Skip to main content Features Implemented
- Solvers
- EOS SPH Solver
- EOS with Splitting
- Iterative EOS with Splitting
- Different Equations of State
- IISPH (2013, Notes)
- Datastructures
- Index search - regular grid
- XYZ, Morton, Hilbert curves
- GPU version with parallel radix sort and reductions
- Experiments
- Oscillation frequencies over stiffness
- Error over stiffness
- Stability over viscosity, timestep size
- Aliasing and stability improvements with jitter
- Initialization
- Oblique and regular grid initialization
- Jittered initialization
- Non-uniform mass, uniform density solver (-> influence on viscosity?)
- GUI/Visualization
- Off-thread, GPU accelerated rendering
- Atomic parameters
- SPH “number density” using OpenGL quads and alpha blending
- Miscellanous
- arbitrary non-uniform boudnary handling (-> tests?)
- Viscosity-based adhesion
- GPU implementation of EOS SPH
IISPH v2
- 2013 Paper version more numerically stable than Notes version?
Improving Robustness
- minimum/maximum timestep relevant (stability vs. shock problems)
- higher \(\epsilon\) for \(A_{ii} \neq 0\)
- maximum compression over average compression as convergence criterion (but higher iteration count)
- lower \(\omega\)
- higher (minimum) iteration count
- jittered boundaries and spacing not a multiple of \(h\) - less porous
Observations
- iteration count coupled to compression:
Stability
- problems seem largely related to boundary quality
- occur mostly at the start (better equilibration? remove outliers?)
- also, velocity field is noisy:
Improvements?
- other solver?
- regularization?
- stochastic behaviour?
- preconditioning?
- other initialization?
- better parameters \(\epsilon, \omega, \text{miniter}, \dots\)
Other Questions
- meaning of diagonally dominant and non-zero entries
- micropolar vorticity model / use vorticity to integrate on curve?