Diagonal dam break simulation with the shallow water equations. Here one side is initialised to 0.1m and the other to 0.01m. The main wave becomes oriented in the lengthwise direction but there is some interaction with the spanwise wave. Fortran code here github.com/garcfd/shallow…

converted the Lax-Wendroff shallow water equations to an Euler solver, but its too oscillatory, the convergence is poor, and there is some weird discontinuity going on at the inlet... apart from that its 'great'

2D Lax-Wendroff solver at Mach 0.9 runs on GPU

Mach 0.9 flow on GPU takes only 2minutes to do 200k iterations. Interesting 2D case which takes a while to stabilize. No smoothing used but there is a source term needed to combat the instabilities. Code here: github.com/garcfd/lax-wen…

I always enjoy the opportunity to look at density gradients in #Paraview using the python calculator filter, and specify something like this for the expression: log10(mag(gradient(dens)+0.001))

a comparison of M=0.9 (upper) and M=0.8 (lower)

Mach 0.95 case runs in 1 minute on dicehub

wow - I've never seen a convergence plot like that before - I wonder what happened at ~2500...?

Mach 0.95 - I extended the domain to 500x200 cells, and ran it longer for 400k iterations - but it really seemed to converge rapidly when the shock hit the exit plane. I had an if/then condition at the exit plane for both subsonic and supersonic flows.

Mach 0.95 extended domain and duration

Just been using #UFOLBM for vehicle aerodynamics and noticed how the flow field was rather oscillatory. Tried increasing the viscosity (smag), and then tried extending the domain upstream and downstream to resolve the issue. Here are the results, thanks to Jonas Latt for the help

#UFOLBM Ford GT40 at 30m/sec - 96 million cells 200k iterations. Ran on single T4 GPU via dicehub, runtime 15 hours. VTK files saved every 1000 iterations. Here showing eddy viscosity contours. (Re = 1.8million)

#UFOLBM Ford GT40 at 30m/sec - 96 million cells 200k iterations. Ran on single T4 GPU via dicehub, runtime 15 hours. VTK files saved every 1000 iterations. Here showing velocity magnitude contours (Re = 1.8million).

Just made some corrections to #UFOLBM to ensure the new "wall reflection" method runs as fast as it can compared with the "bounce-back" method. Re-running this demo case on dicehub to check the results look ok.

#UFOLBM eddy viscosity contours Ford GT40

#UFOLBM another comparison of bounce-back (top) with wall reflection (bottom). 96M cells, 100k its, 5.5 hours runtime, on single T4 GPU via dicehub

#UFOLBM bounce-back vs wall reflection (eddy viscosity)

Linux is so much fun when you can use simple commands to search a load of files for a single expression: grep -i momentumexchange $(find -name *.h)