Hi Joel, I forwarded this to the list. As far as updating to 1.2, if you post some of your scripts and/or some of the errors your'e getting, we could try and help you make the switch. It may be something trivial.
---------- Forwarded message ---------- From: joelvuillon <[EMAIL PROTECTED]> Date: Sep 24, 2007 11:40 AM Subject: Re: Solving elasticity Equations To: Daniel Wheeler <[EMAIL PROTECTED]> Cc: [EMAIL PROTECTED] ----- Original Message ----- From: "Daniel Wheeler" <[EMAIL PROTECTED]> To: "joelvuillon" <[EMAIL PROTECTED]> Sent: Monday, September 24, 2007 5:02 PM Subject: Re: Solving elasticity Equations > On 9/21/07, joelvuillon <[EMAIL PROTECTED]> wrote: >> Hi Daniel and thank you for your quick answer on the subject. >> Unfortunately, these elasticity equations are solved on unstructured >> hexaedral meshes imported into FiPy. > > That's fine. We are working on implementing a "diffusion tensor" > that will make the "DiffusionTerm" more flexible. For that kind of elasticity problems, il would be fine. > >> Besides, the work is done under FiPy >> 1.1. > > I would recommend switching to 1.2. Why are you using 1.1 and where > and when did you download it? The work on the subject had started in 2006/06 and the starting point was enthon-python2.3-0.9.7.exe FiPy-1.1.win32.zip I have tried a switch to FiPy 1.2 one month ago (2007/08) but I experienced numerical troubles with the modeling. I finally did not follow the switch. > >> I have not try the formulation yet but it would surely gives good >> results. >> Relaxation is also needed in such problems and a balance between the >> implicit and explicit terms has to be found : good results can be >> obtained >> by changing >> >> div.[ mu grad(U) + mu [grad(U)]^T + lambda I tr(grad(U)) ] + F = (0,0) >> >> into relations like : >> >> div.[ DiffCoeff*grad(U) + mu [grad(U)]^T + lambda I tr(grad(U) - >> (3*lambda + >> mu)grad(U)) ] + F = (0,0) where DiffCoeff = 3*lambda + 2*mu has a >> physical >> meaning (hydrostatic compression module). > > Interesting. Please keep me posted on how things proceed with the work. The 3D elasticity equations work well with the relaxation above. The convergence is sometimes long (80 iterations for res = 1e-2). Lake of convergence is found when strong discontinuities on the lamé coefficients (lambda, mu) exist from one point to another. Regards Joel -- Daniel Wheeler
