Currently, JASP throws an error if classical (G)LMMs fail to converge, but does not allow users to adjust optimizer settings. This is a limitation, especially given that optimizer tuning is often required for complex models.

Request: Add optimizer control settings to the classical LMM and GLMM modules, similar to those implemented for meta-analysis. Controls should include (where applicable):

• Optimization method (e.g., nlminb, BFGS, Nelder-Mead, bobyqa, etc.)
• Maximum iterations
• Maximum function evaluations
• Convergence tolerance
• Initial values (where relevant)
• Trust region radius (where relevant)

These options should be surfaced in the Advanced section of UI and passed to the underlying fitting function (e.g., lme4::lmer/glmer), and only be available for classical LMM and GLMM (not Bayesian models).

Motivation: This will allow users to more flexibly resolve convergence issues and match optimizer settings used in other tools/packages.

Context: See meta-analysis modules for example implementation of optimizer controls. This request is for classical LMM and GLMM only, not Bayesian modules.

If possible, allow defaults to match lme4's defaults, but expose all major optimizer settings that impact convergence.

References:

• Conversation and example implementation attached above.
• No optimizer controls currently available, see fitting function for context.

Labels: enhancement, GLMM, LMM, optimizer