[molpro-user] VCI - insufficient memory available

Guntram Rauhut rauhut at theochem.uni-stuttgart.de
Mon Sep 4 10:24:21 CEST 2017


Dear Dennis,

the problem you encountered most likely arises from strongly correlated 
CH-stretching vibrations. According to your output the vibrational 
ground-state calculation succeeded without problems and the next state 
to be computed is a CH-stretching mode. As you have 4 hydrogen atoms in 
your systems, these will be strongly coupled, leading to very low 
leading VCI coefficients. As a consequence very many configurations need 
to be selected to describe this state properly. For example, if the 
selection process has picked more than 50.000 configurations (which 
would be just 5% of the original space) the memory demands for the VCI 
matrix become huge. The newest (still unreleased :(  ) version of Molpro 
avoids this bottleneck by exploiting sparsity techniques. The fact that 
the message provided by Molpro states, that there a just 25 MByte left 
is a result of the iterative nature of the algorithm, which steadily 
increases the memory demands. It does not request the memory for the VCI 
matrix in one piece. Essentially everything looks quite alright to me.

What can you do? Make up your mind, if you really need the CH-stretching 
vibrations. Very often, the fingerprint regions is more important. If 
you do not need them, you can exclude them by using the VIBSTATE program 
(do not forget the USERMODE option). If you really need them, I suggest 
to switch to a calculation on just one core and to provide about 2000 MW 
to the calculation. Most likely this will do it, but definitely will 
take some time. You could also try to reduce the correlation space in 
the VCI calculation by reducing CIMAX and LEVEX keywords. This of course 
will lead to less reliable results, but you will get a "feeling" about 
the the problem. You may start with very small VCI spaces and then you 
steadily increase the correlation space in order to understand what's 
going on.

I believe that this is not a technical problem, but rather a molecule 
specific problem which arises from the strong couplings of the 
CH-stretching modes.
Best wishes,

    Guntram



On 08/29/17 11:36, Dinu, Dennis Florian wrote:
> Dear Molpro-Community,
>
> first of all the SURF/VSCF/VCI programs in molpro perform really nice, 
> the majority of the anharmonic vibrational calculations were 
> unproblematic and all in all the experience was very user-friendly. 
> But now that I switched to a bigger System (C2H4O3), I run into a 
> memory problem, which I cannot solve from the rather brief error output.
>
> Setup: precompiled Molpro 2015.1 / Intel Xeon CPU E5-2643 v3 / 12 
> cores / 256 GB RAM
> The error message:
> --------------------------------------------------------------------------------------------------
> PROGRAM * VCI (Vibrational configuration interaction) Authors: G. 
> Rauhut 2004, M. Neff 2009, D. Oschetzki 2013
>
> Reading potential from record        5600.2
> Reading wave function from record    5750.2
>
> Results of VCI calculation:
>
> Number of configurations:    A`      1259812   ( S:       84 D:     
> 3528  T:    91210  Q:  1164989 )
>                                              A``     1023169 ( 
> S:       21  D:     1722  T:    61740  Q:   959686 )
>
> Fundamentals
>
> Mode         E(abs)      VCISDTQ      IR Intens     Leading 
> term        Type    Dim.      CPU tot.       CPU #
>     0        14548.94 0.00                                   0.9507 
> J         5721       11616.6     11616.6
> insufficient memory available - require 26270376  have
>              25383256
> the request was for real words
> --------------------------------------------------------------------------------------------------
>
> The calculation was started with 512 Megawords per CPU, so the error 
> message, saying that there are just about 25 Megawords available is 
> rather strange for me. I already tried much less (100 Megawords), and 
> much more (1024 Megawords) memory allocation per CPU, but with the 
> same error message.
>
> There are some posts in the mailing-lists concerning this, most of 
> then within electronic structure theory. Here some examples:
> CASSCF calculation: 
> http://www.molpro.net/pipermail/molpro-user/2005-June/001380.html
> CCSD(T) calculation: 
> http://www.molpro.net/pipermail/molpro-user/2009-June/003122.html
> UCCSD(T)-F12 calculation: 
> http://www.molpro.net/pipermail/molpro-user/2011-April/004320.html
> and much more..
>
> The main answers that i found:
> 1) Increase the memory until the calculation works.
> Question: For the CC programs, molpro suggests the amount of memory it 
> needs. Is there a similar output for the VCI progams?
>
> 2) Adjust size of the shmmax kernel parameter in /proc/sys/kernel/shmmax
> Question: For the MPI-2 version of molpro, this parameter should not 
> be of interest.. is that right?
>
> Are there any other suggetions and answers to that problem?
>
> PS: On our large-memory nodes, we have no SWAP. Could that be a 
> problem for the molpro system?
>
> greets
>
> Dennis Dinu
> Institute of Theoretical Chemistry
> University of Innsbruck
>
>
>
>
>
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user

-- 
********************************************************************************

  Apl. Prof. Dr. Guntram Rauhut
  Institut f. Theoretische Chemie
  Universitaet Stuttgart
  Pfaffenwaldring 55
  D-70569 Stuttgart
  Germany

  Tel. :   +49/(0)711/685-64405
  FAX :    +49/(0)711/685-64442
  E-Mail : rauhut at theochem.uni-stuttgart.de
  HTTP :   www.uni-stuttgart.de/theochem/rauhut/

********************************************************************************

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.molpro.net/pipermail/molpro-user/attachments/20170904/a6ae2587/attachment.html>


More information about the Molpro-user mailing list