Yade

Sphere - nonSphere interactions never deleted by Law2_ScGeom_CapillaryPhys_Capillarity

Bug #1628273 reported by Jérôme Duriez on 2016-09-27

This bug affects 1 person

Affects		Status	Importance	Assigned to	Milestone
	Yade	Fix Released	Undecided	Unassigned

Bug Description

Hi,

Let say one performs simulation with Law2_ScGeom_CapillaryPhys_Capillarity and Law2_ScGeom_FrictPhys_CundallStrack.neverErase = 1 (in order to keep menisci between separated particles)

In such case, interactions between sphere and non spherical particles (e.g. boundaries) are not deleted by Law2_ScGeom_FrictPhys_CundallStrack when geometrical contact is lost, and are neither handled (deleted because of no meniscus, in particular) by Law2_ScGeom_CapillaryPhys_Capillarity because Law2_ScGeom_CapillaryPhys_Capillarity simply disregards such interaction [*].

And the user ends up with much more interactions than necessary..

I'm filling a Bug report for the record and possible discussion about desired behaviour (sorry I could not provide a MWE !)

Jerome

[*] https://github.com/yade/trunk/blob/master/pkg/dem/Law2_ScGeom_CapillaryPhys_Capillarity.cpp#L98

Revision history for this message

Bruno Chareyre (bruno-chareyre) wrote on 2016-09-28:

I confirm the bug conceptually (did not reproduce).
Call it memory leak...
Did you realize that incidentaly or have you seen an anomalous growth of interaction number in a simulation?

I wonder how it can be solved without hacks. The notion of multiple physical interactions between two bodies does not fit very well in Yade's picture at the moment.

Bruno

Revision history for this message

Jérôme Duriez (jduriez) wrote on 2016-09-28:

We realized it thanks to a colleague who was working on interaction data, as exported in a text file. However we did not quantify it precisely yet (and I'm now away from office for a couple of weeks)

Revision history for this message

Jérôme Duriez (jduriez) wrote on 2016-10-17:

Hi,

Just to give some figures about this bug:
Random examples of triaxial simulations (with 20000 spherical particles and 6 boxes) show here ~ 45% of sphere-wall interactions that are useless at the end of the simulation. Ie ~ 3% of the (real) interactions total number.

I think checking for penetrationDepth < 0 around l.95 of the code [1], before deciding to "continue", would make sense and solve this bug.
OK ?

Do we take the opportunity to replace the shape test [2] with something else ? (what ? if yes)

Jerome

[1] https://github.com/yade/trunk/blob/master/pkg/dem/Law2_ScGeom_CapillaryPhys_Capillarity.cpp#L95
[2] https://github.com/yade/trunk/blob/master/pkg/dem/Law2_ScGeom_CapillaryPhys_Capillarity.cpp#L98