Speed Up External Superelement Reduction by 30x with AMSR in MSC Nastran

5 Aug 2026 • 3 minute read

MSC Nastran pioneered superelements almost four decades ago and, still today, continues to push the limits of what's possible. Automatic Multilevel Static Reduction (AMSR) is a step in this evolution.

What's a Superelement?

If you are already an MSC Nastran user, you probably have used superelements multiple times by now. Superelements are a powerful modeling technique that can significantly improve computational efficiency for large systems while also helping protect proprietary model details. Think of superelements as a compressed, accurate version of the detailed model. Instead of carrying, let's say every single internal node of a full body-in-white or a subframe, you replace part of the model with an equivalent component that exposes only a set of boundary degrees-of-freedom (A-set) to the rest of the assembly. Internally, the stiffness, mass, and loads are transformed so that, from the outside, the superelement behaves like the full model, but with a fraction of the degrees of freedom and data volume.

To create that superelement, you need to eliminate the interior DOFs while keeping the response at the boundary accurate. Static condensation is a classic way to do this. You start from the full stiffness system, split it into boundary (a) and interior (o) sets, and then solve for the interior DOFs in terms of the boundary DOFs. Substituting back yields a reduced stiffness matrix defined only on the A‑set, plus a transformation that lets you recover interior quantities when needed. Conceptually, you are "folding" the interior of the structure into the boundary so the global system sees a much smaller problem with the same static behavior at the interfaces.

Where AMSR Comes In

For very large external superelements, the "Creation Run" (also known as the "Reduction run") is typically the most time-consuming portion of the entire analysis. Much of the "Creation Run" time is spent computing the static reduction global transformation matrix. If there is a large number of boundary (A-set) points, the whole solution can quickly become computationally intensive.

MSC Nastran can make use of the Automatic Multilevel Static Reduction method, which is designed to compute the global transformation matrix in a very efficient manner, resulting in significant performance improvements.

In static analysis, the global transformation matrix is computed by solving the following system of equations:

System of Equations

K = Stiffness matrix
O = Interior DOF (O-set)
A = Exterior (Boundary) DOF (A-set)
G = Global Boundary Transformation Matrix

This calculation is performed in two steps: a factorization of [Koo ], and a solve step using the lower factor matrix via forward-backward substitution (FBS). When [Koo ] is large, the FBS operation can be very time-consuming.

The new Automatic Multilevel Static Reduction (AMSR) capability computes the [Goa ] matrix and the required reduced quantities (for example, stiffness, mass, or loads) efficiently. The AMSR method employs a domain decomposition to automatically generate sub-components of [Koo], and then performs a multi-level static condensation. This numerical procedure is exact; no approximation is introduced.

Example

An example of improved performance using the AMSR method is presented below. Model and analysis details:

Number of grid points: 2.1 million
Number of global DOF: 12 million
Number of 3-d elements: 1.2 million
A-set size: 36,000
Command line memory specification: mem=max memorymax=500gb

Overall Model Performance: The blue line indicates no AMSR used, whereas the orange line indicates AMSR used in the run. The x-axis numbers represent SMP scaling (Serial, 2, 4, 8, 16, and 32) for the same model.

Summary

In recent large‑scale industrial benchmarks, as well as the case study presented above, we notice that:

Large, expensive FBS module avoided
Speed-up factors in overall runtimes range from 20-40x when AMSR is enabled compared to older workflows without AMSR
Database I/O significantly reduced (in the example above, from 3.4TB to 139GB)
Scratch database required significantly reduced (in the example above, from 1.7TB to 123GB)
Newer MSC Nastran releases deliver similar AMSR performance with fewer threads than earlier versions, thanks to continued optimization.

AMSR does not change what a superelement is or how it behaves in your system model, nor does it change its accuracy. It is still an exact method that changes how quickly and efficiently you can reach your results. MSC Nastran pioneered superelements and, with AMSR, it continues to extend what is practical on real hardware: larger A‑sets, shorter runtimes, and far less I/O and scratch. If external superelements are central to your workflows, AMSR is not just a nice optimization; it is a capability you should actively plan to adopt.

Learn more about Cadence MSC Nastran.