Molcas Forum

MQC

Member

Registered: 2020-12-17

Posts: 5

Restarting DMRGSCF calculation with CheMPS2

Dear all,

I am having some issues with the restarting of a DMRG-SCF calculation employing the CheMPS2 interface.

To provide context for this issue, I was running a DMRGSCF calculation without 3RDM keyword. The calculation was already converged and the DMRG-SCF last iteration was running (obviously with 0 noise prefactor and with a maximum number of sweeps of 40 as it is written in the manual). Unfortunately, the cluster where I was running the calculation crashed, so my job was killed. I wonder if I can fully restart DMRG from the previous step with the specifications of the last iteration, taking into account that I have all the files in the working directory before the job was stopped. I tried CHRestart keyword but the calculation was only partially restarted; also, MXSweep was set to 40 and the noise prefactor to 0 by myself, but the calculation was not fully restarted as I would like to... I have just checked if DMREstart=2 would work for this last one iteration, even though in the manual is explicitly written that it only works when 3RDM keywork is on, so not, it did not work too...

I am state-averaging over 9 roots so the computational cost is quite expensive... I ask you for help for fully restarting the calculation if it is possible. Also, I have consider the option of taking the energies of the last iteration before the crashing, but in order to characterise my states I do not have the natural orbitals as they are not printed until the calculation is completed. I have the JobIph, RasOrb and rasscf.h5 files and the CheMPS2_natorb_MPSn.h5 files, but I do not know any way of obtaining the natural orbitals from this format, so if this can be done, please, I would like to know how if it is not much to ask for...

Thanks in advance.

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Restarting DMRGSCF calculation with CheMPS2

I don't know about restarting, but the natural orbitals for each root should be in the RasOrb.x files (x is the root number). These are text files, so they are easy to read if you just want the MO coefficients and/or occupations. If you want to view the orbitals in 3D, you can open the rasscf.h5 file in Pegamoid (select the root you want once you open the file). Presumably the CheMPS2_natorb_MPSn.h5 also has the natural orbitals, but I don't know its format; however, since it's an HDF5 file, it should be self-documented to some extent, try opening it with some general HDF5 file viewer.

MQC

Member

Registered: 2020-12-17

Posts: 5

Re: Restarting DMRGSCF calculation with CheMPS2

Thank you so much for your soon response, Ignacio! Concerning the RasOrb.x files, the problem is that they were not printed because the calculation did not finish at all. Also, there is a problem with CheMPS2, which I have recently found. I am afraid that with the CheMPS2 module the natural orbitals of each root are not written to the hdf5 file. I first realised this when I tried to use the file with the WFA module in order to obtain NTOs from the hdf5 files and it was not possible. Then, I checked the natural orbitals of this rasscf.h5 both using your program pegamoid (which is a very nice job, by the way) and also penny and only average orbitals for the DMRG calculation were there. This does not happen with QCMaquis, though. However, CheMPS2 does write the natural orbitals to the rasscf.molden.x files, but as it happens with the RasOrb.x, I don't have it as the calculation did not finish.

Thanks again for your quick help and kind regards

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Restarting DMRGSCF calculation with CheMPS2

Just note that the root-specific natural orbitals are never explicitly written to rasscf.h5, it is actually the root-specific density matrix (in the basis of the state-average MOs) that is stored, and the NOs are trivially obtained from there. So, if in any of the files you find something that looks like 9 density matrices, it could be what you need (although it could be in some format not recognized by penny or Pegamoid).

MQC

Member

Registered: 2020-12-17

Posts: 5

Re: Restarting DMRGSCF calculation with CheMPS2

OK, I will check this. Thanks again