Automatic Convergence Assessment for the Scalable Solver

In an effort to continuously evolve the solution efficiency of the Autodesk® CFD, the Scalable Solver is available as a solver option. This solver is designed specifically for large simulations containing on the order of 10 to 50 million elements. Unlike the original Classic Solver, the Scalable Solver enables each process to run on multiple threads as a process/thread hybrid and is designed to optimize cache performance. The solver has shown significant solution time speedup for large models when using multiple compute nodes. The performance for smaller models on a single compute node is comparable to that of the default Classic Solver.

Automatic Convergence Assessment

Solver Auto-stop

Automatic Convergence Assessment determines when a solution is converged (that is, the solution stops changing), and automatically halts the calculation. It examines small and large frequency changes throughout the solution field, and evaluates the local and global fluctuations of each degree of freedom. This removes the guess-work of knowing when a solution is completed.

Convergence Criteria is assessed for each degree of freedom by looking at the last fifty values of min, max and average data. The average data is fitted by a linear regression to yield a slope. The min, max data is used to derive a data range. This range data is used to compute a normalized slope:

A variable is considered converged when the value is less than the convergence criteria.

This approach is best suited for flow fields that attain monotonic convergence. In turbulent flow, monotonic convergence rarely happens, and there is a certain amount of noise in the average data that will never go away. If this noise is greater than your convergence criteria, full convergence will never be obtained.

As a secondary criterion, the RMS behavior is observed over longer iteration intervals. A variable is considered converged when the RMS slopes meet our convergence criteria.

There are four phases that the solver steps through in an analysis to determine convergence:

Start-up Phase

The first fifty (50) iterations, advection scheme ADV 1, is used to establish the basic flow field. After fifty iterations the selected advection scheme is enabled in the solver. This phase is initiated to promote stability early on in the analysis.

Flow Convergence Phase 1

The pressure equation is solved for an exact pressure every 50 iterations, to correct machine roundoff. The exact pressure solver is also invoked if the mass balance has more than five percent error. This method continues until the component velocity and pressure terms meet the convergence criteria (temperature and turbulence quantities have not met the convergence criteria).

Flow Convergence Phase 2

The exact pressure is obtained for every iteration until the component velocity, pressure, and the turbulence terms meet the convergence criteria.

Thermal Convergence Phase

The flow solution is locked, similar to the way that Auto Forced Convection worked with CFD 1. Exact solvers are used to obtain a full steady state solution to the thermal problem. Once the thermal solution has been completed the analysis is fully converged with regards to all degrees of freedom and the analysis will automatically stop.