[opencurrent-users] Issue 10 in opencurrent: NSTest broken (silently)

[openc...@googlecode.com]

Status: Accepted
Owner: andrewhhunter
Labels: Type-Defect Priority-Medium

New issue 10 by andrewhhunter: NSTest broken (silently)
http://code.google.com/p/opencurrent/issues/detail?id=10

NSTest is failing (the multigrid solver isn't converging for some reason,
though it appears to be spurious, as the L2 error is zero!)

Furthermore, due to a missing ASSERT, this failure is going unreported.

[openc...@googlecode.com]

Comment #1 on issue 10 by andrewhhunter: NSTest broken (silently)
http://code.google.com/p/opencurrent/issues/detail?id=10

This is broken (I think) because NSTest runs a
Eqn_IncompressibleNS3D<float> with
max_divergence at its default of 1e-5, which the multigrid solver can't
achieve in
single precision. (L2 error is zero up to rounding error, and the LINF
error, which
is what we're testing convergence against, is still 6e-5 or so.)

The easy fix is modifying the test parameters to solve to 1e-4, which I'm
doing (and
will commit, unless I hear otherwise.) The right fix is probably to set
the default
for <float> divergence to a more reasonable value.