2019-11-11 MercuryDPM Meeting

Hope that you are doing well.

ISSMGE’s TC105 Geomechnics from Micro to Macro chaired by me is running a DEM round robin test.

https://www.issmge.org/news/tc105-dem-round-robin-test

Can you pass this news to our Powders and Grains colleagues, if possible?

Thanks in advance.

Best wishes,

• Hongyang Cheng takes lead, will contact the organisers.

Update: Hongyang contacted the organisers. He will put the description of the test problem on Jira. They will publish the results without linking the codes to the outputs It will be angle of repose measurement for multisphere particles.

• Hongyang will make a JIRA job for this problem.
• @all: We need a volunteer to run the test problem

• Luca
  • When configuring cmake I get the first error, concerning a CMP0057 error.
    I get rid of it by editing the file Source/CmakeModules/MercuryDocumentation.cmake and commenting the first line (cmake_policy(SET CMP0057 NEW)).
• DPM issues.docx

No news, no h.shi-1 only Linux version is tested, need to compile all the settings before move to other systems.

Hao needs to talk to Han this week on the docker before he leaves.

Hao added the MercuryDPM-Trunk docker image for both Ubuntu and Mac. The windows version will be tested later on.

Update: Docker does not work on Windows home, even if HyperV is installed. Need to test on Windows Professional & Enterprise.

1-way coupled code is approx 4000 faster but does not work for non-refinables elements as it needs the extra data

Hao did the test on 4000 particles with new code, the Oomph-part is indeed much faster than the older version, this reduces the total simulation time by approx. 20%.

However, it seems somewhere else is getting heavier and the new run compare to the old run is actually longer. Hao checked with Mitchel on both old and new implementations

and found out the time cost difference on the locate_zeta function in pressure gradient. Think of updating coupling force not every Mercury Timestep to reduce the run time.

• Anthony Thornton and Thomas Weinhart will look at it with Mitchel Post on Wednesday
  

Jinja or Jinja2

Do we want to add a python interface to MercuryDPM in the future?

Would be useful to use in combination with GrainLearning.

We also have the Mercury command line interface, but that cannot pass functions.

Starts April 20; first week we give the MercuryDPM courses (April 27-May 22)

• Thomas Weinhart book Vrijhof

There is a student from Erlangen (attanded DEM8) who might be interested in writing this interface in the MercuryMonth

Youetta made a video:

• Hao: put it on youtube, share privately
• @all Please comment

Profiling shows for dense free-cooling problem that computeforces(p,q) is the main cost, but hgrid and computeWallForces are also significant.

• Julius Mbaziira Subtract time for output from the analysis
  

Output files: Had a discussion on file types: Jens Harting uses HDF5 for LB, but not for DEM. HDF5 will be binary (3x more ocmpact and more accurate than restart files) and can handle parallel output. But Hao and Hong prefer a python script to convert output.