What Is Compressible Multiphase Flow?#

A few bubbles in a water tank are enough to slow a pressure wave. Water and air, fuel droplets in hot gas, the gas pocket from an underwater blast. Two phases, and the flow is multiphase. Add compressibility and each phase brings its own equation of state (EOS). Shocks on the interface reflect, transmit, and refract.

A time step built on the pure-phase sound speed goes wrong at the interface first. Push the probe slider to a gas fraction α=0.5 and the cyan Wood curve drops to about one hundredth of the single-phase sound speed.

At α of 0 or 1 the Wood curve meets the amber frozen line. Only near the middle does Wood hit the floor. That floor stretches the acoustic time scale by about a hundred.

How It Differs from Single-Phase Compressible Flow#

The single-phase Euler system is short.

$$\frac{\partial \mathbf{U}}{\partial t} + \frac{\partial \mathbf{F}}{\partial x} = 0$$

The conserved state is $\mathbf{U} = [\rho, \rho u, E]^T$. One ideal-gas closure, $p = (\gamma - 1)\rho e$, is enough.

Across an interface the EOS changes. Gas uses $\gamma = 1.4$. Liquid uses a stiffened gas.

$$p = (\gamma - 1)\rho e - \gamma p_\infty$$

With γ at 1.4, (γ−1) is 0.4. Only 40 percent of the internal energy ρe shows up as pressure. The liquid law subtracts γ p_∞ from that same term. A mixed cell that inherits an in-between γ drags the coefficient of the shift along with it. Density and energy can look fine while the recovered pressure is not the arithmetic mean of the two neighbors.

The numerical trouble starts when the two equations of state mix near the interface. Averaging conserved variables leaves a non-physical pressure oscillation. A pressure field that looks flat while the residual will not die shows up in single-phase pressure residuals too. Here the cause is the mixed EOS, not a turbulence constant.

Read Mach off the figure. At α=0.5 the mixture sound speed is one hundredth of the pure-phase value. A speed that was Mach 0.01 in the liquid is Mach 1 on the mixture scale. The velocity did not change. The sound speed did, by a factor of a hundred. A CFL based on the liquid sound speed takes steps a hundred times smaller than the mixture needs. A CFL based on the mixture sound speed suddenly coarsens the time step in the interface cells alone.

Why It's Hard: Three Core Difficulties#

1. Pressure Oscillations at the Interface#

After Abgrall's 1994 note, killing this oscillation became the starting move in multiphase numerics.

“

"The history of compressible multiphase numerics is the history of fighting pressure oscillations."

A plain conservative finite-volume update lets γ take an intermediate value at the interface and writes a spurious oscillation.

Two cells can share a pressure and still disagree on ρ and e, because the EOS disagrees. Feed the averaged $\mathbf{U} = [\rho, \rho u, E]^T$ back into either $p = (\gamma - 1)\rho e$ or $p = (\gamma - 1)\rho e - \gamma p_\infty$ and the original pressure does not come back. The pressure error enters the next velocity, and that velocity pushes the density. The oscillation does not die in one step.

2. Variety of Equations of State#

Ideal gas does not describe water. Water wants a Tait EOS or a stiffened gas EOS. High pressure wants the Jones–Wilkins–Lee (JWL) EOS. Reactive flow couples chemistry into the closure. Each EOS needs its own Riemann solver, and the code gets long.

3. Interface Tracking vs. Interface Capturing#

Two ways to hold the interface.

A sharp method keeps the interface as a line. Ghost fluid extrapolates the other side into the Riemann problem. A diffuse 5- or 6-equation model transports the volume fraction α and spreads the interface over a few cells. Phase change is easier on the spread side. A shock hitting the interface head-on keeps a cleaner pressure jump on the sharp side. If γ still mixes, both sides oscillate.

Why It Matters#

The cases that need this are the ones where an interface and a shock share a cell.

• Underwater explosion (UNDEX): ship survivability assessment
• Supersonic combustion: fuel injection and mixing inside a scramjet
• Medical: bubble motion in shock-wave lithotripsy
• Space propulsion: sloshing of cryogenic propellant

A conserved average breaks the interface pressure#

A conserved average is a short implementation. One flux, same as single phase. It comes apart at the interface. γ and p_∞ differ from cell to cell, and averaging only U does not make the recovered pressure match its neighbors. The spike can look like a shock. It is not a physical wave.

If the interface is a real discontinuity, the next decision is where the order lives. High-order compressible discretization on unstructured meshes shows how a single-phase jump sits on a cell face. Before moving that idea into two phases, it is worth seeing how the single-phase Riemann problem puts the pressure back.