betaFeedback

WorksMathematics

Statistical Geometry of Group Actions: Symmetry, Transport, Estimation

Abstract

We study variational inference in the presence of distributional symmetry, focusing on the interaction between two distinct group actions on a sample space: a symmetry action that leaves a reference distribution invariant, and a transport action that generates a parametric variational family by pushforward. By analyzing the algebraic structure of the total transformation group generated by both actions and the induced adjoint representation of the symmetry group on the transport Lie algebra, we establish that the Riemannian gradient of the Kullback–Leibler divergence at the true transport parameter is constrained to lie in the trivial representation of the symmetry group. Consequently, symmetry forces the gradient to vanish along all non-trivial representations. Combining this representation-theoretic constraint with parameter identifiability yields an orthogonal decomposition of the transport Lie algebra into three interpretable subspaces: an identifiability kernel (along which the objective is locally constant), a bias subspace (where the gradient may be nonzero due to model misspecification), and a symmetry-breaking subspace (where the gradient is forced to vanish by symmetry alone). Finally, we characterize the residual bias directions in terms of distributional mismatch between the base model and the standardized target, and derive an explicit Stein-type moment-matching condition that eliminates this bias, extending the location-scale results of (Margossian and Saul, 2025) to general Lie group transformation families.

Introduction

In modern statistical practice, variational inference ([3, 14, 28]) has become a primary methodology for scalable approximate inference, most notably for approximating intractable posterior distributions in Bayesian models. Rather than sampling directly from a complex target distribution pp on a sample space X\mathcal{X}, variational inference casts inference as an optimization problem: one seeks a distribution q∈Qq \in\mathcal{Q} within a tractable variational family that minimizes a statistical divergence—most commonly the Kullback–Leibler (KL) divergence KL⁡(q∥p)\operatorname{KL}(q\|p). In practice, the variational family Q\mathcal{Q} is nearly always misspecified with respect to pp: factorized mean-field families sacrifice dependence structure for tractability, while structured transformation families and normalizing flows ([24, 26]) trade computational cost for richer geometry. Because pp rarely belongs to Q\mathcal{Q}, a foundational statistical question arises: even under model misspecification, when does the variational optimum exactly recover meaningful structural parameters of the target distribution?

Symmetry offers a natural geometric principle to address this question. In classical mathematical statistics, group invariance is foundational: equivariance principles govern the theory of optimal invariant estimators and hypothesis tests ([8, 18, 30]), the Hunt–Stein theorem establishes conditions under which minimax procedures can be chosen to be group-invariant ([4]), and the interplay between sufficiency and invariance is central to statistical decision theory ([12]). In contemporary machine learning, group invariance and equivariance have similarly guided the design of structured neural architectures ([6, 7, 29]). By contrast, the role of distributional symmetry in shaping the optimization landscape of variational inference—specifically, how shared invariances between the variational base distribution and the standardized target distribution constrain stationary points of the variational objective—has only recently begun to receive systematic study.

A striking step in this direction was taken by Margossian and Saul (2025), who demonstrated that variational inference can achieve exact parameter recovery in misspecified location families (generated by translations x↦x+μx \mapsto x+\mu) and location-scale families (generated by affine transformations x↦μ+Axx \mapsto\mu+Ax) whenever the base distribution q0q_0 and the standardized target share a common symmetry. Specifically, they established that the gradient of the KL divergence vanishes at the true location parameter under central reflection symmetry (x↦−xx \mapsto-x), and at the true correlation matrix under elliptical symmetry, while uncovering a persistent scale bias when estimating the full covariance matrix. These findings led the authors to pose the following general question:

“Finally, an intriguing conjecture, also deserving of study, is that for each symmetry of pp, there is a corresponding statistic that VI under mild conditions can exactly recover.”

In subsequent work, Margossian et al. [21] extended these location-scale recovery guarantees from the standard variational objective KL⁡(q∥p)\operatorname{KL}(q\Vert p) to general ff-divergences without requiring log-concavity. In the present paper, we address this conjecture in full generality by developing a Lie-theoretic and representation-theoretic framework for transformation-based variational inference. While our first-order symmetry constraint (Theorem 6.2) relies only on diffeomorphism invariance of the divergence and therefore applies broadly to any invariant loss functional, working explicitly with the KL divergence KL⁡(qg∥p)\operatorname{KL}(q_g\Vert p) allows us to derive exact Stein-type moment-matching conditions that govern the remaining non-stationary directions. Our analysis reveals precisely how the algebraic compatibility between the symmetry group of the distribution and the Lie group of variational transports determines which parameter components are recovered without bias and which remain susceptible to model misspecification.

Concretely, we study the interaction between two group actions on the sample space X\mathcal{X}: a compact symmetry group HH acting by isometries, and a transport Lie group GG acting by diffeomorphisms to generate the variational family. This formulation connects classical statistical transformation models (Eaton, [8]) and Lie group representation theory ([9, 11, 15]) with information geometry ([2, 23]) and Riemannian optimization on manifolds ([1, 5]).

To formalize the estimation problem, we consider variational families obtained by pushing forward a fixed reference distribution q0∈P(X)q_0 \in\mathcal{P}(\mathcal{X}) under the action of GG: if Z∼q0Z \sim q_0, then X=g⋅Z∼qgX=g\cdot Z \sim q_g, yielding the variational optimization problem

min⁡g∈GKL⁡(qg∥p).\min_{g\in G} \operatorname{KL}(q_g\Vert p).

On the data side, we assume that the target distribution pp is generated by transporting an unknown standardized distribution p~\tilde{p} by a true parameter g∗∈Gg^* \in G that we wish to estimate; equivalently, p~\tilde{p} is the pullback of pp under g∗g^*, meaning that if X∼pX\sim p, then the standardized variable ξ=(g∗)−1⋅X\xi=(g^*)^{-1}\cdot X follows p~\tilde{p}. Importantly, we do not assume that p~\tilde{p} coincides with q0q_0, so the variational family QG={qg:g∈G}\mathcal{Q}_G=\{q_g:g\in G\} may be misspecified (p∉QGp\notin\mathcal{Q}_G). Instead, we assume a shared structural symmetry: both the variational base q0q_0 and the standardized target p~\tilde{p} are invariant under the action of the symmetry group HH. Geometrically, this means that q0q_0 and p~\tilde{p} share the same notion of centering or alignment around the origin, and the statistical question is whether this shared invariance suffices to make g∗g^* a stationary point (and local minimizer) of the variational loss g↦KL⁡(qg∥p)g\mapsto\operatorname{KL}(q_g\Vert p).

Main contributions. We characterize the first-order geometry of transport variational inference at g∗g^* through the conjugation action of HH on GG and its linearization, the adjoint representation of HH on the Lie algebra g=TeG\mathfrak{g}=T_eG:

  1. A Lie-theoretic framework for symmetric transport VI (§ 3–§ 4). We formulate a general statistical framework pairing a compact symmetry group HH with an HH-normalized transport Lie group GG. Analyzing the total transformation group K=GHK = GH and the intersection subgroup N=G∩HN = G \cap H allows us to separate parameter identifiability on the coset manifold M=G/N\mathcal{M} = G/N from first-order estimation bias, unifying location, scale, directional, and permutation models under a single geometric theory.

  1. Gradient constraint and orthogonal Lie algebra decomposition (Theorem 6.2 and Proposition 6.3). We prove that the Riemannian gradient ∇GL(g∗)\nabla_G \mathcal{L}(g^*) is constrained to lie in the trivial representation gH={u∈g:Ad⁡hu=u ∀h∈H}\mathfrak{g}^{H} = \{u \in\mathfrak{g} : \operatorname{Ad}_h u = u \ \forall h \in H\} of the adjoint action of HH on g\mathfrak{g}. Combining this invariance with the identifiability structure yields an orthogonal decomposition of the transport Lie algebra,

g=n⊕gbiasH⊕g⊥,\mathfrak{g} = \mathfrak{n} \oplus\mathfrak{g}^{H}_{\mathrm{bias}} \oplus\mathfrak{g}^{\perp},

into an identifiability kernel n\mathfrak{n} (along which the loss is locally constant), a bias subspace gbiasH=gH∩n⊥\mathfrak{g}^{H}_{\mathrm{bias}} = \mathfrak{g}^{H} \cap\mathfrak{n}^{\perp} (consisting of symmetry-preserving perturbations where model mismatch can induce a nonzero gradient), and a symmetry-breaking subspace g⊥=(gH)⊥∩n⊥\mathfrak{g}^{\perp} = (\mathfrak{g}^{H})^{\perp} \cap\mathfrak{n}^{\perp} (along which the gradient is forced to vanish by symmetry alone). Whenever gbiasH={0}\mathfrak{g}^{H}_{\mathrm{bias}} = \{0\}, g∗g^* is guaranteed to be a first-order stationary point regardless of model misspecification between q0q_0 and p~\tilde{p}.

  1. Stein-type moment-matching condition and debiasing recipe (Theorem 6.5). When the bias subspace gbiasH\mathfrak{g}^{H}_{\mathrm{bias}} is nontrivial, we derive an exact necessary and sufficient condition for g∗g^* to be a critical point. Using integration by parts along the infinitesimal generators VuV_u of the group action, we show that the gradient along u∈gbiasHu \in\mathfrak{g}^{H}_{\mathrm{bias}} vanishes if and only if the directional scores of the log-potentials of q0q_0 and p~\tilde{p} match in expectation under q0q_0. This characterization provides a constructive calibration procedure to eliminate estimation bias by rescaling or shifting the base distribution q0q_0 along exp⁡(gbiasH)\exp(\mathfrak{g}^{H}_{\mathrm{bias}}) while preserving its HH-invariance.

Organization. The remainder of the paper is organized as follows. § 2 situates our work within the literature on variational inference, group invariance in statistics, and Riemannian optimization. § 3 introduces transport variational families, HH-conjugate symmetry, and the structural normalization condition, together with six guiding examples that we trace throughout the paper. § 4 analyzes the total transformation group K=GHK = GH, the conjugation homomorphism, and the three resulting identifiability regimes on the quotient space M=G/N\mathcal{M} = G/N. § 5 develops the required differential calculus on GG and introduces the adjoint representation Ad⁡H\operatorname{Ad}_H and its trivial subspace gH\mathfrak{g}^{H}. § 6 presents our main theoretical results: the gradient constraint theorem, the orthogonal decomposition of g\mathfrak{g}, and the moment-matching theorem. § 7 discusses gauge-fixing vs. base calibration, second-order curvature conditions, and extensions. Technical proofs and detailed example derivations are deferred to the Appendices.

Related work

Variational inference and transport-based families. The variational formulation of approximate inference, rooted in statistical physics and convex duality, was brought to prominence in statistics and machine learning by Jordan et al. [14], Wainwright and Jordan [28], and Blei et al. [3]. The development of black-box variational inference and reparameterization gradients ([25]; [26]) subsequently enabled optimization over general continuous families where expectations under qgq_g are evaluated by pushing forward samples Z∼q0Z \sim q_0 through a differentiable transformation x=g(Z)x = g(Z). In the normalizing flow literature ([16]; [24]; [26]), the transformation gg is typically parameterized by a deep composition of general diffeomorphisms. Our framework studies the structured setting in which the transport maps form a finite-dimensional Lie group GG acting on X\mathcal{X}: this encompasses classical parametric families (such as location, scale, and directional families) while endowing the variational parameter space with an explicit Lie algebra that makes global and local symmetry analysis tractable.

Group invariance in mathematical statistics and machine learning. Group actions have long served as a unifying language in theoretical statistics [8, 18, 30]. In classical statistical decision theory, when a family of distributions is closed under a group action, one seeks estimators that are equivariant—such as the Pitman estimator for location or invariant Bayes estimators under right Haar measure—and hypothesis tests that are invariant with respect to nuisance group actions. In those classical works, the statistical model is typically well-specified and the group acts transitively on the parameter space to reduce risk. Our perspective differs in two important respects: first, we study the deterministic optimization landscape of a divergence functional under model misspecification (p≠QGp \ne Q_G); second, the symmetry group HH of the standardized distribution and the transport group GG indexing the variational family need not coincide, and their interaction via conjugation governs which parameters are unbiasedly estimated. In a complementary direction, geometric deep learning and equivariant neural networks [6, 7, 29] build group symmetries directly into function approximators and normalizing flows [17]; rather than constraining the variational family to produce strictly HH-invariant distributions (which would preclude estimating symmetry-breaking parameters g∗g^{*}), our work uses HH-invariant base distributions pushed forward by GG to estimate the unknown transformation g∗g^{*}.

Information geometry and optimization on Lie groups. Differential-geometric approaches to statistical estimation have been extensively studied in information geometry [2, 23], where the Fisher–Rao metric defines a natural Riemannian structure on parametric families, and in optimization on smooth manifolds and Lie groups [1, 5]. When a variational family is indexed by a Lie group GG, left-trivialization identifies the tangent space TgGT_gG at any parameter gg with the Lie algebra g=TeG\mathfrak{g}=T_eG. By equipping g\mathfrak{g} with an Ad⁡H\operatorname{Ad}_H-invariant inner product (guaranteed by the compactness of HH), we connect Riemannian first-order stationarity directly to the decomposition of gg into irreducible representations of HH.

Symmetry in variational inference. The most direct predecessors of the present work are Margossian and Saul [19] and Margossian et al. [21], who proved that location-scale variational inference exactly recovers the target mean under even symmetry and the target correlation matrix under elliptical symmetry across general ff-divergences, and Zellinger and Vergari [31], who extended even-symmetry location recovery to forward KL and α\alpha-divergences without log-concavity assumptions. Complementing these positive results, Margossian et al. [20] established an impossibility theorem for factorized (mean-field) variational approximations of non-factorized targets, proving that a mean-field family can accurately recover at most one of three canonical dispersion summaries (marginal variances, marginal precisions, or generalized variance). This impossibility underscores the necessity of non-factorized group-transport families such as full GL⁡(d)\operatorname{GL}(d) covariance models. Concurrently with our work, Marks et al. [22] developed a global theory of symmetry-induced statistic recovery in variational inference, characterizing when global minimizers inherit target symmetries and applying their framework to von Mises–Fisher distributions on the sphere, while Ghafourpour et al. [10] investigated radial transport families. Our differential-geometric approach is directly complementary: by linearizing the conjugation action of HH on GG via the adjoint representation, we provide an explicit subspace decomposition of the transport Lie algebra g\mathfrak{g} that simultaneously explains both why symmetry-breaking gradient directions vanish and why symmetry-preserving directions (gbiasH\mathfrak{g}_{\mathrm{bias}}^{H}) exhibit estimation bias unless calibrated by moment matching.

Problem description

In parametric variational inference [3, 25, 28], one approximates an intractable target distribution p∈P(X)p \in\mathcal{P}(\mathcal{X}) on a sample space X\mathcal{X} by selecting a candidate distribution qθq_\theta from a parameterized family QΘ={qθ∈P(X):θ∈Θ}\mathcal{Q}_\Theta= \{q_\theta\in\mathcal{P}(\mathcal{X}) : \theta\in\Theta\} that minimizes a divergence functional over θ∈Θ\theta\in\Theta. In this paper, we study structured variational families generated by the smooth action of a transport Lie group GG on X\mathcal{X}, in the presence of a symmetry group HH acting isometrically on X\mathcal{X}. For any g∈Gg \in G (respectively, h∈Hh \in H) and any point x∈Xx \in\mathcal{X}, we denote the group action on the sample space by g⋅x∈Xg \cdot x \in\mathcal{X} (respectively, h⋅x∈Xh \cdot x \in\mathcal{X}).

Viewing both GG and HH through their actions on the common domain X\mathcal{X} serves two fundamental purposes. First, any measurable action on X\mathcal{X} canonically induces a pushforward action on probability measures over X\mathcal{X}: transforming a random variable Z∼q0Z \sim q_0 via X=g⋅ZX = g \cdot Z transports its law to qg=g♯q0q_g = g_{\sharp}q_0. Second, because both GG and HH act as transformations of X\mathcal{X}, they can be composed within the ambient group Aut⁡(X)\operatorname{Aut}(\mathcal{X}) of diffeomorphisms of X\mathcal{X}, even when GG and HH are defined a priori as distinct algebraic groups. As we show throughout the paper, the non-commutativity of these two actions on X\mathcal{X}—captured algebraically by the conjugation of transports in GG by symmetries in HH—is the central mechanism governing parameter identifiability and unbiased estimation in variational inference.

Group transport variational inference

We begin by formalizing variational families generated by pushing forward a fixed reference measure q0q_0 under a Lie group action. From a statistical standpoint, this construction generalizes classical location-scale transformation models [8] to arbitrary smooth manifolds and Lie groups; from a machine learning standpoint, it corresponds to a normalizing flow [24, 26] whose transformation maps are structured as a Lie group GG. Throughout the paper, we restrict attention to distributions admitting strictly positive, smooth (or piecewise smooth, locally Lipschitz) densities with respect to the canonical reference measure on X\mathcal{X} (Lebesgue measure when X=Rd\mathcal{X} = \mathbb{R}^{d}, or the Riemannian volume measure when X\mathcal{X} is a Riemannian manifold), and we identify each probability measure with its density.

Definition 3.1 (Transport variational inference). Let X\mathcal{X} be a smooth data domain (such as Euclidean space Rd\mathbb{R}^{d} or a smooth Riemannian manifold) and let GG be a Lie group acting smoothly and effectively on X\mathcal{X} by diffeomorphisms, written (g,x)↦g⋅x∈X(g,x) \mapsto g \cdot x \in\mathcal{X} (where acting effectively simply means that distinct group elements g≠g′g \ne g' define distinct transformations on X\mathcal{X}). We call GG the parameter transport group. Given a base distribution q0∈P(X)q_0 \in\mathcal{P}(\mathcal{X}) and a target data distribution p∈P(X)p \in\mathcal{P}(\mathcal{X}), both with positive densities on X\mathcal{X}, we define:

  • The transported distribution qg=g♯q0q_g = g_{\sharp}q_0: the law of X=g⋅ZX = g \cdot Z when Z∼q0Z \sim q_0, whose density is given by the change-of-variables formula

qg(x)=q0(g−1⋅x)∣det⁡Jg−1(x)∣,q_g(x) = q_0(g^{-1} \cdot x) \left|\det J_{g^{-1}}(x)\right|,

References

  1. [1]P.-A. Absil, R. Mahony, and R. Sepulchre. Optimization Algorithms on Matrix Manifolds. Princeton University Press, 2008.DOI
  2. [2]S.-i. Amari. Information Geometry and Its Applications, volume 194 of Applied Mathematical Sciences. Springer, 2016.
  3. [3]D. M. Blei, A. Kucukelbir, and J. D. McAuliffe. Variational inference: A review for statisticians. Journal of the American Statistical Association, 112(518):859–877, 2017.
  4. [4]J. V. Bondar and P. Milnes. Amenability: A survey for statistical applications of Hunt–Stein and related conditions on groups. Zeitschrift für Wahrscheinlichkeitstheorie und verwandte Gebiete, 57 (1):103–128, 1981.DOI
  5. [5]N. Boumal. An Introduction to Optimization on Smooth Manifolds. Cambridge University Press, 2023.DOI
  6. [6]M. M. Bronstein, J. Bruna, T. Cohen, and P. Veličković. Geometric deep learning: Grids, groups, graphs, geodesics, and gauges. arXiv preprint arXiv:2104.13478, 2021.arxiv.org/abs/2104.13478
  7. [7]T. Cohen and M. Welling. Group equivariant convolutional networks. In International Conference on Machine Learning, pages 2990–2999, 2016.arxiv.org/abs/1602.07576
  8. [8]M. L. Eaton. Group Invariance Applications in Statistics. Institute of Mathematical Statistics and American Statistical Association, 1989.
  9. [9]W. Fulton and J. Harris. Representation Theory: A First Course, volume 129 of Graduate Texts in Mathematics. Springer, 1991.DOI
  10. [10]L. Ghafourpour, S. Chewi, A. Figalli, and A.-A. Pooladian. Variational inference via radial transport. In International Conference on Artificial Intelligence and Statistics, 2026.arxiv.org/abs/2602.17525
  11. [11]B. C. Hall. Lie Groups, Lie Algebras, and Representations: An Elementary Introduction, volume 222 of Graduate Texts in Mathematics. Springer, 2 edition, 2015.
  12. [12]W. J. Hall, R. A. Wijsman, and J. K. Ghosh. The relationship between sufficiency and invariance with applications in sequential analysis. The Annals of Mathematical Statistics, 36(2):575–614, 1965.DOI
  13. [13]A. Hyvärinen. Estimation of non-normalized statistical models by score matching. Journal of Machine Learning Research, 6:695–709, 2005.DOI
  14. [14]M. I. Jordan, Z. Ghahramani, T. S. Jaakkola, and L. K. Saul. An introduction to variational methods for graphical models. Machine Learning, 37(2):183–233, 1999.
  15. [15]A. W. Knapp. Lie Groups Beyond an Introduction, volume 140 of Progress in Mathematics. Birkhäuser, 2 edition, 2002.
  16. [16]I. Kobyzev, S. J. D. Prince, and M. A. Brubaker. Normalizing flows: An introduction and review of current methods. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(11):3964–3979, 2021.
  17. [17]J. Köhler, L. Klein, and F. Noé. Equivariant flows: exact likelihood generative learning for symmetric densities. In International Conference on Machine Learning, pages 5361–5370. PMLR, 2020.arxiv.org/abs/2006.02425
  18. [18]E. L. Lehmann and J. P. Romano. Testing Statistical Hypotheses. Springer, 3 edition, 2005.
  19. [19]C. C. Margossian and L. K. Saul. Variational inference in location-scale families: Exact recovery of the mean and correlation matrix. In International Conference on Artificial Intelligence and Statistics, volume 258, pages 3466–3474. PMLR, 2025.arxiv.org/abs/2410.11067
  20. [20]C. C. Margossian, L. Pillaud-Vivien, and L. K. Saul. Variational inference for uncertainty quantification: An analysis of trade-offs. Journal of Machine Learning Research, 26(202):1–41, 2025a.arxiv.org/abs/2403.13748
  21. [21]C. C. Margossian, I. E. Rankin, and L. K. Saul. Generalized guarantees for variational inference in the presence of even and elliptical symmetry. arXiv preprint arXiv:2511.01064, 2025b.arxiv.org/abs/2511.01064
  22. [22]D. Marks, D. Paccagnan, and M. van der Wilk. Symmetry guarantees statistic recovery in variational inference. arXiv preprint arXiv:2604.18310, 2026.arxiv.org/abs/2604.18310
  23. [23]M. K. Murray and J. W. Rice. Differential Geometry and Statistics. Chapman & Hall/CRC, 1993.
  24. [24]G. Papamakarios, E. Nalisnick, D. J. Rezende, S. Mohamed, and B. Lakshminarayanan. Normalizing flows for probabilistic modeling and inference. Journal of Machine Learning Research, 22(57):1–64, 2021.
  25. [25]R. Ranganath, S. Gerrish, and D. M. Blei. Black box variational inference. In International Conference on Artificial Intelligence and Statistics, volume 33, pages 814–822. PMLR, 2014.
  26. [26]D. J. Rezende and S. Mohamed. Variational inference with normalizing flows. In International Conference on Machine Learning, pages 1530–1538, 2015.arxiv.org/abs/1505.05770
  27. [27]C. M. Stein. Estimation of the mean of a multivariate normal distribution. The Annals of Statistics, 9 (6):1135–1151, 1981.DOI
  28. [28]M. J. Wainwright and M. I. Jordan. Graphical models, exponential families, and variational inference. Foundations and Trends in Machine Learning, 1(1–2):1–305, 2008.
  29. [29]M. Weiler and G. Cesa. General E(2)-equivariant steerable CNNs. Advances in Neural Information Processing Systems, 32, 2019.arxiv.org/abs/1911.08251
  30. [30]R. A. Wijsman. Invariant Measures on Groups and Their Use in Statistics, volume 14 of Institute of Mathematical Statistics Lecture Notes–Monograph Series. Institute of Mathematical Statistics, 1990.DOI
  31. [31]L. Zellinger and A. Vergari. Even more guarantees for variational inference in the presence of symmetries. arXiv preprint arXiv:2604.21407, 2026.arxiv.org/abs/2604.21407

Paper details

vlnc:W_3x3hgxnb
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents