[sammy2]

I'm trying to implement a simple minmod limiter on my primitive variables.
It (seemingly) works perfectly when advecting a square pulse (u=p=const) but is worse than piecewise constant at Sod's problem (see pics).
Here is how I'm calculating the edge states:

leftDiff = rho[i-1] - rho[i-2];
rightDiff = rho[i] - rho[i-1];
left.rho = rho[i-1] + 0.5 * minmod(leftDiff, rightDiff);
//...

leftDiff = rho[i] - rho[i-1];
rightDiff = rho[i+1] - rho[i];
right.rho = rho[i] + 0.5 * minmod(leftDiff, rightDiff);
//...

Does anyone have any idea where I'm going wrong? I've been stuck on this for weeks.

[mprinkey]

I guess it depends on what you are doing in your "minmod" function. I am used to seeing fluxes interpolated based on:

phi_f = phi_0 + 0.5*psi(r)*(phi_0 - phi_-1)

where r = (phi_1 - phi_0)/(phi_0 - phi_-1)

and for minmod, psi(r) = max(0,min(1,r))

(This all assumes evenly spaced grids. Also, I am used to using slope limiters on unstructured meshes, so my flux-limiter Fu is a little rusty.)

if your minmod performs those same calculations and returns a dimensional value for density, then it should be right. But r in the forms I am used to seeing is dimensionless, as is the return value of minmod(). So, I'd expect a trailing factor of units of density to make it physical in your right.rho and left.rho lines.

[sammy2]

Originally Posted by mprinkey

I guess it depends on what you are doing in your "minmod" function. I am used to seeing fluxes interpolated based on:

phi_f = phi_0 + 0.5*psi(r)*(phi_0 - phi_-1)

where r = (phi_1 - phi_0)/(phi_0 - phi_-1)

and for minmod, psi(r) = max(0,min(1,r))

(This all assumes evenly spaced grids. Also, I am used to using slope limiters on unstructured meshes, so my flux-limiter Fu is a little rusty.)

if your minmod performs those same calculations and returns a dimensional value for density, then it should be right. But r in the forms I am used to seeing is dimensionless, as is the return value of minmod(). So, I'd expect a trailing factor of units of density to make it physical in your right.rho and left.rho lines.

I should have said that I am limiting slopes here not fluxes (also on an evenly spaced grid). Therefore my minmod function is simply

[naffrancois]

I think your problem does not come only from 2nd order reconstruction. Your 1st order solution does not look TVD to me, large oscillations near the shock do not occur when using shock capturing upwind fluxes. What spatial discretisation are you using ?

[FMDenaro]

I agree, Godunov must give a monotonic solution

[sammy2]

Originally Posted by naffrancois

I think your problem does not come only from 2nd order reconstruction. Your 1st order solution does not look TVD to me, large oscillations near the shock do not occur when using shock capturing upwind fluxes. What spatial discretisation are you using ?

I'm using finite volume. What kind of things can I do to find the problem?

[naffrancois]

What Riemann solver are you using ? If it is the exact or Roe flux, maybe you should start with the simplest one: the Rusanov flux. As it seems that the raréfaction and contact are Well captured it may be à bug with the définition of left and right pressure as they are discontinuous across the shock

[sammy2]

Originally Posted by naffrancois

What Riemann solver are you using ? If it is the exact or Roe flux, maybe you should start with the simplest one: the Rusanov flux. As it seems that the raréfaction and contact are Well captured it may be à bug with the définition of left and right pressure as they are discontinuous across the shock

I am using the exact solver found here with their function sampleWaveSolution (with wave_speed = 0.0) which I don't really understand but I couldn't find any simple ones in C/++.

[naffrancois]

Hello,

I haven't looked much the code you are referring to. I can't recommend you taking pieces of code as a black box, the mistake could be in that code, the way you use it or both.

It is worth taking time to learn how the exact solution is built (Toro, Riemann Solvers and Numerical Methods for Fluid Dynamics is one of the references) and how approximate solvers have been derived. This way you will understand the code you mention and find what is wrong.

Cheers,
François

[sammy2]

I have tried another solver which is producing the same results..
It's a bit of a long shot but does anyone recognise what is happening in the Godunov vs the minmod plot?