc unit cell (2 atoms), (24 × 24 × 24) MP k-‐point set: gDFT 1 = 1.9991 (Berry phase)
gDFT 2 = 1.9990 (non-‐eq. GIPAW)
Young et al., PRB 55, 16245 (1997)
Calcula7ng the g-‐tensor – two different methods
Efficient relativistic DFT calculations
18
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
M K K M
EF
E [e
V]
scalar-rel.
scalar-‐rela/vis/c pseudopoten/als: -‐ RelaOvisOc kineOc energy -‐ spin-‐orbit (SO) coupling neglected
Bi(111) bilayer
Efficient relativistic DFT calculations
19
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
M K K M
EF
E [e
V]
full-rel.
scalar-rel.
scalar-‐rela/vis/c pseudopoten/als: -‐ RelaOvisOc kineOc energy -‐ spin-‐orbit (SO) coupling neglected full-‐rela7vis7c approach: large SO coupling effects, but: factor-‐of-‐40 more computaOonal costs
Bi(111) bilayer
Efficient relativistic DFT calculations
20
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
M K K M
EF
E [e
V]
full-rel.
Pauliscalar-rel.
scalar-‐rela/vis/c pseudopoten/als: -‐ RelaOvisOc kineOc energy -‐ spin-‐orbit (SO) coupling neglected alterna7ve rela7vis7c approach: PAW-‐reconstruc7on of SO coupling
Bi(111) bilayer
Efficient relativistic DFT calculations
21
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
M ! K M
EF
E [
eV]
full-rel.ZORA
Pauliscalar-rel.
scalar-‐rela/vis/c pseudopoten/als: -‐ RelaOvisOc kineOc energy -‐ spin-‐orbit (SO) coupling neglected alterna7ve rela7vis7c approach: PAW-‐reconstruc7on of SO coupling
ZORA:
Bi(111) bilayer
Efficient relativistic DFT calculations
22
scalar-‐rela/vis/c pseudopoten/als: -‐ RelaOvisOc kineOc energy -‐ spin-‐orbit (SO) coupling neglected alterna7ve rela7vis7c approach: PAW-‐reconstruc7on of SO coupling
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
M ! K M
EF
E [
eV]
full-rel.ZORA
Pauliscalar-rel.
iden7cal results !
Bi(111) bilayer
23
efficient „reconstruc6on-‐only“ approach:
No -‐k à k scatering @ impurity site
rela7vis7c calcula7ons for large systems (500 atoms) possible:
Efficient relativistic DFT calculations