Searcharxiv⌕ Search

arXiv · 2611.19355

Numerical Integration Instabilities in Likelihood Fits: Mechanisms and Remedies

Abstract

Numerical instabilities in the normalization of probability density functions can produce artificially lower negative log-likelihood (NLL) values than the true minimum of a likelihood fit, even when the resulting fit provides a poorer description of the data. Motivated by anomalously low NLL values found in non-converged fits associated with a recent CMS search for fully charmed tetraquark candidates, we use a pseudo-dataset generated from a likelihood model based on that analysis to investigate how numerical integration can yield unreliable normalization results, how these errors affect likelihood minimization, and how they can be addressed. We find that, even at fixed integration tolerance, the stopping decision of a numerical integrator can change with the fit parameters, introducing discontinuities in the normalization integrals that propagate to the likelihood surface. Such discontinuities can trap the minimizer at a numerical step and yield an artificially lower NLL. Tightening the integration tolerance mitigates these numerical discontinuities, while adaptive integration methods, such as Gauss--Kronrod quadrature, can also improve the numerical stability. A more severe failure occurs when the integration grid cannot adequately resolve a sharply localized contribution. In this case, the integrator may satisfy its estimated-error criterion despite substantially underestimating the normalization, allowing the minimizer to enter a region with an artificially low NLL. Tightening integration tolerances or using alternative adaptive integration methods does not necessarily resolve sharply localized contributions. We demonstrate that a suitable variable transformation can improve their numerical resolution, yielding more accurate normalization integrals and restoring fit convergence.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Manqi Zhu, Gerry Bauer, Kai Yi, Xining Wang. 2026-11-14. Numerical Integration Instabilities in Likelihood Fits: Mechanisms and Remedies. https://searcharxiv.com/abs/2611.19355

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

A Paired Monte Carlo Intensity-Ratio Method for Incorporating Detector Resolution in Amplitude Analysis

Finite detector resolution can bias resonance parameters and fit fractions extracted from amplitude analyses. We present the paired Monte Carlo intensity-ratio method for incorporating resolution effects into amplitude analysis. The method uses ratios of reference intensities evaluated at paired truth-level and reconstructed kinematics to reweight the Monte Carlo normalization and approximately account for detector resolution. A resampling study of simulated $D^+\to K^+π^+π^-π^0$ decays shows that the method substantially reduces resolution-induced shifts in fitted resonance widths. Likelihood evaluation uses intensity calculations and normalization sums, avoiding a response convolution at each data event.

physics.data-an↗

Distinguishing thyroid nodule malignancy using digital palpation

Accurate pre-operative diagnosis of thyroid nodules remains a major clinical challenge. Current diagnostic techniques rely on fine-needle aspiration and demonstrate high rates of inconclusive results, often leading to avoidable thyroidectomies with lifelong implications. Here, we introduce a mechanics-based diagnostic method that leverages Digital Palpation to determine malignancy and identify tissue subtypes in human thyroid nodules. From a cohort of patients undergoing thyroidectomies, we evaluated ex vivo thyroid specimens and processed the resulting mechanical data from 36 tests in 10 thyroid nodules. Principal Component Analysis reveals that these experimental results are distributed within a three-dimensional space of interpretable scalar features reflecting tissue grittiness, strength, and toughness, which map closely to descriptions of malignant and benign nodules given by clinicians. Our results further demonstrate that malignant nodules exhibit distinct mechanical signatures compared to their benign counterparts. This revealed a statistically significant difference (p = 9 x 10^-4) and separability (95% lower-bound Hedges' g = 1.08) between benign and malignant nodules along a direction in principal component space representing effective stiffness. Preliminary clustering patterns were also observed among malignant subtypes, including follicular, oncocytic, and papillary thyroid carcinomas. Together, these results highlight the potential of Digital Palpation as a novel mechanics-based method for tissue characterization and motivate its further development toward diagnostic applications.

physics.data-an↗

Learning transferable event representations for charmed baryon physics at BESIII

Deep learning has become an essential tool in high-energy physics, where the ability to learn transferable event representations can significantly improve model generalization across related physics processes. In this work, we present a Particle Transformer-based framework for learning such representations for charmed baryon physics in the BESIII experiment. The framework is implemented through large-scale pre-training on Monte Carlo simulation samples and subsequent fine-tuning for downstream analyses. Using the production and decays of the charmed baryon $Λ_c^+$ as a benchmark, we develop pre-trained models for both event classification and momentum-direction regression. The classification model learns discriminative event representations for the dominant physics categories, rejecting $96.90\%$ of background events at a signal efficiency of 90.00\%. Across 12 benchmark $Λ_c^+$ decay channels, fine-tuning from the pre-trained model achieves performance comparable to or better than training from scratch, with particularly clear improvements in low-statistics regimes. For the regression task, the pre-trained model improves the average angular resolutions across the same benchmark channels. Further improvement is obtained after fine-tuning in the representative semileptonic decay $Λ_c^+ \to p K^- e^+ ν_e$. In addition, the predicted momentum direction enables reconstruction of the parent kinematics without a recoil-side tag and thereby increases signal efficiency. For this semileptonic decay, the pre-trained regression model increases the signal efficiency by approximately one order of magnitude relative to the double-tag method. This strategy provides a scalable solution for a wide range of physics cases at BESIII and can be extended to other high energy experiments.

physics.data-an↗