Local Correlation from modified localized HF orbitals

Dear developers,
This is a question I asked a long time ago in here,
mrcc.hu/index.php/forum/running-mrcc/386...ocalized-hf-orbitals
Now I started to work in this project.  What I have understood that I need the canonical HF orbitals from MRCC and localize them before running local CCSD. I have read the manual on the scfiguess, orblocguess keywords can be useful. Thus the first step is to obtain the MOCOEF.CAN file. In order to generate the MOCOEF.CAN, VARS and FOCK, I tried this calculation,#
mem=16GB
basis=cc-pvdz
dfbasis_scf=None
dfbasis_cor=None
calc=lccsd
uncontract=on
gauss=cart
verbosity=3
symm=0
core=0
ccprog=mrcc
localcc=2015
###lcorthr=vvtight


unit=angs
geom=xyz
4
opt b3lyp/avdz
H  -1.674794   -0.378923    0.001049
H  -1.673913    0.381594    0.000604
H   1.294239   -0.012657   -0.008105
H   2.054468    0.009987    0.006453

I am using MRCC version 25.1.1 and now showing last few lines of the output file,
 Calculation of two-electron integrals...
 Memory requirement   0.000 Gb/thread,   0.007 Mb/thread,   0.000 Gb total
   3% done.
  18% done.
  32% done.
  64% done.
 100% done.
 CPU time [min]:     0.009                   Wall time [min]:     0.008

 ************************ 2026-02-16 14:12:22 *************************
 Executing drpa...

 Three-index integral transformation
 Allocation of 16.0 Gbytes of memory...
 
 Corrupted file VARS!
 Variable dfipra_cor is missing!
 Please run integ first!
 
 Fatal error in drpa.
 Program will stop.
 
 ************************ 2026-02-16 14:12:23 *************************
                   Error at the termination of mrcc.
 **********************************************************************

I do not know what caused this error. However, there are MOCOE, MOCOEF.CAN, MOCOEF.LOC, FOCK, VARS files generated but all of them are binary. How do I extract the information from them? Can I also obtain the overlap matrix and the HF 1-particle density matrix?

Initially, I am only trying to calculate the CCSD energy obtained by my localization method. I will compare this with canonical CCSD energy. 

Please let me know if you require any further clarification/information from my side.

Thanks and kind regards
Prasanta

Dear Prasanta,

If you want to use an external localization method that is not implemented in MRCC, you can try the following:

1) you run a calc=CCSD ccprog=ccsd calculation
(so these are not good in your attempt: calc=LCCSD, localcc=2015, you do not seem to want local approximations)

2) after this calculation, you need to keep all files and rewrite MOCOEF with your orbitals.

You may try MOKIT to make MOCOEF file format:
mrcc.hu/index.php/forum/running-mrcc/108...m-other-program#1550
Or you will find the MOCOEF format in the wrtmo subroutine of combin.f

3) After MOCOEF is properly rewritten, you can do a second calculation skipping the orbital computation, doing only the MO integral calculation and CCSD. To do that you, you execute ovirt and then ccsd subprograms of the MRCC package in the folder of step 1).

The overlap matrix and the HF 1-particle density matrix are in the OEINT and SCFDENSITIES files.

Best regards,
Peter

Thank you very much Prof. Nagy for your reply. With the help of MOKIT, I have generated the MOCOEF file. 
Could you please correct me if I am wrong to understand your suggestions.
First, I have generated all the files with this MINP file,

#
mem=16GB
basis=cc-pvdz
dfbasis_scf=None
dfbasis_cor=None
calc=ccsd
uncontract=on
gauss=cart
verbosity=3
symm=0
core=0
ccprog=ccsd
###localcc=2015
###lcorthr=vvtight
unit=angs
geom=xyz
4
opt b3lyp/avdz
H  -1.674794   -0.378923    0.001049
H  -1.673913    0.381594    0.000604
H   1.294239   -0.012657   -0.008105
H   2.054468    0.009987    0.006453

Now with MOKIT I have generated the MOCOEF. I have copied all the files from the above calculation and replaced the MOCOEF file from the one generated with MOKIT. Now, the new MINP file to calculate the Local CCSD energies from modified MOCOEF file is below.

#
mem=16GB
basis=cc-pvdz
dfbasis_scf=None
dfbasis_cor=None
calc=lccsd
uncontract=on
gauss=cart
verbosity=3
symm=0
core=0
ccprog=ccsd
localcc=2015
scfiguess=off
###lcorthr=vvtight
unit=angs
geom=xyz
4
opt b3lyp/avdz
H  -1.674794   -0.378923    0.001049
H  -1.673913    0.381594    0.000604
H   1.294239   -0.012657   -0.008105
H   2.054468    0.009987    0.006453

Is this new MINP file right?

Thanking you
Kind regards
Prasanta

Dear Prasanta,

you description of what you want to do is unclear.

If you want to do a conventional CCSD (without any local, ie. LNO approximations), then you have to do what I recommended before. So there is NO second MINP file.
You run step 1), replance MOCOEF in 2) and then you do not execute dmrcc. Instead in step 3) you execute the programs ovirt and then ccsd.

If you want to do LNO-CCSD, ie, using not only localized orbitals but also LNO approximations, then you need to do something else. Namely, you make a properly formatted MOCOEF-type file with you localized orbitals and then you rename this file as MOCOEF.LOC. Then you can use the orblocguess=read option to use this set of localized orbitals for an LNO-CCSD computation. This is more similar to you second MINP, but localcc=2015 still make no sense, why not use the defualt localcc value? And you take out scfiguess=off as in this since dmrcc run the scf has to be executed fully.

Best regards,
Peter

Respected Professor,
Sorry as my intentions were not clear to you. I do not want to perform conventional CCSD. What I want to do is similar to this work,  doi.org/10.1002/jcc.21462 . Except for what they did Local MP2 with GAMESS but I want to do Local CCSD. It would be better if I use with LNO and without LNO for comparison of energies.
If I have understood you correctly, then using localized HF orbitals with LNO approximations, I have to

1. rename the MOCOEF (localized and formatted using MOKIT) as MOCOEF.LOC
2. use the orblocguess=read
3. use localcc=on

Right? Or I have misunderstood something?

In order to use localized HF orbitals without LNO approximation, could you please specify what keywords are required?

Thanking you
Kind regards
Prasanta

Sorry as my intentions were not clear to you. I do not want to perform conventional CCSD. 

If I have understood you correctly, then using localized HF orbitals with LNO approximations, I have to

1. rename the MOCOEF (localized and formatted using MOKIT) as MOCOEF.LOC
2. use the orblocguess=read
3. use localcc=on

Right? Or I have misunderstood something?

> You understand correctly.

In order to use localized HF orbitals without LNO approximation, could you please specify what keywords are required?

> without  LNO approximation is equivalent to conventional CCSD. But you write on the top that you do not want to do conventional CCSD. 
So it is unclear what you want. You can do as a second option what I recommended previously. A third option is that adding to localcc=on also lcorthr=on. This will set the threshold of all LNO approximations to the truncation free setting. This is a very expensive way of computing concentional CCSD via the LNO code. But you can use your orbitals via orblocguess=read in a way you want "localized HF orbitals without LNO approximation".