MRCC crashes in MP2(full) calculation with external RI basis sets

Dear all,

I have a problem with an MP2(full)/cc-pCVTZ calculations that crashes
with an EOF error for DFINV (details in appendix A). See appendix B
for the last lines of regular output that I get.

To fill you in: I wish to do an MP2 calculation with all electrons
correlated, using the cc-pCVTZ basis set as well as RI/DF treatment
for the HF and MP2 parts. Since no density fitting basis sets are
available for either one or the other, I use a large basis (cc-pV5Z-RIFIT
for H, cc-pwCV5Z-RIFIT for C) downloaded from the Basis Set Exchange
for both purposes.

The above problem happens reproducibly with statically linked binaries
downloaded from the MRCC web page, but not with all of them:

version 2016_07_15: runs fine
version 2017_09_25: runs fine
version 2019_02_09: produces above error
version 2020_02_22: produces above error

Those that run fine, give results (-116.529295) very close to the
canonical (non-DF) result with Molpro (-116.529296), as expected.

Any ideas?

Thanks,
best regards,

Dirk Bakowies

---

Appendix A
forrtl: severe (24): end-of-file during read, unit 20, file <FULL_PATH_OFSCRATCH_DIRECTORY>/DFINV

and a number of messages like

Image PC Routine Line Source
drpa 000000000534A2BB Unknown Unknown Unknown
drpa 000000000537B24E Unknown Unknown Unknown
drpa 00000000032600A9 Unknown Unknown Unknown

---

---

Appendix B
Executing drpa...

Three-index integral transformation and DF-MP2 calculation

Allocation of 1600.0 Mbytes of memory...

Number of integral batches: 1
Occupied indices per batch: 11

Transforming integrals of batch 1...

---

Sorry, didn't get the tar file in. Trying again...
Dirk

Dear Dirk,
Please download the patches for mrcc, especially the patch for integ, which fixes this problem.

Dear Mihaly,
thanks. I just tried it, replacing, successively, integ, dmrcc, mrcc of the precompiled versions of your 2020 distribution with the binaries posted in your patches download area. Unfortunately no success. The same error still appears. If instead I used the precompiled version of the 2017 distribution, things work fine, as mentioned initially already. I tried it again and verified that.
Best,
Dirk

Dear Dirk,
Sorry, please download and try integ now.

Dear Mihaly,

yep you found the bug and fixed it. Works well now. Thanks a lot!!! Let's see if we get the other problem (geo opt) fixed too...
Best wishes,
Dirk