Cubic Spline/Verlet simulation Instabilities
Hello everyone,
I'm working on a 2D simulation of a stepped spillway and have encountered instability problems when using the Cubic Spline Kernel and Verlet step algorithm. To the best of my description I would say the simulation simply "exploded" at a certain time (after about 40-50 seconds of simulation time) and send particles all over the domain making it impossible to continue the run. The exact same simulation runs fine with the Wendland kernel and Simplectic step algorithm.
Has anybody encountered the same problems?
One added layer of doubt to me is that I altered a portion of the source code to impose a certain velocity at a certain [x,y,z] position of the flow (as a inflow condition + periodic boundary). What I've altered is inside the JSphGpu::ResizeGpuMemoryParticles rotine in JSphGPU.cpp. The added part has no influence on kernel or time step procedures as far as I can tell (also as I said Wendland and Simpletic work just fine) but maybe there are some unseen (to me) problems.
Also, just as extra info, the experimental data is in good agreement with with the simulation for several different discharges and positions along the spillway (all far from the imposed inflow velocity). I just would like to use Cubic Spline/Verlet to reduce runs times.
Any thoughts?
<?xml version="1.0" encoding="UTF-8" standalone="no"?>
actual | dp | fluid | bound | void
% Note alpha can depend on the resolution. A value of 0.01 is recommended for near irrotational flows.
I'm working on a 2D simulation of a stepped spillway and have encountered instability problems when using the Cubic Spline Kernel and Verlet step algorithm. To the best of my description I would say the simulation simply "exploded" at a certain time (after about 40-50 seconds of simulation time) and send particles all over the domain making it impossible to continue the run. The exact same simulation runs fine with the Wendland kernel and Simplectic step algorithm.
Has anybody encountered the same problems?
One added layer of doubt to me is that I altered a portion of the source code to impose a certain velocity at a certain [x,y,z] position of the flow (as a inflow condition + periodic boundary). What I've altered is inside the JSphGpu::ResizeGpuMemoryParticles rotine in JSphGPU.cpp. The added part has no influence on kernel or time step procedures as far as I can tell (also as I said Wendland and Simpletic work just fine) but maybe there are some unseen (to me) problems.
Also, just as extra info, the experimental data is in good agreement with with the simulation for several different discharges and positions along the spillway (all far from the imposed inflow velocity). I just would like to use Cubic Spline/Verlet to reduce runs times.
Any thoughts?
<?xml version="1.0" encoding="UTF-8" standalone="no"?>
actual | dp | fluid | bound | void
% Note alpha can depend on the resolution. A value of 0.01 is recommended for near irrotational flows.
Comments
Your XML is not shown here since some symbols are not working in this text format.
Verlet time algorithm is only recommended for previous tests since the simulation is faster. However a predictor-corrector step algorithm is suggested for long and stable simulations such as the symplectic one. You should always use this order of integrators as minimum. Verlet is only for tests.
Regards
Regards,