[Paper Review] Beyond the $\sqrt{\mathrm{N}}$ limit of the least squares resolution and the lucky model

Gregorio Landi, Giovanni Landi|arXiv (Cornell University)|Aug 20, 2018

Particle Detector Development and Performance7 references4 citations

TL;DR

This paper proposes a novel fitting approach that achieves linear resolution improvement with the number of detector layers (N), surpassing the conventional √N limit in least squares fitting. By using effective variances derived from center-of-gravity hit positioning—formalized in the 'lucky model'—the method enables linear growth in resolution, validated through Gaussian and realistic silicon microstrip detector simulations.

ABSTRACT

A very simple Gaussian model is used to illustrate a new fitting result: a linear growth of the resolution with the number N of detecting layers. This rule is well beyond the well-known rule proportional to $\sqrt{N}$ for the resolution of the usual fit. The effect is obtained with the appropriate form of the variance for each hit (measurement). The model reconstructs straight tracks with N parallel detecting layers, the track direction is the selected parameter to test the resolution. The results of the Gaussian model are compared with realistic simulations of silicon microstrip detectors. These realistic simulations suggest an easy method to select the essential weights for the fit: the lucky model. Preliminary results of the lucky model show an excellent reproduction of the linear growth of the resolution, very similar to that given by realistic simulations.

Motivation & Objective

• To demonstrate that resolution in least squares fitting can grow linearly with N, the number of detector layers, rather than as √N.
• To address the limitations of standard least squares fitting, which assumes homoscedasticity and fails to exploit detector-specific hit variance information.
• To develop and validate a simplified yet effective fitting model—the 'lucky model'—that replicates the resolution performance of complex likelihood methods with minimal computational cost.
• To provide a practical, experimentally verifiable method for improving track reconstruction resolution in high-energy physics detectors.

Proposed method

• A simple Gaussian model is constructed to simulate straight tracks crossing N parallel detecting layers, with hit positions determined by a center-of-gravity (COG) algorithm.
• Effective variances (weights) for each hit are derived from the COG's statistical properties, particularly its dependence on signal amplitude and strip position.
• The 'lucky model' applies weighted least squares using these derived weights, where higher signal density (better hits) receives greater weight, mimicking optimal fitting behavior.
• The schematic model, a simplified version of the full likelihood method, is used to initialize and validate the fitting process, focusing on effective variances to avoid convergence issues.
• The method is tested using both idealized Gaussian simulations and realistic silicon microstrip detector simulations, comparing resolution performance across different N values.
• The linear resolution growth is quantified by fitting track direction and measuring the standard error of the estimate as a function of N.

Experimental results

Research questions

• RQ1 Can resolution in track reconstruction grow linearly with the number of detector layers, N, rather than as √N, under realistic conditions?
• RQ2 What is the role of detector-specific hit variance estimation in achieving superior resolution beyond the √N limit?
• RQ3 How does the 'lucky model'—a simplified weighted least squares approach based on hit signal amplitude—compare to complex likelihood methods in resolution performance?
• RQ4 What experimental conditions are necessary to verify the linear resolution growth in a real test beam setup?
• RQ5 To what extent can the center-of-gravity algorithm's statistical properties be leveraged to improve fitting accuracy without full likelihood computation?

Key findings

• The Gaussian model demonstrates that resolution improves linearly with N, achieving a resolution that scales as 1/N, in stark contrast to the standard √N scaling.
• The schematic model, which uses effective variances extracted from the full likelihood probability distributions, reproduces the linear resolution growth observed in complex simulations.
• The 'lucky model'—a weighted least squares fit using weights derived from the COG histogram—achieves resolution performance within 10% of the schematic model, despite its simplicity.
• The linear growth in resolution is robust even when using the three-strip COG (COG3), though the resolution is reduced by approximately 50% compared to COG2 due to noise and discontinuities.
• The functional form of the inverse variance (σ⁻¹) as a function of hit position η matches the average MIP signal shape, confirming the physical basis of the lucky model's weights.
• Experimental verification is feasible with a high-precision test beam (divergence < 10⁻⁵ rad), where the difference in fitted track direction between full and reduced layer sets should show a parabolic increase under the lucky model, unlike the small increase expected under standard fitting.