The ABINIT Users Mailing List ( CLOSED )

["Ludwig, Christian" <ludwigc@uni-mainz.de>]

Hello,

I am using abinit for parallel calculations on a cluster of x86 nodes. The 
nodes have 4 and 8 processors and I am using one or multiple nodes for one 
job. Sometimes everything works fine, but sometimes I get the following.

After some Broyden Steps I get the message 
"At Broyd/MD step   3, gradients are converged : 
  max grad (force/stress) = 9.3796E-04 < tolmxf= 1.0000E-03 ha/bohr (free 
atoms)"
This is the last thing written to abinit.out. When I look into the log-file, 
after this message I find

From Broyden minimization, approx. inverse Hessian(ndim,ndim):

     1
  2.61406366996692E-03  1.57678343240617E-03  0.00000000000000E+00
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
...
...
...
    13
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
  2.61406366996692E-03  1.57678343240617E-03  0.00000000000000E+00
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
  0.00000000000000E+00  0.00000000000000E+00  0.00000000000000E+00
forrtl: Resource temporarily unavailable
forrtl: severe (38): error during write, unit 6, file stdout
Image              PC                Routine            Line        Source    
         
abinip             00000000014249FA  Unknown               Unknown  Unknown
abinip             0000000001423BFA  Unknown               Unknown  Unknown
abinip             00000000013CC4AA  Unknown               Unknown  Unknown
abinip             000000000138B932  Unknown               Unknown  Unknown
abinip             000000000138AF4E  Unknown               Unknown  Unknown
abinip             00000000013C3F3E  Unknown               Unknown  Unknown
abinip             00000000013C1C7A  Unknown               Unknown  Unknown
abinip             00000000012825C3  Unknown               Unknown  Unknown
abinip             0000000001281F33  Unknown               Unknown  Unknown
abinip             000000000051E24D  Unknown               Unknown  Unknown
abinip             000000000041FD4C  Unknown               Unknown  Unknown
abinip             000000000040E2F5  Unknown               Unknown  Unknown
abinip             00000000004053F0  Unknown               Unknown  Unknown
abinip             000000000040030E  Unknown               Unknown  Unknown
abinip             000000000142D65E  Unknown               Unknown  Unknown
abinip             0000000000400229  Unknown               Unknown  Unknown
p4_11598:  p4_error: net_recv read:  probable EOF on socket: 1
p1_4097:  p4_error: net_recv read:  probable EOF on socket: 1
bm_list_11557: (125868.613281) wakeup_slave: unable to interrupt slave 0 pid 
11534

It looks like the connection between master and node is lost. But why does 
this happen regularly and _always_ after convergence is reached and the 
Inverse Hessians are written to the log-file?

Since the calculations are converged, I want to use the data. Is there a way 
to extract energy and str_relax.out from an incomplete file?

Thanks for your help.
Cheers,
Christian