Science › Physics1 min read

Physicists propose MEP emulator that predicts observables directly from observables, tested on 28O

The multiparameter eigenvalue problem emulator skips parameter estimation, the costly step of fitting theoretical models to data, and also yields predictive probability distributions.

Physicists propose MEP emulator that predicts observables directly from observables, tested on 28O
Photo: innovanews.ru

A new surrogate model, the multiparameter eigenvalue problem (MEP) emulator, is described in Physical Review Letters. According to the authors, it connects existing emulators and makes predictions directly from observables to observables aps.orgPhysical Review Lettersaps.orgOpen source ↗. The method was tested on a simulation of a one-dimensional lattice and on the nucleus 28O.

Why parameters are a bottleneck

Physicists often fit theoretical models to experimental data in order to predict phenomena that have not yet been observed. Realistic models contain many numerical quantities, called parameters, that cannot be measured directly. Researchers have to estimate these parameters before computing other observables.

This step demands large computing resources. It also makes it difficult to determine how uncertainties in the parameters propagate into the final predictions. According to the description of the work, estimating parameters accounts for most of the computing effort in such studies.

What the MEP emulator does

The MEP emulator predicts unknown observables from relations between known data, without introducing parameters and without estimating them aps.orgPhysical Review Lettersaps.orgOpen source ↗. The authors show that it can be trained with data from two other types of emulators: eigenvector continuation and parametric matrix model emulators.

A simple simulation on a one-dimensional lattice confirmed the performance of the emulator. According to the published description, the method reproduces complex behaviour that conventional computational approaches handle with difficulty.

As a nuclear-physics example, the authors used 28O. They demonstrate that the predictive probability distribution of the target observables can be obtained straightforwardly with the new emulator. The description of the work adds that this allows fast large-scale calculations and a systematic assessment of the uncertainty of predictions.

Sources · 2

Found a mistake?
  1. ↩

Read also

Up next

The next story opens automatically

More
Follow us

Type at least two letters — results appear instantly

↑↓to choose↵to openescto close