[molpro-user] "Expectation value and eigenvalue not identical"

[Dan Haxton]

This question has been asked in previous postings, in
2002, 2003, and 2005:

http://www.molpro.net/molpro-user/archive/all/msg01443.php
http://www.molpro.net/molpro-user/archive/all/msg00490.php
http://www.molpro.net/molpro-user/archive/all/msg00702.php

but it does not seem to have gotten a reply.

I'm doing a MRCI calculation with version 2006.1 that
requires high precision in the asymptotic region.  I
have found that I must set the "thrdlp" threshold
option in order not to exclude contracted configurations
and produce a result that is smooth as a function of
geometry.  E.g., (c2v geometry)


{ ci; maxiter,100,100; core,1,0,0,0; closed,1,0,0,0;
occ, 6,2,2,0; wf,9,4,1; state,1,3; ref,2; ref,1; ref,3;
thresh,pnorm=0.0, thrdlp=0.0; save,7510.2; }


setting thrdlp=0.0 like this leads to a number of contracted
configurations that does not change with geometry, and smooth
results for certain states.  (Low but nonzero values of thrdlp
give similar results.)  But, for other states, the CI gives
the following sort of error message,


  ITER. STATE  ROOT      NORM       CORR.ENERGY   TOTAL ENERGY   ENERGY
CHANGE       DEN1      VAR(S)    VAR(P)      TIME
     .
     .
   (snip)
     .
     .
 ?EXPECTATION VALUE AND EIGENVALUE NOT IDENTICAL STATE=1  EIG= 
-55.23478091  EXPECT.=   -55.23478065
   97     1     1     1.04936859    -0.15208975   -55.23478091    
.00000000    -0.00000005  0.52D-08  0.11D-07   309.48

 ?EXPECTATION VALUE AND EIGENVALUE NOT IDENTICAL STATE=1  EIG=  
55.23478091  EXPECT.=   -55.23478065
   98     1     1     1.04936859    -0.15208975   -55.23478091   
0.00000000    -0.00000005  0.52D-08  0.11D-07   312.70

 ?EXPECTATION VALUE AND EIGENVALUE NOT IDENTICAL STATE=1  EIG=  
55.23478091  EXPECT.=   -55.23478065
   99     1     1     1.04936859    -0.15208975   -55.23478091   
0.00000000    -0.00000005  0.52D-08  0.11D-07   315.94

 ?EXPECTATION VALUE AND EIGENVALUE NOT IDENTICAL STATE=1  EIG=  
55.23478091  EXPECT.=   -55.23478065
  100     1     1     1.04936859    -0.15208975   -55.23478091   
0.00000000    -0.00000005  0.52D-08  0.11D-07   319.18


and then exits with no convergence.  The calculation has clearly
converged, by which I mean, the numbers have stopped changing
with the iterations.  But, it appears that the columns VAR(S)
and VAR(P) are not correctly computed...?  In the above output,
the convergence thresholds are nearly reached, but at other
geometries, the VAR(S) and VAR(P) columns reach a steady state
long before reaching the convergence criterion THRVAR.  Because
the convergence criterion seems to be that only ONE of the
thresholds THRDEN or THRVAR must be satisfied -- as opposed to
both -- I may not simply set THRVAR to a large value.

How may I fix this error?

Thanks for any replies

Dan Haxton
Postdoc, JILA, CU-Boulder, USA