2006/12/29, Stephane Del Pino <[EMAIL PROTECTED]>:
Le jeudi 21 décembre 2006 09:15, Emilio Melero García a écrit: > Dear Stephane, Dear Emilio. > thank you so much for your help and patience. I will certainly increase the > time step. It is a very good solution even with the memory leak since i do > not need to go foward in time so much (no more than 13 days, which are only > 936 iterations with dt=1200). > > By the way, and i understand it is not the place to ask these, and that you > might be very busy to answer, the stability condition that you mention and > the residual of the conjugate gradient is due to the problem i am trying to > solve or is inherent to the PDE solver you have implemented?, and if it is > problem dependent, could you hint me on how did you evaluate the residual > and the stability? First the residual: - it is just written by ff3d when the default verbosity (3) is used. - for your problem, it is always small because your time step so small that the difference between u(n+1) and u(n) is very small... Second concerning the stability: The time scheme you use [u(n+1)-div(dt*grad(u(n+1))) = u(n)] is unconditionally stable. This means that you can choose the time step arbitrarily... > And finally, i am trying to make my way into all this > world of FEM problems and solvers, but i still have a notion that i might > be resorted to implement my own very specific FEM code, any good book you > would recommend to learn FEM methods and implementations? Well there are many good books for finite elements. I recommend the book from Ciarlet for the basis of the theoritical point of view. For the implementation, there are a lot that depend on the physic you want to simulate... Anyway ff3d might be the solution to your problem: can you describe the specificities you are facing? Also, some people have already done it: you can modify ff3d to fit your needs and if that can be useful for others you can then distribute your changes, ff3d is distributed under the GPL... > Please, as i said i understand this might be not the place to ask these > things but i have felt that people in these list read the e-mails and > answer when they can. So please, consider this a question to the people in > the list. No, I think it is a right place to ask this kind of question! Feel free for that. > Thank you very much from a newbie. You are welcome! Best regards, Stephane. PS: sorry for this late answer, due to holidays ;-) _______________________________________________ ff3d-users mailing list [email protected] http://lists.nongnu.org/mailman/listinfo/ff3d-users
Dear Stephane, Thank you very much for your reply. Since you ask for the details of my problem i will give them to you, and of course to everyone in the list in case they know of someone sharing my interests. I mentioned in a previous mail that my problem concerns crystallisation but i will extend now a little bit. I would like to model and simulate a counterdiffusion crystalisation experiment. In this kind of experiments what one has are two solutions separated by a porous media. In my case the solutions are of a protein which i want to crystallize, and of a precipitant agent which is usually a much smaller molecule than the protein. Due to the difference in size one can consider the protein solution esentially frozen, with the precipitant molecules diffusing through the porous media into the proteing solution. So far is just solving a diffusion problem which is how far i have managed to get with freefem3d (apart from that memory leak i had with the code, which i have thought maybe it is because "merge" and "union" may have some difference at the end¿?¿?, i will try to check into that). I would also like to include a gravity driven convection term at some time. However, for the future, i would like to implement a more complex model. What happens when the precipitant molecules diffuse through the protein solution is that the local solubility of the protein is changed (diminished), since it is a function of the concentration of protein and precipitant. When the solubility, or better, the supersaturation (a related function, depending of solubility, concentration of protein and of precipitant) reaches a certain level there is a probability of a crystal seed being created, that depends on how much is the supersaturation above the stablished level. So far i could do also this with freefem3d since it only involves evaluation of function defined inside a domain, or functions created from "femfunctions" (of course it would help a lot if the random number generator from C++ was implemented in freefem3d). However, the problem arise when i want to consider that my newborn crystals are physical objects, through which diffusion can not occur for example. To model this i would need to be able to redefine the domain (and remesh it) in run time, being able to create or erase elements, even simple ones like boxes or spheres, from it for example. I don´t know if someone has encountered such a need, and i would like to contact anyone that could point me to an appropriate tool. Unfortunately i think my programming skills are somehow below of what it takes to modify, for example, the ff3d code. Although if there is no other solution... I was wondering about interfacing with other more flexible mesh generators, or reading files created with other CAD tools. How are the capabilities of ff3d in this area? Well, i think is enough for one mail. Sorry if it became too long. Best regards and best wishes of a happy new year to you all!! Emilio -- ############################################### Dr. Emilio Melero García Laboratorio de Estudios Cristalográficos, IACT Instituto de parasitología Lopez Neyra, P.T. Ciencias de la Salud, E-18100 Armilla, Granada, España Correo elec. :emilio_arroba_lec_._csic_._es Tfno: 958181621, lab 107 ###############################################
_______________________________________________ ff3d-users mailing list [email protected] http://lists.nongnu.org/mailman/listinfo/ff3d-users
