yade-users team mailing list archive

[Bruno Chareyre <bruno.chareyre@xxxxxxxxxxx>]

Based on my test, ScGeom doesn't work in PB. It is (I think) because it 
needs relative velocity, which is not defined correctly across periods.
I have a few ideas to solve that though, and some uncommited code (see 
**)...
It is a pitty that we can't use the same law in periodic and finite BCs 
at the moment. On one hand, there is no sphere-box, hence no 
TriaxialTest, with Dem3DOF (facets and planes for boundaries is bugged 
for some reasons). On the other hand, ScGeom can't be used in periodic BCs.

> just a couple of quick questions about PB.
> 1) Are you following Cundall's algorithm?
I don't know Cundall's papers on this. Any reference?
> 2) I see the variable velGrad being a matrix, why?
It is the gradient of velocity. It is a tensor, just like strain tensor 
and any grad(vector).
It is the matrix dH/dt in classical litterature on periodicity in 
molecular dynamics.
> 3) I understand it is possible to set the variable 
> homotheticCellResize as equal to 0, 1 or 2. I was reading Cundall's 
> paper and I remember he was updating displacements and relative 
> velocities at contact according to the strain rate. Is this the same 
> as done in Newton Integrator?
(**) Is not done properly. We need to adjust relative velocities, which 
should not be too hard using the shift2 vector in interaction dispatcher 
(should be something like velGrad*shift2). It needs to consider the 
ordering of the sequence contactLaw/PeriIsoCopressor/Newton carefully, too.
I plan to fix that soon, if I can find time... In the end, the case 
homotheticCellResize=1
should disapear and everything will be done via velocity impulses 
(homothetic=2), for now it makes things better but not perfect.
By the way, I reply to your questions because vellGrad and cellResize is 
the only small part of periodic BC's that is not due uniquely to Vaclav 
(they were introduced by me).

Bruno