pith. sign in

arxiv: 1907.04212 · v1 · pith:DPHBUV2Lnew · submitted 2019-07-06 · 🧮 math.RT · cs.LG· stat.ML

On a method to construct exponential families by representation theory

Pith reviewed 2026-05-25 01:53 UTC · model grok-4.3

classification 🧮 math.RT cs.LGstat.ML
keywords exponential familyrepresentation theoryhomogeneous spacegeneralized inverse Gaussianinformation geometry
0
0 comments X

The pith

Theorems 1 and 2 determine when the parameter map is injective and when distinct representation pairs produce the same exponential family on a homogeneous space.

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

A prior construction builds an exponential family on the homogeneous space G/H from a representation V of G together with an H-fixed vector v0. The paper proves two theorems that settle when the map sending each parameter theta to the density p_theta is one-to-one and when two different pairs (V, v0) and (V', v0') generate identical families. In the special case where G is the positive reals, H is trivial, and V is a two-dimensional representation, the resulting family coincides with the generalized inverse Gaussian distributions.

Core claim

Theorems 1 and 2 answer when the correspondence theta to p_theta is injective and when distinct pairs generate the same family. For the case (G, H) = (R>0, {1}) with a representation on R^2, the family obtained is essentially the generalized inverse Gaussian distribution.

What carries the argument

The pair (V, v0) consisting of a representation V of G and an H-fixed vector v0 in V, which generates the exponential family on G/H.

If this is right

  • When the injectivity conditions hold, each theta corresponds to a distinct member of the family.
  • The equivalence criterion identifies which representation pairs can be regarded as producing one and the same family.
  • The generalized inverse Gaussian distribution is recovered exactly as the family generated by the indicated representation on R^2.

Where Pith is reading between the lines

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

  • The classification supplies a practical test for whether a given exponential family admits multiple distinct realizations through different representation pairs.
  • The results make it possible to list, up to equivalence, all families obtainable from representations of a fixed group G on a given homogeneous space.

Load-bearing premise

The construction from the cited prior work produces a valid exponential family on G/H for any given pair (V, v0).

What would settle it

An explicit pair (V, v0) for which the map theta to p_theta fails to be injective although the conditions of Theorem 1 hold, or two pairs that produce different families although they satisfy the equivalence criterion of Theorem 2.

read the original abstract

Exponential family plays an important role in information geometry. In arXiv:1811.01394, we introduced a method to construct an exponential family $\mathcal{P}=\{p_\theta\}_{\theta\in\Theta}$ on a homogeneous space $G/H$ from a pair $(V,v_0)$. Here $V$ is a representation of $G$ and $v_0$ is an $H$-fixed vector in $V$. Then the following questions naturally arise: (Q1) when is the correspondence $\theta\mapsto p_\theta$ injective? (Q2) when do distinct pairs $(V,v_0)$ and $(V',v_0')$ generate the same family? In this paper, we answer these two questions (Theorems 1 and 2). Moreover, in Section 3, we consider the case $(G,H)=(\mathbb{R}_{>0}, \{1\})$ with a certain representation on $\mathbb{R}^2$. Then we see the family obtained by our method is essentially generalized inverse Gaussian distribution (GIG).

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 answers two questions about the representation-theoretic construction of exponential families on G/H from pairs (V, v0) introduced in the authors' prior work: (Q1) when the map θ ↦ p_θ is injective (Theorem 1) and (Q2) when distinct pairs (V, v0) and (V', v0') generate the same family (Theorem 2). Section 3 treats the case (G, H) = (R>0, {1}) with a representation on R^2 and claims the resulting family is essentially the generalized inverse Gaussian distribution.

Significance. If the underlying construction is valid, the theorems supply concrete criteria for injectivity and equivalence of families generated by this method, which may help classify exponential families arising from representation theory. The explicit link to the generalized inverse Gaussian in Section 3 connects the approach to a well-studied distribution in statistics.

major comments (2)
  1. [Theorems 1 and 2] Theorems 1 and 2 presuppose without re-derivation or independent verification that the construction of arXiv:1811.01394 already yields a valid normalized probability density p_θ on G/H; this assumption is load-bearing for both the injectivity and equivalence claims.
  2. [Section 3] Section 3 asserts that the (R>0, {1}) case with the given representation on R^2 produces (essentially) the generalized inverse Gaussian, but supplies no explicit integration check or normalization computation to confirm the densities are well-defined and match the GIG form.
minor comments (1)
  1. Notation for the parameter space Θ and the precise form of the exponential family densities should be stated explicitly before Theorems 1 and 2 to make the statements self-contained.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed review and constructive suggestions. We address each major comment below and will incorporate clarifications in a revised version.

read point-by-point responses
  1. Referee: [Theorems 1 and 2] Theorems 1 and 2 presuppose without re-derivation or independent verification that the construction of arXiv:1811.01394 already yields a valid normalized probability density p_θ on G/H; this assumption is load-bearing for both the injectivity and equivalence claims.

    Authors: The current manuscript is a direct follow-up to arXiv:1811.01394, where the construction of the normalized densities p_θ is established. Theorems 1 and 2 address the subsequent questions of injectivity and equivalence under that construction. To make the paper more self-contained, we will add a short paragraph in the introduction recalling the normalization step from the prior work, including the relevant reference and a brief outline of why the integral converges. revision: yes

  2. Referee: [Section 3] Section 3 asserts that the (R>0, {1}) case with the given representation on R^2 produces (essentially) the generalized inverse Gaussian, but supplies no explicit integration check or normalization computation to confirm the densities are well-defined and match the GIG form.

    Authors: In Section 3 we derive the unnormalized density explicitly from the representation and observe that its functional form coincides with that of the GIG family after a change of parameters. We will strengthen this by adding the explicit evaluation of the normalizing integral in the revised manuscript, confirming it matches the known GIG constant and thereby verifying that the densities are well-defined. revision: yes

Circularity Check

1 steps flagged

Theorems 1-2 and GIG claim rest on validity of self-cited prior construction without re-derivation

specific steps
  1. self citation load bearing [Abstract]
    "In arXiv:1811.01394, we introduced a method to construct an exponential family P={p_θ} on a homogeneous space G/H from a pair (V,v0). ... Then the following questions naturally arise: (Q1) when is the correspondence θ↦p_θ injective? (Q2) when do distinct pairs (V,v0) and (V',v0') generate the same family? In this paper, we answer these two questions (Theorems 1 and 2). Moreover, in Section 3, we consider the case (G,H)=(R>0,{1}) with a certain representation on R^2. Then we see the family obtained by our method is essentially generalized inverse Gaussian distribution (GIG)."

    Theorems 1 and 2 characterize injectivity of θ↦p_θ and equivalence of families only after assuming the prior self-cited method already defines valid probability densities on G/H. The GIG claim in Section 3 similarly rests on that un-rederived construction step rather than an independent integration or normalization check supplied here.

full rationale

The paper's central results (Theorems 1 and 2 answering Q1/Q2, plus the Section 3 GIG identification) presuppose that the method from the authors' own prior arXiv:1811.01394 produces valid exponential families p_θ on G/H. This is a load-bearing self-citation: the new theorems characterize properties of that construction but do not independently verify or re-derive its normalization, domain, or density status. The specific (R>0, {1}) case yielding GIG likewise inherits the prior step. This matches self_citation_load_bearing with partial circularity (score 6), while the theorems still add independent content on injectivity and equivalence.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no information on free parameters, background axioms, or new entities introduced in the proofs of Theorems 1 and 2.

pith-pipeline@v0.9.0 · 5720 in / 1035 out tokens · 32244 ms · 2026-05-25T01:53:38.004626+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel echoes
    ?
    echoes

    ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

    We answer these two questions (Theorems 1 and 2). Moreover, in Section 3, we consider the case (G,H)=(ℝ>0,{1}) with a certain representation on ℝ². Then we see the family obtained by our method is essentially generalized inverse Gaussian distribution (GIG).

  • IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean absolute_floor_iff_bare_distinguishability unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    Theorem 1. … the following three conditions are equivalent: (i) The correspondence Θ ∋ θ ↦→ p_θ ∈ P is injective. (ii) There does not exist ξ ∈ V∨∖{0} such that f_ξ ∈ log Ω0(G,H).

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.