wrong ovos-ccsdt gradient

Dear mrcc users/developers,

When using  to speed up ccsdt calculations, the final dipole & molecular gradient seem to be totally wrong. Take H2 for example, these are the results:

[with ovos]
 Molecular gradient [au]:
    1  H      0.0449345831     0.0004366843     0.2871843142
    2  H     -0.0449345831     0.0004366843    -0.2871843142
 Dipole moment [Debye]:        0.02476652
   x=     -0.00024495  y=     -0.02438433  z=      0.00432724

while the reasonable ones are [no use of ovirt]:
 Molecular gradient [au]:
    1  H      0.0000000000     0.0000000000     0.0000531176
    2  H      0.0000000000     0.0000000000    -0.0000531176
 Dipole moment [Debye]:        0.00000000
   x=      0.00000000  y=      0.00000000  z=     -0.00000000

The computed energies by ovos scheme are very close to those without truncation of virtual space though, and the equilibrium H2 distance is verified by scanning the distances.

I suspect that the code fails to transform from optimized virt space to MO space. This issue exists for both 2019 & 2020 version of binary mrcc. Any idea?

Best,
Bing

PS: here is my input
=======================
mem=120GB
mpitasks=1
basis=cc-pCV6Z
#scfiguess=mo #restart
scftype=rhf
calc=ccsdt
ccprog=mrcc
ovirt=ovos
ovosnorb=50
dens=2
charge=0
mult=1
geom=xyz
2
Exp.
H 0. 0. 0.
H 0. 0. 0.7414
 

Dear Bing,
Unfortunately, CC gradients are not implemented with the OVOS approximation.

Thanks for your prompt reply!

Maybe it helps to make mrcc display some warning message or simply terminate when both ovos & dens=2 are given as input...

Best,
Bing

Can one trust the resulting 2RDM? It seems that energy from 2RDM (see below) in the case of ovirt=ovos is correct. But the associated dipole moment is totally wrong. I'm puzzled! Any idea?

ovirt=ovos; ovosnorb=50
Total CCSDT energy [au]: -1.174253395035
Total energy from RDM [au]: -1.174253395062
Dipole moment [Debye]: 0.02476652

ovirt=off
Total CCSDT energy [au]: -1.174345651956
Total energy from RDM [au]: -1.174345652024
Dipole moment [Debye]: 0.00000000

Best,
Bing

Yes, you can. The 1RDM and 2RDM are the correct unrelaxed RDMs. The properties are wrong.

Good to know!

BTW, Is there any keyword control for computing unrelaxed rdm2 only, without carrying on to property calc in the case of dens=2?

best,
Bing