[BUG] *Different Results for Two TO Settings Files that Should Give Same Results*

[bfregly]

Describe the bug
I am running torque-driven TO cases for UF Subject 1 using different settings files.

I ran one case where I removed all arm coordinates from the states but accidentally left generalized_coordinate_tracking and generalized_speed_tracking set to true for those coordinates. See settings file, all inputs, and results are in TORSS2Dv7_Different_380iters_48core.zip.

After realizing my goof, I reran the same torque-driven TO case except with generalized_coordinate_tracking and generalized_speed_tracking set to false for the arm coordinates. See settings file, all inputs, and results are in TORSS2Dv9_Different_354iters_48core.zip. The two torque-driven TO run converged to slightly different solutions, even though they should have converged to exactly the same solution. For both runs, the arm coordinates show RMS errors of 0.0000, so arm motion appears to have been prescribed properly in both runs.

I used Maggie's old 48-core PC workstation for both runs. Also, to speed up the solution process, both runs use only about 1/3 of the gait cycle covering right single leg support (RSS).

To Reproduce

1. Run the torque-driven TO case in TORSS2Dv7_Different_380iters_48core.zip.
2. Run the torque-driven TO case in TORSS2Dv9_Different_354iters_48core.zip.
3. Compare the results - they are slightly different, as clearly shown by the different numbers of iterations requires for the two cases, even though the results should be identical.

Expected behavior
The two torque-driven TO runs should produce identical results. If coordinates are not included in the states_coordinate_list, then including cost function terms related to these non-existent coordinates should not change the solution.

Screenshots
If applicable, add screenshots to help explain your problem.

Desktop (please complete the following information):

• Windows 10
• OpenSim Version: 4.5
• MATLAB Version: 2024a
• NMSM Pipeline Version: 1.3.2 dev branch with one-line change to calcModeledVerticalGroundReactionForce.m to eliminate truncating contact element height above a pre-set value (this file is attached as a text file, since GitHub won't upload m files).

Additional context
Add any other context about the problem here.

calcModeledVerticalGroundReactionForce.txt
TORSS2Dv7_Different_380iters_48core.zip
TORSS2Dv9_Different_354iters_48core.zip