Topic Options
#35894 - 08/23/16 09:23 AM Minimization with block and PME
RobArbon Offline
Forum Member

Registered: 11/02/15
Posts: 17
Hi all,

I've solvated and neutralised my protein and water system and I'm using PME for my electrostatics. I'm trying to use Block to turn off some VDW interactions, in conjunction with some adiabatic mapping. I'm leaving out the adiabatic mapping here for parsimony.

My input and output files are attached. I'm using C41a2 on a Mac OSX Yosemite.

Problem: Whenever I try and minimize my system the coordinates do not change but the total energy (in particular the EWKSum) blows up (to 'infinity' eventually). When I turn off Block, everything works fine.

With Block I'm defining a region of 3 atoms in which the VdW interactions are set to zero. The scaling between this block and the 'outer' block is set to 1. The total charge on the group is -1. The overall charge on the system is 0.0

Things I've tried but the energy still blows up:
1. Initializing BLOCK before and after my crystal/image setup.
2. Keeping the BLOCK setup the same except changing the scaling from 0 to 1 for the VdW interactions with the region.
3. Increased the Cutoff from 30 to 40 to 60 in my crystal build command.
4. Used ABNR instead of SD.

I've read the latest docs on Block which say that the 'image' command should work with Block but the 'Crystal' command might not. I don't particularly understand this as I thought the two went hand in hand.

Any help would be greatly appreciated.

Thanks,

Rob

EDIT - I just noticed 'fswitch' was specified for the electrostatic interactions. Taking that out didn't change anything however.

EDIT - looked again at the previous thread on this, here I misread the date and assumed it was 10 years ago and had been fixed by now. The latest docs do say it should work however.

EDIT - just tried using the slower Standard Ewald summation for the non-bonded interactions. I haven't fully tested it but it appears to work.

Code:
set kppa 0.34
calc kmx int( @kppa*@xdim )   !XDIM = x-dim for crystal.
calc kmy int( @kppa*@ydim )
calc kmz int( @kppa*@zdim )
NBONDED inbfrq -1 imgfrq -1 nbxmod 5 -                                              ! C36 DEFAULTS 
        atom cdiel - !fshift -                                                         ! use atom based electrostatics, C36 DEFAULTS
        vatom vdistance vfswitch -                                                  ! C36 DEFAULTS   
        elec ewald kappa @kppa spline nopm kmxx @kmx kmxy @kmy kmxz @kmz   -        ! standard Ewald SUMMATION
        cutnb 14.0 ctofnb 12.0 ctonnb 10.0 -                                        ! C36 DEFAULTS  
        cutim 14.0 -                                                                ! must be same as cutnb
        eps 1.0 e14fac 1.0 wmin 1.5                                                 ! C36 DEFAULTS   



Attachments
pme_test.txt (253 downloads)
pme_test_out.txt (261 downloads)



Edited by RobArbon (08/23/16 11:40 AM)

Top
#35897 - 08/26/16 05:20 PM Re: Minimization with block and PME [Re: RobArbon]
rmv Offline

Forum Member

Registered: 09/17/03
Posts: 8379
Loc: 39 03 48 N, 77 06 54 W
I polled some of the CHARMM developers, and there was skepticism about whether or not BLOCK was actually compatible with PME, despite this statment in block.doc:

Code:
   BLOCK was modified so that it works with the Ewald (simple and PME)
method of CHARMM. The Syntax and Function of BLOCK module are unchanged. 

There are some issues with Ewald, and what you've done above may be a little problematic; the kmx, etc. values may be so large that the calculation will be very slow, and no value was provided for the KSQMAX option.
_________________________
Rick Venable
computational chemist


Top
#35902 - 08/29/16 11:41 AM Re: Minimization with block and PME [Re: RobArbon]
rmv Offline

Forum Member

Registered: 09/17/03
Posts: 8379
Loc: 39 03 48 N, 77 06 54 W
Further comments from developers indicate PME and BLOCK may have worked before the conversion to Fortran95 (which became public with the c36b1 release). As noted in the earlier post, the issue appears to arise with two or more blocks, and the problem went unnoticed as the test case used only a single block. I believe the probability of a fix has gone up a bit.
_________________________
Rick Venable
computational chemist


Top
#35903 - 08/29/16 04:32 PM Re: Minimization with block and PME [Re: RobArbon]
lennart Offline

Forum Member

Registered: 09/25/03
Posts: 4742
Loc: ~ 59N, 15E
We have used BLOCK with PME for standard free energy calculations (three blocks, with lamdad-dependent scale factors as per block.doc; we see no crashes, and we obtain results that are very close to PERT/PME (within ~0.2 kcal/mol for changes of ~1.5 kcal/mol). There are a number of testcases for this type of application, also for use with OpenMM.
_________________________
Lennart Nilsson
Karolinska Institutet
Stockholm, Sweden

Top
#35921 - 09/02/16 06:57 AM Re: Minimization with block and PME [Re: RobArbon]
RobArbon Offline
Forum Member

Registered: 11/02/15
Posts: 17
Thanks both for your input, especially about the KSQMAX, I'd missed that, I thought it had a default value based on the other inputs. I'll correct that.

I'm creating another simple test case and will report back.

Top

Moderator:  BRBrooks, John Legato