On Apr 10, 2006, at 4:43 AM, hzhatlboro wrote:
(1) With respect to my question (2), as far as I understand, you said dbdpsi=d(beta)/d(psi). However, referring to page 65, A=alpha**2*c*(1+c*beta)*[d(Phi)/d(psi)], and the program on page 66, A=alpha**2*c*(1+c*beta)*dbdpsi, I suspect that dbdpsi=[d(Phi)/d (psi)], but I can not see that [d(Phi)/d(psi)]=-N*2*Phi/(1+PhiSq).
I see your point and I don't know. Dan Wheeler coded up that example and he's out for a week or so.
Other than that, I have greater faith in the code than I do in the math printed there.
(2) With respect to my question (5), I tried to leave the program running for about 16 hours and the program reached 7640 steps, however, I still can not see anything in the two figures on my screen.
I don't know what to tell you. If I define "anything" as "it's not a circle", then I can see signs of anisotropy from very early. If you drop the mesh down to 100x100, you can see the beginning of side- branches at about 1000 steps (it should happen at pretty much the same time for 500x500; I just don't want to wait that long).
I haven't done a detailed error analysis, but if you change the ExplicitDiffusionTerm to an ImplicitDiffusionTerm, you can raise the time step at least an order of magnitude and still get reasonable looking answers.
Could you please have a look and see if there is anything wrong in the program I sent to you previously?
I don't see anything obviously wrong with your script.
(3) You mentioned Ryo Kobayashi’s Fortran code and I am wondering whether his Fortran code is accessible somewhere to the public?
Nope. You'd need to contact Ryo about that.
Just for interest, for this particular program, how long it takes for his Fortran code to run?
Chapter 5 "Efficiency" is a head-to-head comparison of FiPy against Ryo's code.
-- Jonathan E. Guyer, PhD Metallurgy Division National Institute of Standards and Technology <http://www.metallurgy.nist.gov/>
