Memory issue in CP-QCISD when obtaining BCC natural orbitals

[Rob Lavroff]

Dear Molpro team,

I am looking to obtain coupled cluster natural orbitals and their occupations to help me in picking a MCSCF active space. Usual CCSD and DCSD will not converge due to the multiconfigurational nature of the system, but BCCD and BDCD will converge for me. The trouble is that Brueckner CC natural orbitals are obtained using CP-QCISD rather than CP-CCSD, and there appears to be some issue here where CP-QCISD runs out of memory no matter how much is given to it.

To test this, I set up a calculation on water with an extremely large amount of memory (80 gigawords, the most a single core on the cluster I'm using will allow). My input is copied below. Unfortunately with this much memory, I still get the error below, which is the same as what I get with a "reasonable" amount of memory for a water molecule. The workaround I'm using is to do BDCD/BCCD, save the orbitals, do a 0-iteration HF, then do CCSD/DCSD and get the natural orbitals from there (CP-CC does not have this memory issue). But it would be nice to confirm that these NO occupations are similar to what the Brueckner calculation would give directly. Could you please help me investigate this issue?

The input

***,h2o
memory,80G

basis=vdz
symmetry,nosym

geometry={
O,, 0.000000,    0.000000,    0.465140
H,, 0.000000,    1.415022,    1.558516
H,, 0.000000,   -1.415022,    1.558516
}
hf
{bdcd
natorb,record=2140,print=30}

The error

" !BDCD total energy                   -76.241635552693

  SCS-BDCD total energy               -76.257081592833
 (using cOS= 1.050 cSS= 1.150)

 Approximate expectation values using Brueckner orbitals

 !BDCD(ref) STATE 1.1 Dipole moment     0.00000000     0.00000000     0.76827218
 Dipole moment /Debye                   0.00000000     0.00000000     1.95275310

 Timing summary (sec):

 STEP                 CPU(USER)    SYS     CPU(TOT)    WALL
 Transformation          0.01      0.01      0.02      0.07
 CCSD iterations        20.29      0.00     20.29      0.01

 Program statistics:

 Available memory in ccsd:             79999999931
 Min. memory needed in ccsd:                 16978
 Max. memory used in ccsd:                   18360
 Max. memory used in cckext:                 47994 ( 8 integral passes)

 Starting CP-QCISD iterations.     Author: G. Rauhut (1999)


 NOT ENOUGH MEMORY IN QCIDIP. ONLY    6 OUT OF   10 K-OPERATORS FIT

 GLOBAL ERROR fehler on processor   0"

[Rob Lavroff]

Just an update that NOs for Brueckner and ODCD calculations are not available (because analytic gradients and the corresponding effective DMs are not available for these methods) and there is now error handling to inform the user of this in the next version