[molpro-user] surprisingly slow mp2 gradient calculations with molpro

[Toon Verstraelen]

Hi,

Related to the resource allocation, we wonder whether the memory 
allocation is working properly. The memory that has to be specified with 
the -m parameter appears to be the memory per cpu, e.g. when we want 
molpro to use 12GB of ram in total for a job that runs on 8 cpu's, we 
start the job with the following command:

molpro -n8 -m201326592 molpro.in

Where 201326592 words correspond to 1.5GB of memory. Is this the correct 
behavior of the -m option? I could not immediately find this in the 
manual, but I might have overlooked it.

best regards,

Toon


David T Moore wrote:
> Hi Toon,
> 
> I wonder if this could be a problem with resource allocation.  An MP2
> gradient calculation requires much more memory and disk than a single
> point energy calculation.  If your default memory and disk settings are
> high, then this is likely not the problem, but you can try explicitly
> setting them to maximum values for the system.
> 
> It might also be interesting to compare the results from "that other
> program" with comparable resources.
> 
> Hope this helps,
> 
> Dave Moore
> 
> On Wed, March 14, 2007 12:55 am, Toon Verstraelen wrote:
>> Dear Molpro users,
>>
>> We are currently testing a compilation of molpro on our ibm p575
>> machines (AIX 5.3). Most calculations run fine but when we try to run an
>> mp2 gradient calculation, molpro keeps running without further output
>> after the mp2 energy is reported in the output file. This is the input
>> file. It is just a test job, molpro.in
>>
>> punch,molpro.pun,rewind
>> geomtyp=xyz
>> geometry={
>> include,geometry.xyz
>> }
>> basis,STO-3G
>> SET,symmetry=0
>> hf
>> basis,6-311++G**
>> SET,symmetry=0
>> hf
>> mp2
>> force
>> pop
>>
>>
>> this is geometry.xyz
>>
>>     30
>>
>>   H     3.559515    -3.602968     0.480477
>>   H     0.000000     0.000000     0.000000
>>   H     2.231618    -0.126968    -2.973653
>>   H     2.164509    -5.199926    -2.981076
>>   H    -1.525048    -1.769387    -2.541999
>>   H    -0.283606    -4.942521    -0.158983
>>   O     3.211113    -4.202985    -0.172099
>>   O     1.622107    -5.196719    -2.205522
>>   O     0.578705    -4.569284     0.135543
>>   O     1.634017    -2.537234    -1.790187
>>   O    -0.170479    -0.889723    -0.455256
>>   O     1.332520    -0.013415    -2.763644
>>   O    -0.707978    -1.494727    -2.920517
>> Si     1.754767    -3.998427    -0.991862
>> Si     0.592276    -1.284170    -1.911350
>>   H     3.097785    -1.009335     6.285552
>>   H     0.000000     0.000000     1.682236
>>   H     2.076392     0.668372     5.305142
>>   H     3.170994    -3.996060     3.617948
>>   H     3.143732     0.626618     2.427053
>>   H     1.640463    -4.632139     6.450378
>>   O     3.293025    -1.899958     6.099096
>>   O     2.990040    -4.078557     4.551136
>>   O     1.253281    -3.746910     6.474897
>>   O     1.340477    -2.068433     4.077228
>>   O    -0.065984    -0.094797     2.623509
>>   O     2.188434     0.431384     4.374149
>>   O     2.529447    -0.016530     2.124618
>> Si     2.135106    -2.835983     5.351803
>> Si     1.429452    -0.586945     3.296122
>>
>>
>> With another very famous quantum chemistry package this job runs for 70
>> minutes while molpro was still running after 48 hours on the same
>> machine, although the HF and mp2 energy part in molpro is fast. We had
>> the chance the run the same job with molpro on yet a larger cluster,
>> where it took 30 hours to complete the molpro job, which is extremely
>> long compared to the 70 minutes. We believe that this must be a bug in
>> molpro or an unpractical default setting that causes the long run time.
>> Are there other users that experience similar problems? Is there a
>> work-around?
>>
>>
>>     15
>>
>>   H    -0.059728    -0.411120    -0.618055
>>   H     0.514182    -3.191541     3.083949
>>   H    -3.101425    -2.004277     4.342903
>>   H     1.902183    -0.100208     2.828450
>>   H    -2.666575    -4.089570     1.406482
>>   H    -0.884653     1.977845     1.925963
>>   O     0.215520     0.237492     0.010073
>>   O     1.373611    -0.721015     2.351302
>>   O    -0.387441     1.342727     2.416665
>>   O    -1.240360    -1.242104     1.727592
>>   O    -0.323932    -3.579847     2.881498
>>   O    -2.167500    -1.994261     4.202866
>>   O    -2.905062    -3.375850     1.976998
>> Si    -0.036196    -0.116254     1.642964
>> Si    -1.630889    -2.527372     2.695378
>>
>>
>> I have also tried direct-mp2 and different basis sets (6-31++g** and
>> cc-pVDZ), but the symptoms are exactly the same.
>>
>>
>> best regards,
>>
>> Toon Verstraelen
>>