Relaxing Parameter Optimization Convergence Tolerance value automatically based on error difference in Parameter Identification analysis

Hi all,
I am running parameter identification analysis using the gait static c3d file.
However, the analysis fails if the value of [ParameterOptimizationConvergenceTol = 0.0001;], which is default value. If I relax this parameter manually to a certain value like 0.057, then only the analysis converges.
Is it good practice to relax this parameter value?
If yes, how can I set the model to converge such that when the difference in AbsSumKinError in two consecutive step length is less than a certain value like 0.0001 rather than based on ParameterOptimizationConvergenceTol?

Hi Rajanprasad460,

Relaxing the convergence criteria that much does not make sense. Remember that this is the tolerance on the optimality conditions (KKT conditions), which also include the errors on the kinematic joint constraints. In your case, with a tolerance of 0.057, you are allowing errors up to 5,7 cm on joint translations. For most applications, that is too much.

I think you instead have to identify why it fails. There are typically a couple of places to look: 1) the scaling of the model is far from the scaling of your subject and you should improve the scaling manually. 2) The initial guess on some local marker coordinates are far off and should be manually improved.

Best regards

Hi Michael,

Thank you for your response.

I will try the suggested steps.

Rajan Prasad