"Model Convergence and Optimizing Runtimes with HydroGeoSphere" - Aquanty Webinar
A recording of our August 16th, 2023 webinar focused on building high-quality HydroGeoSphere models that are more likely to converge on a solution, and optimizing those models and numerical criteria to reduce model runtimes. Here we review a number of different numerical concepts with a focus on practical guidance on how to approach convergence optimization.
The first half of this webinar is delivered by Brayden McNeill, who discusses general advice for building HydroGeoSphere models. We review the importance of building a strong conceptual model and the data associated with different HGS processes, the potential pitfalls associated with poorly conceptualized boundary conditions, the benefits of modeling in stages, and introduce several time-saving commands.
The 2nd half of the webinar is delivered by Dr. Killian Miller, with a focus on optimizing the numerical performance of your models. Dr. Miller reviews the various solver methods and options available to HGS users (finite difference vs finite element; control volume vs standard finite difference; solver acceleration techniques; underrelaxation factor; dual node vs common node formulation; etc.). We also investigate solver outputs to help you understand how the model is performing, reviewing simulation progress output, the summary of nonlinear iteration and adaptive timestepping summary. Finally, we provide recommendations for optimizing convergence of different types of problems and numerical setting recommendations for different types of problems.
Best practices for model convergence
Building conceptual models
Data requirements and preparation; Developing boundary conditions
The importance of a high-quality triangular element mesh
A staged modelling workflow
Start simple, then introduce more complexity; “Steady-state” vs transient flow
Understanding model results
Reviewing water balance: error, changes in storage, fluxes
General tips and tricks
Time-saving commands at each stage of the modelling workflow
Model runtime optimization
Review of solver methods and options
Finite difference vs finite element; Control volume vs standard finite difference; Solver acceleration techniques; Upstream weighting factor, central weighting; Underrelaxation factor; Dual node vs common node formulation
Review of solver output, i.e. understanding phgs.exe runtime output
Simulation progress output; Summary of nonlinear iteration; Adaptive timestepping summary
Recommendations for optimizing convergence of different types of problems
Numerical setting recommendations for different types of problems
How to read runtime output, which variables to monitor, how to respond and improve runtime
Review debug.control and restart functionality