pith. sign in

arxiv: 2604.12725 · v1 · submitted 2026-04-14 · 🧮 math.ST · cs.LG· math.AG· math.DG· stat.TH

On Higher-Order Geometric Refinements of Classical Covariance Asymptotics: An Approach via Intrinsic and Extrinsic Information Geometry

Malik Amir , Sourangshu Ghosh This is my paper

Pith reviewed 2026-05-10 14:11 UTC · model grok-4.3

classification 🧮 math.ST cs.LGmath.AGmath.DGstat.TH
keywords higher-order asymptoticsFisher informationinformation geometrycurved modelssingular modelsresolution of singularitiesFisher-Rao metricHellinger discrepancy
0
0 comments X

The pith

An n^{-2} geometric correction refines the asymptotic covariance of first-order efficient estimators in curved models.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a curvature-aware refinement to the classical asymptotic covariance of efficient estimators. It views the parametric family as a Riemannian manifold immersed in L2 space and derives an n^{-2} correction term that accounts for intrinsic curvature, extrinsic immersion effects, and higher-order probabilistic information. This matters for understanding finite-sample behavior in curved models such as mixtures and latent variable models, where first-order Fisher information alone is insufficient. The correction is invariant and vanishes for exponential families. The approach extends the geometric analysis to singular models through resolution of singularities.

Core claim

The central claim is that the covariance of score-root, first-order efficient estimators is given by n^{-1}I(θ)^{-1} plus an n^{-2} correction governed by the tensor P_ij. This tensor decomposes canonically into three parts: an intrinsic Ricci-type contraction of the Fisher-Rao curvature tensor, an extrinsic Gram-type contraction of the second fundamental form, and a Hellinger discrepancy tensor. The full correction is coordinate-invariant, the extrinsic term is positive semidefinite, and it is identically zero for full exponential families. For singular models, under an additive normal crossing assumption, resolution of singularities yields a resolved metric and a curvature-based covariance

What carries the argument

The tensor P_ij governing the n^{-2} covariance correction, which decomposes into intrinsic Ricci-type contraction, extrinsic Gram-type contraction of the second fundamental form, and Hellinger discrepancy tensor.

If this is right

  • The correction vanishes for full exponential families, recovering classical asymptotics exactly at this order.
  • The extrinsic term is positive semidefinite, implying larger finite-sample variances in curved models.
  • The framework yields curvature-based covariance expansions on resolved spaces for singular models.
  • It provides geometric diagnostics of weak identifiability tied to the log canonical threshold.
  • It suggests curvature-aware principles for regularization and optimization.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Adjusted confidence intervals or estimators incorporating the P tensor could improve finite-sample accuracy.
  • The Hellinger term may link this expansion to other higher-order divergence-based refinements.
  • Monte Carlo experiments in mixtures or manifold-constrained models can directly test the predicted rate.
  • The same geometric decomposition might extend to risk expansions or optimization trajectories in inference.

Load-bearing premise

Suitable regularity and moment assumptions together with the additive normal crossing assumption for resolution of singularities in singular models.

What would settle it

Numerical computation of sample covariance for an efficient estimator in a specific curved family at large n, checking whether the difference from n^{-1}I^{-1} + n^{-2}P matches o(n^{-2}).

read the original abstract

Classical Fisher-information asymptotics describe the covariance of regular efficient estimators through the local quadratic approximation of the log-likelihood, and thus capture first-order geometry only. In curved models, including mixtures, curved exponential families, latent-variable models, and manifold-constrained parameter spaces, finite-sample behavior can deviate systematically from these predictions. We develop a coordinate-invariant, curvature-aware refinement by viewing a regular parametric family as a Riemannian manifold \((\Theta,g)\) with Fisher--Rao metric, immersed in \(L^2(\mu)\) through the square-root density map. Under suitable regularity and moment assumptions, we derive an \(n^{-2}\) correction to the leading \(n^{-1}I(\theta)^{-1}\) covariance term for score-root, first-order efficient estimators. The correction is governed by a tensor \(P_{ij}\) that decomposes canonically into three parts, an intrinsic Ricci-type contraction of the Fisher--Rao curvature tensor, an extrinsic Gram-type contraction of the second fundamental form, and a Hellinger discrepancy tensor encoding higher-order probabilistic information not determined by immersion geometry alone. The extrinsic term is positive semidefinite, the full correction is invariant under smooth reparameterization, and it vanishes identically for full exponential families. We then extend the picture to singular models, where Fisher information degenerates. Using resolution of singularities under an additive normal crossing assumption, we describe the resolved metric, the role of the real log canonical threshold in learning rates and posterior mean-squared error, and a curvature-based covariance expansion on the resolved space that recovers the regular theory as a special case. This framework also suggests geometric diagnostics of weak identifiability and curvature-aware principles for regularization and optimization.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript develops a coordinate-invariant refinement of the asymptotic covariance for first-order efficient estimators in regular parametric models by immersing the parameter space as a Riemannian manifold with the Fisher-Rao metric into L^2 via the square-root density map. Under regularity and moment assumptions, it derives an n^{-2} correction tensor P_{ij} to the leading n^{-1} I(θ)^{-1} term, canonically decomposed into an intrinsic Ricci-type contraction of the curvature tensor, an extrinsic Gram-type contraction of the second fundamental form, and a Hellinger discrepancy tensor. The correction is shown to be reparameterization-invariant, positive semidefinite in its extrinsic part, and to vanish for full exponential families. The framework is extended to singular models by invoking resolution of singularities under an additive normal crossing assumption, yielding a resolved metric whose curvature recovers the regular theory, along with discussion of the real log canonical threshold and geometric diagnostics for weak identifiability.

Significance. If the central derivations hold, the work would advance higher-order asymptotics by supplying a geometrically decomposed, invariant n^{-2} correction that separates intrinsic curvature, extrinsic embedding effects, and additional probabilistic information. This could improve understanding of finite-sample deviations in curved exponential families, mixtures, and latent-variable models where first-order Fisher asymptotics are known to be insufficient. The vanishing on full exponential families provides a useful consistency check, and the singular-model extension via resolution of singularities, if substantiated, offers a pathway to curvature-aware regularization and identifiability diagnostics in non-regular settings.

major comments (2)
  1. In the section extending the framework to singular models, the n^{-2} covariance expansion on the resolved space is derived under the additive normal crossing assumption after resolution of singularities. The manuscript provides no explicit verification that this assumption holds for the motivating examples (finite mixtures, latent-variable models). If the assumption fails for these families, the claimed recovery of the regular theory as a special case and the finiteness of the Hellinger discrepancy term after blow-up do not follow, rendering the extension unsupported.
  2. The well-definedness of the Hellinger discrepancy tensor after resolution of singularities is not demonstrated. Since this term is one of the three canonical components of P_{ij} and is required for the curvature-based expansion to be finite, its behavior under the blow-up must be addressed for the singular-model claim to be load-bearing.
minor comments (1)
  1. The precise statement of the regularity and moment assumptions needed for the regular-case derivation of the n^{-2} term could be collected in a single proposition or remark for easier reference.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough reading and for identifying key points that strengthen the presentation of the singular-model extension. We address each major comment below, acknowledging where the manuscript is currently incomplete and outlining the revisions we will make.

read point-by-point responses

  1. Referee: In the section extending the framework to singular models, the n^{-2} covariance expansion on the resolved space is derived under the additive normal crossing assumption after resolution of singularities. The manuscript provides no explicit verification that this assumption holds for the motivating examples (finite mixtures, latent-variable models). If the assumption fails for these families, the claimed recovery of the regular theory as a special case and the finiteness of the Hellinger discrepancy term after blow-up do not follow, rendering the extension unsupported.

    Authors: We agree that the manuscript states the singular-model results under the additive normal crossing assumption without supplying explicit verification for the motivating examples. This assumption is the standard technical hypothesis under which Hironaka's resolution theorem produces a manifold with normal crossings, allowing the resolved metric and curvature quantities to be well-defined. While the assumption is known to be satisfiable for broad classes of algebraic models (including many finite mixtures after suitable blow-ups, as indicated in the algebraic statistics literature), the current text does not demonstrate this for the specific families mentioned. In the revised manuscript we will add a dedicated paragraph clarifying that the extension is conditional on the assumption, citing the relevant resolution theorems, and noting that verification for concrete mixture and latent-variable models is an important direction for follow-up work. This will make the scope and limitations of the claim explicit. revision: yes

  2. Referee: The well-definedness of the Hellinger discrepancy tensor after resolution of singularities is not demonstrated. Since this term is one of the three canonical components of P_{ij} and is required for the curvature-based expansion to be finite, its behavior under the blow-up must be addressed for the singular-model claim to be load-bearing.

    Authors: This observation is correct: the manuscript defines the Hellinger discrepancy tensor in the regular setting and asserts that the same decomposition persists on the resolved space, but does not supply a separate argument establishing that the tensor remains finite and well-defined after the blow-up. In the revision we will insert a short technical subsection proving that, under the additive normal crossing assumption, the pull-back of the square-root density map yields a Hellinger discrepancy that is locally bounded on the resolved manifold. The argument proceeds by expressing the discrepancy in terms of the monomial coordinates furnished by the resolution and showing that the normal-crossing condition prevents the appearance of non-integrable singularities at the exceptional divisors, thereby guaranteeing that the n^{-2} term stays finite and the curvature expansion recovers the regular case when the singularity is absent. revision: yes

Circularity Check

0 steps flagged

Derivation from Fisher-Rao immersion geometry is self-contained with no load-bearing self-reference

full rationale

The claimed n^{-2} covariance correction is obtained by direct computation from the Riemannian structure of the parametric family equipped with the Fisher-Rao metric and its immersion into L^2 via the square-root density map; the tensor P_{ij} is explicitly decomposed into contractions of the curvature tensor, second fundamental form, and Hellinger discrepancy, all defined from the given geometry rather than from any fitted parameter or output quantity. The singular-model extension proceeds by assuming an additive normal-crossing condition after resolution of singularities and then recovering the regular case as a special instance; this is an external modeling hypothesis, not a self-definitional or self-cited reduction. No equations in the abstract or described chain equate a derived object to a fitted input or to a prior result whose only justification is the present paper itself.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard regularity conditions from asymptotic statistics and an algebraic-geometry assumption for handling singularities; no free parameters or new postulated entities are introduced in the abstract.

axioms (2)
  • domain assumption Suitable regularity and moment assumptions on the parametric family
    Invoked as prerequisite for the n^{-2} expansion in both regular and singular cases.
  • domain assumption Additive normal crossing assumption for resolution of singularities
    Required to describe the resolved metric and recover the regular theory as a special case.

pith-pipeline@v0.9.0 · 5624 in / 1458 out tokens · 61017 ms · 2026-05-10T14:11:52.475041+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages

  1. [1]

    194, Springer, Tokyo, 2016

    Shun-ichi Amari,Information geometry and its applications, Applied Mathematical Sciences, vol. 194, Springer, Tokyo, 2016

    work page 2016

  2. [2]

    191, American Mathematical Society and Oxford University Press, Providence, RI and Oxford, 2000, Translated by Daishi Harada

    Shun-ichi Amari and Hiroshi Nagaoka,Methods of information geometry, Translations of Mathematical Mono- graphs, vol. 191, American Mathematical Society and Oxford University Press, Providence, RI and Oxford, 2000, Translated by Daishi Harada

    work page 2000

  3. [3]

    Nihat Ay, Jürgen Jost, Hông Vân Lê, and Lorenz Schwachhöfer,The intrinsic geometry of statistical models, Information Geometry, Springer, Cham, 2017, pp. 185–239

    work page 2017

  4. [4]

    ,Parametrized measure models, Information Geometry, Springer, Cham, 2017, pp. 121–184

    work page 2017

  5. [5]

    Barndorff-Nielsen and David R

    Ole E. Barndorff-Nielsen and David R. Cox,Inference and asymptotics, Chapman and Hall, London, 1994

    work page 1994

  6. [6]

    Bates and Donald G

    Douglas M. Bates and Donald G. Watts,Relative curvature measures of nonlinearity, Journal of the Royal Statistical Society. Series B (Methodological)42(1980), no. 1, 1–16

    work page 1980

  7. [7]

    Bhattacharyya,On some analogues of the amount of information and their use in statistical estimation, Sankhy¯ a8(1946), 1–14

    A. Bhattacharyya,On some analogues of the amount of information and their use in statistical estimation, Sankhy¯ a8(1946), 1–14

    work page 1946

  8. [8]

    Bickel, Chris A

    Peter J. Bickel, Chris A. J. Klaassen, Ya’acov Ritov, and Jon A. Wellner,Efficient and adaptive estimation for semiparametric models, Johns Hopkins University Press, Baltimore, MD, 1993

    work page 1993

  9. [9]

    L. L. Campbell,An extended čencov characterization of the information metric, Proceedings of the American Mathematical Society98(1986), no. 1, 135–141

    work page 1986

  10. [10]

    N. N. Cencov,Statistical decision rules and optimal inference, Translations of Mathematical Monographs, vol. 53, American Mathematical Society, Providence, RI, 1982, Translated by the Israel Program for Scientific Transla- tions

    work page 1982

  11. [11]

    Harald Cramér,Mathematical methods of statistics, Princeton Landmarks in Mathematics and Physics, Princeton University Press, Princeton, NJ, 1999, Originally published in 1946

    work page 1999

  12. [12]

    6, 1189–1242

    Bradley Efron,Defining the curvature of a statistical problem (with applications to second order efficiency), The Annals of Statistics3(1975), no. 6, 1189–1242

    work page 1975

  13. [13]

    3, 793–803

    Shinto Eguchi,Second order efficiency of minimum contrast estimators in a curved exponential family, The Annals of Statistics11(1983), no. 3, 793–803

    work page 1983

  14. [14]

    Kass and Paul W

    Robert E. Kass and Paul W. Vos,Geometrical foundations of asymptotic inference, John Wiley & Sons, New York, 1997

    work page 1997

  15. [15]

    Kay,Fundamentals of statistical signal processing

    Steven M. Kay,Fundamentals of statistical signal processing. vol. I: Estimation theory, Prentice Hall, Englewood Cliffs, NJ, 1993

    work page 1993

  16. [16]

    Lauritzen,Statistical manifolds, Differential Geometry in Statistical Inference, IMS Lecture Notes– Monograph Series, vol

    Steffen L. Lauritzen,Statistical manifolds, Differential Geometry in Statistical Inference, IMS Lecture Notes– Monograph Series, vol. 10, Institute of Mathematical Statistics, Hayward, CA, 1987, pp. 163–216

    work page 1987

  17. [17]

    Lehmann and George Casella,Theory of point estimation, 2nd ed., Springer, New York, 1998

    Erich L. Lehmann and George Casella,Theory of point estimation, 2nd ed., Springer, New York, 1998

    work page 1998

  18. [18]

    1, 77–93

    Paul Marriott,On the local geometry of mixture models, Biometrika89(2002), no. 1, 77–93

    work page 2002

  19. [19]

    Radhakrishna Rao,Information and the accuracy attainable in the estimation of statistical parameters, Bulletin of the Calcutta Mathematical Society37(1945), no

    C. Radhakrishna Rao,Information and the accuracy attainable in the estimation of statistical parameters, Bulletin of the Calcutta Mathematical Society37(1945), no. 3, 81–91

    work page 1945

  20. [20]

    van der Vaart,Asymptotic statistics, Cambridge Series in Statistical and Probabilistic Mathematics, vol

    Aad W. van der Vaart,Asymptotic statistics, Cambridge Series in Statistical and Probabilistic Mathematics, vol. 3, Cambridge University Press, Cambridge, 1998

    work page 1998

  21. [21]

    25, Cambridge University Press, Cambridge, 2009

    Sumio Watanabe,Algebraic geometry and statistical learning theory, Cambridge Monographs on Applied and Computational Mathematics, vol. 25, Cambridge University Press, Cambridge, 2009. centre de recherche du chu de l’université de montréal centre de recherche du chu sainte-justine société québécoise de l’intelligence artificielle en médecine Email address:...

    work page 2009