Skip to content

GitLab

  • Projects
  • Groups
  • Snippets
  • Help
    • Loading...
  • Help
    • Help
    • Support
    • Community forum
    • Submit feedback
    • Contribute to GitLab
  • Sign in
fleur
fleur
  • Project overview
    • Project overview
    • Details
    • Activity
    • Releases
  • Repository
    • Repository
    • Files
    • Commits
    • Branches
    • Tags
    • Contributors
    • Graph
    • Compare
  • Issues 51
    • Issues 51
    • List
    • Boards
    • Labels
    • Service Desk
    • Milestones
  • Operations
    • Operations
    • Incidents
  • Packages & Registries
    • Packages & Registries
    • Container Registry
  • Analytics
    • Analytics
    • Repository
    • Value Stream
  • Wiki
    • Wiki
  • Members
    • Members
  • Collapse sidebar
  • Activity
  • Graph
  • Create a new issue
  • Commits
  • Issue Boards
  • fleur
  • fleurfleur
  • Issues
  • #486

Closed
Open
Opened Aug 21, 2020 by stichri@stichri

EELS calculation issue(s)

Hello everyone,

I'm trying to compute the K edge in an EELS spectrum of a Be crystal given this structural input to compare the result with a reference signal.

To that end, I installed fleur(_MPI) by cloning the release branch of the git repository. I'm reasonably sure I successfully compiled the binary because I am able to run the SCF loop for the Be crystal on a single HPC node (providing 48 physical CPU cores) with 6 MPI instances and 8 OMP threads without any apparent issues, see the complete SCF calculation files.

As indicated in the documentation, I modified the inp.xml file (that was successfully used in the SCF calculation) by adding coreSpec="T" to the default <output dos="F" band="F" vacdos="F" slice="F" mcd="F"> tag and added the corresponding element

  <coreSpectrum eKin="200.0" atomType="1" lmax="2" edgeType="K"
         eMin="-1.0" eMax="30.0" numPoints="78" nqphi="10"
         nqr="10" alpha_Ex="0.001" beta_Ex="0.100" 
         I_initial="155" verbose="T">
    <edgeIndices> 1 </edgeIndices>
  </coreSpectrum>

Note this includes changed values for eKin, edgeType, eMax, numPoints, alpha_Ex, beta_Ex as well as <edgeIndices> to reflect this reference signal's (meta)data. Because I have absolutely no EELS experience whatsoever, I'm not sure all these choices are correct though, particularly the choice of <edgeIndices> is based on the relation between spectroscopic notation and quantum numbers as summarized in the table on page 436, Appendix D of "Electron Energy-Loss Spectroscopy in the Electron Microscope " by Egerton.

Executing fleur_MPI with the resulting input however results in an early termination of the code with the output containing 5 instances of

corespec_init> found csi%I0 <= 0 ! ... STOP !

Assuming this indicates 5 of the 6 MPI instances fail to initialize the data properly, I decided to run just 1 MPI instance and 48 OMP threads, which results in a successful initialization and a bad MPI termination with exit code 11 later on. However, there is lots of additional data given in the output of the corresponding calculation, including some suspicious fort.x files up to 38 MB in size.

Further testing results in similar output for calulations with lmax="3" and lmax="8" (the latter effectively not limiting lmax compared to the earlier input in inp.xml), although the fort.39 files in both tests seem to simply keep growing until I kill the jobs at around 1.4 GB of output. Both calculations however also result in fort.36 files wich are seemingly meant to contain the spectra as a column of energy and intensity values for lmax="3" and lmax="8", respectively.

So I'm wondering, is there a fatal flaw in how or what input is provided or is there a problem with the code? If I can obtain any additional useful data, I'll happily provide that as well ASAP.

Any help is greatly appreciated, thanks in advance!

Assignee
Assign to
None
Milestone
None
Assign milestone
Time tracking
None
Due date
None
Reference: fleur/fleur#486