[gmx-users] Re to Berk Hess (pull code core dump)

[Daniel Santos]

Hi Berk,

Thank you very much for your kind reply.

> Which Gromacs version are you using?
:-)  VERSION 3.2.1  (-:
 
> There were problems with constraint pulling in 3.2, but these have been fixed in 3.2.1.
Yes, I know.

> I have tested the constraint pulling in 3.2.1, but not in combination with a dynamic reference group.
The problem seams to be there.
As I said, all goes well until it segmentation faults (all properties seams to be well behaved). I just checked that in one of the files generated (I use a queue system so I think that this is the standart error output file), one can read:

Step 252141, time 504.282 (ps)  LINCS WARNING
relative constraint deviation after LINCS:
max 1893.144043 (between atoms 2663 and 2664) rms 40.527763
bonds that rotated more than 30 degrees:
 atom 1 atom 2  angle  previous, current, constraint length
      1      2   90.7    0.1230 219.1096      0.1230
      2      3   93.5    0.1340  62.6208      0.1340
     14      2   94.0    0.1390  58.5084      0.1390
      3      4   95.6    0.1340  21.9224      0.1340
      4      5  106.6    0.1390   5.9581      0.1390
      4     12  112.4    0.1390   5.2224      0.1390
      5      6  101.4    0.1390   1.5593      0.1390
      6      7  121.0    0.1330   0.3058      0.1330
      6     10  107.7    0.1390   0.3725      0.1390
      7      8   88.7    0.1330   0.2270      0.1330
      8      9   89.1    0.1330   0.4649      0.1330
      9     10   87.1    0.1330   0.4419      0.1330
     10     11  123.5    0.1390   0.5343      0.1390
     11     12  153.6    0.1390   0.8874      0.1390
     12     13  111.7    0.1390   4.7773      0.1390
     13     14   92.6    0.1390  29.7670      0.1390
   2658   2659   48.9    0.1530   0.2136      0.1530
   2659   2660   85.7    0.1530   0.5328      0.1530
   2660   2661   90.8    0.1530   3.8225      0.1530
   2661   2662   89.6    0.1530  17.3407      0.1530
   2662   2663   90.0    0.1530 286.8617      0.1530
   2663   2664   90.0    0.1530 289.8040      0.1530


Atoms 1 and 2 belong to the molecule inside the bilayer (in the middle) and atoms 2663 and 2664 are terminal atoms in the lipid carbon chain (near the molecule's insertion point).

The question is: how the bond lengths can just explode? to large integration step? but I'm using a typical 2 fs time-step for a typical united-atom system!
Or it can be due to the readjustment of the distance between the two reference groups, were the atoms can be to near and then the forces become two strong and the Lincs algorithm can't adjust the distances to the correct values. So this can be a problem with the methodology used for this system...


Thank you very much again!

Any comment/help will be very much appreciated.

Sincerely,
Daniel dos Santos