[molpro-user] parallel problem using 2010 version and NON-ZERO P-SPACE GRADIENT

[lydia]

Hi,
For my mashine, it has 36GB memory for one node, and one node has 12
process. It was built by open-mpi 1.4.1 with intel compiler.
My memory set of  200*8*10=16 GB does not exceed the limitation.
My jobs was hanging on the first Multi module with zero interation as follows:

"1PROGRAM * MULTI (Direct Multiconfiguration SCF)       Authors: P.J.
Knowles, H.-J. Werner (1984)     S.T. Elbert (1988)


 Number of closed-shell orbitals: 31 ( 31 )
 Number of active  orbitals:      12 ( 12 )
 Number of external orbitals:    298 (298 )

 State symmetry 1

 Number of electrons:    11    Spin symmetry=Doublet   Space symmetry=1
 Number of states:        2
 Number of CSFs:     339768   (731808 determinants, 731808 intermediate states)

 Molecular orbitals read from record     2100.2  Type=RHF/CANONICAL (state 1.1)

 Wavefunction dump at record             2140.2

 Convergence thresholds  0.10E-01 (gradient)  0.15E-05 (energy)
0.10E-02 (step length)

 Weight factors for state symmetry  1:    0.50000   0.50000

 Number of orbital rotations:   13186     ( 372 Core/Active 9238
Core/Virtual   0 Active/Active 3576 Active/Virtual)
 Total number of variables:   1476802


 ITER. MIC  NCI  NEG     ENERGY(VAR)     ENERGY(PROJ)   ENERGY CHANGE
   GRAD(0)  GRAD(ORB)   GRAD(CI)     STEP       TIME
"
By the way, the scalability for the parallel jobs is not good as I
expected. For example, 8 or 10 process job is not twice faster than 4
processors. Do you have some ideas about how to improve the
efficiency?

All the best,
yang


On Tue, Oct 12, 2010 at 12:52 PM, Manhui Wang <wangm9 at cardiff.ac.uk> wrote:
> Hello,
>
> Regarding question(1), please be aware that you request at least about
> 200*8*10=16GB on that node, while you request less memory when you run
> the job with 9 or less processes. For more details about the
> memory in Molpro, please refer to previous discussions, eg:
> http://www.molpro.net/pipermail/molpro-user/2010-April/003723.html
> http://www.molpro.net/pipermail/molpro-user/2010-April/003718.html
>
>
> Best wishes,
> Manhui
>
>
> lydia wrote:
>> Hi all,
>>  I have two questions to ask for your help.
>> (1) I installed the newest version 2010 molpro package to do some
>> parallel calcualtions. In the input file, the memory was set to
>> "memory,200,m", no symmetry,341 basis function. If more than 10 ways
>> (i.e.,processors) are used for the parallel runing, the job is hanging
>> there without any error message in the output. If I tried with less
>> than 10 ways, such as 2,4,6,8,9 ways, it always work. I am confused
>> why the high-way parallel jobs do not work? Does this related to the
>> memory set or  other parameters?
>>
>> (2) In my output file, it shows " NON-ZERO P-SPACE GRADIENT. LARGEST
>> VALUE:    0.36D+01
>>  Probably the p-space is not closed under symmetry operations. This
>> can happen if the initial orbitals are not symmetry adapted or if there is
>> root-flipping between states of different symmetry. Try different
>> p-space threshold, or select
>> p-space configurations manually.
>>  The check can be bypassed by setting thresh,pspace=thr to a larger value.
>>  Non-zero p-space gradient elements smaller than thr are ignored."
>>  How to deal with this problem?
>>  I tried to incease the pspace to 5, (the default pspace is 0.4), it
>> still doesnot work.
>> Waiting for your reply. Thanks.
>>
>> All the best,
>> yang
>> _______________________________________________
>> Molpro-user mailing list
>> Molpro-user at molpro.net
>> http://www.molpro.net/mailman/listinfo/molpro-user
>
> --
> -----------
> Manhui  Wang
> School of Chemistry, Cardiff University,
> Main Building, Park Place,
> Cardiff CF10 3AT, UK
> Telephone: +44 (0)29208 76637
>