CHARMM Development Project Forums User Discussion & Questions Molecular Dynamics temp drift with "gbmv"

Hello,

i am using GBMV with dynamics and I see a huge drift in temp(rise) when i use time step of .002 with shake. Re-running with .002 only leads to a small drift but i am afriad if i run longer i might still see it rising. What do you suggest to avoid this temp jump w/o larger time step?

thanks, preeti
____
set ctofnb = 12.0
calc ctonnb = @ctofnb - 2.0
calc cutnb = @ctofnb + 2.0
calc cutim = @cutnb

scalar wmain set 0.0 sele all end

NBOND atom switch cdie vdw vswitch -
ctonnb @ctonnb ctofnb @ctofnb cutnb @cutnb cutim @cutim

GBMV BETA -20 EPSILON 80 DN 1.0 watr 1.4 GEOM -
TOL 1e-8 BUFR 0.5 Mem 10 CUTA 20 HSX1 -0.125 HSX2 0.25 -
ALFRQ 1 EMP 1.5 P4 0.0 P6 8.0 P3 0.70 ONX 1.9 OFFX 2.1 -
WTYP 2 NPHI 38 SHIFT -0.102 SLOPE 0.9085 CORR 1

open read card unit 10 name prot-eq.pdb
read coor pdb unit 10
close unit 10

shake fast bonh tol 1.0e-8 para

open unit 10 write file name "dyna-gbmv.dcd"
OPEN UNIT 21 write card name "dyna-gbmv.rest"

dynamics strt verl nstep 5000 time 0.001 imgfrq 10 iprfrq 200 -
inbfrq 10 nprint 200 ihtfrq 0 ieqfrq 0 ihbfrq 0 iunwri 21 -
iuncrd 10 nsavc 1000 nsavv 0 -
firstt 300.0 finalt 300.0 twindh 1.0 twindl -1.0 -
iasors 1 iasvel 1 ichecw 1 tstruc 300.0 ntrfrq 100 teminc 0 -
atom switch cdie vdw vswitch ctonnb @ctonnb -
ctofnb @ctofnb cutnb @cutnb cutim @cutim

close unit 10
close unit 21

You mention 2 fs, but the input posted uses 1 fs (timestep 0.001); what happens with 1 fs?

It's best to use the default SHAKE tolerance, and not to reduce.

If you see timestep based effects, using a smaller timestep is the recommended course of action.

Rick,

I have attached a plot of temp drift.

From earlier post, i realized two things. One use of "UPDATE 1" and another use of "Langevin" dynamics.

1) Can I still use 2fs step and shake with GBMV and langevin dynamics?
2) Use of Langevin for dynamics run and not equilibration should i use "IEQFRQ=0"?
3) SHould i run equilibration and dynamics,both using langevin dynamics or can i just begin a new dynamics run without restarting from equilibrated run for GBMV?

Thank you for your help, preeti

DYNA LANGVEIN LEAP @STATUS NSTEP @NSTEP TIMESTEP @TSTEP -
IPRFRQ 1000 IHTFRQ 0 IEQFRQ 100 NTRFRQ 200 INBFRQ -1 IMGFRQ -1 -
IHBFRQ 0 -
IUNREA -1 IUNWRI 14 IUNCRD 15 IUNVEL -1 KUNIT -1 -
NPRINT 100 NSAVC 10 NSAVV -1 -
ILBFRQ 1000 RBUFF 0.0 ISVFRQ 10000 -
FIRSTT 300.00 FINALT 300.0 TEMINC 0.0 TBATH 300.0 -
TWINDH 10.0 TWINDL -10.0 TSTRUC 300.0 -
IASORS 1 ISCVEL 0 IASVEL 1 ICHECW 1 ECHECK 99999999.0

What are red and green lines on the plot?

I'm not familiar with "UPDATE 1"; but then again I'm not very familiar with GBMV, either. Have you considered using another of the many implicit water models in CHARMM, such as SCPISM?

SHAKE BONH PARAM should always be used with a 2 fs timestep.

There should not be a problem using Langevin dynamics with implicit solvent; it provides T control, and the use of low friction (SCALAR FBETA SET 2.0) accelerates barrier crossings (torsional state changes) w/o affecting populations. You may have to run some test simulations to determine if a 2 fs timestep is reasonable for your case or not.

For Langevin dynamics, IHTFRQ, IEQFRQ, and ICHECW should all be zero; these represent a different T control method.

ECHECK should not be so large; even ECHECK 1000. may be excessive.

ISVFRQ is not really needed; it's a legacy feature from early on when computers used to crash a lot more often.

Dear Rick,

I did re-attach the plot to make it better.

However, after use of update 1, i am getting better energy conservation.

But, I will also try another implict solvent you suggested. When i run parallel charmm on my supercmputers, i find use of "isvfreq" necessary. If i don't specify it and if the job terminates on exceeding the run time limit, it dosen't write the restart file until the required number of steps have finished(or dynamics run is finished).

I am currently using "fbeta=10" from another test file. You recommend, "fbeta=2" is better?

thanks for your help,
preeti

I do recommend "SCALAR FBETA SET 2.0" for torsional sampling with Langevin dynamics.

I'm not still not sure what you mean by "update 1"