[C. Okubo]

Hi!

People using adjointShapeOptimizationFoam, have you ever noticed that the adjoint velocity vectors of the very first line of cells are usually pointing to an akward direction? I mean, I know adjoint velocity is not phisically something we can interpret, but this behavior is really strange to me. I would expect something more continuous. Seems like some discontinuity, source or something is acting. Please, see figure 1 of an example (this is the first iteration of an example I am running).

The left faces are the inlet of my domain (the only faces with arrows) and vectors are adjoint velocity field. Note that, for the next cells (second column from left to right), things seem to be back to expected behaviour. So, when this velocities are weird (as the picture), as sensitivities are scalar product of both fields (primal and adjoint), the method starts to block the entrance cells and that is where the problem starts: adjoint field diverges and I can't get a clean solution.

The original code removes the Adjoint Transpose Convection at inlet cells, but this is not exactly OK and I am trying to not use this.

Note: all examples shown here were generated with the ATC being calculated in the whole domain, so I am not removing it.

Something interesting I found out is that, when I use relaxation factors below, I get the result from figure 1.
pa = 0.3
Ua = 0.7

But, if I invert them:
pa = 0.7
Ua = 0.3

Then, I have figure 2, which is more according to what is expected.

Even though I am being able to have good results using cell limiters on grad(Ua), I think the best I could have until now is with the use of the flow rate inlet BC and relaxation factors higher for pa than Ua. Please, see BC example below and figure 3 of the same case (without grad limiter or removal of ATC).

inlet
{
type flowRateInletVelocity;
massFlowRate 0.01;
value uniform (0 0 0);
rhoInlet 1.0;
extrapolateProfile yes;
}

This last combination (relaxation factor + flow rate at inlet) gives a clean solution without removal of ATC.

I would really appreciate any comments of the more experienced people on this.

Best Regards,

Okubo

[C. Okubo]

Forgot the figures!