new progress and functionalities in pwmat
TRANSCRIPT
Topics
Accuracy, speed and size
More complete function combination
Electrochemical methods
Nonadiabatic dynamic methods
Charge transfer and electron-phonon coupling
Charge patching method
FSM + Escan
GW methods
U calculation in LDA+U
Machine learning
Modules
Accuracy
Phonon spectrum: Cu2ZnSnO4, both using HSE for phonon calculation
Same phonon spectrum using HSE
cell relaxation (need larger Ecut, or stress correction procedure)
System Code Psp Ecut(eV) XC U(Ni) Kpoints a b c
LiNiO2
PWmat SG15 952 LDA+U
2.6eV
8 4 2
2.84 5.95 10.24
VASP PAW 520 2.81 5.91 10.24
Accuracy
Size and Speed
(4) Optimize the GPU memory usage.
For PBE, LDA calculation, can calculate system > 2000 atoms!
体系 截断能 K point xc SCF time
Si512 50Ry Gamma HSE 1.6 Hour
Si2016 30Ry Gamma PBE 13 Hour
All on one Mstation!
Iteration steps
Convergence error
Rho_error
V_error
30Ry Ecut, 6-13 hours for SCF convergence on one Mstation
Size and Speed
2016 Si atoms, 8064 electron
Biological systems (protein pocket with
a drug molecule).
Large supercell ! (a lot of empty space).
Another example for biological system ( 1640 atoms)
One Mstation (4 GPU)
Size and Speed
Relax convergence
We have improved the convergence, and identify the parameters:
193个原子,927个电子的复杂体系
这个结构用CG的方法很难收敛,这里我们采用PCG的方法进行结构优化
体系在第71个离子步达到收敛标准,用时5541秒
(1) Ecut2=4Ecut; (2) Larger Rcut (especially for LDA+U); (3) higher SCF convergence
More complete function combination
HSE
LDA+U
HSE
SOC
LDA+U
SOC
HSE
SOC
LDA+U
Yes Yes Yes Yes
Note: no stress for spin=22 and spin=222;
no force for LDA+U when spin=22 or 222;
SPIN XCFUNCTIONAL PRECISION PSEUDOPOTETIAL
SOC 22,222 LDA,PBE,HSE SINGLE,DOUBLE NCPP
LDA+U 1,2,222 LDA,PBE,HSE SINGLE,DOUBLE NCPP
Electrochemical methods
Grand canonical
Fixed electrode potential
Calculations.
Poisson Boltzmann Eq.
For the solvent model.
Fix Fermi energy SCF.
Energy Correction term.
Nonadiabatic MD (NAMD)
A density matrix based new formalism for NAMD,
Including: coherence decay time and detail balance
NAMD is done as a post process after a AIMD
rt-TDDFT with Boltzmann factor
A mean field treatment to restore detailed balance
can treat problems like hot carrier cooling
occupied
states
Time (fs)
Orb
ita
l e
ne
rgy (
eV
)
Charge transfer calculations
(b) Model-II
(a) Model-I
0.6nm
1.2nmSi HfO2
SiO2-1
3
211
21
2
1
2
3
Si/SiO2-1 SiO2/HfO2-2
HfO2-1
(c) Model-III: Amorphous
0.6nmSpecial module for charge transfer calc.
Realistic amorphous oxide model
HSE, yield accurate band alignment
Marcus theory for transfer transfer
Besides:
New formalism for multi-phonon process
Procedure to correct anharmonic effect
Charge patching method
Easier to use charge motif generation procedure
More flexible ways to generate the motifs
Adjustable tolerance
CPM is an extremely powerful method to study nanosystems with >10,000 atoms
Can be used to study:
(1) Moire’s pattern
(2) Defect level
(3) Charge transfer
(4) Nanostructures
(5) Amorphous system
(6) For >10,000 atom systems
Charge patching method
CPM accuracy example: Moire’s pattern (graphene + h-BN)
CPM DFT
State 884
Eigen energy Difference: 7.5 meV
Automatic generate many motifs
Charge patching method
Example: using CPM to
study defect level coupling
in 2D material, and the hopping
Transport.
(user provided example)
FSM and Escan (20,000 atoms!)
This follows the CPM, to calculate the electronic structure
Defect level energy and wave function, using CPM Hamiltonian
21,952 Si atoms. One Mstation, 10 hours
GW calculations: YAMBO
Yes, we can do GW calculations using YAMBO
Easy step by step procedure (module) for how to calculate YAMBO with Pwmat
Easy to install (run script) on Mstation
Using the linear respond and Koopmans’ approach to calculate U
A module to calculate U in LDA+U
A new section in atom.config
DeepMD + PWmat
Machine Learning
PWmat
Han Wang, Lin Feng Zhang, et al, Computer Physics Communications,228(2018)