Finite-Volume Conservation Laws

Finite-volume methods evolve cell integrals rather than point values. Their central strength is discrete conservation: the flux leaving one cell is exactly the flux entering its neighbor, even when the continuum solution contains a shock. Conservation does not by itself guarantee accuracy or physical admissibility. Reconstruction, the Riemann flux, source discretization, primitive recovery, and time stepping must be tested as one scheme.

Required background. Ordinary Differential Equations supplies semidiscrete updates and time integration.

Helpful background. Formulations of General Relativity provides the geometric evolution systems and source terms, while Partial Differential Equations and Green Functions supplies conservation laws, characteristics, and weak solutions.

Conservation by flux cancellation

For

$$\partial_tU+\partial_xF(U)=S(U,x),$$

define the cell average $\bar U_i=\Delta x_i^{-1}\int_{I_i}U\,dx$. Integrating the equation gives the semidiscrete update

$$\frac{d\bar U_i}{dt} =-\frac{\hat F_{i+1/2}-\hat F_{i-1/2}}{\Delta x_i}+\bar S_i.$$

Summing over cells makes every interior numerical flux cancel pairwise. Only boundary fluxes and sources change the total. This telescoping property survives discontinuities because it follows from the integral form, not from differentiability of $U$.

High-resolution methods reconstruct left and right interface states, apply a limiter or nonlinear stencil, and feed those states to an exact or approximate Riemann solver. The relation among conservative variables, characteristic waves, and shock-capturing updates is developed for relativistic hydrodynamics in Martí and Müller 2003, §§ 2.1–2.3, Eqs. 1–30.

Compare at least two fluxes on the same reconstruction. A two-wave HLL-type flux is robust but diffuses contact structure; a solver resolving the contact can sharpen density while depending on more detailed characteristic information. Harten, Lax, and van Leer 1983, § 4(f), Eq. 4.27 explains why the split-flux construction cannot exactly preserve a stationary contact. This comparison should hold the timestep, limiter, and primitive recovery fixed. Otherwise a sharper profile cannot be attributed to the Riemann solver. Record estimated signal speeds as output, because an underestimated extremal speed can destroy stability even when the algebraic flux formula is implemented correctly.

Relativistic states and admissibility

In special relativistic hydrodynamics, primitive variables may be $(\rho,p,v)$, with Lorentz factor $W=(1-v^2)^{-1/2}$. A common conservative state contains rest-mass density, momentum, and energy combinations such as

$$D=\rho W, \qquad S=\rho hW^2v, \qquad \tau=\rho hW^2-p-D,$$

for specific enthalpy $h$. With $c=1$, the ideal-gas closure used below is

$$p=(\Gamma-1)\rho\epsilon, \qquad h=1+\epsilon+\frac{p}{\rho} =1+\frac{\Gamma p}{(\Gamma-1)\rho}.$$

The numerical update acts on $(D,S,\tau)$, but fluxes require primitives, so nonlinear primitive recovery is part of every timestep.

For this one-dimensional special-relativistic system, the flux vector is

$$F(U)=\begin{pmatrix}Dv\\Sv+p\\S-Dv\end{pmatrix}, \qquad U=\begin{pmatrix} D\\ S\\ \tau \end{pmatrix}.$$

The energy flux is $S-Dv$ because $\tau$ excludes rest-mass energy. Including rest mass in the evolved energy changes that component and must be translated together with its recovery equations. The sound speed and acoustic characteristic speeds are

$$c_s^2=\frac{\Gamma p}{\rho h}, \qquad a_\pm=\frac{v\pm c_s}{1\pm v c_s}.$$

For interface states $U_L,U_R$, choose signal bounds $s_L\le0\le s_R$ that enclose the Riemann fan. The HLL flux is

$$\widehat F_{\rm HLL}= \frac{s_RF(U_L)-s_LF(U_R)+s_Ls_R(U_R-U_L)}{s_R-s_L}.$$

It follows by integrating conservation over the two-wave approximate fan. When all signals travel right or left it reduces respectively to $F(U_L)$ or $F(U_R)$; it also gives $F(U)$ when $U_L=U_R=U$. These are useful algebraic checks before the shock-tube test. The physical characteristic structure is given in Martí and Müller 2003, §§ 2.1–2.3, Eqs. 1–30. Taking extrema of endpoint characteristic speeds is an estimate, not a universal proof that nonlinear shock speeds are enclosed; a code must document its speed prescription.

Admissibility requires at least $\rho\gt 0$, $p\gt 0$, and $\lvert v\rvert\lt 1$. Reconstruction in conservative or characteristic variables can create an inadmissible interface state even when both neighboring cells are physical. Positivity-preserving fallback and a documented atmosphere policy may be necessary; silently clipping a state changes conservation and must be recorded.

Shock-tube benchmark

Use the ideal-gas equation of state above with $\Gamma=5/3$ on $0\leq x\leq1$, outflow ghost cells at both ends, final time $t=0.4$, and an initial discontinuity at $x=0.5$:

$$(\rho,p,v)_L=(1,1,0), \qquad (\rho,p,v)_R=(0.125,0.1,0).$$

Run $N=200,400,800$ equal cells with piecewise-linear primitive-variable reconstruction, the monotonized-central limiter, third-order strong-stability-preserving Runge–Kutta time stepping, and Courant number $0.4$. Compare the two-wave HLL flux with an exact relativistic Godunov flux while holding every other choice fixed, and compare both cell-average solutions with exact cell averages from the relativistic Riemann solution. Report density, pressure, velocity, and conserved variables, together with $L^1$ error, total conserved quantities, minimum density and pressure, maximum speed, and primitive-recovery failures. The wave pattern and exact-solver construction provide stronger evidence than visual curve agreement.

Use the three grids to compute pairwise observed orders at the same physical time and with the same cell-average norm. Smooth portions may show the reconstruction order, but a shock limits global $L^1$ convergence to approximately first order, while a numerically spreading contact can converge more slowly. These reduced rates are expected and do not imply that the higher-order reconstruction is inactive in smooth regions. LeVeque explains reconstruction, limiters, Riemann solvers, and loss of order at discontinuities in Finite Volume Methods for Hyperbolic Problems, Chapters 10–12 and 17.

Curved spacetime and source balance

In curved coordinates, evolve the properly densitized conserved variables and geometric fluxes dictated by the continuum formulation. Metric derivatives often appear as sources. A naive discretization can destroy a stationary equilibrium through imperfect flux–source cancellation. Add a smooth advection benchmark and, where applicable, a known stationary relativistic state to test well balancing separately from shock capture.

At mesh refinement boundaries, refluxing restores conservation by replacing coarse fluxes with the sum of fine fluxes. Without it, local solutions may look convergent while the global budget drifts. Conservation histories and boundary flux integrals must therefore accompany field norms.

Reproducible environment and failure criteria

Pin the implementation revision, compiler, precision, equation of state, conservative-variable definition, reconstruction, limiter, Riemann solver, flux formula, primitive-recovery tolerance, atmosphere or floor, boundary conditions, time integrator, mesh, and Courant factor. A clean rerun exports cell averages, interface fluxes, conservation budgets, exact-reference errors, refinement results, and every fallback count.

The validity regime assumes a hyperbolic physical state and a Riemann solver appropriate to the equation set. Failure modes include negative density or pressure, superluminal primitives, unlimited oscillations near the discontinuity, inconsistent interface fluxes, a CFL violation, and conservation error hidden by floors. Reject apparent high-order convergence measured only away from the shock, or agreement based on cell-center interpolation that ignores cell averages.

Check your understanding

A run conserves total mass to roundoff but develops negative pressure next to the shock. Has discrete conservation validated the solution?

Solution

No. Flux cancellation validates one structural property, while negative pressure violates physical admissibility and may make primitive recovery undefined. Diagnose reconstruction and the Riemann flux, apply a documented positivity-preserving fallback, and repeat the refinement and exact-solver comparison. Conservation and admissibility are separate acceptance conditions.

References

• Harten, A., Lax, P. D., and van Leer, B. (1983), “On Upstream Differencing and Godunov-Type Schemes for Hyperbolic Conservation Laws,” SIAM Review 25, 35–61. DOI.
• LeVeque, R. J. (2002), Finite Volume Methods for Hyperbolic Problems, Cambridge University Press. DOI.
• Martí, J. M., and Müller, E. (2003), “Numerical Hydrodynamics in Special Relativity,” Living Reviews in Relativity 6, 7. DOI.