← All papers
First page of Closed-form fractional radial links for elliptical Mahalanobis discriminant analysis

Closed-form fractional radial links for elliptical Mahalanobis discriminant analysis

Serhii Zabolotnii

math.ST Jul 7, 2026 · v2 stat.ME stat.ML
Structural algebra (LLR bridge, GAM membership, Gaussian collapse to QDA, non-affinity of Student-t generator) is verified in a sorry-free Lean 4 development.
We study binary classification when the two classes are elliptical with a shared radial generator $g$. The log-likelihood ratio is then an additive function of the two squared Mahalanobis radii, with radial link $\varphi=\log g$. For Gaussian classes $\varphi$ is affine and the rule reduces to quadratic discriminant analysis (QDA); otherwise the Bayes link is not the identity. Instead of fitting this link with a spline smoother, we derive its family from the within-class law of the radius and estimate a finite fractional-power projection. We prove that the link is identifiable from the radius law, that the plug-in estimator is $\sqrt{n}$-consistent and asymptotically normal under finite-moment conditions, and that the induced classifier is asymptotically Bayes-optimal in an iterated sieve limit. For $d\ge2$ and non-nested class parameters, the identity link suffices only for affine (Gaussian) generators. The bridge, GAM membership and the affine direction of this dichotomy are verified in Lean 4. We compare against the global Mahalanobis-GAM of Ghosh et al. (2025), re-implemented with mgcv REML splines at equal input budget. On two UCI benchmarks the derived link ties with the fitted GAM. On six financial series from three asset classes, with validation that respects temporal dependence, it is never significantly worse than the fitted GAM: better on CAD/USD and tied on the other five. It beats QDA on the three heaviest-tailed series (oil $[+0.023,+0.069]$, S&P 500 $[+0.039,+0.125]$, JPY/USD $[+0.009,+0.047]$) and ties elsewhere. A rate simulation with a closed-form ground truth reproduces the $\sqrt{n}$ rate and, under covariance heterogeneity, QDA's risk floor against the derived link's vanishing excess risk. The derived link thus matches a tuned GAM without smoothing-parameter selection and improves on QDA where the generator departs from the Gaussian.

For binary classification with elliptical classes sharing a radial generator, the Bayes log-likelihood ratio is an additive function of the two squared Mahalanobis radii, with link equal to the log generator. The open question is whether this non-identity link should be fitted nonparametrically, as in Ghosh et al.'s Mahalanobis-GAM, or derived in closed form.

The link is derived from the within-class law of the radius and estimated by a finite fractional-power basis (powers 1, 0.5, 1.5) inside regularized logistic regression. Identifiability, sqrt(n)-consistency, asymptotic normality and iterated-sieve Bayes-optimality are proved on paper. The bridge identity, GAM membership, the affine direction of the Gaussian/QDA dichotomy, and the non-affinity of the Student-t generator are machine-checked in Lean 4 (EllipticalUnification.lean). The method is compared against an mgcv REML spline re-implementation of the Ghosh GAM and against QDA.

On two UCI datasets the derived link ties the fitted GAM. On six financial series it is never significantly worse than the GAM, is better on CAD/USD, and beats QDA on the three heaviest-tailed series. A simulation with closed-form ground truth reproduces the sqrt(n) rate.

seriesexc. kurt.kun - ghoshkun - qda
CAD/USD9.3[+0.004,+0.009][-0.010,+0.017]
JPY/USD9.1[-0.003,+0.001][+0.009,+0.047]
WTI oil64.9[-0.003,+0.007][+0.023,+0.069]
S&P 50016.8[-0.025,+0.007][+0.039,+0.125]
EUR/USD2.5[-0.001,+0.003][-0.016,+0.013]
Paired bootstrap 95% CIs of accuracy difference on financial series (>0 favours derived link)