Page 387 - Process Modelling and Simulation With Finite Element Methods
P. 387
374 Process Modelling and Simulation with Finite Element Methods
vector calculus, since much of the higher theory was developed to treat
intentionally idealized models. Nevertheless, the basics of vector calculus are
necessary to understand what such pde engines do, and how they do it. Theory
simultaneously becomes more important in some aspects - dealing with
complexities that are still beyond computability, proposing physical models that
are amenable to numerical computation - but also less necessary for “run-of-the-
mill” applications. Theorists should be challenged that they must remain ahead
of the game to still be relevant practioners due to the advent of general purpose
solvers like FEMLAB.
Epilogue (WBJZ)
As an epilogue to this book, I have found that MATLAB programming is not
really essential to modeling with FEMLAB. The GUI, with experience, serves
for most purposes rather well. I believe that many experienced FEMLAB users
are amazed at the flexibility I have teased out of the FEMLAB GUI. So
MATLAB programming is really essential in only a few cases: (1) Massively
parallel parameter space studies; (2) Non-PDE models of nonlocal, discrete
coupling; (3) Moving boundaries; (4) Getting the detailed data out for post-
processing. This list is probably not complete. I am not sure that FEMLAB
developers envisaged people would do (1)-(3) with their tools. (4) should be
addressed in later editions of FEMLAB. An I10 wizard of some sort to read in
data into initial conditions or functions and to write output into standard formats
would go a long way in removing the need to write your own MATLAB
code. Until then, information concentrated in this Appendix on the
MATLABIFEMLAB interface, and sprinkled throughout the book in worked
examples, will prove invaluable to the budding expert user.
AS End Note: Platform Dependence of Meshes
I could not think of a good place to put this note. As you might have noticed, the
book was developed under both FEMLAB 2.2 and FEMLAB
2.3LCS/2.3/2.3a/2.3b, with either MATLAB 5.316.116.5, while using both
Microsoft Windows and linux operating systems. This is bad news in that your
MAT-files are not necessarily compatible across versions and platforms. So the
advice here is to save your models as m-files for cross platform transfer. But in
many cases, it is the results of the computations that are required to cross
platforms. For instance, for us it was useful to do long computations on our best
linux workstation as background jobs, then save the solution (the whole fem or
xfem structure) and read it into the MATLAB workspace, and upload the
relevant parts into the FEMLAB GUI (see all those import options on the file
menu). For several of our models, we found that the same sequences of m-file

