2020-8-25 Meeting notes Dariel

Attendees

Dariel gives presentation.

Thomas' notes:

Segregation strength plot: would look better with x-axis log; should have segregation strength 0 at AR=1
Clarify labels 1=spheres 2=elongated
averaging region is very deep; data. like kinetic stress should not be averaged over inhomogeneous regions
density fraction: some asymmetry in x
Observation: spheres always have higher kinetic stress ratio (s_k^1/s_k) than 0.5, even higher than (rho^1/rho), ratio is increasing with more extreme AR. But: you are averaging over inhomogeneous region in z
d_z=1 characteristic axis perpendicular to surface, d_z increases with AZ more extreme
Ant notes upcoming paper comparing volume vs shape segregation, and paper with Mueller Christoph
A_eff: uses average angle, thus not accurate; better integrate over psd.

Goal of the paper:

Next steps:

Plot R_eff=f(AR) for all AR
Compare with the Hill Tan theory (see Deepak's paper) by looking at s_k^1/s_k and s_c^1/s_c vs phi^1/phi (using non-averaged values). Use local values, not averages over big regions
problem: s_c cannot be computed b/c of nan's. THis is likely b/c SuperQuadricParticle::getInteractionWith does return an interaction if the radii overlap, not if the particles are in contace (Solved)

problem: s_c cannot be computed b/c of nan's. THis is likely b/c SuperQuadricParticle::getInteractionWith does return an interaction if the radii overlap, not if the particles are in contac (Solved).
Casually, Jurgen Besten emailed me reporting the same issue yesterday.
There was a bug in getContactWithSuperQuad; it used to create a contact even when it did not exist.
The spherical particles always have more kinetic energy in relation to the density than non-spherical particles no matter the direction where the segregation occurs. Thus, changes in contact stress gradient depending on the aspect ratio must be the cause of segregation direction flipping. @Dariel Hernández Delfin will show that in next meeting.
@Dariel Hernández Delfin should create smoother colormap fields using more time steps in the coarse-graining calculation. The segregation index will be more accurate.

Problem: Even after the last fix, there are still nan/inf values in the fstat files. These nan's in the fstat files cause nan's in the stat files.
Solution: @Dariel Hernández Delfin: Check that the nan values in the fstat files are for contacts with zero force (i.e. they can be ignored). Remove them from the fstat file (e.g. sed -i 's/nan/0/g' *.fstat). Then your stat files should be without nan's
Problem: f^nu-phi^nu vs phi^nu plots are incorrect: f^nu-phi^nu should be 0 at phi^nu equal to 0 and 1, and the plots for spheres and ellipsoids should be point-symmetric (because f^sp+f^el=1, phi^sp+phi^el=1).This must be a bug
Solution: @Dariel Hernández Delfin should find the bug.
Dariel showed me plots that show a quite clear correlation between segregation strength and contact overstress f_c^nu-phi^nu, not with kinetic-overstress.
To show this clearly: @Dariel Hernández Delfin plot overstress f^nu-phi^nu as a function of aspect ratio (for both kinetic and contact). You could limit the data to phi^nu values between 0.4 and 0.6 to get reliable data for such a plot.
Problem: The overstress behaviour seems different near the drum wall
Solution: @Dariel Hernández Delfin color the data taken from near the wall differently, so we can distinguish.
Problem: Raul cannot see the notes I think
Solution: @Thomas Weinhart Get Raul to sign up to confluence.

Plots are now symmetric. Data shows two competing mechanisms, f^c-f^k and f^k-phi.
We should add Chute data
Suggestion for Mercury by Raul: estimate contact point of superquadrics with spherocylinder
Suggestion by Raul: You can define quite general shapes using spherocylinders with a variable radius.

Note: Deepaks paper predicts segregation even in the non-sheared regions; there is a shear-rate dependency missing; so we should only consider the sheared flowing layer. (There is a draft paper by Deepak showing dependence on I)
V_r shows no flip in segregation directions; but, see above, it should be averaged only over the flowing layer.

@Dariel Hern├índez Delfin redo the plot. Also replot numerator and denominator separately.