Update on linear-response TDDFPT codes
Iurii Timrov
SISSA – Scuola Internazionale Superiore di Studi Avantazi, Trieste, Italy
Q UANTUM ESPRESSO developers’ meeting
14 January 2015
1/15
Outline
1. Absorption spectroscopy for finite systems: turboTDDFT
code
2. Electron Energy Loss Spectroscopy for periodic solids:
turboEELS code
3. Inelastic Neutron Scattering in periodic solids
2/15
Outline
1. Absorption spectroscopy for finite systems: turboTDDFT
code
2. Electron Energy Loss Spectroscopy for periodic solids:
turboEELS code
3. Inelastic Neutron Scattering in periodic solids
3/15
Developers’ of the turboTDDFT code∗
I Stefano Baroni (2004 - present), coordinator.
I Ralph Gebauer (2004 - present)
I Brent Walker (2004 - 2006)
I Dario Rocca (2006 - 2012)
I Baris Malcioglu (2008 - 2010)
I Simon Binnie (2010 - 2012)
I Xiaochuan Ge (2010 - 2014)
I Iurii Timrov (2010 - present)
I Tommaso Gorni (2014 - present)
∗
The timing is approximate.
4/15
Structure of the turboTDDFT code
I turboTDDFT is the open-source code for a calculation of the absorption
spectra of finite systems containing up to several hundreds of atoms.
I turboTDDFT is based on the Time-Dependent Density Functional
Perturbation Theory (TDDFPT).
I There are two flavours of the turboTDDFT code:
I turbo_lanczos.x
I turbo_davidson.x
Both codes give the same absorption spectrum but have their Pros and
Cons (next slide).
Both codes are residing in the same directory TDDFPT/src and are
using many same routines in TDDFPT, PH, and PW.
5/15
Pros & Cons
Liouville-Lanczos approach Casida-Davidson approach
( turbo_lanczos.x ) ( turbo_davidson.x )
Solution method: Computes an Solution method: Computes
absorption coefficient directly without eigenvalues of the Liouvillian by
computing eigenvalues of the diagonalizing it, and then computes
Liouvillian, using the Lanczos the absorption coefficient from them.
recursion algorithm.
, Spectrum is computed at once in / Many eigenvalues are required to
the large frequency range (only one compute spectrum in the large energy
Lanczos recursion is needed). range.
, No empty states (use of DFPT , No empty states (use of DFPT
techniques) techniques)
, No large-matrix inversions / Large-matrix diagonalization
(Lanczos recursion instead) (Davidson-like solver) ⇒ high RAM
/ No information about individual , Information about individual
electronic transitions electronic transitions
6/15
Restrictions of the turboTDDFT code
I Linear-response regime.
I Use of only the Γ point (gamma_only) and supercells.
k = 0 or many k points algorithm seems working well, but nobody have
ever performed an extensive testing. It may be used at your own risk!
I Supports NC and US pseudopotentials. No PAW.
I Hybrid functional are supported, but only with NC PP’s (S. Binnie).
I Adiabatic XC kernels only.
I Only unpolarized systems are allowed (nspin=1). Collinear (nspin=2)
and noncollinear (noncolin=.true.) cases are not supported.
I No relativistic effects: spin-orbit coupling is not implemented.
I Parallelization: PWs, bands, task groups (only NC PP’s, experimental).
No OpenMP.
7/15
New features of the turboTDDFT code (I)
What new was done during 2014?
I Implemented a subroutine (compute_d0psi_rs) which calculates
P̂c r ϕv (r) in the real space. This can be used only for finite systems!
This is need when hybrid functionals are used, because the usual trick
with the commutator [Ĥ, r̂] misses a contribution from the exact
exchange potential. (X. Ge)
I Tried to implement [V̂EXX , r̂] for hybrid functionals using 1) spherically
truncated Coulomb potential, and 2) Martyna-Tuckerman partitioning of
the Coulomb potential, but have not succeeded so far to agree with the
real-space method mentioned above. (I. Timrov)
I Martyna-Tuckerman periodic-boundary correction for charged
molecules is now supported. (I. Timrov)
Recall: The supercell size must be at least two times larger than the
molecule size!
8/15
New features of the turboTDDFT code (II)
What new was done during 2014?
I turboTDDFT has been interfaced with the Environ module. (I. Timrov)
I. Timrov, O. Andreussi, A. Biancardi, N. Marzari, S. Baroni, J. Chem.
Phys. (2015), in press. −→ poster
What does it give? ⇒ A possibility to compute optical absorption
spectra of molecules in various solvents (e.g. water) using the revised
self-consistent continuum solvation model of O. Andreussi et al. (JCP
136, 064102 (2012)).
However, the Environ module is not publically distributed yet
(coordinated by O. Andreussi).
Ongoing work:
I Speeding up TDDFPT calculations with optimally reduced PW basis
sets. (T. Gorni, I. Timrov, S. Baroni) −→ poster
9/15
Outline
1. Absorption spectroscopy for finite systems: turboTDDFT
code
2. Electron Energy Loss Spectroscopy for periodic solids:
turboEELS code
3. Inelastic Neutron Scattering in periodic solids
10/15
turboEELS code (I)
I Developers: Iurii Timrov, Nathalie Vast, Ralph Gebauer, Stefano Baroni
I turboEELS was created during I. Timrov’s PhD thesis at the École
Polytechnique (Paris) during 2009-2013, in strong collaboration with
SISSA and ICTP.
I. Timrov’s PhD thesis:
https://pastel.archives-ouvertes.fr/pastel-00823758
See also: I. Timrov, N. Vast, R. Gebauer, and S. Baroni, PRB 88,
064301 (2013).
I turboEELS is not yet public. Is planned to be public during next months
(will be re-implemented in the SVN version of Quantum ESPRESSO on
http://qe-forge.org/gf/project/q-e/).
I. Timrov, N. Vast, R. Gebauer, and S. Baroni, Comput. Phys.
Commun., in preparation.
11/15
turboEELS code (II)
I turboEELS is the code for a calculation of the EELS (IXSS) spectra at
q 6= 0 of periodic solids, containing up to several hundreds of atoms in
the unit cell.
I turboEELS ⇒ calculation of plasmons and single-particle excitations.
I turboEELS is based on the Liouville-Lanczos approach to TDDFPT.
I turboEELS resides in the directory TDDFPT/src, and is strongly linked
to turboTDDFT, PH, and PW.
12/15
Restrictions of the turboEELS code
I Linear-response regime.
I Uses general k points algorithm and allows finite transferred momenta
(q 6= 0).
I Supports NC and US pseudopotentials. No PAW.
I No hybrid functionals.
I Adiabatic XC kernels only.
I Unpolarized (nspin=1) and noncollinear (noncolin=.true.) cases
are allowed. Collinear case (nspin=2) is not supported.
I Relatifictic effects: spin-orbit coupling is supported (only with NC PP’s).
Application to Bi: I. Timrov, N. Vast, R. Gebauer, and S. Baroni,
in preparation.
I Symmetry is exploited (small group of q).
I Parallelization: PWs, k points. No OpenMP.
13/15
Outline
1. Absorption spectroscopy for finite systems: turboTDDFT
code
2. Electron Energy Loss Spectroscopy for periodic solids:
turboEELS code
3. Inelastic Neutron Scattering in periodic solids
14/15
Magnons
I Developers: Tommaso Gorni, Iurii Timrov, Andrea Dal Corso, Ralph
Gebauer, Stefano Baroni
I The project has been started recently. Aim: calculations of magnons
(spin waves) at finite q, which can be observed in INS experiments.
I Use of the Liouville-Lanczos approach to TDDFPT. It must be extented
to magnetic systems (no time-reversal symmetry).
I Noncollinear case (noncolin=.true.)
I +q and −q perturbations must be treated simultaneously (not present
in PH), or, equivalently, time-dependent Kohn-Sham equation and the
complex-conjugate one must be solved simultaneously.
Work in progress...
15/15
