Liquid Si - MLFF - Revision history

Karsai: /* Structral properties of the force field */

2022-04-15T15:13:44Z

Structral properties of the force field

Singraber: /* Obtaining a more accurate force field */

2021-09-30T14:37:03Z

Obtaining a more accurate force field

← Older revision		Revision as of 14:37, 30 September 2021
Line 280:		Line 280:

	After running the calculation, we inspect the last instance of the errors in the {{TAG\|ML_LOGFILE}} by typing:		After running the calculation, we inspect the last instance of the errors in the {{TAG\|ML_LOGFILE}} by typing:
	grep ~~BEE~~ ML_LOGFILE		grep ERR ML_LOGFILE
	The last few lines should have values close to:		The last few lines should have values close to:
	ERR 675 5.10937061E-03 1.46895065E-01 2.50094941E+00		ERR 675 5.10937061E-03 1.46895065E-01 2.50094941E+00

Singraber: /* Structral properties of the force field */

2021-09-30T14:00:47Z

Structral properties of the force field

← Older revision		Revision as of 14:00, 30 September 2021
Line 226:		Line 226:
	[[File:PC MLFF vs AI 3ps.jpg\|400px]]		[[File:PC MLFF vs AI 3ps.jpg\|400px]]

	We see that pair correlation is quite well reproduced although the error in the force of 0.~~247664~~ eV/<math>\AA</math> shown above is a little bit too large. This error is maybe too large for accurate production calculations (usually an accuracy of approximately 0.1 eV/<math>\AA</math> is targeted), but since the pair correlation function is well reproduced it is perfectly fine to use this on-the-fly force field in the time-consuming melting of the crystal.		We see that pair correlation is quite well reproduced although the error in the force of ~0.242 eV/<math>\AA</math> shown above is a little bit too large. This error is maybe too large for accurate production calculations (usually an accuracy of approximately 0.1 eV/<math>\AA</math> is targeted), but since the pair correlation function is well reproduced it is perfectly fine to use this on-the-fly force field in the time-consuming melting of the crystal.

	=== Obtaining a more accurate force field ===		=== Obtaining a more accurate force field ===

Singraber: /* Creating the liquid structure */

2021-09-30T12:54:01Z

Creating the liquid structure

← Older revision		Revision as of 12:54, 30 September 2021
Line 174:		Line 174:
	# ERR		# ERR
	# ERR nstep ......... MD time step or input structure counter		# ERR nstep ......... MD time step or input structure counter
	# ERR rmse_energy ... RMSE of energies		# ERR rmse_energy ... RMSE of energies (eV atom^-1)
	# ERR rmse_force .... RMSE of forces		# ERR rmse_force .... RMSE of forces (eV Angst^-1)
	# ERR rmse_stress ... RMSE of stress		# ERR rmse_stress ... RMSE of stress (kB)
	# ERR ######################################################################		# ERR ######################################################################
	# ERR nstep rmse_energy rmse_force rmse_stress		# ERR nstep rmse_energy rmse_force rmse_stress

Karsai: /* Download */

2021-09-01T07:25:58Z

Download

← Older revision		Revision as of 07:25, 1 September 2021
Line 303:		Line 303:

	== Download ==		== Download ==
	[[Media:~~MLFF Liquid Si~~.tgz\| MLFF_Liquid_Si_tutorial.tgz]]		[[Media:MLFF_Liquid_Si_tutorial.tgz\| MLFF_Liquid_Si_tutorial.tgz]]

	{{Template:Machine learning force field - Tutorial}}		{{Template:Machine learning force field - Tutorial}}

	[[Category:Examples]]		[[Category:Examples]]

Karsai at 07:24, 1 September 2021

2021-09-01T07:24:57Z

← Older revision		Revision as of 07:24, 1 September 2021
Line 303:		Line 303:

	== Download ==		== Download ==
	[[Media:MLFF Liquid Si.tgz\| ~~MLFF_Liquid_Si~~.tgz]]		[[Media:MLFF Liquid Si.tgz\| MLFF_Liquid_Si_tutorial.tgz]]

	{{Template:Machine learning force field - Tutorial}}		{{Template:Machine learning force field - Tutorial}}

	[[Category:Examples]]		[[Category:Examples]]

Karsai: /* Obtaining a more accurate force field */

2021-08-27T10:08:46Z

Obtaining a more accurate force field

← Older revision		Revision as of 10:08, 27 August 2021
Line 238:		Line 238:
	{{TAGBL\|ML_LMLFF}} = .TRUE.		{{TAGBL\|ML_LMLFF}} = .TRUE.
	{{TAGBL\|ML_ISTART}} = 0		{{TAGBL\|ML_ISTART}} = 0
	{{TAGBL\|NSW}} = ~~10000~~		{{TAGBL\|NSW}} = 1000
	If you run have resources to run in parallel, you can reduce the computation time further by adding k point parallelization with the {{TAG\|KPAR}} tag.		If you run have resources to run in parallel, you can reduce the computation time further by adding k point parallelization with the {{TAG\|KPAR}} tag.
	We use a denser k-point mesh in the {{TAG\|KPOINTS}} file		We use a denser k-point mesh in the {{TAG\|KPOINTS}} file

Karsai: /* {{TAG|INCAR}} */

2021-08-27T10:07:10Z

← Older revision		Revision as of 10:07, 27 August 2021
Line 117:		Line 117:

	#Machine learning paramters		#Machine learning paramters
	{{TAGBL\|~~ML_FF_LMLFF~~}} = .TRUE.		{{TAGBL\|ML_LMLFF}} = .TRUE.
	{{TAGBL\|~~ML_FF_ISTART~~}} = 0		{{TAGBL\|ML_ISTART}} = 0

	~~;{{TAG\|ML_FF_LMLFF}} = ''.TRUE.'': switches on the machine learning of the force field~~
	~~;{{TAG\|ML_FF_ISTART}} = 0: corresponds to learning from scratch~~

	== Calculation ==		== Calculation ==

Karsai: /* {{TAG|POSCAR}} */

2021-08-27T10:06:41Z

← Older revision		Revision as of 10:06, 27 August 2021
Line 10:		Line 10:

	In this example we start from a 64 atom super cell of diamond-fcc Si (the same as in {{TAG\|Liquid Si - Standard MD}}):		In this example we start from a 64 atom super cell of diamond-fcc Si (the same as in {{TAG\|Liquid Si - Standard MD}}):
	~~Si cubic diamond 2x2x2 super cell of conventional cell~~		Si_CD_2x2x2
	5.43090000000000		5.43090000000000
	2.00000000 0.00000000 0.00000000		2.00000000 0.00000000 0.00000000

Karsai: /* Obtaining a more accurate force field */

2021-08-27T10:03:09Z

Obtaining a more accurate force field

← Older revision		Revision as of 10:03, 27 August 2021
Line 263:		Line 263:
	We immediately see that the errors for the forces are significantly lower than in the previous calculation with only one k point. This is due to the less noisy ab initio data which is easier to learn.		We immediately see that the errors for the forces are significantly lower than in the previous calculation with only one k point. This is due to the less noisy ab initio data which is easier to learn.

	To understand how the force field is learned, we inspect the {{TAG\|ML_ABN}} file containing the training data. In the beginning of this file, you will find information about the number of reference structures		To understand how the force field is learned, we inspect the {{TAG\|ML_ABN}} file containing the training data. In the beginning of this file, you will find information about the number of reference structures for training
	1.0 Version		1.0 Version
	**************************************************		**************************************************
Line 275:		Line 275:
	382		382

	We will ~~monitor how much~~ training ~~data is added to~~ a ~~structure after each learning cycle~~ and ~~what impact this has on~~ the ~~accuracy~~ of the ~~force field~~. ~~First we will perform a continuation run keeping~~ the ~~Bayesian threshold fixed~~		We will further continue the training a 1000 MD steps and see how the number of training structures and the number of local reference configurations change.
			Do the following steps:
	cp {{TAGBL\|ML_ABN}} {{TAGBL\|ML_AB}}		cp {{TAGBL\|ML_ABN}} {{TAGBL\|ML_AB}}
	cp {{TAGBL\|CONTCAR}} {{TAGBL\|POSCAR}}		cp {{TAGBL\|CONTCAR}} {{TAGBL\|POSCAR}}
	and set the following {{TAG\|INCAR}} ~~tags~~		and set the following {{TAG\|INCAR}} tag
	{{TAGBL\|ML_ISTART}} = 1		{{TAGBL\|ML_ISTART}} = 1
	~~{{TAGBL\|ML_CTIFOR}} = ''x''~~
	~~where ''x'' is the threshold of the Bayesian error obtained from the {{TAG\|ML_LOGFILE}} by typing~~
	~~grep BEEF ML_LOGFILE~~
	and using the value for the threshold (6th column) in the last entry. By setting a good estimate for the threshold several ab initio steps are skipped in the first few steps, which would be required to determine the threshold automatically. '''Mind''': When continuation runs are performed for different crystal structures, the last previous threshold for the Bayesian error might be too large leading to premature skipping of ab initio steps. This is particular relevant when studying the liquid phase, first, and applying its threshold to the solid phase. In that case it is safer to use the default value for {{TAG\|ML_CTIFOR}} = <math>10^{-16}</math>.

	After running the calculation, we inspect the last instance of the errors in the {{TAG\|ML_LOGFILE}}~~, which should look similar to~~		After running the calculation, we inspect the last instance of the errors in the {{TAG\|ML_LOGFILE}} by typing:
	~~====================================================================================================~~		grep BEE ML_LOGFILE
	~~Information on error estimations~~		The last few lines should have values close to:
	--~~--------------------------------------------------------------------------------------------------~~		ERR 675 5.10937061E-03 1.46895065E-01 2.50094941E+00
	~~Error in energy (eV atom^~~-1~~): 0~~.~~006252~~		ERR 861 5.10937061E-03 1.46895065E-01 2.50094941E+00
	~~Error in force~~ ~~(eV Angst^~~-1~~): 0~~.~~168549~~		ERR 924 5.10937061E-03 1.46895065E-01 2.50094941E+00
	~~Error in stress (kB):~~ 2.~~606332~~		ERR 942 5.01460989E-03 1.47816836E-01 2.47329693E+00
	~~Bayesian error (eV Angst~~-1~~): 0~~.~~037569 0~~.~~101357~~		ERR 1000 5.01460989E-03 1.47816836E-01 2.47329693E+00
	~~Spilling factor (~~-~~): 0~~.~~001114 0~~.~~020000~~		We see that the accuracy has changed slightly. We also look at the ML_ABN file and the number of reference structures for training should increase compared to the run before
	~~====================================================================================================~~		1.0 Version
	We see that the accuracy has ~~not~~ changed ~~significantly and the threshold for the Bayesian error is set to the one specified in the {{TAG\|INCAR}} file~~. ~~Looking~~ at the ~~actual errors we observe that the threshold was exceeded only a few times. You can find all these instances with the command~~
	~~grep "Bayesian error (eV Angst-1):" ML_LOGFILE \| awk '{if ($5 > $6) { print $5 }}'~~
	~~Consequently only very few structures were added to the ab initio data in the {{TAG\|ML_ABNCAR}}~~ file~~. The entry for~~ the reference structures ~~and basis sets~~ should ~~look similar~~ to
	**************************************************		**************************************************
	The number of configurations		The number of configurations
	--------------------------------------------------		--------------------------------------------------
	65		99
	~~...~~		Also the number of local reference configurations (basis sets) increases compared to the previous calculation
	**************************************************		**************************************************
	The numbers of basis sets per atom type		The numbers of basis sets per atom type
	--------------------------------------------------		--------------------------------------------------
	~~407~~		665
	From this observation, we conclude either that the force field is already accurate enough or that the threshold is set too high. In the latter case, the machine would not conduct ab initio calculations because the force field is judged accurate enough according to the threshold.

	~~We compare this behavior to the case of using the default value for {{TAG\|ML_FF_CTIFOR}}. We continue from the existing structure and force field~~
	~~cp {{TAGBL\|CONTCAR}} {{TAGBL\|POSCAR}}~~
	~~cp {{TAGBL\|ML_ABNCAR}} {{TAG\|ML_ABCAR}}~~
	~~Remove the {{TAG\|ML_FF_CTIFOR}} tag from the {{TAG\|INCAR}} file.~~

	The default value for {{TAG\|ML_FF_CTIFOR}} is practically zero, so we expect more structures to be added. We inspect the {{TAG\|ML_ABNCAR}} to investigate how the number of structures and the basis set size changed
	~~**************************************************~~
	~~The number of configurations~~
	~~--------------------------------------------------~~
	~~160~~
	~~...~~
	~~**************************************************~~
	~~The numbers of basis sets per atom type~~
	~~--------------------------------------------------~~
	~~726~~
	~~Even though the number of structure datasets and the basis set size increased, the accuracy of the force field did not improve significantly~~
	~~====================================================================================================~~
	~~Information on error estimations~~
	~~----------------------------------------------------------------------------------------------------~~
	~~Error in energy (eV atom^-1): 0.005950~~
	~~Error in force (eV Angst^-1): 0.162064~~
	~~Error in stress (kB): 2.628567~~
	~~Bayesian error (eV Angst-1): 0.029723 0.061348~~
	~~Spilling factor (-): 0.000406 0.020000~~
	~~====================================================================================================~~

	Ideally, one should continue learning until no structures need to be added to the training data and basis set anymore. In practice, the prediction of the Bayesian error exhibits numerical inaccuracies so that an ab initio calculation is conducted from time to time even if the force field is accurate enough. ~~Hence, in this tutorial, we stop at this point because further~~ addition of ~~training~~ data ~~does~~ not ~~improve~~ the accuracy on the training ~~structures compared to~~ the ~~ab initio~~ data.		Ideally, one should continue learning until no structures need to be added to the training data and basis set anymore. Very often this can take up to 100ps depending on the material and conditions. In practice, the prediction of the Bayesian error exhibits numerical inaccuracies so that an ab initio calculation is conducted from time to time even if the force field is accurate enough. So measuring only the decreasing frequency of addition of new data is not sufficient to know when to finish. One should also look at the accuracy of the force on the training data and more importantly on the accuracy on some test data that is outside of the training sets.

	== Download ==		== Download ==

← Older revision		Revision as of 10:08, 27 August 2021
Line 238:		Line 238:
	{{TAGBL\|ML_LMLFF}} = .TRUE.		{{TAGBL\|ML_LMLFF}} = .TRUE.
	{{TAGBL\|ML_ISTART}} = 0		{{TAGBL\|ML_ISTART}} = 0
	{{TAGBL\|NSW}} = ~~10000~~		{{TAGBL\|NSW}} = 1000
	If you run have resources to run in parallel, you can reduce the computation time further by adding k point parallelization with the {{TAG\|KPAR}} tag.		If you run have resources to run in parallel, you can reduce the computation time further by adding k point parallelization with the {{TAG\|KPAR}} tag.
	We use a denser k-point mesh in the {{TAG\|KPOINTS}} file		We use a denser k-point mesh in the {{TAG\|KPOINTS}} file