I am reading the paper, http://math.mit.edu/~gs/cse/codes/mit18086_navierstokes.pdf

I want to solve an optimization problem using this Navier-Stokes code as constraining PDE. Now I am trying to add an external force parameterized some input parameter to the method. The method solves the Navier-Stokes equation in 3 steps, 1) nonlinear term, 2) linear term, 3) pressure correction. $$\text{(1)} \quad \frac{U^*-U^n}{\Delta t} = -((U^n)^2)_x - (U^nV^n)_y$$ $$\text{(2)} \quad \frac{U^**-U^*}{\Delta t} = \frac{1}{Re}(U^{**}_{xx}+U^{**}_{yy})$$ $$\text{(3)} \quad U{n+1} = U^{**} - \Delta t \nabla P^{n+1}$$ (written only for U)

Should I add the external force term to the step 1 like this? $$\text{(1)} \quad \frac{U^*-U^n}{\Delta t} = -((U^n)^2)_x - (U^nV^n)_y +f $$

But then, will the external force corresponds to the input be modified to satisfy the incompressible condition at step 3?

If I give some physically impossible external force term (like too much force), the method will automatically decrease it to satisfy the incompressible condition?


1 Answer 1


I would not bother too much. Chorin's method is known to be a consistent scheme, which means, that for smaller time-steps it will approximate the actual solution. Checkout Gresho's/Sani's book Incompressible Flow and the Finite Element Method. Vol. 2: Isothermal Laminar Flow there is the convergence theory and also Chorin's method with a forcing term.

So, just put your force into $(1)$ or $(2)$ -- there is no difference.

The last step $(3)$ is a projection step that projects your tentative velocity $U^{**}$ to a divergence-free corrected velocity. It will remove certain parts of $f$, namely those that entered the gradient-part of $U^{**}$. In fact, if $f=\nabla \rho$, where $\rho$ is suitable scalar function, then (in theory) $f$ should never appear in the velocity and then, at best, also not in $U^{**}$.

So, to address your last points... It's not about too much force, it is about the direction of the force (whether it is a gradient for example). And Chorin's method will project out the gradient parts up to the consistency error.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.