Strange "drpa" memory issue

Dear Developers,

I am running inputs as described below.
If the reference determinant is konstructed from Kohn-Sham orbitals (scftype=rks, dft=(arbitary functional)) and I write dens=1, the program is terminated as:

Executing drpa...
 Three-index integral transformation
 Allocation of 878.9 Gbytes of memory...
 Fatal error in exec drpa.
 Program will stop.

However, there is certainly enough memory, as the same calculations run perfectly if scftype=rhf (and dens keyword is omitted). There are also no problems if dens=0.
The memory issue ONLY appears if dens=1 AND scftype=rks.
Decreasing the memory limit does not help.
What could be the reason?

Best,
Zsolt

Sample input file:

# energy

scftype=rks
dft=hse06
dens=1
scfmaxit=500
scfiguess=sad
calc=ccs  # or any other CC level
core=frozen
ccmaxit=300
mem=900000MB
basis=cc-pvdz
charge=0
mult=1
unit=angs
geom=xyz
2

B        0.000000    0.000000   -0.748417
N        0.000000    0.000000    0.534583

 

Dear Zsolt,
Unfortunately, density calculations using CC methods with KS references are currently not possible.

I am surprised to hear that because I have already successfully run such calculations.
For example, the input below works fine (as you can see, the only difference is ccprog=mrcc, the removal of density fitting and the correlation of core electrons.)
So it is certainly not true (according to my experience) that these calculations are not possible.
The question is, why doesn't it work with frozen core and density fitting?

# energy

scftype=rks
dft=hse06
dens=1
ccprog=mrcc
dfbasis_scf=none
dfbasis_cor=none
scfiguess=mo
scfmaxit=500
scfdamp=on
scflshift=off
calc=ccsd
core=corr
ccmaxit=300
mem=100000MB
basis=cc-pvdz
charge=0
mult=1
unit=angs
geom=xyz
2

B        0.000000    0.000000   -0.748417
N        0.000000    0.000000    0.534500




 

Dear Zsolt,

As wrote by Mihály, CC density calculations are not implemented properly with KS reference (the orbital relaxation term is only available with HF reference). So from your previous ccprog=mrcc calculation with KS reference, only the unrelaxed CC densities can be useful.

Since CC densities are not implemented with ccprog=ccsd, this causes the problem in your original post (thus it has not much to do with frozen-core or DF).

For now, the recommendation for CC densities is ccprog=mrcc and HF reference.

Best wishes,
Peter