yade-dev team mailing list archive

[noreply@xxxxxxxxxxxxx]

------------------------------------------------------------
revno: 4041
committer: bchareyre <bruno.chareyre@xxxxxxxxxxxxxxx>
timestamp: Mon 2017-05-15 18:31:11 +0200
message:
  Introduce a tolerance on overlap determination (in periodic BC only), avoiding false negative due to round-off errors. (usage in a separate commit).
modified:
  pkg/common/InsertionSortCollider.hpp


--
lp:yade
https://code.launchpad.net/~yade-pkg/yade/git-trunk

Your team Yade developers is subscribed to branch lp:yade.
To unsubscribe from this branch go to https://code.launchpad.net/~yade-pkg/yade/git-trunk/+edit-subscription

=== modified file 'pkg/common/InsertionSortCollider.hpp'
--- pkg/common/InsertionSortCollider.hpp	2016-06-13 07:37:27 +0000
+++ pkg/common/InsertionSortCollider.hpp	2017-05-15 16:31:11 +0000
@@ -197,6 +197,7 @@
 		((bool,allowBiggerThanPeriod,false,,"If true, tests on bodies sizes will be disabled, and the simulation will run normaly even if bodies larger than period are found. It can be useful when the periodic problem include e.g. a floor modelized with wall/box/facet.\nBe sure you know what you are doing if you touch this flag. The result is undefined if one large body moves out of the (0,0,0) period."))
 		((bool,sortThenCollide,false,,"Separate sorting and colliding phase; it is MUCH slower, but all interactions are processed at every step; this effectively makes the collider non-persistent, not remembering last state. (The default behavior relies on the fact that inversions during insertion sort are overlaps of bounding boxes that just started/ceased to exist, and only processes those; this makes the collider much more efficient.)"))
 		((int,targetInterv,50,,"(experimental) Target number of iterations between bound update, used to define a smaller sweep distance for slower grains if >0, else always use 1*verletDist. Useful in simulations with strong velocity contrasts between slow bodies and fast bodies."))
+		((Real,overlapTolerance,1e-7,,"Tolerance on determining overlap. In rare cases different parts of the code can inconsistently lead to different results in terms of overlap, with false negative by spatialOverlapPeri possibly leading to nasty bugs in contact detection (false positive are harmless). This tolerance is to avoid false negative, the value can be understood as relative to 1 (i.e. independent of particle size or any other reference length). The default should be ok."))
 		((Real,updatingDispFactor,-1,,"(experimental) Displacement factor used to trigger bound update: the bound is updated only if updatingDispFactor*disp>sweepDist when >0, else all bounds are updated."))
 		((Real,verletDist,((void)"Automatically initialized",-.5),,"Length by which to enlarge particle bounds, to avoid running collider at every step. Stride disabled if zero. Negative value will trigger automatic computation, so that the real value will be *verletDist* × minimum spherical particle radius; if there are no spherical particles, it will be disabled. The actual length added to one bound can be only a fraction of verletDist when :yref:`InsertionSortCollider::targetInterv` is > 0."))
 		((Real,minSweepDistFactor,0.1,,"Minimal distance by which enlarge all bounding boxes; superseeds computed value of verletDist when lower that (minSweepDistFactor x verletDist)."))