{"sections":[{"heading":"Establish a baseline","steps":["Start with the smallest representative geometry and record a completed baseline run, versions, mesh settings, element count, time and the quantities of interest.","Refine one setting at a time while holding the physical geometry, materials, ports and extraction model fixed.","Compare the relevant output changes and check mesh aspect ratios, connectivity and solution-quality diagnostics.","Stop treating a result as converged if refinement introduces topology changes, failed elements or incomplete solver stages. Investigate those changes first."]},{"heading":"Read the quality indicators","paragraphs":["The manual warns that poor segment aspect ratios can degrade accuracy and stability. It also distinguishes field/mesh quality from the quality of a lumped-model fit: neither element count nor one small residual proves the complete result.","Historical support includes slow dense capacitance meshes, many-hole layouts and version-specific mesh defects. It does not establish one mesh setting or promised speedup for every design."]},{"heading":"Controlled worked comparison","paragraphs":["The linked mesh-settings guide works through three independent SegmentSize cases on the same complete MQS cuboid strip. Reducing 2 to 1 to 0.5 um increases filaments from 1878 to 7216 to 27216, while L1 moves from 3.896479063 to 3.872526244 to 3.851454130 pH. The successive changes are about 0.61853% and 0.54712%, using the finer result as denominator.","This is measured mesh dependence, not convergence acceptance. The finer case remains further from the fixed 3.9 pH design target despite smaller fit error. Keep design, fit, solver and mesh-change criteria separate. The worked guide records time/memory methods, unchanged baseline recovery and their limits."]},{"heading":"For a performance report","paragraphs":["Provide the version, OS/architecture, available memory, element count, first failing stage and a controlled comparison. Use generalised or approved example inputs when sharing the problem."]}],"canonical_url":"https://www.sun-magnetics.com/knowledge-base/inductex-mesh-convergence","url":"/knowledge-base/inductex-mesh-convergence","known_issues":[],"search_record":{"execution":true,"relatedProducts":[],"version_tags":["InductEx 8 preview"],"text":"Check mesh convergence when a run is slow or results change Use controlled refinement and solution-quality checks. More elements alone do not establish a more accurate or stable extraction. InductEx 8 draft guidance; version-specific mesh controls Troubleshooting Establish a baseline Start with the smallest representative geometry and record a completed baseline run, versions, mesh settings, element count, time and the quantities of interest. Refine one setting at a time while holding the physical geometry, materials, ports and extraction model fixed. Compare the relevant output changes and check mesh aspect ratios, connectivity and solution-quality diagnostics. Stop treating a result as converged if refinement introduces topology changes, failed elements or incomplete solver stages. Investigate those changes first.  Read the quality indicators The manual warns that poor segment aspect ratios can degrade accuracy and stability. It also distinguishes field/mesh quality from the quality of a lumped-model fit: neither element count nor one small residual proves the complete result. Historical support includes slow dense capacitance meshes, many-hole layouts and version-specific mesh defects. It does not establish one mesh setting or promised speedup for every design.  Controlled worked comparison The linked mesh-settings guide works through three independent SegmentSize cases on the same complete MQS cuboid strip. Reducing 2 to 1 to 0.5 um increases filaments from 1878 to 7216 to 27216, while L1 moves from 3.896479063 to 3.872526244 to 3.851454130 pH. The successive changes are about 0.61853% and 0.54712%, using the finer result as denominator. This is measured mesh dependence, not convergence acceptance. The finer case remains further from the fixed 3.9 pH design target despite smaller fit error. Keep design, fit, solver and mesh-change criteria separate. The worked guide records time/memory methods, unchanged baseline recovery and their limits.  For a performance report Provide the version, OS/architecture, available memory, element count, first failing stage and a controlled comparison. Use generalised or approved example inputs when sharing the problem. ","id":"inductex-mesh-convergence","keywords":"","kind":"Troubleshooting","title":"Check mesh convergence when a run is slow or results change","product":"InductEx"},"version":"8 draft guidance; version-specific mesh controls","id":"inductex-mesh-convergence","related":["/knowledge-base/inductex-negative-resistance","/knowledge-base/inductex-example-result-differences","/knowledge-base/inductex-mesh-size-cost","/knowledge-base/inductex-gmsh-import","/knowledge-base/inductex-package-wirebond"],"downloads":[],"publication_status":"published","numerical_output":null,"provenance":{"execution_checks":[{"routes":["cli"],"scope":"Four fresh normal licensed CLI cases: fixed MQS cuboid strip with SegmentSize 2, 1 and 0.5 micrometres, plus baseline restoration/repeat. Fourteen checks cover input isolation, completed outputs, actual counts, self-reported costs, residuals and repeat identity","result":"passed","articles":["inductex-mesh-size-cost","inductex-mesh-convergence"],"purpose":"controlled-mqs-segmentation-cost-refinement-and-repeat","version":"Windows catalog InductEx 8.0.1 and bundled TTH 1.1.50, 3 November 2026; exact hashes retained","id":"VAL-025","records":["four fresh copied-input runs","separate current detailed mesh/solver artifacts","baseline-repeat with exact mesh and inductance recovery","exact authored PowerShell blocks replayed with four additional checks"],"limit":"One sequential sweep on this Windows host, not a benchmark, convergence acceptance, memory guarantee or universal scaling law. TTH/InductEx process peaks are separately reported, not simultaneous combined memory. Height slicing, physical calibration, GUI and final distributed-package applicability remain unqualified. EQS is excluded.","observed":[{"InductEx_peak_MiB_reported":32.5,"msh_nodes":16038,"job_seconds_reported":4.139,"gmres_iterations_high":51,"SegmentSize_um":2,"TTH_seconds_reported":1.117,"gmres_residual_high":0.0000627,"solver_version":"1.1.50","filaments":1878,"msh_elements":3780,"design_deviation_percent":-0.0902804358974274,"cli_wall_seconds":4.77,"TTH_peak_MiB_reported":49.8,"completed_exit_code":0,"case":"coarse","L1_H":3.896479063e-12,"svd_relative_error_percent":0.000170625,"gmres_iterations_low":51},{"InductEx_peak_MiB_reported":43.8,"msh_nodes":61482,"job_seconds_reported":4.56,"gmres_iterations_high":52,"SegmentSize_um":1,"TTH_seconds_reported":4.056,"gmres_residual_high":0.0000878,"solver_version":"1.1.50","relative_change_using_finer_denominator":0.006185321284035776,"filaments":7216,"msh_elements":14476,"design_deviation_percent":-0.7044552820512862,"cli_wall_seconds":4.893,"TTH_peak_MiB_reported":158.1,"completed_exit_code":0,"case":"medium","L1_H":3.872526244e-12,"svd_relative_error_percent":0.000129005,"gmres_iterations_low":52},{"InductEx_peak_MiB_reported":103.7,"msh_nodes":231622,"job_seconds_reported":15.464,"gmres_iterations_high":56,"SegmentSize_um":0.5,"TTH_seconds_reported":14.86,"gmres_residual_high":0.0000815,"solver_version":"1.1.50","relative_change_using_finer_denominator":0.005471209908970145,"filaments":27216,"msh_elements":54516,"design_deviation_percent":-1.2447658974358955,"cli_wall_seconds":15.81,"TTH_peak_MiB_reported":606.7,"completed_exit_code":0,"case":"fine","L1_H":3.85145413e-12,"svd_relative_error_percent":0.00000664474,"gmres_iterations_low":56},{"InductEx_peak_MiB_reported":32.6,"msh_nodes":16038,"job_seconds_reported":1.455,"gmres_iterations_high":51,"SegmentSize_um":2,"TTH_seconds_reported":0.997,"gmres_residual_high":0.0000627,"solver_version":"1.1.50","filaments":1878,"msh_elements":3780,"design_deviation_percent":-0.0902804358974274,"cli_wall_seconds":1.795,"TTH_peak_MiB_reported":49.8,"completed_exit_code":0,"case":"baseline-repeat","L1_H":3.896479063e-12,"svd_relative_error_percent":0.000170625,"gmres_iterations_low":51}]}],"figure_evidence":[],"review_notes":["Support history informed this draft. Current-source guidance has been checked; a fresh walkthrough on the intended distributed release and product-owner review remain pending.","VAL-025 records fourteen scoped CLI checks and a baseline repeat. A declared numerical stopping criterion, physical calibration and final-package applicability remain pending."],"scope":"Recorded inputs/builds only. Complete walkthrough and final-package acceptance remain separate.","verification_requirements":{"gui_method":"not-required-for-this-answer","status":"review-incomplete","illustrations":"not-required-for-this-answer","required":["technical-review","release-applicability","numerical-interpretation","recovery"],"legacy_walkthrough_field":"pending retained; not evidence of a complete scenario"},"sources":[{"id":"support:SUP-009","section":"Generalised editorial finding; outcome qualification retained"},{"id":"inductex:docs/manual/ix_output.tex","section":"Mesh quality; SVD solution quality"},{"id":"inductex:docs/manual/ix_ldf.tex","section":"SegmentSize and modelling parameters"},{"id":"validation:inductex-mesh-cost-windows-20261103","section":"VAL-025 fixed-input refinement, residual/design distinctions and repeat; no convergence acceptance"}]},"title":"Check mesh convergence when a run is slow or results change","type":"Troubleshooting","markdown_url":"/_functions/kbResource?path=markdown%2Finductex-mesh-convergence.md","summary":"Use controlled refinement and solution-quality checks. More elements alone do not establish a more accurate or stable extraction.","product":"InductEx"}