vmcopt

Description

This command manages the VMC optimization. It internally calls turborvb.x for the calculation, and readalles.x for the averaging in the postprocess.

Synopsis

turbogenius vmcopt [ACTION] [OPTIONS]

ACTION is one or any combination of -g (generate an input file), -r (run a program), or -post (perform postprocess). It is mandatory.

turbogenius vmcopt --help

This command shows the list of available options.

Options

general option

This option affects all actions.

option

default value

description

-log TEXT

INFO

Specify log level. The argument is DEBUG, INFO, or ERROR.

generate (-g) options

These options affect the generation of the input file. The correspondence between the options and the input parameters are described in the note section.

option

default value

description

-vmcoptsteps INTEGER

1000

Specify vmcoptsteps

-steps INTEGER

20

Specify steps per one iteration

-bin INTEGER

1

Specify bin_block

-warmup INTEGER

0

Specify warmupblocks

-nw INTEGER

-1

Specify num_walkers. If omitted, it is set to the number of MPI processes.

-maxtime INTEGER

3600

Specify maxtime

-optimizer TEXT

lr

Specify optimizer, sr or lr

-learn FLOAT

0.35

Specify learning_rate

-reg FLOAT

0.001

Specify regularization

-opt_onebody

false

flag for opt_onebody

-opt_twobody

false

flag for opt_twobody

-opt_det_mat

false

flag for opt_det_mat

-opt_jas_mat

false

flag for opt_jas_mat

-opt_det_basis_exp

false

flag for opt_det_basis_exp

-opt_jas_basis_exp

false

flag for opt_jas_basis_exp

-opt_det_basis_coeff

false

flag for opt_det_basis_coeff

-opt_jas_basis_coeff

false

flag for opt_jas_basis_coeff

-opt_structure

false

flag for opt_structure

-strlearn FLOAT

1.0e-6

Specify str_learning_rate

-twist

false

flag for twist_average

-kpts INTEGER…

[0, 0, 0, 0, 0, 0]

Specify Monkhorst-Pack grids and shifts, [nkx,nky,nkz,kx,ky,kz]

-num_opt_param INTEGER

0

Specify the number of optimized parameters. 0 means all the parameters are optimized.

postprocess (-post) options

These options affect the postprocess.

option

default value

description

-optwarmup INTEGER

1

Specify optwarmupsteps

-plot

false

flag for plotting graph

-interactive

false

flag for interactive plotting graph
-plot

This flag specifies whether to plot the history of optimization parameters. If enabled, the plots are generated in the parameter_graphs/ directory.

-interactive

When this flag is enabled, the program waits for the user to press a key before showing the next plot.

Environment variables

TURBOVMC_RUN_COMMAND

This specifies the execution command used in the VMC optimization. The default is turborvb-serial.x. If the MPI parallel version is to be used, set the variable, for example, to mpiexec -n 16 turborvb-mpi.x, where mpiexec is the MPI launch command, and -n 16 specifies the number of MPI processes (16 in this case).

TURBOCOPYJAS_RUN_COMMAND

This specifies the path to the copyjas.x command. It is used during the postprocess when the twist average is considered.

Input and output files

action

input

output

rename

-g

  • fort.10

  • pseudo.dat

  • datasmin.input

  • vmcopt_genius_cli.pkl

-r

  • fort.10

  • pseudo.dat

  • datasmin.input

  • vmcopt_genius_cli.pkl

  • datasmin.input

  • forces.dat

  • fort.11

  • fort.12

  • fort.12.fn

  • out_min

  • parminimized.d

  • randseed.000000

  • turborvb.scratch/

-post

  • out_min

  • parminimized.d

  • vmcopt_genius_cli.pkl

  • Average_parameters.dat

  • ave_temp/

  • average_story.d

  • fort.10_bak

  • out_readalles_for_averages_for_average

  • plot_energy_and_devmax.png

  • run_local.sh

  • story.d

  • parameter_graphs/

Notes

Corresponding input parameters

The correspondence between the options and the input parameters in datasmin.input is summarized as follows.

turbogenius option

section

paramter

vmcoptsteps (-vmcoptsteps)

steps (-steps)

&simulation

ngen = vmcoptsteps × steps

steps (-steps)

&optimization

nweight

num_walkers (-nw)

&simulation

nw

maxtime (-maxtime)

&simulation

maxtime

bin_block (-bin)

&optimization

nbinr

warmupblocks (-warmup)

&optimization

iboot

learning_rate (-learn)

&optimization

tpar

regularization (-reg)

&optimization

parr

num_opt_param (-num_opt_param)

&optimization

npbra

optimizer (-optimizer)

opt_basis_coeff (-opt_det_basis_coeff or -opt_jas_basis_coeff)

&simulation

itestr4 (optimization number; see below)

opt_onebody (-opt_onebody)

opt_twobody (-opt_twobody)

  • &parameters iesd

  • &optimization iesdonebodyoff

  • &optimization iesdtwobodyoff

(see below)

opt_onebody (-opt_onebody)

opt_jas_mat (-opt_jas_mat)

  • &parameters iesfree

  • &optimization twobodyoff

(see below)

opt_det_mat (-opt_det_mat)

  • &parameters iessw

(see below)

opt_det_basis_exp (-opt_det_basis_exp)

opt_det_basis_coeff (-opt_det_basis_coeff)

  • &parameters iesup

(see below)

opt_jas_basis_exp (-opt_jas_basis_exp)

opt_jas_basis_coeff (-opt_jas_basis_exp)

  • &parameters iesm

(see below)

twist_average (-twist)

kpoints (-kpts)

(see below)

opt_structure (-opt_structure)

str_learning_rate (-strlearn)

  • &parameters ieskin

  • &optimization idyn

  • &optimization tion

  • &dynamic temp

  • &dynamic iskipdyn

  • &dynamic maxdev_dyn

  • &simulation ngen

(see below)

optimizer and optimized terms

optimization number depends on optimizer (-optimizer) and opt_basis_coeff (-opt_det_basis_coeff or -opt_jas_basis_coeff):

opt_basis_coeff

optimizer=lr

optimizer=sr

ON

-8

-5

OFF

-4

-9

iesd, iesfree, iessw, iesup, iesm, iesdonebodyoff, iesdtwobodyoff, and twobodyoff depend on the optimization flags:

opt_onebody

opt_twobody

iesd

iesdonebodyoff

iesdtwobodyoff

ON

ON

1

.false.

.false.

ON

OFF

1

.false.

.true.

OFF

ON

1

.true.

.false.

OFF

OFF

0

.false.

.false.

opt_onebody

opt_jas_mat

iesfree

twobodyoff

ON

ON

1

.false.

ON

OFF

1

.true.

OFF

ON

1

.false.

OFF

OFF

0

.false.

opt_det_mat

iessw

ON

1

OFF

0

opt_det_basis_exp

opt_det_basis_coeff

iesup

ON

ON

1

ON

OFF

1

OFF

ON

1

OFF

OFF

0

opt_jas_basis_exp

opt_jas_basis_coeff

iesm

ON

ON

1

ON

OFF

1

OFF

ON

1

OFF

OFF

0

kpoints

When twist_average is 0 or False, no additional parameters concerning kpoints are set. When twist_average is 1 or True, i.e. Monkhorst-Pack algorithm is enabled, the following parameters are set, where kpoints is an array of integers containing nkx, nky, nkz, kx, ky, and kz:

section

paramter

value

&parameters

yes_kpoints

.true.

&kpoints

kp_type

1

nk1, nk2, nk3

nkx, nky, nkz

k1, k2, k3

kx, ky, kz

skip_equivalence

.true.

double_kpgrid

.true.

When twist_average is 2, i.e. the user-defined parameters are used, the following parameters are set:

section

paramter

value

&parameters

yes_kpoints

.true.

&kpoints

kp_type

2

nk1

length of kpoints_up or kpoints_dn

double_kpgrid

.true.

kpoints should contain two arrays kpoints_up and kpoints_dn, each holds an array of 4-component arrays having [kx, ky, kz, wkp].

KPOINTS section is added to the input file datasmin.input.

structure optimization

When opt_structure is enabled, the following parameters concerning the structural optimization are set:

section

paramter

value

&parameters

ieskin

1

&optimization

idyn

5

tion

str_learning_rate (-strlearn)

&dynamic

temp

0.0

iskipdyn

5

maxdev_dyn

6.0

&simulation

ngen

vmcoptsteps \(\times\) steps \(\times\) 5 (iskipdyn)