From: Erik Nordgren (nordgren_at_sas.upenn.edu)
Date: Fri Feb 03 2017 - 14:05:23 CST
Hi all,
I just noticed that there seems to be a drastic difference in the energy
output reported in the logs for two jobs run on the very same system, one
using NAMD 2.9 and the other with NAMD 2.11. Seems like it must just be a
change in how the energy values are reported, at least during minimization,
but I was unable to dig up any documentation or description of such.
As an example, I ran two instances of a simulation of a membrane/protein
system for 1000 steps of minimization and then 10k steps of dynamics, with
the exact same starting coordinates (PDB), same structure (PSF), same force
field parameter files, and all the same NAMD configuration options, both
using 64 cores on a Linux cluster, but ran one with NAMD 2.9 and the other
with version 2.11. I realize that two "identical" simulations won't agree
*exactly* due to randomness of Langevin dynamics, rounding error, etc...
but what I see can't be attributed to that. Here are a few lines of energy
output excerpted from the two logs -- at step zero, at step 1 of
minimization, at step 1000 of minimization, and at step 0 of dynamics:
With NAMD 2.9:
ETITLE: TS BOND ANGLE DIHED IMPRP
ELECT VDW BOUNDARY MISC
KINETIC TOTAL TEMP POTENTIAL TOTAL3
TEMPAVG
ENERGY: 0 40562.3800 156368.0743 125759.6833 3270.7073
-1166882.5822 -8171.6672 16030.7769 0.0000
0.0000 -833062.6278 0.0000 -833062.6278 -833062.6278
0.0000
ENERGY: 1 40469.1518 156280.6164 125755.1820 3267.5845
-1167021.5620 -8185.6547 16029.4913 0.0000
0.0000 -833405.1908 0.0000 -833405.1908 -833405.1908
0.0000
ENERGY: 1000 47980.3143 61233.5700 104550.3725 612.7848
-1536797.0824 23311.5138 638.5428 0.0000
0.0000 -1298469.9843 0.0000 -1298469.9843 -1298469.9843
0.0000
ENERGY: 1000 8374.3601 35906.4000 104548.8197 612.0375
-1404946.8200 20647.6005 638.9072 0.0000
0.0000 -1234218.6949 0.0000 -1234218.6949 -1234230.5841
0.0000
With NAMD 2.11:
ETITLE: TS BOND ANGLE DIHED
IMPRP ELECT VDW BOUNDARY
MISC KINETIC TOTAL TEMP
POTENTIAL TOTAL3 TEMPAVG
ENERGY: 0 40541.8443 156365.8565 125752.1995
3270.6998 -1166882.8307 -8172.1404 16030.7650
0.0000 0.0000 -833093.6060 0.0000 -833093.6060
-833093.6060 0.0000
ENERGY: 1 18190.6290 123471.3949 123824.6905
1946.1441 -1170595.7197 -12505.9502 15396.3245
0.0000 0.0000 -900272.4868 0.0000 -900272.4868
-900272.4868 0.0000
ENERGY: 1000 8313.6243 35641.1338 104237.5781
613.8773 -1390124.0655 -10.3182 578.5605
0.0000 0.0000 -1240749.6098 0.0000 -1240749.6098
-1240749.6098 0.0000
ENERGY: 1000 8313.6243 35641.1338 104237.5781
613.8773 -1390124.0655 -10.3182 578.5605
0.0000 0.0000 -1240749.6098 0.0000 -1240749.6098
-1240752.5355 0.0000
What seems really strange is that under NAMD 2.9, several of the energies
suddenly and dramatically change upon the start of dynamics (but before
anything has actually moved), but not at the start of minimization;
meanwhile the reverse appears to be true under NAMD 2.11 (i.e., a sudden
change in energies at the first step of minimization, but not at the start
of dynamics). Once dynamics gets started, both systems again agree within
reasonable expectations -- except for the VdW energy...
There is probably a simple/obvious explanation for what I'm seeing... but
unfortunately it isn't obvious to me, so I'll appreciate any insight,
thanks!
Erik
-- C. Erik Nordgren, Ph.D. Department of Chemistry University of Pennsylvania
This archive was generated by hypermail 2.1.6 : Sun Dec 31 2017 - 23:21:02 CST