ffc team mailing list archive

["Alessio Quaglino" <alessio@xxxxxx>]

Yes, the final line of the expression should have *_trial instead of
*_basis. Ok thanks, can I call some specialized function of BLAS for the
matrix product using dolfin or do I have to define a child matrix class
derived from uBlasSparseMatrix?

Thanks,
Alessio


> I'm not sure I follow your notation completely, but no you can't write
> something in FFC that is the product of two integrals. You need to
> compute each integral separately (or find a clever way to rewrite it as
> a single integral over the domain).
>
> /Anders
>
> Alessio Quaglino wrote:
>> I was confusing with the notation not distinguishing between S as a
>> function or as its coefficients. What I'm thinking is to write first the
>> unknowns S and u using the basis functions (indicated with the name of
>> the
>> test function belonging to the same space):
>>
>> S = S_coeff psi_trial
>> u = u_coeff phi_trial
>>
>> so that we have:
>>
>> A = dot(S, grad(phi)) =
>> = S_coeff (psi_trial, grad(phi)) =
>> = dot(u, psi) * (psi_trial, grad(phi)) =
>> = u_coeff * dot(phi_basis, psi) * (basis_psi, grad(phi))
>>
>> the problem is that the two terms have first to be integrated and then
>> multiplied. Would that be possible?
>>
>> Thanks,
>> Alessio
>>
>>
>>> I don't see how to do this. As I understand, you have two matrices and
>>> you want to compute their product. So you want to compute a sum of
>>> products of integrals. I don't know how to write this as a single
>>> integral (other than as a "double integral" over the square of the
>>> domain).
>>>
>>> What is the quantity you want to compute in the end? Can you write it
>>> as
>>> a single integral?
>>>
>>> /Anders
>>>
>>>
>>> Alessio Quaglino wrote:
>>>> I have a question about matrix multiplication in FFC. Say I want to
>>>> measure a certain quantity S in respect to a test function psi
>>>> belonging
>>>> to the same finite space:
>>>>
>>>> S = dot(u, psi) (1)
>>>>
>>>> and then I want to use this measure to assemble the bilinear form
>>>> where
>>>> phi is taken from the same finite space of u:
>>>>
>>>> A = dot(S, grad(phi)) (2)
>>>>
>>>> this is equivalent to assemble the bilinear form (1) obtaining the "m
>>>> x
>>>> n"
>>>> matrix S and then the bilinear form (2) obtaining the "n x m" matrix
>>>> A.
>>>> Then what I need is to perform the multiplication A*S getting a "n x
>>>> n"
>>>> matrix. Can I write A in such a way that this is done directly in the
>>>> form? Do I have to define a trilinear form?
>>>>
>>>> Thanks,
>>>> Alessio Quaglino
>>>>
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> FFC-dev mailing list
>>>> FFC-dev@xxxxxxxxxx
>>>> http://www.fenics.org/mailman/listinfo/ffc-dev
>>
>>
>> _______________________________________________
>> FFC-dev mailing list
>> FFC-dev@xxxxxxxxxx
>> http://www.fenics.org/mailman/listinfo/ffc-dev
>