COMSOL rrecently held a User Conference at the Hyatt Regency in Cambridge Mass. Oct. 22 - 24. The program was interesting, with themes such as Multiphysics, Quantum Mechanics, and Fluid-Structure Interactions and minicourses such as Introduction to COMSOL Multiphysics, Turbulent Fluid Flow, and PDE Modeling Introduction.

There were various fascinating Keynote Sessions. For example, Hur Koser, assistant professor at Yale University, discussed "Magnetic Liquids for Lab-on-a-Chip and Rapid Diagnostics Applications." His talk covered ferrofluids, colloidal suspensions of tiny ferromagnetic particles in aqueous or oil-based media. They are used as drug-delivery agents, and provide an attractive alternative to mechanical components in labs-on-a chip. He focused on FE studies of the fluids using the multiphysics solvers in COMSOL software.

......................

It’s helpful to know some of the lingo when talking about multiphysics. This involves systems with more than one physical phenomena, such as temperature, heat flow, and electromagnetic fields. John Dunec of COMSOL says a “coupling variable” basically takes data from one part of a model and puts it in another and often links simple and complex systems. Modeled systems comprise a “domain,” or a group of geometries. For example, a model of a cup of coffee would be a cup domain with a coffee subdomain. Lastly, “application modes” are basically the equations that are built into the software.

He says physical phenomena can be modeled in three ways:

-- Divide everything into a point cloud. This then becomes a problem similar to solving a big resistor network.
-- Finite volume: Divide everything into points, but also say something about the energy representation of each.
-- Finite elements: What is going on is represented by polynomial equations.

I took an introductory hands-on course for the software. The instructor explained COMSOL Multiphysics is used to describe the world in partial and ordinary differential equations, while COMSOL Script provides a programming language that talks to C, Fortran, Java, Spice, and Excel. The basic equation is called a Coefficient Form equation, a second-order partial differential equation. Most linear systems fit into this form. Any time there is a second-order derivative in time, it is a wave equation. In a steady-state system, the time derivitaves are zero. Another kind of equation includes the General Form equation, which is more compact than the others.

Attendees at a User Forum on Numerical Methods discussed what to do when a problem won’t converge. The first thing to do, they say, is check the physics. Does the model behave like the physics says it should? Are the equations appropriate? Are the numbers the correct magnitude? Things to then try include play with the mesh sizes (use bigger ones), use a parametric solver, and maybe refine a mesh in a certain area.

The group also discussed some of the different solvers. 2D or linear solvers are known as sparse direct solvers, or Gaussian solvers. They are limited to 60,000 degrees of freedom. 3D solvers are the default for iterative solvers. A new one called the Pardiso solver allows parallel modeling on multiple processors.

Another topic: how do you know simulation results are realistic? This involves both verification and validation, says a COMSOL engineer. Verification answers the question, “Am I building the model right?” In other words, does it follow the laws of physics? Validation answers, “Am I building the right model.” He says the process is a lot like chess -- built on experience and art.

This message has been edited. Last edited by: Leslie Gordon,

Hi There,
I am a user of ANSYS simulation software and find it very near the goal of any realistic results. However there is no gain without pain so there should be a correlation results with some testing. so no simultaion is going to give you the perfect results. there you are!.

Now you have me curious: What kind of systems are you studying using ANSYS software? And what kind of testing do you perform? Do you always back-up your simulations with real-world testing? Thanks.....