adi + matlab
Now I am studying about the 3D ADI-FDTD method in PML to do the simulation of EM. Because, it is spent so much calculating time to do the simulation by using FDTD.
But I find it is so difficult to write the Matlab code.
I check the previous topics, but no one have,
now, two years have passed, Is anyone finished the code.
please help me.
-------3D ADI-FDTD in PML Matlab code
Thank you a lot.
word,pdf,M-form OK!
my Email: tianye821220@yahoo.co.jp
What paper/book did you use?
I tried it some years ago but the computation time for each cell was quite a bit larger
than for standard FDTD. For large timesteps which made ADI-FDTD faster the error
was getting to big.
For me optimizing the speed of my FDTD code (and parallelizing it) turned out to be
more useful in terms of speed.
In addition it is more difficult to add other features (like dispersive materials) to
ADI-FDTD than to standard FDTD
So if speed is your only issue I would suggest trying to speed up my fdtd code first.
mr iyami,if possible than please upload ur codes here
sorry I cannot (commercial code). In any case it is one of the older
ADI methods. I believe there are better (lower error) ADI methods
available now.
But let me repeat I believe ADI is of limited use. There are model where
it may work well but I would only go through the pain of implementing it/
adapting someone else's code if I were fairly sure that the application
fits ADI well.
thank you mr iyami,for ur valuable suggestion,i want to just know,what is the actual meaning of ADI-FDTD,i think u know this better,if possible please refer some site or book by which i clearify my doubt.
This is for Iyami - what is your opinion about recent trend of researches on implicit FDTD (CN, LOD etc). Are they fit for real-life applications or what is their future? Thank you very much.
for smruti:
ADI = alternating direction implicit
in standard fdtd you get the equations:
E^{n+1} = c_1*E^n + c_2 DEL x H^n
H^{n+1} = ... E^n
which means you compute the new E-field from old E and H values.
That is to say the equation is explicit.
Now if (some) of the values on the right hand side were also step n+1
values then you can run into trouble because ADI would need to update
E before H and vice versa. The equations are implicit because you cannot
write down simple update equations. The only solution which remains is
to use matrices.
ADI is a clever way to mix explicit and implicit steps to
1) get unconditionally stability form the implicit equations and
2) rather nice (tridiagonal) matrices for the update step.
The one big problem (classical) ADI has is that at material interfaces bigger timesteps
can lead to pretty big errors
As for references: there is a section on ADI in Taflove. Also Namiki's papers which
somebody else posted a short while ago are pretty accessible I think
for confi999
I have not tried them yet. The vague impression I got is that
a) CN may have mostly solved the large error at material interfaces at larger time step so it is
definitely worthwhile investigating
b) LOD is just a more efficient way to compute ADI (I have not read any paper yet only abstracts).
So I would try CN before LOD but mostly the question is what model you have. If you have lots of
fine geometrical details then I would not use these methods. However if you have rather simple
models with large uniform regions then I would give them a try.
they probably are but the real difficulty is to determine whether they are appropriate for a
concrete model.
for iyami-
thanks,i will check the chapter or section of taflov.
Thank you very much iyami.
mr iyami,
if possible than please upload adi-fdtd 1d written in matlab code,for understanding the process clearly.
actually if possible than download it from www.pubn.com,because i cnt download that,and i really want to see the code,please help me.