My Community

Support forum for VASP
https://vrruwv.vasp.at/

Questions about MLFF settings

https://vrruwv.vasp.at/viewtopic.php?t=18887

Page 1 of 1

Questions about MLFF settings

Posted: Wed Mar 22, 2023 2:07 pm
by jun_yin2
Hi everyone, there are some questions about settings in MLFF which confuses me.
(1) https://www.vasp.at/wiki/index.php/Best ... rce_fields In this practice, the INCAR in Example part set POMASS of H as 8.0. I want to know that if I change the default POMASS in INCAR, do we need to change the POMASS of H in POTCAR? or After I changed it in INCAR, the old POMASS in POTCAR will be overwritten?
(2) Also in the above-mentioned website, VASP official suggest us to change the POMASS of light element like it did in the INCAR posted. I have questions about which elements can be defined in 'light elements' in this case. Because I noticed that the POMASS of O element which is can be regarded as a type of light element is maintained constant with POTCAR but H changed. So I hope someone could give me a hint.
(3) How to generate RANDOM_SEED? Is there some programs or I could just randomly type a number there?

Re: Questions about MLFF settings

Posted: Wed Mar 22, 2023 2:52 pm
by henrique_miranda
(1) You don't need to change the POTCAR file, in fact, there is no standard usage of VASP that requires you to modify the POTCAR file (https://www.vasp.at/wiki/index.php/POTCAR). The value of POMASS you set in the INCAR will replace the value of POMASS read from the POTCAR in your VASP run.
(2) Notice that increasing the mass of H is a `trick` to allow you to use a larger time step (POTIM) in the MD simulation. By increasing the mass you reduce the vibration frequency which means you can sample the movement of the ions with a larger time step. This is justified if you are not interested in the particular way in which the H atoms are moving, otherwise, you need to use a small time step (POTIM). You can in principle estimate the time step you need in your MD simulation from the frequency of vibration of the lighter atoms you have in the simulation. Now, what constitutes a `light element` and whether changing its mass is a reasonable approximation really depends on the material and the properties you are interested in.
(3) RANDOM_SEED is written to the output of an MD run in VASP. You can use that value in the INCAR of another run to reproduce the calculation.

Re: Questions about MLFF settings

Posted: Thu Mar 23, 2023 3:30 am
by jun_yin2
henrique_miranda wrote: ↑Wed Mar 22, 2023 2:52 pm (1) You don't need to change the POTCAR file, in fact, there is no standard usage of VASP that requires you to modify the POTCAR file (https://www.vasp.at/wiki/index.php/POTCAR). The value of POMASS you set in the INCAR will replace the value of POMASS read from the POTCAR in your VASP run.
(2) Notice that increasing the mass of H is a `trick` to allow you to use a larger time step (POTIM) in the MD simulation. By increasing the mass you reduce the vibration frequency which means you can sample the movement of the ions with a larger time step. This is justified if you are not interested in the particular way in which the H atoms are moving, otherwise, you need to use a small time step (POTIM). You can in principle estimate the time step you need in your MD simulation from the frequency of vibration of the lighter atoms you have in the simulation. Now, what constitutes a `light element` and whether changing its mass is a reasonable approximation really depends on the material and the properties you are interested in.
(3) RANDOM_SEED is written to the output of an MD run in VASP. You can use that value in the INCAR of another run to reproduce the calculation.
Thanks to your reply! I have known about the POMASS setting.
What is more, I `grep` RANDDOM_SEED in OUTCAR and it showed several results:
RANDOM_SEED = 644577723 0 0
RANDOM_SEED = 644577723 2376 0
RANDOM_SEED = 644577723 3564 0
RANDOM_SEED = 644577723 4752 0
RANDOM_SEED = 644577723 5940 0
RANDOM_SEED = 644577723 7128 0
RANDOM_SEED = 644577723 8316 0
RANDOM_SEED = 644577723 9504 0
RANDOM_SEED = 644577723 10692 0
RANDOM_SEED = 644577723 11880 0
RANDOM_SEED = 644577723 13068 0
RANDOM_SEED = 644577723 14256 0
RANDOM_SEED = 644577723 15444 0
RANDOM_SEED = 644577723 16632 0
RANDOM_SEED = 644577723 17820 0
RANDOM_SEED = 644577723 19008 0
RANDOM_SEED = 644577723 20196 0
RANDOM_SEED = 644577723 21384 0
RANDOM_SEED = 644577723 22572 0
RANDOM_SEED = 644577723 23760 0
RANDOM_SEED = 644577723 24948 0
RANDOM_SEED = 644577723 26136 0
RANDOM_SEED = 644577723 27324 0
RANDOM_SEED = 644577723 28512 0
RANDOM_SEED = 644577723 29700 0
RANDOM_SEED = 644577723 30888 0
RANDOM_SEED = 644577723 32076 0
RANDOM_SEED = 644577723 33264 0
RANDOM_SEED = 644577723 34452 0
RANDOM_SEED = 644577723 35640 0
RANDOM_SEED = 644577723 36828 0
RANDOM_SEED = 644577723 38016 0
RANDOM_SEED = 644577723 39204 0
RANDOM_SEED = 644577723 40392 0
RANDOM_SEED = 644577723 41580 0
RANDOM_SEED = 644577723 42768 0
RANDOM_SEED = 644577723 43956 0
RANDOM_SEED = 644577723 45144 0
RANDOM_SEED = 644577723 46332 0
RANDOM_SEED = 644577723 47520 0
RANDOM_SEED = 644577723 48708 0
RANDOM_SEED = 644577723 49896 0
RANDOM_SEED = 644577723 51084 0
RANDOM_SEED = 644577723 52272 0
RANDOM_SEED = 644577723 53460 0
RANDOM_SEED = 644577723 54648 0
RANDOM_SEED = 644577723 55836 0
RANDOM_SEED = 644577723 57024 0
RANDOM_SEED = 644577723 58212 0
RANDOM_SEED = 644577723 59400 0
RANDOM_SEED = 644577723 60588 0
If I want to reproduce this calculation, should I use the first value RANDOM_SEED = 644577723 0 0 ?

Re: Questions about MLFF settings

Posted: Thu Mar 23, 2023 7:37 am
by jun_yin2
Hi, I have another question about error. In https://www.vasp.at/wiki/index.php/Best ... rce_fields, VASP said that "Typical the fitting errors should be a few meV/atom for the energies and 30-100 meV/Angstrom for the forces at temperature between 300-1000 K. Errors that are slightly above these values can be potentially be acceptable, but those calculations should be thoroughly checked for correctness." What does this "fitting errors" refer to? Is the ERR in ML_LOGFILE or some other errors.

What is more, if this error means the ERR in ML_LOGFILE. Whether it means the RMSE (grep ERR in force in ML_LOGFILE) in figure is too big? Because I am rookie in VASP_MLFF, I am strange that why this RMSE curve suddenly rise after 10000 step and does not become flat anymore. I put my INCAR POSCAR KPOINTS ICONST here. Hope you can help me.

INCAR

Code: Select all

Startparameters for MD
# Gloabal paramters
ISTART = 0             # Read existing wavefunction =1
ICHARG = 2             # Default=2 if ISTART=0. Non-self-consistent: GGA/LDA band structures, =11
LWAVE =.FALSE.         # Write WAVECAR or not


# Parallelisation
LPLANE = .FALSE.       # Real space distribution; supercells 
NCORE = 16              # Max is no. nodes; don't set for hybrids##
# NSIM = 1             # Divides k-grid into separate groups
KPAR = 4               # Divides k-grid into separate groups
# NPAR = 4             # Divides k-grid into separate groups

# Electronic relaxation
PREC = Normal          # Precision level, Normal or High
LREAL = Auto           # Auto for molecule containing for than 20 ionics
ENCUT = 550            # Cut-off energy for plane wave basis set, in eV, default=max ENMAX in POTCAR,for bulk structure relaxation, ENCUT=1.3 * default
NELM = 60              # Max SCF steps
NELMIN = 5             # MIN SCF steps
EDIFF = 1E-05          # SCF energy convergence, this number is quite precise
ISPIN = 1              # Closed shell DFT
IVDW = 11              # DFT-D3 method of Grimme with zero-damping function

# Ionic relaxation
EDIFFG =  -0.02        # Ionic convergence; eV/AA^3
LDIOPL = .TRUE.        # Switch on/off dipole correction
IDIPOL = 3             # =3 means Dipole correction on z direction
ISMEAR = 0             # Gaussian smearing; metals:1
SIGMA = 0.05           # Smearing value in eV; metals:0.2
ISYM = 0               # =0 for MD

# Molecuar dynamics
IBRION = 0             # =0 for molecular dynamics
NSW = 20000            # MAX Inonic relaxation steps
POTIM = 1.5            # time step in fs
ISIF = 3               # NpT ensemble
MDALGO = 3             # Langevin thermostat, =1 for Anderson thermostat, =2 for Nose-Hoover thermostat
TEBEG = 100             # Starting temperature
TEEND = 450            # Ending temperature
LANGEVIN_GAMMA = 10.0 10.0  # 
LANGEVIN_GAMMA_L = 3.0 #
PMASS = 100            # These three tags related to Langevin thermostat
PSTRESS = 0.001        # Sets the external pressure in kB
POMASS = 16.0 8.0      # 8.0 for hydrogen atom, 16.0 was default in O_POTCAR, this array order should be consistent with POSCAR and POTCAR

# Machine learning force field
ML_LMLFF = .TRUE.      # Switch on/off MLFF
ML_ISTART = 0          # Get a MLFF from scrach
KPOINTS

Code: Select all

Gamma-point only
 0
Monkhorst Pack
 2 2 2
 0 0 0
POSCAR

Code: Select all

water_box
1.0
       12.9541997910         0.0000000000         0.0000000000
        0.0000000000        13.1023998260         0.0000000000
        0.0000000000         0.0000000000        11.4561004639
    O    H
   65  130
Direct
     0.665939987         0.530420005         0.148420006
     0.882219970         0.610669971         0.971570015
     0.705410004         0.340270013         0.388969988
     0.680060029         0.351000011         0.824930012
     0.092320003         0.417329997         0.965269983
     0.317580014         0.219779998         0.801919997
     0.893769979         0.295139998         0.973190010
     0.383679986         0.228650004         0.108599998
     0.369870007         0.565289974         0.950060010
     0.449559987         0.946179986         0.982330024
     0.029880000         0.781859994         0.774469972
     0.665650010         0.826250017         0.774020016
     0.705799997         0.044170000         0.441560000
     0.603640020         0.824119985         0.279119998
     0.458180010         0.910690010         0.532090008
     0.884239972         0.705439985         0.544340014
     0.090690002         0.057370000         0.890290022
     0.962149978         0.550509989         0.148389995
     0.965330005         0.120180003         0.133870006
     0.891799986         0.087360002         0.704010010
     0.296820015         0.860939980         0.275830001
     0.558210015         0.615530014         0.446029991
     0.706560016         0.176200002         0.117729999
     0.506039977         0.139100000         0.704150021
     0.481920004         0.478780001         0.686299980
     0.306019992         0.345779985         0.476339996
     0.462289989         0.208069995         0.440329999
     0.931270003         0.291170001         0.726270020
     0.218280002         0.753350019         0.599820018
     0.843840003         0.469520003         0.586579978
     0.070859998         0.397260010         0.330839992
     0.508130014         0.759890020         0.636950016
     0.878859997         0.251010001         0.370929986
     0.356240004         0.700659990         0.875790000
     0.212420002         0.698069990         0.983870029
     0.927640021         0.738290012         0.252590001
     0.201509997         0.097640000         0.660120010
     0.255470008         0.901080012         0.806770027
     0.227459997         0.448720008         0.710990012
     0.602440000         0.879490018         0.514829993
     0.070540003         0.540520012         0.793600023
     0.346320003         0.352959991         0.687759995
     0.550650001         0.101010002         0.436360002
     0.229369998         0.138119996         0.145600006
     0.668240011         0.597930014         0.933929980
     0.367320001         0.499579996         0.206760004
     0.210339993         0.743830025         0.365700006
     0.168019995         0.581070006         0.095190004
     0.638239980         0.052540001         0.965380013
     0.801750004         0.784739971         0.111550003
     0.136790007         0.327080011         0.769050002
     0.596040010         0.277069986         0.133169994
     0.769630015         0.941649973         0.106540002
     0.677820027         0.692070007         0.591899991
     0.515190005         0.758379996         0.084339999
     0.079790004         0.471410006         0.471749991
     0.032740001         0.419340014         0.708130002
     0.421469986         0.713710010         0.509519994
     0.477829993         0.460709989         0.430269986
     0.744769990         0.844579995         0.413170010
     0.505879998         0.343540013         0.863279998
     0.078599997         0.102090001         0.440129995
     0.849449992         0.405550003         0.373600006
     0.867219985         0.436260015         0.194519997
     0.007480000         0.639389992         0.419330001
     0.665480018         0.561869979         0.225290000
     0.595080018         0.508260012         0.136429995
     0.948490024         0.590939999         0.938719988
     0.893830001         0.678810000         0.002529979
     0.711019993         0.356759995         0.306459993
     0.633809984         0.354849994         0.408270001
     0.648249984         0.379339993         0.755219996
     0.732169986         0.400889993         0.846759975
     0.124229997         0.351020008         0.953059971
     0.090709999         0.425339997         0.049640059
     0.296970010         0.172480002         0.863030016
     0.382809997         0.247779995         0.829089999
     0.889180005         0.353260010         0.920679986
     0.857630014         0.240679994         0.932500005
     0.325989991         0.271710008         0.085730001
     0.430720001         0.231629997         0.042550001
     0.371210009         0.520519972         0.017729998
     0.382120013         0.520099998         0.884169996
     0.419149995         0.892849982         0.934369981
     0.394039989         0.967149973         0.034190059
     0.998449981         0.831510007         0.722360015
     0.995909989         0.793099999         0.849059999
     0.717040002         0.783309996         0.736429989
     0.681429982         0.821210027         0.856800020
     0.760380030         0.076590002         0.486490011
     0.698530018         0.086900003         0.372649997
     0.553820014         0.771000028         0.260729998
     0.564740002         0.873510003         0.324719995
     0.387430012         0.932410002         0.518450022
     0.457230002         0.882260025         0.610490024
     0.874040008         0.773959994         0.574800014
     0.857230008         0.708419979         0.465220004
     0.129099995         0.006420000         0.846459985
     0.104489997         0.040950000         0.971589983
     0.999360025         0.518999994         0.212689996
     0.987529993         0.620329976         0.146500006
     0.963329971         0.176390007         0.189229995
     0.981459975         0.060710002         0.181250006
     0.881380022         0.098410003         0.787100017
     0.826080024         0.105630003         0.668720007
     0.327089995         0.918079972         0.233879998
     0.265760005         0.819029987         0.215220004
     0.489300013         0.631009996         0.474620014
     0.546620011         0.572459996         0.378160000
     0.744090021         0.114299998         0.137490004
     0.701200008         0.174490005         0.033089999
     0.567669988         0.153050005         0.749890029
     0.493400007         0.066639997         0.715680003
     0.534309983         0.504710019         0.739340007
     0.511229992         0.415899992         0.655719995
     0.279320002         0.389149994         0.414440006
     0.348670006         0.391039997         0.523239970
     0.387439996         0.210759997         0.445879996
     0.485450000         0.202169999         0.520789981
     0.990920007         0.261110008         0.687929988
     0.872770011         0.268620014         0.679740012
     0.145040005         0.738950014         0.608370006
     0.251599997         0.687460005         0.610080004
     0.773150027         0.448599994         0.602129996
     0.841949999         0.543619990         0.590789974
     0.040010002         0.330960006         0.315450013
     0.143110007         0.382330000         0.346269995
     0.573409975         0.725860000         0.621439993
     0.486790001         0.733749986         0.712639987
     0.889249980         0.178580001         0.385149986
     0.897610009         0.283549994         0.444209993
     0.385069996         0.766979992         0.895470023
     0.401679993         0.674430013         0.815249979
     0.182170004         0.630410016         0.977180004
     0.201900005         0.728680015         0.907459974
     0.994480014         0.736859977         0.291150004
     0.943719983         0.751869977         0.171149999
     0.259070009         0.124650002         0.615270019
     0.178929999         0.038190000         0.616169989
     0.327730000         0.881389976         0.811140001
     0.239559993         0.926689982         0.884370029
     0.301160008         0.448040009         0.727060020
     0.222920001         0.453170002         0.626420021
     0.619270027         0.843200028         0.443289995
     0.659330010         0.863969982         0.567279994
     0.078210004         0.531090021         0.709869981
     0.051040001         0.473479986         0.822579980
     0.301730007         0.375319988         0.624469995
     0.413489997         0.381660014         0.668839991
     0.601300001         0.149250001         0.465999991
     0.532400012         0.059680000         0.503769994
     0.190490007         0.159099996         0.214110002
     0.241679996         0.065640002         0.157419994
     0.656069994         0.554970026         0.866110027
     0.642939985         0.557829976         0.999379992
     0.301200002         0.476889998         0.237399995
     0.377920002         0.566229999         0.242100000
     0.283030003         0.750180006         0.385540009
     0.182699993         0.812219977         0.376239985
     0.101219997         0.586560011         0.057020001
     0.161029994         0.522539973         0.146779999
     0.677150011         0.101219997         0.918820024
     0.690429986         0.007360000         0.997820020
     0.776960015         0.716970026         0.131819993
     0.865429997         0.772419989         0.068870001
     0.075360000         0.291900009         0.741509974
     0.156990007         0.370440006         0.704060018
     0.590690017         0.350230008         0.145999998
     0.550180018         0.263850003         0.067690000
     0.839489996         0.922680020         0.128739998
     0.746420026         0.885940015         0.056979999
     0.741770029         0.728389978         0.607680023
     0.673349977         0.690190017         0.507200003
     0.547990024         0.701020002         0.123369999
     0.442009985         0.741859972         0.085320003
     0.024960000         0.424450010         0.493510008
     0.046769999         0.519630015         0.419129997
     0.043979999         0.365759999         0.650810003
     0.980279982         0.391739994         0.759980023
     0.401679993         0.723850012         0.428479999
     0.360280007         0.732420027         0.553789973
     0.426750004         0.409179986         0.450100005
     0.491100013         0.495959997         0.503440022
     0.752049983         0.770889997         0.418810010
     0.778410017         0.870500028         0.483020008
     0.500429988         0.402980000         0.812860012
     0.566510022         0.307509989         0.834879994
     0.018030001         0.119879998         0.394309998
     0.135279998         0.136639997         0.400950015
     0.892690003         0.457650006         0.409130007
     0.812359989         0.375360012         0.438820004
     0.925989985         0.450040013         0.244900003
     0.890959978         0.382270008         0.142849997
     0.966490030         0.679629982         0.365130007
     0.033890001         0.688970029         0.474979997

ICONST

Code: Select all

LA 1 2 0
LA 1 3 0
LA 2 3 0
LR 1 0
LR 2 0
LR 3 0
S  1  0  0  0      0    0    0
S  0  1  0  0      0    0    0
S  0  0  1  0      0    0    0
S  0  0  0  13.0 -12.0  0.0  0
S  0  0  0  11.0  0.0 -12.0  0
S  0  0  0  0.0  11.0 -13.0  0

Re: Questions about MLFF settings

Posted: Thu Mar 23, 2023 12:52 pm
by henrique_miranda
Yes, at each MD step, you need to get new random numbers for the thermostat so the state of the random number generator is updated.
The RANDOM_SEED is reported so that if you want you can restart the MD calculation with the random number generator in the same state.
If you want to reproduce this calculation then you should use the first value:
RANDOM_SEED = 644577723 0 0

Note that if you don't set the RANDOM_SEED then a random value is used based on the system clock.
You only need to set the RANDOM_SEED in the input file if you want to reproduce a previous run.

Re: Questions about MLFF settings

Posted: Fri Mar 24, 2023 2:04 am
by jun_yin2
henrique_miranda wrote: ↑Thu Mar 23, 2023 12:52 pm Yes, at each MD step, you need to get new random numbers for the thermostat so the state of the random number generator is updated.
The RANDOM_SEED is reported so that if you want you can restart the MD calculation with the random number generator in the same state.
If you want to reproduce this calculation then you should use the first value:
RANDOM_SEED = 644577723 0 0

Note that if you don't set the RANDOM_SEED then a random value is used based on the system clock.
You only need to set the RANDOM_SEED in the input file if you want to reproduce a previous run.
Very thanks to your reply! I also posted another question about fitting errors above. Would you mind helping me? Very thanks to you.

Re: Questions about MLFF settings

Posted: Fri Mar 24, 2023 7:48 am
by henrique_miranda
Regarding your second question, it looks to me like the error in the forces is too high.
It does not make sense that the error starts increasing linearly at around the 10000 step.
[EDIT: the error is not linearly increasing, instead there is only a calculation of the error when an ab initio calculation is done at around 10000 and then again at 20000 which yields a larger value (see post of victor_robinson bellow)]

You might consider trying the INCAR for water on this page
https://www.vasp.at/wiki/index.php/Best ... ds#Example

Re: Questions about MLFF settings

Posted: Fri Mar 24, 2023 7:58 am
by jun_yin2
henrique_miranda wrote: ↑Fri Mar 24, 2023 7:48 am Regarding your second question, it looks to me like the error in the forces is too high.
It does not make sense that the error starts increasing linearly at around the 10000 step.
[EDIT: the error is not linearly increasing, instead there is only a calculation of the error when an ab initio calculation is done at around 10000 and then again at 20000 which yields a larger value (see post of victor_robinson bellow)]

You might consider trying the INCAR for water on this page
https://www.vasp.at/wiki/index.php/Best ... ds#Example
Very thanks to your reply. All settings in INCAR are same as the website mentioned except the ENCUT. I will try again to use much higher ENCUT. Also, I want to know in which range can the error value of RMSE be regarded as normal?

Re: Questions about MLFF settings

Posted: Fri Mar 24, 2023 9:20 pm
by victor_robinson
It looks like the error jumps (rather than linearly rises) between 10,000 and 20,000 steps as there is a new ab initio calculation at step 20,000. I would guess that more training is required if you want this error to be lower.

Re: Questions about MLFF settings

Posted: Mon Mar 27, 2023 7:42 am
by jun_yin2
victor_robinson wrote: ↑Fri Mar 24, 2023 9:20 pm It looks like the error jumps (rather than linearly rises) between 10,000 and 20,000 steps as there is a new ab initio calculation at step 20,000. I would guess that more training is required if you want this error to be lower.
Very thanks to your reply. Do you know what the exact meaning of 'fitting errors' in the 'Typical the fitting errors should be a few meV/atom for the energies and 30-100 meV/Angstrom for the forces at temperature between 300-1000 K. Errors that are slightly above these values can be potentially be acceptable, but those calculations should be thoroughly checked for correctness.' from this VASP wiki. Is it refers to Bayesian errors or the Root mean square error?

Re: Questions about MLFF settings

Posted: Thu Mar 30, 2023 2:10 am
by victor_robinson
It sounds like that refers to the 'exact' error of the ML potential, which would be the RMSE error (different between the potential and DFT).
Victor

Re: Questions about MLFF settings

Posted: Wed Apr 12, 2023 8:05 am
by jun_yin2
victor_robinson wrote: ↑Thu Mar 30, 2023 2:10 am It sounds like that refers to the 'exact' error of the ML potential, which would be the RMSE error (different between the potential and DFT).
Victor
very thanks yo your reply Victor. I believe you are right. I read some papers using MLFF. When considering accuracy and quality, they would post the RMSE curve. As a result, I think the ''error' here should refer to RMSE error.

Tieuyan Bian
All times are UTC
Page 1 of 1

Powered by phpBB® Forum Software © phpBB Limited