Experimental Validation
The decisive test of a model is comparison to measurement, with quantified uncertainty on both the simulation and the experiment.
Comparing to Reality
Validation is the confrontation of a model with the physical world. A verified code is run for a configuration that has been measured, and its predictions are compared to the data. The comparison is only meaningful when both sides carry uncertainty: the experiment has measurement error and setup uncertainty, and the simulation has numerical error and input uncertainty. Agreement within combined uncertainty is the standard; agreement of central values alone is not.
The Validation Experiment
A validation experiment differs from a discovery experiment. It is designed to test a model, which means the boundary conditions, material properties, and inputs the model needs must all be measured, not assumed. An experiment that measures only the output while leaving the inputs uncertain cannot cleanly test a model, because a discrepancy could be blamed on unknown inputs.
The Comparison Metric
- Define the quantity of interest before comparing, so the test is not chosen to flatter the model.
- Combine experimental and numerical uncertainty into a validation uncertainty band.
- Report the discrepancy and its uncertainty, not a binary pass or fail.
Hierarchy and Extrapolation
Real systems are usually validated in pieces. Unit-physics experiments test individual phenomena; subsystem experiments test their coupling; the full system is often beyond direct test. Confidence in the full-system prediction is then an extrapolation supported by the validated pieces, and the extrapolation itself carries uncertainty that must be stated honestly.
For fusion designs that are not yet built, including the Hyperion breeder and the burner concepts, validation rests on comparison to existing devices, dedicated experiments, and validated component physics. No hardware net-gain claim is made before first-of-a-kind operation, precisely because that claim can only be validated by the machine itself. Naming the boundary between validated physics and design projection is part of honest validation.