Running the Tests

The usual way to operate TestInterp is to use the ``series'' test function with the maximum order precisely equal to the order of the interpolator. At least for polynomial interpolators (which includes all the present Cactus interpolators), if the interpolator is working properly this should give interpolation results which exactly equal the test function, up to floating-point roundoff error. The difference between the interplation results and the test function (or between the query and numerical-perturbation Jacobians) is thus a measure of the error in the computation. This thorn prints out various norms (taken over all the interpolation points or Jacobian elements) of these errors.

Each test in the test suite corresponds to a single Cactus run, and tests a single interpolator, order, and derivative (if the interpolator supports derivatives), but tests all these Cactus data types together. The results are split out for each of the 3 subsets and for the Jacobian tests (see below).

The run/InterpGV/, run/InterpLocal/, and run/InterpLocalUniform/ directories all have makefiles with targets ``1d'', ``2d'', and ``3d'' to run the corresponding tests and report the error norms.

Each makefile also has a target ``testsuite'' to run the tests with a slightly different set of parameters, and create a Cactus testsuite in the test/ directory.

Finally, the run/try/ directory contains various parameter files for ad-hoc interpolator tests. Jacobian.par is a good one to start with if you want to poke around...