Attendees

Discussion items

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 Hern├índez Delfin will show that in next meeting.
Dariel Hern├índez 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 Hern├índez 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 Hern├índez 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 Hern├índez 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 Hern├índez 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.