Outer Boundaries and Cauchy–Characteristic Matching

A Cauchy simulation ends at an artificial timelike outer boundary unless the domain is compactified, hyperboloidal, or coupled to a characteristic exterior. Boundary data can inject constraints and gauge disturbances or reflect outgoing radiation back into the strong-field region. Pushing the boundary farther away delays these errors but does not identify or remove them. A defensible algorithm controls incoming characteristic fields, measures reflection and constraint flux, and varies both radius and boundary order. Cauchy–characteristic matching adds an exterior evolution, but its interface must itself be verified.

Helpful background. Formulations of General Relativity supplies characteristic fields and continuum boundary systems, Ordinary Differential Equations supplies stability and convergence, Computational Relativity supplies shared evidence standards, and Black-Hole Dynamics and Collapse supplies strong-field applications.

Incoming fields determine boundary data

At the outer surface with outward spatial normal $s_i$, diagonalize the principal symbol into characteristic combinations $u^{\hat\alpha}$ with normal speeds $v_{\hat\alpha}$. Numerical data may be prescribed only for incoming fields. Those fields divide into physical radiation, gauge, and constraint sectors. A boundary treatment should control incoming constraints, choose gauge data consistently, and approximate absence of incoming physical radiation without overconstraining outgoing fields.

Simple Sommerfeld conditions applied component by component ignore this structure. They can work approximately for weak spherical waves but may violate constraints or reflect nonspherical modes. Radiation-controlling conditions often constrain an incoming Weyl scalar such as $\Psi_0$. Buchman and Sarbach 2007, §IV.B, Eq. (89), Figures 1–2 and §IV.C, Eq. (96) derives a hierarchy whose reflection decreases with boundary radius and order.

To see both the utility and the limitation of a Sommerfeld condition, take an outgoing spherical scalar wave in flat spacetime, $u(t,r)=f(t-r)/r$, outside its source. It satisfies

$$(\partial_t+\partial_r)(ru)=0, \qquad \partial_tu+\partial_ru+\frac{u}{r}=0.$$

The $u/r$ term accounts for spherical dilution; omitting it even reflects part of this simple monopole. Higher multipoles and curved-background backscatter contain additional radial structure, so the same condition is no longer exactly transparent. For a pulse emitted near radius $r_s$, reflected at $R$, and monitored at $r_m$, flat-space radial propagation predicts a return delay $(R-r_s)+(R-r_m)$ in $c=1$ units. Increasing $R$ by $\Delta R$ shifts that return by $2\Delta R$, a useful diagnostic distinct from resolution-dependent dispersion. In curved coordinates use the actual characteristic travel times, not this flat estimate.

Measure reflection instead of assuming absorption

Send a weak, known $(\ell,m)=(2,2)$ pulse toward boundaries at several radii $R$ and with several absorbing orders. Extract outgoing and reflected amplitudes on interior worldtubes. For a first radiation-controlling condition in its asymptotic regime, the amplitude reflection scales roughly as $(kR)^{-4}$ for wave number $k$; the precise coefficient and higher-order scaling depend on the system and condition. Demonstrate the measured law rather than quote it outside its assumptions.

Monitor incoming constraint characteristic fields and an energy estimate or boundary flux. Move $R$ while keeping the physical grid resolution in the wave zone fixed. The return time should shift by the causal travel time. Use at least three resolutions so numerical dispersion is separated from physical-boundary reflection. A low reflected waveform with growing constraints is not a successful boundary.

For a stationary Schwarzschild or Kerr run, the physical incoming-radiation target is zero. Compare horizon invariants and constraints across radii. Boundary contamination may first appear in gauge fields and only later reach the horizon, so retain full time series.

Cauchy–characteristic matching

Matching exchanges data in both directions across a timelike worldtube: a Cauchy interior supplies geometric data to a characteristic exterior, which returns the incoming information required by the Cauchy system. This differs from one-way characteristic extraction. The interface must reconcile coordinates, tetrads, variables, interpolation, time steps, and constraint data.

A suitable benchmark evolves a weak Teukolsky wave whose Cauchy and characteristic representations are independently known. Measure the mismatch of induced worldtube data, returned characteristic fields, constraints, and waveform at null infinity. Refine both grids and the interpolation order independently. Recent matching experiments show important progress but also system-dependent stability questions. Ma et al. 2024, §§II–V describes a fully relativistic three-dimensional implementation that returns the physical incoming degrees of freedom through the matched Weyl scalar; its Cauchy gauge and constraint sectors still use their own boundary conditions. It therefore does not by itself verify all-sector matching or prove that the coupled continuum problem is well posed.

Environment and failure criteria

Pin the evolution formulation, characteristic normalization, null-tetrad orientation and normalization, outer radius and shape, boundary operators and order, penalty or ghost-zone implementation, gauge and constraint data, worldtube map, interpolation, Cauchy and characteristic resolutions, time interpolation, precision, and code revisions. A clean interface could be boundary-test --config l2-pulse.toml --output run/, emitting all characteristic fields, constraint fluxes, reflected amplitudes, mismatch norms, sweeps in $R$ and resolution, and environment metadata. It specifies a rerun record without claiming that code is supplied here.

Reject a treatment if reflection does not follow its predicted regime, constraints enter through the boundary, convergence fails at the interface, or results depend on unreported interpolation filters. Other failure modes include classifying a characteristic speed with the wrong normal, freezing a physical incoming field in a curved background, worldtube gauge mismatch, and weakly hyperbolic subproblems that appear stable only at coarse resolution.

Diagnostic

Doubling $R$ reduces the reflected quadrupole amplitude by about sixteen while constraint influx remains unchanged and nonconvergent. Has the outer boundary passed because the radiation scaling matches $(kR)^{-4}$?

Solution

No. The physical radiation sector shows the expected leading reflection scaling, but the constraint sector still injects an unresolved error. A valid boundary must control both. Inspect incoming constraint characteristics and their discrete penalty, refine the boundary closure, and require convergent constraints together with the radiation result before using the boundary in a strong-field run.

References

• Buchman, L. T., and Sarbach, O. C. A. (2007). “Towards absorbing outer boundaries in general relativity.” Classical and Quantum Gravity 24, S307–S326. arXiv.
• Ma, S., Moxon, J., Scheel, M. A., et al. (2024). “Fully relativistic three-dimensional Cauchy–characteristic matching for physical degrees of freedom.” Physical Review D 109, 124027. DOI. Open PDF.