Frequently asked questions.
I am about to install DIRAC on a 64-bit machine, should I compile using 64-bit integers or 32-bit integers?
DIRAC runs smoothly (with few exceptions; see below) on 64-bit platforms using either 32-bit or 64-bit integers.
It is easier to install for 32-bit integers because MPI and math libraries are typically built for 32-bit integers. Therefore DIRAC is by default built for 32-bit integers.
If you decide to build for 32-bit integers, you will not be able to:
- Safely allocate more than 16 GB of memory per cpu-core
- Run LUCITA
- Run really, really large Coupled-Cluster calculations
If you want to be able to run the above calculations, you have to build for 64-bit integers.
If you decide to build for 64-bit integers, you have to make sure that:
- The math library you link to can handle 64-bit integers
- MPI library is built for 64-bit integers
Otherwise DIRAC will stop.
Running Jobs with DIRAC
Parallel run information
My shell environment variables are not forwarded to the compute nodes. What do i have to do?
Assuming you are using our submit script pam to submit your parallel calculation you can forward any environment variable, e.g., LD_LIBRARY_PATH, to the slave-node shell by adding to you pam command line:
- ./pam --mpiarg="-x LD_LIBRARY_PATH" (for OpenMPI)
- ./pam --mpiarg="-envall" (for IntelMPI and MPICH(2))
In the latter case ALL environment variables will be forwarded in one shot whereas for OpenMPI you may need to provide the most essential environment variables individually. These may be for a typical parallel Dirac run:
General run information
How can I restart from the MO coefficients file DFCOEF?
Simply copy DFCOEF to the scratch directory using the pam script. If DIRAC finds DFCOEF in the scratch directory it will automatically try to restart from it.
Can I move the DFCOEF file from one machine to another and restart from it?
In general no, unless it is the same architecture.
What you can do instead is to export the DFCOEF in a machine independent format using the .PCMOUT keyword. This creates a file called DFPCMO. Copy DFPCMO to the other machine, make it avaliable in the scratch directory using the pam script, then restart from it using .PCMOIN.
Note: if you are running a calculation in linear symmetry you will find an indication in the DIRAC output (.PCMOUT run) of whether you need the keyword .PCMOIN or .PCMOI3 for the restart. This is connected to the formatted writing of DFCOEF and the space needed to store the mj-value of each spinor.
I tried to restart but get the error "Incompatible number of basis functions" - what does it mean?
This probably means that you have changed the basis set. Presently DIRAC cannot restart from DFCOEF if the basis set changes (more correctly if matrix dimensions change).
I would like to use the 2c-molecular-mean-field Hamiltonian (X2Cmmf) for my post-HF (Coupled-Cluster [CC], Fock-Space CC, KR-MCSCF or KR-CI) calculations - what do i need to consider?
The 2c-molecular-mean-field Hamiltonian (X2Cmmf) is at present (Dirac11 release Nov 2011) available for the following methods:
- Fock-space Coupled Cluster (including the intermediate Hamiltonian version)
The remaining modules are not supported at the Dirac11 release date but patches are planned in the course of 2012:
The X2Cmmf Hamiltonian scheme can be used for the Dirac-Coulomb as well as the Dirac-Coulomb-Gaunt Hamiltonian (add .GAUNT to **HAMILTONIAN).
As follows we will consider the necessary inputs for an X2Cmmf calculation where we perform the SCF + post-SCF calculation in one shot or two separate runs:
- one-step procedure:
- add .X2Cmmf to the **HAMILTONIAN keyword section:
that's it. After the SCF calculation the program will proceed with the transformation to 2-components prior to the post-HF steps.
- two-step procedure (EXPERT mode!):
1. SCF step. Add .X2Cmmf to the **HAMILTONIAN keyword section:
and save the file DFCOEF (keyword --outcmo with our default submit script pam)
2. post-HF step. Remove .X2Cmmf from the **HAMILTONIAN section (non-intuitive (!) step which will be solved in Dirac12) and add the following keywords to the **MOLTRA and namelist RELCCSSD input section.
**MOLTRA .INTFL2 1 0 0 0 .INTFL4 1 0 0 0 ... ... &CCSORT USEOE=T, NORECMP=T &END
You are now ready to restart your post-HF calculation but do not forget to copy your MO-coefficient file DFCOEF to your scratch directory, i.e. add --incmo to your pam command line.