does anybody have experience in pseudospectral methods in matlab? help me thanks

I have some experience in pseudospectral methods, but not in matlab...

does this helps?

Cheers,Patrick

I have experience both in F90 and MATLAB. Email me for help. -- Sujit

The problem is I computed fft and ifft using the matlab inbuilt functions. When I am doing the differentiation of the function and inverse transforming it. I get real and imaginary values. But my input is only real I am wondering how can real input fater diff.. and doing ifft will give imaginary values. please help..

I suggest you post this question to the comp.soft-sys.matlab newsgroup. There's loads of DSP gurus on there who'll help for free.

Its easy:

MATLAB does a Forward FFT by the typical real to complex algorithm. Similarly you can do n-dimensional complex to complex forward transforms using fft().

To do an inverse transform, use ifft(). ifft() can do a complex to complex and complex to real inverse transform. If the data on inverting is purely real, the MATLAB algorithm may return small imaginary parts which can be zeroed out.

If you want me to go into details about how modes are stored, let me know.

--Sujit

Hi,

You got some answers concerning matlab, concerning Fourier pseudospectral methods:

Use FFT where there is a place for the real and complex part of the vector. Put the function in the real part and put 0 (zero) in the complex one. Do your FFT forward and then multiply the results by -ij, where i=sqrt(-1) and j is the discrete index number. This means that the complex value (obtained from the FFT) will now be real and that the real value (obtained from the FFT) will now be complex. Then put these back in you vector, taking care of putting the new complex and the new real in their appropriate place in the input vector. Then carried out an inverse FFT of that input vector and consider only the real part of it as your Fourier derivative. It is very important to take care of identifying correctly what is j in your program (subroutine) since you have to multiply by -j between the two FFT, and will you have also to divide by N (the number of points; k=0,1,..N/2) to normalize properly. This is if you work on a [0:2pi] segments.

I am copying here below a FORTRAN subroutine for what I just wrote for Fourier pseudospectral solver, using the Fortran FFT

I hope this helps, Patrick

subroutine pdfrnt2(f,df,if) integer i,n,if,j,it,nt,ir,nr include 'input.par' parameter (n=nt,nr=2**ir) real*8 f(n),df(n),c2(2,n),c1(2,n) real*8 fil(0:n/2),cut(0:nr),dn common /filter/cut,fil dn=dfloat(n) do i=1,n c1(1,i)=f(i)/dn ---> real value divided by N c1(2,i)=0.d0 ---> imaginary part end do call DFOURT(c1,n,1,-1,0,0.d0) ---> FFT is return in c1 do i=1,n/2+1 j=i-1 c2(1,i)=-dfloat(j)*c1(2,i) ---> multiply by -ij c2(2,i)=dfloat(j)*c1(1,i)---> multiply by -ij end do do i=n/2+2,n j=i-n-1 c2(1,i)=-dfloat(j)*c1(2,i)---> multiply by -ij c2(2,i)=dfloat(j)*c1(1,i)---> multiply by -ij end do call dfourt(c2,n,1,1,1,0.d0)---> back FFT into c1 do i=1,n df(i)=c2(1,i)---> here is the Fourier derivative end do return END

How to do differentiation is there any command in matlab. I am getting the complex values of this diffrentiation(of real values) it is not possible. I don not know where is the error. Do you have literature pf pseudospectral methods in matlab. I want to disscuss this issue in deep. thanks prasat

Although I don't understand what your question is, this is what I think you should be doing.

=>Take a discrete real function (1D, 2D, 3D, etc)

=>Take it into Fourier Space by MATLAB function fftn()

=>To differentiate a function in Fourier Space simply multiply each coefficient by - i * (wavenumber in the direction of the derivative)

=>To get back the differentiated function on Physical space use MATLAB function ifftn() and neglect any imaginary part returned.

Hope this helps.

A quick web surf gave me this link:

http://www.onid.orst.edu/~weidemaj/differ.html

Or how about:

http://web.comlab.ox.ac.uk/projects/...ectra/eigtool/

Hai sujit, Whya should I neglect the complex part after getting the diff in physical space. Lets suppose that I have neglected it but then How will I confirm that my diffentiation is correct. That is my basic questions Prasat

MATLAB's documentation (type : help ifftn) says that the FFT works such that the same function ifftn does both the inverse complex-to-complex and complex-to-real-symmetric transforms.

In case of the latter, you get back complex vectors (matrices) with very small imaginary parts, which you can throw away.