ABINIT Discussion Forums

Hi, when I run a ddk calculation (rfelfd 2 or rfddk 1, same thing) in the manner of test 5 of section 5 (v5, t05.in) in order to compare finite difference eigenvalues with ddk eigenvalues, I find a huge difference between the output from the .out file and the output in the EIG file. For example, run...

Hi, when I execute runtests.py with the -i option I keep getting a KeyError failure as follows: ../../tests/runtests.py -i 'rfddk' OUTPUT: FortranCompiler: intel None Traceback (most recent call last): File "../../tests/runtests.py", line 479, in <module> sys.exit(main()) File "../../...

most probably scalar relativistic as that is the default of the fhi98 generator.

Hi, we would need to know all about your computer system, build, and input file to make progress on this question.

Joe

you might post your input file. often input file errors cause failures that appear when an mpi function is called, although the errors aren't really themselves mpi related. Also your second run seems to make reference to Cuda libraries, are you running on a gpu machine? you could compile with gpu di...

a 95 atom cell with no symmetry is simply a large calculation in the DFT world. you can speed it up by running on many cores in parallel and taking advantage of parallelization over k points, bands, and fft planes as described in the documentation for paral_kgb and autoparal key words.

please try PAW sets from the new JTH set you can find at http://www.abinit.org/downloads/PAW2

In PAW, norm-conservation of the wavefunctions is relaxed, so <psi|psi> is computed with the overlap operator S as <psi|psi> = <\tilde{psi}|S|\tilde{psi}> where |\tilde{psi}> are being computed by abinit. Note that also to get the physical density out for plotting in PAW, you need to use prtden 2 or...

I think you are trying to force abinit to use spin orbit coupling in a way that is inconsistent with the LDA+U calculation. Try removing the nspden, nsppol, and nspinor variables from your input file, but leave the pawspnorb and usepawu parts and see how it goes. If you still have trouble, post or s...

yes it is

berry_linmin is an orphan and can be ignored. I was trying to carve it out of cgwf but had problems and never finished that task. I will remove it from the tree, and might come back to this aspect of cleaning up cgwf at a later date.

Joe

Is it even a well-defined concept, to compute the free energy in this case? Don't imaginary frequencies imply an unstable configuration, therefore out of equilibrium, therefore standard formula for F not applicable?

First, the output file will contain a section at the top with an estimate of how much memory the calculation should take. Look for the line that begins "P This job should need less than..." Secondly, in my experience, 32 atoms would be a huge calculation to attempt to run on a laptop. In m...

The acoustic contributions to thermodynamic properties are detailed nicely in Ashcroft and Mermin chapter 23.

I don't see why you would exclude acoustic modes--all modes contribute to thermodynamic properties, acoustic modes more than most exactly because of their low frequencies.

OK, I figured out my problem--when adding a new test it is necessary also to do ../../tests/runtests.py --regenerate to re-make the database. The instructions do say this but it is buried at the bottom as an "additional note". It would be better to make it step 6 of the instruction list. A...

Hi, I'm trying to add a new test (t70.in in v7) but the system doesn't find it. I've added t70.in to tests/v7/Input, added t70.out to tests/v7/Refs, and added a line to __init__.py in the tests/v7 directory. But running the following: ../../tests/runtests.py v7[70] from the tests subdirectory of my ...

Currently the code requires the occupation for a given band to be identical at every k point, so I'm not sure whether it will work for the ferromagnetic insulator case. Have you tried it with nspinor 2 (rather than nsppol 2)?

Yes, berryopt works for nsppol /= 1 although not for parallel processing.

I suspect it's not the type of spin-order that's a problem here, it's that your system may be metallic. The entire berryphase theory is predicated on the assumption that the system is an insulator, and that you are only treating the occupied bands.

Oh, I see I misunderstood your question. The post-processing tool cut3d can do what you want, I believe--see the help file http://www.abinit.org/documentation/hel ... _help.html item 1.b.

I think prtdos 3 in combination with prtdosm 1 will do what you're looking for.

Yes, I saw that you are not using mpirun, and that's exactly why abinit is only running on one processor. Like any other parallelized program, abinit to run in parallel must be explicitly launched in the parallel environment with the desired number of processors. mpirun is the program that takes car...

make multi multi_nprocs=8 means that it will use 8 processors to build abinit (which is faster than using 1 processor to build abinit). When you RUN abinit, you have to tell it how many processors to use, typically (if you are using MPI as your parallel environment) through mpirun -np 8 abinit < foo...

Probably not any time soon, I'm afraid--as far as I know, no one is working on this aspect at the moment. I or perhaps Eric Bousquet would probably be the likely ones to do it, and as for me, it wouldn't address any problems my lab is currently studying so it doesn't rise very high up on my to-do li...