[jameswilson620]

Hello all,

I would like to know how to solve the following equation when m_o < 0

## CODE BELOW##
dimensionedScalar D_dt(transportProperties.lookup("D_dt"));

fvScalarMatrix m_o_diffuseEqn
(
D_dt*fvm::laplacian(m_o_diffuse) == fvm::Sp(1,m_o_diffuse) - m_o
);
m_o_diffuseEqn.solve();
## CODE ABOVE##

where m_o is known (volScalarField) and m_o_diffuse will be determined. D_dt is simply a constant with units of m^2

I should say this solves just fine when m_o > 0. The code still compiles and solves when m_o < 0 the problem is that m_o_diffuse is identical to m_o i.e. nothing happened.

I've tried fvm::SuSp and many combinations of the source term handling features but OF doesnt seem to recognize anything:

error: no matching function for call to ‘SuSp(int, Foam::volScalarField&)’

I wouldnt be posting this if I haven't already tried applying information ive read in other posts. I simply cannot figure this one out!

Thanks, James

[hjasak]

You've got sign problem.

First, your equation should look like this


fvScalarMatrix m_o_diffuseEqn
(
- D_dt*fvm::laplacian(m_o_diffuse) == m_o - fvm::Sp(1,m_o_diffuse)
);
m_o_diffuseEqn.solve();

That's the same, but Sp and SuSp expect a positive diagonal (and yours was negative).

I would also consider solving for the variable of opposite sign, to make sure it remains positive. This definitely gets me to a known territory (elliptic relaxation). I think the negative variable should also work, but the meaning of source and sink terms etc. then becomes a bit weird.

Hrv

[jameswilson620]

Excellent! I will try this configuration. Also, another problem is that m_o can be both positive and negative. The implementation is not yet clear to me if the field has both mixed positive and negative values.

Perhaps a superposition of the negative portion on the RHS and the positive values either on the LHS or in an SuSp(1,m_o_positive) on the RHS of the equation you've provided? This is of course, where the summation of the split field results in the original field.

Thank you,

James

[jameswilson620]

Amendment to last post:

The superposition of the two requires zeros in certain location s such that the sum of the two fields forms the original field. I would use either

+eps on the LHS instead of 0

And - eps for the other.

Would this work to eliminate divide by zero errors?