Some general doubts

Dear MRCC developers,

First of all, congratulations for the code.

1) I want to know if densities printed in the CCDENSITIES file are the final relaxed density matrices in MO basis, that is same matrices used to evaluate properties. In case that these are not, where I can print the relaxed 1 and 2 order density matrices.

2) When running a multireference CC the core keyword applies in the same way as in single reference methods ? CCDENSITIES file contain the corresponding relaxed density matrices in MO basis for the multireference calculation ?

3) In case that CCDENSITIES are the matrices I want, where can I modified it ? My program for analizing the calculation need these data in chemist notation and a some slightly modified format, in this way I could have a common format instead o checking in my program each kind of format provide by each code. My code do the properties analisis so i dont need the prop excution part, can I request only density matrices withouth runing prop code ?

4) I dont have access to molpro program to use casscf orbitals instead of canonical hartree-fock, could it be possible to interface mrcc code for example to a FCIDUMP file without much complications ? In that case with a little of help for program work flow and structure, I think I could do it. For example I use pyscf ( github.com/sunqm/pyscf ) which allow me to do very general active space selection and transformations on orbitals, like for example -http://pubs.acs.org/doi/abs/10.1021/acs.jctc.7b00128 . Also this code write FCIDUMP files used for other external codes.

Thanks in advance.
Best regards,
Jose Luis

Dear Jose,
1) These are unrelaxed MO densities, please see the discussion at
www.mrcc.hu/index.php/forum/running-mrcc/67-relaxed-ccdensities
about how to print out relaxed densities.
2) The core keyword functions in the same way for both single- and multi-reference calculations. The CCDENSITIES file contains the unrelaxed MO density also in the MR case.
3) The orbital relaxation contribution is calculated by prop, so if you need relaxed densities, you should hack the code as described above.
4) The integrals are stored in a format very similar to that of the FCIDUMP file, and probably you could easily set up an interface. The problem is that we cannot calculate the orbital relaxation for MCSCF orbitals, and only the unrelaxed densities could be computed.

Dear Mihaly,

Thanks for all the info. I have some doubts about these unrelaxed matrices, when orbital relaxation is included, it means that Hellman-Feynman in satisfied ? What are exactly the differences ? By simple inspection to the CCDENSITIES files, the values sean like a correlated density, for example occupations are not 2 for my closed shell calculation. I have some own code that calculate unrelaxed densities for CCSD taking only the amplitudes but not the lagrange multipliers and ocupations are the same as in Hartre-Foock reference only non-diagonal elements are include. So I am a little bit confused about what realy means unrelaxed.

Could you instruct me a little bit how to bypass the scf procedure to directly read integrals in MO basis from a external code ?

Best regards,
Jose Luis

Dear Jose,
Unrelaxed density means that you do not add the orbital relaxation contribution, that is, the response of the orbitals to the perturbation is ignored. If this is the case, a property is evaluated as an expectation value, and the Hellmann-Feynman theorem is not satisfied when the corresponding property integrals are contracted with the unrelaxed density matrices. In CC theory the unrelaxed density is calculated using a biorthogonal formalism, see, e.g., dx.doi.org/10.1063/1.1668632 for a general derivation. In that paper eq. 9 defines the unrelaxed density, while eq. 8 should be used if we want to evaluate a property as a derivative. Obviously, the derivative of the second-quantized Hamiltonian includes the derivatives of MO coefficients, the contribution of which should also be considered in this case. See, e.g., page 2628 in dx.doi.org/10.1063/1.460915 for the treatment of orbital relaxation.

To use external MO integrals, you should write a MINP file including the iface keyword, e.g, iface=molpro, and other keyword you want (mem, calc, ...). The MO integrals should be saved to the file fort.55. Please find attached two files describing the format of the fort.55 file for restricted and unrestricted calculations.

Dear Mihaly,

Thanks for all the info.

I have modifed the code to print the MO densities and also a plugin to pyscf. Now the pyscf integrals are working correctly but when using the dens=2 keyword the code abort with the following error

************************ 2017-10-16 21:01:11 *************************
Executing prop...

Allocation of 32.0 Gbytes of memory...
Corrupted file VARS!
Variable nbasis is missing!
Please run integ first!

Fatal error in prop.
Program will stop.

************************ 2017-10-16 21:01:11 *************************
Error at the termination of mrcc.
**********************************************************************

I have more or less clear the differences between relaxed and unrelaxed. As a not expertise in the field, it is "wrong" to do energy partitions (I use density matrices to integrated it in real space theories) using the unrelaxed densities in case I use pyscf casscf canonical orbitals even if the energy is reproduced ?

Best,
Jose Luis

Dear Jose,
Please note that prop calculates the orbital relaxation contribution, and if you use MO integrals from an external code, the orbital relaxation contribution cannot be calculated. In this case you can only use the unrelaxed densities stored in the CCDENSITIES file.

You can use the unrelaxed density matrices. As I mentioned, it corresponds to evaluating properties as expectation values, while using relaxed densities corresponds to evaluating properties as derivatives. These are different approaches, and in the full CI limit they give identical properties.