Flamedisx

Fast likelihood analysis in more dimensions for xenon TPCs.

Flamedisx aims to increase the practical number of dimensions and parameters in likelihoods for liquid-xenon (LXe) detectors, which are leading the field of direct dark matter detection.

Traditionally, particle physicists compute signal and background models by filling histogram 'templates' with high-statistics Monte Carlo (MC) simulations. However, the LXe model can also be computed with a series of (large) matrix multiplications, equivalent to the integral approximated by the MC simulation. Using TensorFlow makes this computation differentiable and GPU-scalable, so it can be used practically for fitting and statistical inference.

The result is a better sensitivity, since the likelihood can use all observables, and more robust fits, because using simultaneous correlated nuisance parameters no longer requires challenging interpolation and template morphing.

Getting started

To get started, Launch our tutorial on Colaboratory, or view it statically on GitHub or ReadTheDocs.

Our paper gives a detailed description of Flamedisx, and compares Flamedisx quantitatively to traditional template-based methods.

If you want all the details, see the Flamedisx Documentation and our Notebooks repository.

FlameNEST

Since version 2.0.0, flamedisx includes an implementation of electronic and nuclear recoil models from the Noble Element Simulation Technique. To use this, use sources from the fd.nest subpackage, e.g. fd.nest.ERSource. See the flameNEST paper for a detailed description and validation.

As of April 2024, we implement NEST version 2.3.0, which was released November 2021 (see #249).