[Molpro-user] Segmentation fault with FCI

[Peter Knowles]

This calculation will never run on your machine - it is too big. The  
size of the full CI is more than 4.1 * 10^9 and the program has to  
allocate two vectors of this length plus some other stuff, so it would  
not even run in 64 Gbyte. Even if you had the memory, it would take an  
extremely long time to run.

Peter

On 1 Oct 2008, at 13:27, Lorenzo Lodi wrote:

> I tried to run a frozen-core full-CI calculation on water in a Cs
> geometry using the aug-cc-pVDZ basis set.
> The machine used was a 3.4GHz Pentium IV PC with 3GB of ram; I used
> MOLPRO 2006.1 under Linux 32bit.
> The input is as follows:
>
> ***, Calculation FCI of water
> MEMORY, 250, M
> r1=1.81, r2=2.00, theta=104.5
> geometry={O;  H1,O,r1;  H2,O,r2,H1,theta}
> basis=aug-cc-pVDZ
> {RHF}
> FCI
>
> The computation stopped after 6.6h with this error:
>
> 0Hamiltonian diagonaliser entered at time      0.83
> Received signal 11 Segmentation violation
>
> I was monitoring the memory usage using top and at the moment of the
> crash it was of exactly 1GB. As I had given MOLPRO 2GB this doesn't
> immediately strike me as an error due to lack of memory.
>
> Searching the user forum archive with the keywords <FCI segmentation>
> finds two threads ("FCI calculation for H3" and "error in molpro")  
> which
> may describe similar (or identical) problems.
> I could not read these messages because the links given by the search
> engine do not work (they used to in the past), and furthermore it is
> apparently not possible (any more) to see the whole list of past
> messages (whatever happened to the archive web site?!)
>
> Lorenzo Lodi
>
> NB: in the FCI program the number of determinants is just a string of
> asterisks:
>  Strings:                       46411   44979       0       0       0
>      0       0       0
>                                 46411   44979       0       0       0
>      0       0       0
>  Determinants:               ****************       0       0       0
>      0       0       0
>
>  Load integrals                0.3 sec
>
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user



--
Prof. Peter J. Knowles
School of Chemistry, Cardiff University, Main Building, Park Place,  
Cardiff CF10 3AT, UK
Telephone: +44 29208 79182 Fax: +44 2920874030 Email KnowlesPJ at Cardiff.ac.uk
WWW http://www.cardiff.ac.uk/chemy/contactsandpeople/academicstaff/knowles.html


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.molpro.net/pipermail/molpro-user/attachments/20081001/78b962fb/attachment.html>