[molpro-user] Changing the convergence threshold for FCI calculation

[Bastiaan J. Braams]

I would like advice about how to modify the convergence threshold for
an FCI calculation.  I am working on the CH2 molecule, and when using
the vtz or avtz basis all calculations fail.  When using the vdz basis
I obtain answers, but I can see that the calculation iterates many
times beyond a result that would satisfy me.

Here is a sample input.

***,SP ENERGY FOR H2C1 singlet
memory,84,m
punch,mpsp.pun,new
geomtyp=xyz
geometry
z;mass
 3
  null
 H      0.32378690   1.07256700   0.23934060
 H     -1.09156370   0.02483350  -0.31998320
 C      0.06398140   0.01112250   0.00672000
end
basis=vdz
rhf;wf,charge=0,sym=1,spin=0
fci
---

A fragment of output follows (terminated by -----------------------)

[...]

1PROGRAM * FCI (Full CI)     Author: P.J. Knowles, 1984

[...]

 Symmetry:                   1
 Trial vector:                  0.00
 Result vector:                 0.00
 Maximum iterations:           35
 Convergence threshold:         0.0000100
 RHS vector:                    0.00
 Hamiltonian shift:             0.0000000
 Output threshold:              0.0500000
 Options:                        0
 Number of roots:             1

[...]

0Hamiltonian diagonaliser entered at time      0.28
0Initial configuration generated:
       1      1.0000000    -38.8759613

 It  Tr    CPU  Convergence   Energy       Pop     dE(next)

  1   1   65.6  1.00000000   -38.87596134
+                                         0.00 -0.15591787 -0.15590860
  2   1  126.4  0.26533724   -39.00592383
+                                         0.03 -0.01455227 -0.01428755
  3   1  198.4  0.07323397   -39.01771922
+                                         0.07 -0.00216144 -0.00207235
  4   1  282.3  0.04111007   -39.01957171
+                                         0.10 -0.00041959 -0.00038686
  5   1  375.7  0.01436230   -39.01998285
+                                         0.14 -0.00010137 -0.00009012
  6   1  480.7  0.00973387   -39.02008870
+                                         0.18 -0.00003209 -0.00002752
  7   1  597.4  0.00433518   -39.02012031
+                                         0.23 -0.00001094 -0.00000938
  8   1  725.6  0.00232195   -39.02013078
+                                         0.26 -0.00000321 -0.00000274
  9   1  862.9  0.00113916   -39.02013371
+                                         0.27 -0.00000084 -0.00000072
 10   1 1003.4  0.00049931   -39.02013447
+                                         0.27 -0.00000021 -0.00000018
 11   1 1142.4  0.00027325   -39.02013466
+                                         0.27 -0.00000005 -0.00000004
 12   1 1282.3  0.00012684   -39.02013470
+                                         0.27 -0.00000001 -0.00000001
 13   1 1421.8  0.00006938   -39.02013471
+                                         0.27  0.00000000  0.00000000
 14   1 1562.8  0.00003321   -39.02013472
+                                         0.28  0.00000000  0.00000000
 15   1 1705.6  0.00001761   -39.02013472
+                                         0.28  0.00000000  0.00000000
 16   1 1847.5  0.00000788   -39.02013472

-----------------------

As I said, this was on a vdz basis, and on a vtz basis it never
reaches its tolerance limits.  I would be happy to see the calculation
stop when the energy change is below 1.0e-6, or even 1.0e-5.

Presumably the relevant parameter is the "convergence threshold",
which is 1.0e-5 according to the output.  So how do I change this
parameter?  I tried to insert the command

gthresh,civec=1.d-2

before my geometry specification, but this had no effect.

Bas Braams
--
Bastiaan J. Braams
braams at mathcs.emory.edu
Emory University, Atlanta, GA