Studies and Solvers Updates

COMSOL Multiphysics® version 6.0 includes improved mesh adaptation for curl element formulations, performance improvements for cluster computations, and Component Mode Synthesis for modal reduction. Learn about all of the updates relating to studies and solvers below.

Improved Mesh Adaptation for Curl-Element Formulations

The mesh adaptation method has been extended with a new interpolation error estimator. This makes more efficient mesh adaptation possible for the curl-element formulations used in the RF Module. Both curl elements of type 1 and 2 are supported. A new Frequency Domain, RF Adaptive Mesh study makes the workflow much easier when setting up mesh adaptation for modeling microwave and millimeter-wave antennas and circuits.

A microstrip patch antenna model in the Thermal color table showing the adapted mesh. The adapted mesh of the RF Module microstrip patch antenna inset model using the new RF Adaptive Mesh study.
A 1D plot with a green and blue line with S11 on the y-axis and frequency on the x-axis. The RF Module microstrip patch antenna inset model using the new RF Adaptive Mesh study. The S11 parameter for the initial and final mesh.

Performance Improvements for Cluster Computing

The core machinery for finite element assembly has been improved for cluster computing. Data locality and mesh element partitioning has been improved, both contributing to an overall improved performance for cluster computing. These improvements apply to both multigrid and domain decomposition preconditioners.

A 1D plot with blue, yellow, and green lines with time on the y-axis and nodes on the x-axis. CPU-time comparison between version 5.5, 5.6, and 6.0 for a mesh-refined version of the Ahmed body model (6.3M DOFs) when using the new Nested Dissection partitioning method and all cores on each node are used by multithreading. On 16 nodes, the best performance is obtained by running 2 processes on each node. The solution time is then 3223 s when using Multigrid and 1707 s when using Domain Decomposition. As a comparison, on a single core, the solution time is 87,757 s using Multigrid.
A 1D plot with blue, yellow, and green lines with time on the y-axis and nodes on the x-axis. Memory comparison between version 5.5, 5.6, and 6.0 for a mesh-refined version of the Ahmed body model when using the new Nested Dissection partitioning method and all cores on each node are used by multithreading.

New Coarse Grid Method for Domain Decomposition

It is now possible to create a coarse grid correction with the Nicolaides method. This method can be used to construct a very aggressive coarsening down to just a few unknowns. The method can be used to strike the balance between robustness and performance for the domain decomposition solver, where a standard coarsening method can be too expensive of an alternative and where running without a coarse grid does not work.

New Domain Decomposition Method for Pressure Acoustics

It is now possible to solve large-scale pressure acoustics (Helmholtz problems) with the Domain Decomposition (Schwarz) method. This method is using the Shifted Laplace method together with the same absorbing boundary conditions for the internal overlapping boundaries as is used for the nonoverlapping Schur method. The advantage with this method is that multigrid can be used as a domain solver and a coarse grid is not needed for the domain decomposition method.

New Algebraic Multigrid Methods

The standard, or classical algebraic, multigrid method has been extended with a new coarsening method called Parallel modified independent set. This method supports cluster computing. Furthermore, the Auxiliary-Space AMG method has been extended to support complex-valued curl-element formulations.

Improvements to Batch and Cluster Computing

The Batch and Cluster Computing study types, including the sweep variants Batch Sweep and Cluster Sweep, have been improved. For Batch and Cluster Computing, you can now synchronize the solutions and accumulated probe tables from the batch process. A new cleanup method is available that removes synchronization data for the external process or processes. These new methods are enabled, by default, for the study types Batch and Batch Sweep.

New Ways to Generate Default Plots

You can now generate, or reset, the default plots before solving. Right-click on any study and choose either Show Default Plots or Reset Default Plots.

New Approximate Schur Complement Method for the Vanka Solver

The Vanka solver is extended with a new approximate factorization method for its matrix blocks. When using the Block solver method Direct, stored factorization, there is now an option to Use approximate factorization that is using a Schur complement approximation for larger blocks. This method can save significant memory and CPU time for large blocks, as encountered in, for example, large 3D fluid flow models with the fully developed inflow boundary conditions. This method is available both from the Vanka solver as well as from the SCGS solver with the Vanka option enabled.

Craig–Bampton Method for Model Reduction

Component mode synthesis, based on the Craig–Bampton method, is now available in the Structural Mechanics Module, Multibody Dynamics Module, and Rotordynamics Module. The Craig–Bampton method is used to extend the Modal method for Model Reduction, which can now have model inputs in the constraints. The method takes so-called constraint modes as input, as an extension of the standard modal basis. These modes can be computed in a separate standard Stationary or Parametric Sweep study. Furthermore, modal-reduced-order models now support both a stateless and a stateful form. The stateful form exposes the reduced system of equations to the COMSOL equation machinery, which makes coupling of reduced models to other equations easier. As a side effect of this, the modal degrees of freedom become available for postprocessing or further modeling.

A model with a foundation and rotor showing the von Mises stress. Stress of the foundation and rotor from a component mode synthesis simulation using the Rotordynamics Module.
A closeup view of a 1D plot with green and blue lines with displacement on the y-axis and time on the x-axis. Comparison of the vertical displacement of the right end of the platform between full and reduced model.

Reuse of Solution for Parametric Eigenvalue Problems

It is now possible to select the start vectors when solving parametric eigenvalue problems. This functionality can save computational time for eigenvalue problems with solutions that are smoothly varying with a change in the parameter. Examples can be found in the Semiconductor Module, in models that use the Schrödinger-Poisson interface. With the Eigenvalue search method set to Manual, you can often reduce the number of iterations by at least 50%.

Time-Explicit Multiphysics dG-FEM Hybrid Method

It is now possible to solve for elastic waves coupled to a piezoelectric domain using a time-explicit discontinuous Galerkin-FEM hybrid method. The piezoelectric part of the problem is formulated with the finite element method and is coupled to elastic waves formulated with the time-explicit nodal discontinuous method. The method can handle large elastic domains and is applicable to the design of ultrasound-based medical devices.

An angle beam model showing the pressure and shear waves in the Twilight color table. The pressure and shear waves of an angle beam model computed using the Piezoelectric Waves, Time Explicit multiphysics interface.
A 1D plot with a green and blue line showing the terminal voltage of an angle beam model. Application of the Piezoelectric Waves, Time Explicit multiphysics interface in an angle beam nondestructive testing (NDT) setup.