yade-users team mailing list archive

[Bruno Chareyre <bruno.chareyre@xxxxxxxxxxx>]

Writing the 6x6 stiffness matrix of each body is like examining 
sub-blocs of the global stiffness matrix of the system (which is very 
sparse and can be put as a thin band matrix). Read PFC3D doc, or this 
paper for a special case where there is no rotational inertia 
http://dx.doi.org/10.1061/(ASCE)0733-9399(2005)131:7(689).
Better discuss in my office because I don't understand all questions 
below. A mistake on a sqrt(2) factor is always possible, though I would 
be surprised.

Bruno

Jerome Duriez a écrit :
> Hello,
>
> I'm focusing on the equations used in GlobalStiffnessTimeStepper. And 
> I would like to link all this with what I've understood in the 1D case 
> of mass spring systems (with the help for example of a former mail of 
> Bruno, quoted in 
> http://www.mail-archive.com/yade-users@xxxxxxxxxxxxxxxxxxx/msg00583.html, 
> I did not find the original mail sorry)
>
> The calculation is here apparently based on the diagonal terms, let us 
> say Kd (=stiffness.x/y/z() in the code), of a stifness matrix defined 
> for each body, matrix which takes into account all contacts involving 
> this body. Because, I imagine, that this stifness matrix will appear 
> in the matrixial write of a recursive scheme which would be equivalent 
> to our integration "algorithm".
>
> - So is the condition for the stability of the sheme in 3D coming from 
> the same considerations about stability of recursive sheme as 
> explained former by Bruno ? But by replacing the condition about this 
> real alpha (alpha < 1) by one condition about the diagonal terms of a 
> matrix ?
>
> - If yes, why can we keep this sqrt(2) here in GSTS, while in 1D it 
> was explained that this sqrt(2) has to disappear to take into account 
> opposite motions between bodies ?
>
> "FYI" all these questions because I want to write my kind GSTS for 
> UDEC, and for now UDEC computes a timestep only by Dt = k * sqrt( 
> Min(Mass of elements) / Max(Stifnesses of contact) ), with k = 0.2 
> generally. So, with few theory behind it seems, but with a smaller 
> timestep at the end (is that also your feeling ?)...
>
> And few more technical questions :
> - Chiara : in my use of UDEC my rigidities are also proportionnel to 
> sqrt(u). So I'd be glad to know how you use GSTS in your simulations : 
> wich values of timestepSafetyCoefficient, and timeStepUpdateInterval 
> do you use in your simulations ?
>
> - Vaclav : would it be possible for you to send me also a copy of your 
> thesis (or the paragraphs concerning time steps) ? Note that I finally 
> found this reference 
> http://www.emeraldinsight.com/Insight/viewContentItem.do?contentType=Article&hdAction=lnkpdf&contentId=878199 
> (but I could not get the article)
>
> - Bruno (?) : I'm very sorry but I could not find in the code what 
> prevented GSTS to do its job (I saw all this active or isActivated() 
> but I did not find where these things really act)
>
> Thanks,
>
> Jerome


--
_______________
Bruno Chareyre
Associate Professor
Grenoble INP
Lab. 3SR
BP 53 - 38041, Grenoble cedex 9 - France
Tél : 33 4 56 52 86 21
Fax : 33 4 76 82 70 43
________________