Fwd: ramd seed

From: Francesco Pietra (francesco.pietra_at_accademialucchese.it)
Date: Thu Apr 07 2011 - 08:41:08 CDT

Vlad:
Perhaps it may be useful to know about another (immaterial) error of
ramd/namd exiting from a long simulation (Debian amd64; ts=1fs; rigid
bonds water only):

 ENERGY: 340650 858.1960 1433.4781 578.9458
114.4911 -125473.7392 10324.2822 0.0000
0.0000 21191.4674 -90972.8787 307.3651
-112164.3460 -90939.9722 307.9922 77.3879
207.4763 329214.3808 176.9608 174.6795

WRITING EXTENDED SYSTEM TO RESTART FILE AT STEP 340650
WRITING COORDINATES TO RESTART FILE AT STEP 340650
FINISHED WRITING RESTART COORDINATES
The last position output (seq=340650) takes 0.013 seconds, 28235.674
MB of memory in use
WRITING VELOCITIES TO RESTART FILE AT STEP 340650
FINISHED WRITING RESTART VELOCITIES
The last velocity output (seq=340650) takes 0.003 seconds, 28235.674
MB of memory in use
TCL: EXIT: 340650 > MAX DISTANCE LIGAND COM - PROTEIN COM REACHED
TCL: EXIT: 340650 > LIGAND EXIT EVENT DETECTED: STOP SIMULATION
TCL: EXIT: 340650 > EXIT NAMD
TCL: couldn't fork child process: not enough memory
FATAL ERROR: couldn't fork child process: not enough memory
    while executing
"exec kill -9 $process"
    (in namespace eval "::RAMD" script line 9)
    invoked from within
"namespace eval ::RAMD {

 #***** Terminate NAMD if the ligand has exited from the protein
 if { $exitFlag == 1 } {
  print "EXIT: $timeStep > MAX DI..."
    (procedure "calcforces" line 2)
    invoked from within
"calcforces"
------------- Processor 0 Exiting: Called CmiAbort ------------
Reason: FATAL ERROR: couldn't fork child process: not enough memory
    while executing
"exec kill -9 $process"
    (in namespace eval "::RAMD" script line 9)
    invoked from within
"namespace eval ::RAMD {

 #***** Terminate NAMD if the ligand has exited from the protein
 if { $exitFlag == 1 } {
  print "EXIT: $timeStep > MAX DI..."
    (procedure "calcforces" line 2)
    invoked from within
"calcforces"

[0] Stack Traceback:
  [0:0] CmiAbort+0x7b [0xba4b8d]
  [0:1] _Z8NAMD_diePKc+0x62 [0x534112]
  [0:2] _ZN15GlobalMasterTcl9calculateEv+0x294 [0x850a0e]
  [0:3] _ZN12GlobalMaster11processDataEPiS0_P6VectorS2_S2_S0_S0_S2_S0_S0_S2_+0x71
 [0x8423c1]
  [0:4] _ZN18GlobalMasterServer11callClientsEv+0x8ee [0x846872]
  [0:5] _ZN18GlobalMasterServer8recvDataEP20ComputeGlobalDataMsg+0x6f7
 [0x845a33]
  [0:6] _ZN10ComputeMgr21recvComputeGlobalDataEP20ComputeGlobalDataMsg+0x12
 [0x5bd300]
  [0:7] _ZN18CkIndex_ComputeMgr48_call_recvComputeGlobalData_ComputeGlobalDataMsgEPvP10ComputeMgr+0xf
 [0x5bd2eb]
  [0:8] CkDeliverMessageFree+0x21 [0xae2693]
  [0:9] _Z15_processHandlerPvP11CkCoreState+0x70b [0xae153b]
  [0:10] CsdScheduleForever+0xa5 [0xbab7a7]
  [0:11] CsdScheduler+0x1c [0xbab3a8]
  [0:12] _ZN7BackEnd7suspendEv+0xb [0x53cbd5]
  [0:13] _ZN9ScriptTcl3runEv+0x105 [0xa2eeeb]
  [0:14] _Z18after_backend_initiPPc+0x4fd [0x538585]
  [0:15] main+0x3a [0x538052]
  [0:16] __libc_start_main+0xe6 [0x2b963b0481a6]
  [0:17] _ZNSt8ios_base4InitD1Ev+0x52 [0x533a0a]
Fatal error on PE 0> FATAL ERROR: couldn't fork child process: not enough memory
    while executing
"exec kill -9 $process"
    (in namespace eval "::RAMD" script line 9)
    invoked from within
"namespace eval ::RAMD {

 #***** Terminate NAMD if the ligand has exited from the protein
 if { $exitFlag == 1 } {
  print "EXIT: $timeStep > MAX DI..."
    (procedure "calcforces" line 2)
    invoked from within
"calcforces"

The message "TCL: couldn't fork child process: not enough memory" must
be spurious as the shared-memory machine has plenty of memory even for
the matrices of ab initio calculations.

cheers
francesco

---------- Forwarded message ----------
From: Vlad Cojocaru <vlad.cojocaru_at_mpi-muenster.mpg.de>
Date: Mon, May 31, 2010 at 8:13 PM
Subject: Re: ramd seed
To: Francesco Pietra <francesco.pietra_at_accademialucchese.it>

Ok ! Thanks for the explanation. If RAMD alone suggests multiple
routes, and there is no other data to prove this its of course
difficult, especially if the results do not converge after a number of
simulations.

Vlad

Francesco Pietra wrote:

On Mon, May 31, 2010 at 6:52 PM, Vlad Cojocaru
<vlad.cojocaru_at_mpi-muenster.mpg.de> wrote:

Francesco,

I am not sure I am following you ..
You say that RAMD was showing you only 1 path although you know there
are more than one ? Did I misunderstood something??

No, I was talking about three different proteins.

(1) In the past, with an ion channel, I got a single path by using ramd.

(2) More recently, with a globular protein, I got multiple paths with
ramd. And this agreed with both experiments and unbiased MD.

(3) Now, with another globular protein I am getting - preliminarily -
multiple paths (multiple ligands, I am working on each at a time just
to explore). Here there is no experiment nor unbiased MD simulation
with which to compare. Ramd is alone, except, of course calculating
PMF or some other free-energy landscape (difficult to carry out with a
large protein).

Sorry, for the unclear previous message.

francesco

francesco

In this case, you can try to increase the perturbation and see you still
get a single pathway ...

If RAMD keeps on showing multiple paths and the results do not converge
at all, you'd probably need to backup your results by doing ssome free
energy calculations along the route identified with RAMD

Vlad

Francesco Pietra wrote:

Vlad:
Thanks. More or less, this is what I did recently with a system prone
to multiple paths. In that case, however, I had firm benchmarks:
experimental location of spots along the pathways and unbiased
molecular dynamics. Ramd - cleaned by using statistics - mostly
agreed.

In present case, ramd is alone in dealing with multiple paths. When
these are really many, the problem is difficult. My aptitude is still
to thank ramd for showing that there are multiple paths (I have in the
past investigated an ion channel where ramd constantly indicated a
single path by any choice of seed at the lest acceleration). Though,
to arrive at sound science it will be hard. Here, thorough MD could
not be carried out by any means.

francesco

On Mon, May 31, 2010 at 12:26 PM, Cojocaru, Vlad
<vlad.cojocaru_at_mpi-muenster.mpg.de> wrote:

Francesco,

If you change the seed with only 1 unit, you will have a completely different
trajectory.
Just plot the vector directions generated by the RAMD at different time steps
(and written in the output file) and you will see what I mean. They will be
completely different.

If you change the seed drastically as you say, there is no other effect than
if you change it with 1 unit.

If multiple pathways are possible in your case, run 10 trajectories with
different seeds (no matter whether you change with 1 unit or 10000 units).
Plot the distribution of pathways. Run an additional set of 5 simulations and
see if the distribution changes .... If it does, you don't have convergence.
If it doesn't you might have convergence. You can do this exercise with very
short trajectories to test it.

Vlad

-----Original Message-----
From: chiendarret_at_gmail.com on behalf of Francesco Pietra
Sent: Mon 5/31/2010 12:14 PM
To: Cojocaru, Vlad
Subject: ramd seed

Hi Vlad:
While I am trying to understand if the project I alluded to semms
worth while to be pursued, I noticed that to explore the possibility
of different paths I have to change drastically the ramd seed.

I noticed that before, though, the statistics is still limeted, so
that I would like to know what your experience was about the ramd
seed.

Changing drastically the seed means, for example, from 14257 to 1315.
On small changes, say from 14257 to 14255, change of pathway is much
less likely.

Also, shoud the seed be a prime number?

thanks

francesco pietra

-- 
Dr. Vlad Cojocaru
Max Planck Institute for Molecular Biomedicine
Department of Cellular and Developmental Biology
Roentgenstrasse 20
48149 Muenster
Tel: +49-251-70365-324
Fax: +49-251-70365-399
e-mail: vlad.cojocaru[at]mpi-muenster.mpg.de

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:23:48 CST