Studies of algorithm generation with large language models for energy reconstruction in germanium detectors

A toy Monte Carlo study of energy reconstruction for an inverted-coaxial point-contact germanium detector. All programs were written by a commercial large language model in a dialogue with the author, who proposed the methods and judged the results; no data of any experiment were used. The study compares the trapezoid, cusp and optimum linear filters with filters and charge-collection corrections found by symbolic regression with the single operator eml(x, y) = exp(x) − ln(y), with a polynomial and a neural network as controls, and reports learning curves and tests against artificial sharpening of peaks. Two further tests are included: eml expressions distilled from a neural network, and the energy obtained from a small causal transformer that predicts the next sample of the pulse. The note lists what the model implemented at once and where it was wrong. The complete code and the numerical results are attached. Text and figures: CC BY 4.0; code: MIT licence (see README).

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-17
DOI
https://doi.org/10.5281/zenodo.24318076
Primary Topic
Particle Detector Development and Performance
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Studies of algorithm generation with large language models for energy reconstruction in germanium detectors

M. Misiaszek
3 links
Zenodo (CERN European Organization for Nuclear Research)
Particle Detector Development and Performance
article

Studies of algorithm generation with large language models for energy reconstruction in germanium detectors

M. Misiaszek
article en

Abstract

A toy Monte Carlo study of energy reconstruction for an inverted-coaxial point-contact germanium detector. All programs were written by a commercial large language model in a dialogue with the author, who proposed the methods and judged the results; no data of any experiment were used. The study compares the trapezoid, cusp and optimum linear filters with filters and charge-collection corrections found by symbolic regression with the single operator eml(x, y) = exp(x) − ln(y), with a polynomial and a neural network as controls, and reports learning curves and tests against artificial sharpening of peaks. Two further tests are included: eml expressions distilled from a neural network, and the energy obtained from a small causal transformer that predicts the next sample of the pulse. The note lists what the model implemented at once and where it was wrong. The complete code and the numerical results are attached. Text and figures: CC BY 4.0; code: MIT licence (see README).

Zenodo (CERN European Organization for Nuclear Research)
Jagiellonian University (PL)
Particle Detector Development and Performance
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon