Discussion Forum

I'm trying to get a simple turbulent fluid problem to converge. The geometry involves a water jet entering a larger cavity.

With a stationary analysis, the result appears to be converging relatively smoothly for some number of iterations, but then suddenly begins diverging. I've attached an image of one such convergence graph; in many cases the divergence is much more dramatic with error values jumping up and down by huge amounts every iteration (it often looks like a sawtooth wave in the diverging region). The error is generally something like this:

Failed to find a solution.
In Velocity u, Pressure p:
Divergence of the linear iterations.
There was a warning message from the linear solver.
Ill-conditioned preconditioner. Increase factor in error estimate.
Returned solution is not converged.

I strongly suspect that there's some oscillation such that there is no steady state solution. Because of that, I've attempted to use a time-dependent study instead of stationary, but the time-dependent study has EXTREMELY high error values (orders of magnitude higher than the stationary error values) and those errors never converge for as long as I've left it run. These convergence plots are discontinuous, and values jump randomly instead of forming a continuous line.

For comparison, the first iteration on the segregated solver for the stationary study has both u/p and turbulence error values less than 1, which then decrease over time (until they suddenly and violently diverge). The time dependent study, on the other hand, starts with the u/p error above 10^9.

I've attached the .mph file; any help is appreciated.