Open schmie opened 9 years ago
Comment by miguel on 5 Aug 2013 08:08 UTC Unfortunately, there are some stability considerations in the lattice-Boltzmann algorithm that don't allow the choice of arbitrary timesteps unless the spatial discretisation is modified accordingly. Timm wrote a very good document describing this (attached to the ticket). I suggest you read it at some point this week and discuss it with him or me.
Why do you need to increase the number of cyclesteps? If you want to run longer simulations, keep cyclesteps constant and increase the number of cycles.
I hope this helps. Let me know if you have anymore questions.
Comment by miguel on 5 Aug 2013 08:11 UTC The attachment is too heavy for Trac. I just e-mailed it to you.
Reported by david on 2 Aug 2013 14:17 UTC The error:
!! Maximum relative density difference allowed 5.0% was violated: 11.9% !! !! Simulation was unstable !!
Code is crashing for runs of the code both with SSE enabled and disabled. They occur when input data's config file is changed to use large timesteps. To recreate this problem: Attempt to run the code(both with SSE ON or OFF) after modifying the input_data config file for 128k data points. Change the cyclesteps="1000" to cyclesteps="16000". Run in parallel on 16 cores for D3Q27 geometry. The problem occurs when cyclesteps is large, and stops if cyclesteps is smaller (running with 2000 was succesful).