Integrator V&V

Topics related to QA and V&V of GMAT features

Integrator V&V

Postby shughes » Sun Mar 11, 2012 1:49 pm

Overview

The only significant issue appears to be a performance problem for BS and perhaps ABM. The test coverage for integrator settings is good. However, we need to test integrators in more dynamics regimes, with more complex force models, using forward and backwards propagation. Existing tests primarily test two-body numerical propagation in GMAT against MATLAB semi-analytic Kepler propagator that does not suffer from round off error over time.

Docs and Links


Summary of Existing Issues

Issue: In script you can set MinStep = MaxStep. In GUI, this throws an error.
Recommendation: Remove this validation test from the GUI. We allow this setting at the user's risk in the script and so we should allow in the GUI also. We implemented StopIfAccuracyIsViolated field for this reason.

Issue: Hidden parameters on BulirschStoer Integrator appear to affect performance.
Recommendation: We need find reasonable values for these and then either expose and document as only scriptable, or hide and leave hard coded. By expose, I mean they should be included in GetGeneratingString. Currently they are "read only".

Issue: Integrator numeric and performance , GMT-540, GMT-2390, GMT-2428. At least one may be a user config problem, however, the BS and ABM integrators peform poorly for many orbit cases in simple tests and I suspect there are some tweaks we need to make to the integrators perform better. Note that there is C++ code for a BS integrator submitted to MATLAB file exchange. We could compare against that implementation to see if there are better default settings for the BS integrator.
Recommendation. Fix the performance problems because they occur in too many orbit types to be ignored.

Issue: ABM integrator field names are different in GUI and script. The Input Range error message also uses the GUI sytnax.
Recommendation: Rename in the GUI and change error message syntax. Changing label shouldn't change the text box name for Test Complete.

Issue: InitialStepSize field does not use standard input range error message for invalid data. Set to zero to see this behavior.
Recommendation: Fix it. Trivial

Issue: Ranges for some fields such as accuracy accept values arbitrarily close to "bad" values. Accuracy of 1 for example or MaxStep of 1e-300. Should we force users to input more sensible input for these?

Issue: Are there any special range checks required for BS fields?
shughes
 
Posts: 443
Joined: Mon Jun 09, 2008 6:27 pm

Return to QA and V&V

Who is online

Users browsing this forum: No registered users and 0 guests

cron