The initial conditions for the exterior solution were fixed by requiring that the interior sources be stationary prior to some fixed time. As a result, the exterior analytic solution is causal in the sense that it is stationary in the past of some null cone. This effectively introduces a condition that eliminates extraneous incoming radiation from the system in a physically plausible way and determines the exterior solution uniquely. An overlap region was introduced between the numerical interior and the analytic exterior. In the overlap, the numerical solution was matched to the causal analytic solution, resulting in an evolution that was everywhere causally meaningful.

This is the physically correct approach to a system which is stationary prior to a fixed time but is nontrivial to generalize, say, to the problem of radiation from an orbiting binary. Anderson and Hobill first tackled the 1D model problem of an oscillator coupled to a spherically symmetric, flat space, massless scalar field. The numerical results were in excellent agreement with the exact analytic solution (which could be obtained globally for this problem).

They extended the model to include spherical scalar waves propagating in a spherically symmetric curved background. This introduces backscattering which obscures the concept of a purely outgoing wave. No exact solution exists to this problem so that an approximation method was necessary to determine the exterior analytic solution. The approximation was based upon an expansion parameter controlling the amount of backscatter. In the 0th approximation, the scale parameter vanishes and the problem reduces to the flat space case which can be solved exactly. The flat space Green function was then used to generate higher order corrections.

A key ingredient of this perturbative scheme is that the wave
equation is solved in retarded null coordinates (*u*,
*r*) for the curved space metric, so that the proper causality is
built into the Green function at each order of approximation. The
transformation from null coordinates (*u*,
*r*) to Cauchy coordinates (*t*,
*r*) is known analytically for this problem. This allows a simple
match between the null and Cauchy solutions at the boundary of
the Cauchy grid. Their scheme is efficient and leads to
consistent results in the region that the numerical and analytic
solutions overlap. It is capable of handling both strong fields
and fast motions.

Later, a global, characteristic, numerical study of the
self-gravitating version of this problem, by Gómez and Winicour,
confirmed that the use of the true null cones is essential in
getting the correct radiated waveform [77]. For quasi-periodic radiation, the phase of the waveform is
particular sensitive to the truncation of the outer region at a
finite boundary. Although a perturbative estimate would indicate
an
*O*
(*M*
/
*R*) error, this error accumulates over many cycles to produce an
error of order
in the phase.

Anderson and Hobill proposed that their method be extended to
general relativity by matching a numerical solution to an
analytic 1/
*r*
expansion in null coordinates. However, the only
analytic-numerical matching schemes that have been implemented in
general relativity have been based upon perturbations of a
Schwarzschild background using the standard Schwarzschild time
slicing [1,
3,
2,
142]. It would be interesting to compare results with an
analytic-numeric matching scheme based upon the true null cones.
However the original proposal by Anderson and Hobill has not been
carried out.

Characteristic Evolution and Matching
Jeffrey Winicour
http://www.livingreviews.org/lrr-2001-3
© Max-Planck-Gesellschaft. ISSN 1433-8351 Problems/Comments to livrev@aei-potsdam.mpg.de |