Scanning the landscape of axion dark matter detectors: applying gradient descent to experimental design

TL;DR

This paper introduces a novel gradient descent approach to optimize axion dark matter detector designs, demonstrating its effectiveness in outperforming human-designed experiments and opening avenues for more complex future designs.

Contribution

It applies gradient descent to experimental design, providing a proof of concept for optimizing axion detectors and surpassing traditional design methods.

Findings

Gradient descent can optimize detector properties effectively.

Optimized detectors outperform initial human-designed experiments.

Potential for exploring complex multi-dimensional detector designs.

Abstract

The hunt for dark matter remains one of the principal objectives of modern physics and cosmology. Searches for dark matter in the form of axions are proposed or underway across a range of experimental collaborations. As we look to the next generation of detectors, a natural question to ask is whether there are new experimental designs waiting to be discovered and how we might find them. Here we take a new approach to the experimental design procedure by using gradient descent techniques to search for optimal detector designs. We provide a proof of principle for this technique by searching 1D detectors varying the bulk properties of the detector until the optimal detector design is obtained. Remarkably, we find the detector is capable of out-performing a human designed experiment on which the search was initiated. This opens the door to further gradient descent searches of more complex 2D and 3D designs across a wider variety of materials and boundary geometries of the detector. There is also an opportunity to use more sophisticated gradient descent algorithms to complete a more exhaustive scan of the landscape of designs.