Integral in parallel

General Discussion
1

Hello,

When computing an integral in parallel, I know one needs to scale things and do a global reduction. There are examples in previous posts (for example here) with the integral of 2 vectors, say:

Vh<complex> scaledVecu = Vecu;
for[i, v : A.D] scaledVecu[] *= v;
complex wrong = int2d(Th)(scaledVecu*AnotherVec);
complex reduce;
mpiAllReduce(wrong, reduce, mpiCommWorld, mpiSUM);
if(mpirank == 0)
        cout << reduce << endl;

My question is: if there are more than 2 vectors in the integral, should the rescaling still apply to only one of them?

Thank you.

2

This is in fact independent of the number of components. Keep in mind that in FreeFEM, whenever you do u[], it selects all degrees of freedom. So whether u is a function defined on a fespace Vh(Th, [Pk, Qk, Tk, Zk, Wk]) or fespace Vh(Th, Pk), the syntax from your snippet will indeed scale all values of scaledVecu (proper things to do).

3

Thank you for your reply. Sorry, my question was not clear: I was not talking of the number of components (for instance [ux] vs [ux,uy] vs [ux,uy,uz] etc.), but the number of terms that are multiplied in the integrand: int2d(Th)(u*v) vs int2d(Th)(u*v*w).

4

I see. Only one of the term has to be scaled.

5

Great. Thank you very much.