It could be me getting lost due to a lack of technical knowledge, but how are the errors described and quantified in the test cases? I am speaking in the context of Dr. Smith's following lecture notes,
Description Rather Than Specification}
Test cases are often phrased as Expected = Calculated
In scientific software, you generally should not test for equality
Absolute error within tolerance
Relative error within tolerance
If comparing matrices or vectors, consider using norms of residual
Even a specific tolerance often doesn't make sense in a scientific context
Often your plan should be to describe the error rather than prescribe.
Plot of error versus problem size, or condition number, or ...
Consider summarizing multiple tests with the infinity norm of the relative
error (or similar).
@liziscool
It could be me getting lost due to a lack of technical knowledge, but how are the errors described and quantified in the test cases? I am speaking in the context of Dr. Smith's following lecture notes,