[molpro-user] CASSCF Converges on Athlon, Fails on Intel P4

Li, Jun Jun.Li at pnl.gov
Thu May 13 01:08:59 BST 2004


Joseph,

The slight numerical difference on the Athlon and P4 might be causing
what you saw.  Even though it is not generally the case, but your
molecule might get converged with more MCSCF iterations.  In fact, this
kind of numerical problems might happen even on the same machine under
some circumstances.  

I tried your molecule with 11 iterations and it got converged to 1e-6
(see below).

Jun  Li



 ITER. MIC  NCI  NEG     ENERGY(VAR)     ENERGY(PROJ)   ENERGY CHANGE
GRAD(0)  GRAD(ORB)   GRAD(CI)     STEP       TIME
  
   1   90   32    0    -547.03057272    -547.10702891   -0.07645618
0.19252296 0.00011045 0.22630187  0.12D+01      0.94
   2   34   42    0    -547.10516036    -547.10722947   -0.00206911
0.05331612 0.00000060 0.00116733  0.20D+00      1.90
   3   44   42    0    -547.10724012    -547.10724010    0.00000003
0.00051998 0.00000000 0.00031419  0.59D-02      2.88
   4   44   42    0    -547.10724011    -547.10724051   -0.00000040
0.00000941 0.00000001 0.00144092  0.19D-02      3.85
   5   51   42    0    -547.10724071    -547.10723926    0.00000146
0.00009339 0.00000000 0.00370414  0.30D-02      4.83
   6   50   42    0    -547.10724022    -547.10727147   -0.00003125
0.00032489 0.00000001 0.02154316  0.15D-01      5.82
   7   53   42    0    -547.10728673    -547.10727240    0.00001433
0.00098588 0.00000001 0.02303741  0.47D-02      6.78
   8   49   42    0    -547.10729852    -547.10771889   -0.00042038
0.00066603 0.00000010 0.03035252  0.76D-01      7.78
   9   41   42    0    -547.10782141    -547.10797400   -0.00015260
0.00669151 0.00000001 0.00296160  0.26D-01      8.71
  10   37   39    0    -547.10797529    -547.10797602   -0.00000074
0.00027396 0.00000004 0.00019212  0.27D-02      9.66
  11   31   21    0    -547.10797603    -547.10797603   -0.00000000
0.00000435 0.00000021 0.00002356  0.19D-03     10.35
   
 ** WVFN ****  CONVERGENCE REACHED, FINAL GRADIENT:  0.61D-06




-----Original Message-----
From: Joseph Lane [mailto:jlane at alkali.otago.ac.nz] 
Sent: Wednesday, May 12, 2004 4:04 PM
To: molpro-user at molpro.net
Subject: [molpro-user] CASSCF Converges on Athlon, Fails on Intel P4


Good morning,

I am unable to get the following input file to converge on a P4 machine,

however I have no problems running it using athlon architecture. We are 
running 2002.6 on both machines.

  ***,SO2
  mem, 40, M
  gprint,orbitals=13,civector
  r1=1.431 ang, theta=119.3
  geometry={Z;
    S;
            O1,S,r1;
    O2,S,r1,O1,theta}
  basis=AVDZ

  OCC=[15,4];
  CLOSED=[11,2];

  hf

  casscf;
  WF, 32,1,0; State,2;
  WF, 32,2,2;
  WF, 32,2,0; State,2;

The first output is for the P4 machine, the second for the athlon


<    1   90   32    0    -547.03057257    -547.10726685   -0.07669429

0.19252516 0.00001838 0.19861877  0.12D+01      2.48
<    2   34   42    0    -547.10524588    -547.10723025   -0.00198437

0.05385009 0.00000024 0.00073630  0.19D+00      4.75
<    3   43   42    0    -547.10723951    -547.10725256   -0.00001305

0.00044552 0.00000004 0.01231582  0.12D-01      7.19
<    4   44   42    0    -547.10725850    -547.10723638    0.00002212

0.00038639 0.00000005 0.01342121  0.27D-01      9.66
<
<  *** NO CONVERGENCE ***
---

>    1  100   32    0    -547.03057257    -547.10636764   -0.07579507

> 0.19252516 0.00020290 0.31723526  0.12D+01      3.20
>    2   42   42    0    -547.10493860    -547.10722842   -0.00228983

> 0.05227518 0.00000001 0.00148456  0.21D+00      5.82
>    3   43   42    0    -547.10724233    -547.10724190    0.00000044

> 0.00071918 0.00000000 0.00336102  0.94D-02      8.77
>    4   43   42    0    -547.10724284    -547.10724236    0.00000048

> 0.00009874 0.00000001 0.00447644  0.87D-02     11.57
>    5   42   42    0    -547.10724353    -547.10724009    0.00000344

> 0.00012048 0.00000000 0.00026444  0.38D-02     14.41
>    6   44   24    0    -547.10724009    -547.10724010   -0.00000001

> 0.00000605 0.00000000 0.00019460  0.21D-03     16.67

>  ** WVFN ****  CONVERGENCE REACHED, FINAL GRADIENT:  0.13D-04

Any suggestions would be most appreciated

Regards

Joseph Lane




More information about the Molpro-user mailing list