Weakly Compressible SPH

Weakly compressible SPH as introduced by Monaghan (1994). This formulation relies on a stiff equation of state that generates large pressure changes for small density variations.

TrixiParticles.WeaklyCompressibleSPHSystemType
WeaklyCompressibleSPHSystem(initial_condition,
                            density_calculator, state_equation,
                            smoothing_kernel, smoothing_length;
                            viscosity=nothing, density_diffusion=nothing,
                            acceleration=ntuple(_ -> 0.0, NDIMS),
                            buffer_size=nothing,
                            correction=nothing, source_terms=nothing,
                            surface_tension=nothing, surface_normal_method=nothing,
                            reference_particle_spacing=0.0))

System for particles of a fluid. The weakly compressible SPH (WCSPH) scheme is used, wherein a stiff equation of state generates large pressure changes for small density variations. See Weakly Compressible SPH for more details on the method.

Arguments

Keyword Arguments

  • viscosity: Viscosity model for this system (default: no viscosity). See ArtificialViscosityMonaghan or ViscosityAdami.
  • density_diffusion: Density diffusion terms for this system. See DensityDiffusion.
  • acceleration: Acceleration vector for the system. (default: zero vector)
  • buffer_size: Number of buffer particles. This is needed when simulating with OpenBoundarySPHSystem.
  • correction: Correction method used for this system. (default: no correction, see Corrections)
  • source_terms: Additional source terms for this system. Has to be either nothing (by default), or a function of (coords, velocity, density, pressure, t) (which are the quantities of a single particle), returning a Tuple or SVector that is to be added to the acceleration of that particle. See, for example, SourceTermDamping. Note that these source terms will not be used in the calculation of the boundary pressure when using a boundary with BoundaryModelDummyParticles and AdamiPressureExtrapolation. The keyword argument acceleration should be used instead for gravity-like source terms.
  • surface_tension: Surface tension model used for this SPH system. (default: no surface tension)
  • surface_normal_method: The surface normal method to be used for this SPH system. (default: no surface normal method or ColorfieldSurfaceNormal() if a surface_tension model is used)
  • reference_particle_spacing: The reference particle spacing used for weighting values at the boundary, which currently is only needed when using surface tension.
source

Equation of State

The equation of state is used to relate fluid density to pressure and thus allow an explicit simulation of the WCSPH system. The equation in the following formulation was introduced by Cole (1948) (pp. 39 and 43). The pressure $p$ is calculated as

\[ p = B \left(\left(\frac{\rho}{\rho_0}\right)^\gamma - 1\right) + p_{\text{background}},\]

where $\rho$ denotes the density, $\rho_0$ the reference density, and $p_{\text{background}}$ the background pressure, which is set to zero when applied to free-surface flows (Adami et al., 2012).

The bulk modulus, $B = \frac{\rho_0 c^2}{\gamma}$, is calculated from the artificial speed of sound $c$ and the isentropic exponent $\gamma$.

An ideal gas equation of state with a linear relationship between pressure and density can be obtained by choosing exponent=1, i.e.

\[ p = B \left( \frac{\rho}{\rho_0} -1 \right) = c^2(\rho - \rho_0).\]

For higher Reynolds numbers, exponent=7 is recommended, whereas at lower Reynolds numbers exponent=1 yields more accurate pressure estimates since pressure and density are proportional (see Morris, 1997).

When using SummationDensity (or DensityReinitializationCallback) and free surfaces, initializing particles with equal spacing will cause underestimated density and therefore strong attractive forces between particles at the free surface. Setting clip_negative_pressure=true can avoid this.

TrixiParticles.StateEquationColeType
StateEquationCole(; sound_speed, reference_density, exponent,
                  background_pressure=0.0, clip_negative_pressure=false)

Equation of state to describe the relationship between pressure and density of water up to high pressures.

Keywords

  • sound_speed: Artificial speed of sound.
  • reference_density: Reference density of the fluid.
  • exponent: A value of 7 is usually used for most simulations.
  • background_pressure=0.0: Background pressure.
  • clip_negative_pressure=false: Negative pressure values are clipped to 0, which prevents spurious surface tension with SummationDensity but allows unphysical rarefaction of the fluid.
source
TrixiParticles.StateEquationIdealGasType
StateEquationIdealGas( ;sound_speed, reference_density, gamma, background_pressure=0.0,
                       clip_negative_pressure=false)

Equation of state to describe the relationship between pressure and density of a gas using the Ideal Gas Law.

Keywords

  • sound_speed : Artificial speed of sound.
  • reference_density : Reference density of the fluid.
  • gamma : Heat-capacity ratio
  • background_pressure=0.0 : Background pressure.
  • clip_negative_pressure=false: Negative pressure values are clipped to 0, which prevents spurious surface tension with SummationDensity but allows unphysical rarefaction of the fluid.
source

Density Diffusion

Density diffusion can be used with ContinuityDensity to remove the noise in the pressure field. It is highly recommended to use density diffusion when using WCSPH.

Formulation

All density diffusion terms extend the continuity equation (see ContinuityDensity) by an additional term

\[\frac{\mathrm{d}\rho_a}{\mathrm{d}t} = \sum_{b} m_b v_{ab} \cdot \nabla_{r_a} W(\Vert r_{ab} \Vert, h) + \delta h c \sum_{b} V_b \psi_{ab} \cdot \nabla_{r_a} W(\Vert r_{ab} \Vert, h),\]

where $V_b = m_b / \rho_b$ is the volume of particle $b$ and $\psi_{ab}$ depends on the density diffusion method (see DensityDiffusion for available terms). Also, $\rho_a$ denotes the density of particle $a$ and $r_{ab} = r_a - r_b$ is the difference of the coordinates, $v_{ab} = v_a - v_b$ of the velocities of particles $a$ and $b$.

Numerical Results

All density diffusion terms remove numerical noise in the pressure field and produce more accurate results than weakly commpressible SPH without density diffusion. This can be demonstrated with dam break examples in 2D and 3D. Here, $δ = 0.1$ has been used for all terms. Note that, due to added stability, the adaptive time integration method that was used here can choose higher time steps in the simulations with density diffusion. For the cheap DensityDiffusionMolteniColagrossi, this results in reduced runtime.

density_diffusion_2d
Dam break in 2D with different density diffusion terms
density_diffusion_3d
Dam break in 3D with different density diffusion terms

The simpler terms DensityDiffusionMolteniColagrossi and DensityDiffusionFerrari do not solve the hydrostatic problem and lead to incorrect solutions in long-running steady-state hydrostatic simulations with free surfaces (Antuono et al., 2012). This can be seen when running the simple rectangular tank example until $t = 40$ (again using $δ = 0.1$):

density_diffusion_tank
Tank in rest under gravity in 3D with different density diffusion terms

DensityDiffusionAntuono adds a correction term to solve this problem, but this term is very expensive and adds about 40–50% of computational cost.

API

TrixiParticles.DensityDiffusionAntuonoType
DensityDiffusionAntuono(initial_condition; delta)

The commonly used density diffusion terms by Antuono (2010), also referred to as δ-SPH. The density diffusion term by Molteni (2009) is extended by a second term, which is nicely written down by Antuono (2012).

The term $\psi_{ab}$ in the continuity equation in DensityDiffusion is defined by

\[\psi_{ab} = 2\left(\rho_a - \rho_b - \frac{1}{2}\big(\nabla\rho^L_a + \nabla\rho^L_b\big) \cdot r_{ab}\right) \frac{r_{ab}}{\Vert r_{ab} \Vert^2},\]

where $\rho_a$ and $\rho_b$ denote the densities of particles $a$ and $b$ respectively and $r_{ab} = r_a - r_b$ is the difference of the coordinates of particles $a$ and $b$. The symbol $\nabla\rho^L_a$ denotes the renormalized density gradient defined as

\[\nabla\rho^L_a = -\sum_b (\rho_a - \rho_b) V_b L_a \nabla_{r_a} W(\Vert r_{ab} \Vert, h)\]

with

\[L_a := \left( -\sum_{b} V_b r_{ab} \otimes \nabla_{r_a} W(\Vert r_{ab} \Vert, h) \right)^{-1} \in \R^{d \times d},\]

where $d$ is the number of dimensions.

See DensityDiffusion for an overview and comparison of implemented density diffusion terms.

source
TrixiParticles.DensityDiffusionFerrariType
DensityDiffusionFerrari()

A density diffusion term by Ferrari (2009).

The term $\psi_{ab}$ in the continuity equation in DensityDiffusion is defined by

\[\psi_{ab} = \frac{\rho_a - \rho_b}{2h} \frac{r_{ab}}{\Vert r_{ab} \Vert},\]

where $\rho_a$ and $\rho_b$ denote the densities of particles $a$ and $b$ respectively, $r_{ab} = r_a - r_b$ is the difference of the coordinates of particles $a$ and $b$ and $h$ is the smoothing length.

See DensityDiffusion for an overview and comparison of implemented density diffusion terms.

source
TrixiParticles.DensityDiffusionMolteniColagrossiType
DensityDiffusionMolteniColagrossi(; delta)

The commonly used density diffusion term by Molteni (2009).

The term $\psi_{ab}$ in the continuity equation in DensityDiffusion is defined by

\[\psi_{ab} = 2(\rho_a - \rho_b) \frac{r_{ab}}{\Vert r_{ab} \Vert^2},\]

where $\rho_a$ and $\rho_b$ denote the densities of particles $a$ and $b$ respectively and $r_{ab} = r_a - r_b$ is the difference of the coordinates of particles $a$ and $b$.

See DensityDiffusion for an overview and comparison of implemented density diffusion terms.

source