Pith. sign in

REVIEW 2 major objections 3 minor 1 cited by

Space-time divergence lemmas and optimal non-reversible lifts of diffusions on Riemannian manifolds with boundary

T0 review · 2 major / 3 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read On locally convex manifolds, non-reversible lifts achieve the optimal square-root speedup of relaxation time with explicit constants.

desk verdict Strong extension of the divergence lemma with a real regularity gap in the high-mode construction; main results likely salvageable but the central theorem overclaims. read the letter →

arxiv 2412.16710 v1 pith:NUIGOEGF submitted 2024-12-21 math.PR math.APmath.FA

Andreas Eberle , Francis Lörler This is my paper · ORCID
classification math.PRmath.APmath.FA MSC 60J2560J3558J6558J0535H10
keywords non-reversibleliftsspace-timedivergencelemmahypocoercivityLangevindynamicsrandomizedHamiltonianMonteCarloRiemannianmanifoldswithboundarylocalconvexityReillyformula
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper proves that on a locally convex Riemannian manifold with boundary and a lower, not necessarily positive, curvature bound, every zero-mean function on the time-space cylinder $[0,T]\times M$ decomposes as $f=\partial_t h-Lg$, with explicit quantitative bounds on $h$ and $g$ and explicit constants $c_0(T)$ and $c_1(T)$. The constants have the order conjectured for the Euclidean case: $c_0(T)=2T^2+43/m$ and $c_1(T)=290+991/(mT^2)+43\max(1/m,T^2/\pi^2)\rho$. From this divergence lemma the paper derives space-time Poincar\'e inequalities and exponential decay for randomized Hamiltonian Monte Carlo and Langevin dynamics with reflection at the boundary. The upshot is that these non-reversible lifts are $C$-optimal: they achieve the square-root reduction of relaxation time up to a constant factor, so on a convex domain of diameter $d$ they relax in time of order $d$ rather than $d^2$.

What carries the argument

The load-bearing object is the generalized Reilly formula, an integration-by-parts identity for the weighted Bochner formula whose boundary terms involve the scalar second fundamental form $h$. Under local convexity $h(v,v)\le 0$ on $\partial M$, the boundary terms drop, and the formula yields the Hessian bound $\|\nabla^2u\|^2_{L^2(\mu)}\le \|Lu\|^2_{L^2(\mu)}+\rho\|\nabla u\|^2_{L^2(\mu)}$ for functions with Neumann boundary conditions. That bound is what makes the space-time divergence lemma quantitative: solving $(\partial_t^2+L)u=f$ and controlling second derivatives requires controlling the Hessian in terms of $L$ alone. The second ingredient is a spectral decomposition of $L$, with the space of space-time harmonic right-hand sides split into high and low modes and symmetric and antisymmetric parts; the low symmetric modes are the delicate case that controls the order of $c_1(T)$.

What would settle it

Take the flat interval $M=[0,d]$ with uniform measure and the low symmetric mode $f(t,x)=\cos(2\pi t/T)\sin(\pi x/d)$; solve the one-dimensional equations $f=\partial_t h-Lg$ explicitly with $h(0)=h(T)=0$ and $g(0)=g(T)=0$, and check whether the inequalities (17) and (18) hold with the stated $c_0(T)$ and $c_1(T)$. A contradiction would falsify the divergence lemma; equality would indicate the constants are sharp in this geometry.

Watch

Extended reading notes

Core claim

The central discovery is Theorem 5: under the paper's Assumptions 1 and 2, the space-time divergence equation $f=\partial_t h-Lg$ is solvable for any mean-zero $f$, with $h$ and $g$ vanishing at the time endpoints and $g(t,\cdot)$ in the domain of $L$ for almost every time $t$, and with the explicit constants $c_0(T)=2T^2+43/m$ and $c_1(T)=290+991/(mT^2)+43\max(1/m,T^2/\pi^2)\rho$ in the bounds (17) and (18). The proof splits the space of right-hand sides into space-time harmonic functions and their orthogonal complement, then treats symmetric and antisymmetric, high- and low-mode components separately. The paper feeds this lemma into a space-time Poincar\'e inequality for the lifted dynamics, yielding exponential decay of time-averages with rate $\nu=\gamma/(\gamma^2C_0(T)+C_1(T))$, and hence relaxation-time bounds for randomized Riemannian Hamiltonian Monte Carlo and Riemannian Langevin dynamics with specular reflection. With the choice $T=\pi/\sqrt{m}$ and refresh rate or friction $\gamma$ proportional to $\sqrt{m}$, these bounds give $C$-optimality of both lifts whenever the curvature parameter $\rho$ is at most a constant multiple of $m$, and in particular a relaxation time of order $d$ for a convex domain of diameter $d$.

Load-bearing premise

The argument collapses if the boundary is not locally convex: it needs the scalar second fundamental form $h(v,v)\le 0$ on $\partial M$ so that the boundary terms in the Reilly formula can be discarded, and without that no Hessian bound as in Corollary 4 is available.

Editorial extensions

If this is right

  • On a locally convex domain of diameter $d$ in Euclidean space, reflected Langevin dynamics with critical friction and randomized HMC with critical refresh rate have relaxation time of order $d$, whereas the underlying reversible diffusion has Poincar\'e constant of order $d^2$.
  • The general lower bound for second-order lifts, which says no lift can relax more than a square root faster than the reversible process, is attained up to a constant factor by these two dynamics.
  • The explicit constants convert the space-time Poincar\'e inequality into non-asymptotic exponential decay of time-averages, so the relaxation-time bounds are directly usable in quantitative sampling estimates.
  • The results cover Riemannian manifolds with possibly empty boundary and with a lower bound $\rho\ge 0$ on $\mathrm{Ric}+\nabla^2U$; when $\rho$ is of order $m$, the optimality constants remain bounded.
  • For the class of potentials satisfying $\rho\le cm$, randomized HMC and Langevin dynamics with reflection are $C$-optimal lifts with explicit constants $C$ linear in $\sqrt{1+c/9}$.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper states that the constants in the divergence lemma are surely not optimal, so a direct numerical check of the lemma on simple convex domains such as an interval, disk, or ball could reveal how far the values 43, 290, and 991 can be improved.
  • Since the authors note that the discrete-spectrum assumption can be relaxed, the same divergence lemma should extend to non-compact settings where Assumption 2 fails, with spectral projections replaced by functional calculus.
  • Remark 2 extends the Reilly formula to manifolds with corners, so the relaxation-time bounds should carry over to convex polytopes in $\mathbb{R}^d$, a case relevant to volume computation and constrained sampling.
  • The divergence lemma is stated for continuous-time generators, but the same $h,g$ construction should transfer to discretized or numerically integrated versions of HMC, where reflection at the boundary and refreshment schedules would need separate analysis.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 3 minor

Summary. This paper proves a quantitative space-time divergence lemma on product domains [0,T] x M for weighted Riemannian manifolds with boundary satisfying a local convexity condition, a lower Bakry-Emery curvature bound, and a Poincare inequality. The lemma asserts that every zero-mean f on [0,T] x M decomposes as f = d_t h - L g with explicit constants c0(T) = 2T^2 + 43/m and c1(T) = 290 + 991/(mT^2) + 43 max(1/m, T^2/pi^2) rho, with h and g satisfying Dirichlet conditions in time and g(t,.) in Dom(L). The authors then derive a space-time Poincare inequality and use the second-order lift framework to show that randomized Hamiltonian Monte Carlo with refreshments and Langevin dynamics with reflection are C-optimal lifts of overdamped Langevin diffusion on this class of manifolds, with relaxation time of order 1/sqrt(m) (hence order d for a domain of diameter d). The proof rests on a generalized Reilly formula for weighted manifolds with boundary, a spectral decomposition into high/low and symmetric/antisymmetric modes, and explicit estimates for each mode.

Significance. If the theorem were proved as stated, this would be a significant contribution: the constants are explicit, there is no fitting or post-hoc selection, and the extension to curved state spaces with reflecting boundary is new. The generalized Reilly formula and the careful tracking of constants are valuable, and the optimality results are structurally sound given the divergence lemma and the published lift framework [25]. However, the central regularity and full-Hessian bounds in Theorem 5 have a genuine gap, so the main result is currently conditional on a repair. The subsequent optimality results are convincing only after Theorem 5 is corrected or weakened in the way suggested below.

major comments (2)
  1. [Section 3, proof of Theorem 5, Cases 3 and 4] The functions g_k constructed in Cases 3 and 4 do not have the regularity asserted in (14). For t in [0,T/2], w_k(t) = u_k(t) - v_k'(t), where v_k(t) = phi_k(t) integral_0^t u_k(s) ds and phi_k(t) = (alpha_k t - 1)^2 1_{[0, alpha_k^{-1}]}(t). Since phi_k is only C^1 and identically zero for t >= alpha_k^{-1}, w_k is continuous and piecewise C^1, but w_k' has a jump at t = alpha_k^{-1}: the left limit is u_k'(alpha_k^{-1}) - 2 alpha_k^2 integral_0^{alpha_k^{-1}} u_k(s) ds and the right limit is u_k'(alpha_k^{-1}), and the integral is strictly positive for u_k(t) = e^{-alpha_k t} - e^{-alpha_k(T-t)}. Hence the distributional second derivative w_k'' contains a nonzero Dirac mass at alpha_k^{-1}, so g_k = alpha_k^{-2} w_k e_k is not in H^2([0,T]) and g is not in H^{2,2}_D(mu). Thus (14) is not proved for the constructed g. The theorem needs either a genuinely H^2 construction (for example, smoother cutoffs) or an explicit weakening of the regularity class.
  2. [Section 3, Eq. (18) and Eq. (19)] The proof controls ||d_t h||, ||grad h||, ||d_t grad g|| and the spatial Hessian ||grad^2 g|| via Corollary 4; these are the components of the reduced space-time gradient of X = (-h, grad g). However, (18) bounds ||grad grad g||^2_{L^2(mu)} with grad = (d_t, grad), which includes ||d_t^2 g||^2. No estimate for ||d_t^2 g|| is provided anywhere in Cases 1-4 or the H^perp case, and by the previous comment the constructed high-mode g has d_t^2 g containing a Dirac mass, so the L^2 norm is infinite. Therefore the asserted bound (18) with the constant c1(T) from (19) is not established. Since Theorem 15 uses Theorem 5 through Lemma 14, this gap is load-bearing; if the applications only require the reduced norm, Theorem 5 should be restated accordingly and the proof should say so explicitly.
minor comments (3)
  1. [Section 3, Case 4 heading] The heading 'Let f in Hl,s' should read 'Let f in Hh,s', since the symmetric low modes Hl,s were already treated in Case 2 and the text of Case 4 concerns symmetric high modes.
  2. [Lemma 14] The proof is deferred entirely to [25, Lemma 22] with the remark that it works in the Riemannian setting. Because the present setting includes Riemannian manifolds with boundary and the lemma is used to transfer the divergence lemma to the tangent bundle, please include at least a sketch verifying (35) and (36), in particular the treatment of the boundary terms.
  3. [Theorem 15 proof] The norm notation 'L^2(mu), H^{-1}(kappa)' and 'L^2(mu, H^{-1}(kappa))' is inconsistent and contains misplaced parentheses; it should be L^2(lambda tensor mu; H^{-1}(kappa)).

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the quantitative divergence lemma (Theorem 5) is proved in-paper from the generalized Reilly formula and explicit spectral mode estimates; the paper's self-citations to the authors' prior lift paper [25] supply published framework lemmas used as inputs, not conclusions assumed, so the central derivation has independent content.

full rationale

The central new result, Theorem 5, is derived inside the paper: Lemma 1 (generalized Reilly formula) is stated and proved in Appendix A.4; Assumptions 1 and 2 (local convexity h <= 0, curvature lower bound -rho, Poincare constant 1/m, discrete spectrum) are hypotheses, not fitted outputs; and the constants c0(T) = 2T^2 + 43/m and c1(T) = 290 + 991/(mT^2) + 43 max(1/m, T^2/pi^2) rho in (19) come from an explicit mode decomposition (low/high, symmetric/antisymmetric) with the cutoff beta = 2 chosen at the end to optimize the displayed bounds. No parameter is fitted to a data subset and then renamed a prediction, so the fitted-input pattern does not apply. The application chain (Theorem 5 -> Theorem 15 -> Theorem 16 -> Corollary 17) does rely on the authors' prior paper [25] for the second-order-lift definition, the lower bound trel(P) >= (1/2) sqrt(2) sqrt(trel(P)) cited as [25, Theorem 11], the space-time property of lifts quoted as [25, Lemma 18], Lemma 14 'as in [25, Lemma 22]', Theorem 16(i) 'identical to that of [25, Theorem 17]', and [25, Lemma 8] for the T-average-to-relaxation-time conversion. These are self-citations, but they are load-bearing only as published, peer-reviewed, parameter-free lemmas that do not assume Theorem 5 or the optimality conclusion; they are not invoked to forbid alternatives or to import a uniqueness theorem. Under the rubric such citations count as real evidence, so the score is 2 rather than 0. Separately, and for the record (this is a correctness concern, not circularity): in Cases 3 and 4 of the proof of Theorem 5, the constructed w_k = (1 - phi_k)u_k - phi_k' integral u_k has a first derivative with a jump at t = alpha_k^{-1}, since -phi_k'' integral_0^t u_k ds contributes a jump of size -2 alpha_k^2 integral_0^{alpha_k^{-1}} u_k(s) ds < 0; hence w_k'' contains a Dirac mass and w_k is not in H^2([0,T]), so the asserted regularity g in H^{2,2}_D(mu) in (14) is not established by the given construction, and the bound (18) on the full space-time Hessian is not derived. A referee should require this gap to be fixed, but the derivation still does not reduce to its own inputs, so it does not raise the circularity score.

Assumptions & free parameters 2 free parameters · 8 assumptions · 0 invented entities

No free parameters are fitted to data. The constants m and ρ are inputs from Assumption 1; T is later chosen as π/√m and γ as an explicit function of m and ρ to optimize bounds. These are design choices, not fits. The axioms list the geometric and analytic assumptions and the standard background results (Bochner formula, spectral theorem, Reilly formula) used in the proofs.

free parameters (2)
  • Time horizon T = π/√m in Corollary 17
    Chosen to balance the two terms in the relaxation-time bound trel ≤ 2√(C0 C1) + T; not fitted to data.
  • Refresh rate γ / friction parameter = sqrt((1163+3964/π²)m + 172ρ)/(4π²+86)
    Chosen to maximize the decay rate ν = γ/(γ²C0+C1); a design choice optimizing the explicit bound.
assumptions (8)
  • standard math Generalized Bochner-Lichnerowicz-Weitzenböck formula: ½ L|∇u|² = |∇²u|² + ⟨∇u,∇Lu⟩ + (Ric+∇²U)(∇u,∇u)
    Cited to Bakry-Emery [7]; integrated by parts to derive the generalized Reilly formula (Lemma 1).
  • standard math Generalized Reilly formula (Lemma 1) for weighted manifolds with boundary
    Proved in Appendix A.4. Yields the Hessian bound in Corollary 4.
  • standard math Spectral theorem and eigenfunction expansion for the self-adjoint operator L with discrete spectrum
    Used in the proof of Theorem 5 to decompose the space-time harmonic space H0 into high and low modes.
  • domain assumption Assumption 1(i): local convexity, h(v,v) ≤ 0 on ∂M
    Needed to drop boundary terms in the Reilly formula, giving the Hessian bound.
  • domain assumption Assumption 1(ii): Ric + ∇²U ≥ -ρ for some ρ ≥ 0
    Controls the curvature term in the Hessian bound and appears in the constants.
  • domain assumption Assumption 1(iii): Poincaré inequality with constant m^{-1}
    Provides the spectral gap of L, defines the inverse G, and sets the base relaxation time 2/m.
  • domain assumption Assumption 2: spectrum of L is discrete
    Enables the eigenfunction basis used in the divergence lemma proof; the authors note it can be relaxed via [16].
  • domain assumption Well-posedness of reflected Langevin dynamics on Riemannian manifolds
    Used in Section 4.1; the paper cites Euclidean results [12,13] and [30] for the invariant measure.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Space-time divergence lemmas and optimal non-reversible lifts of diffusions on Riemannian manifolds with boundary." pith.science (2026). https://pith.science/paper/NUIGOEGF

@misc{pith2026241216710,
  author       = {Pith},
  title        = {Pith review of: Space-time divergence lemmas and optimal non-reversible lifts of diffusions on Riemannian manifolds with boundary},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NUIGOEGF}},
  note         = {Machine review of arXiv:2412.16710}
}
abstract

Non-reversible lifts reduce the relaxation time of reversible diffusions at most by a square root. For reversible diffusions on domains in Euclidean space, or, more generally, on a Riemannian manifold with boundary, non-reversible lifts are in particular given by the Hamiltonian flow on the tangent bundle, interspersed with random velocity refreshments, or perturbed by Ornstein-Uhlenbeck noise, and reflected at the boundary. In order to prove that for certain choices of parameters, these lifts achieve the optimal square-root reduction up to a constant factor, precise upper bounds on relaxation times are required. A key tool for deriving such bounds by space-time Poincar\'e inequalities is a quantitative space-time divergence lemma. Extending previous work of Cao, Lu and Wang, we establish such a divergence lemma with explicit constants for general locally convex domains with smooth boundary in Riemannian manifolds satisfying a lower, not necessarily positive, curvature bound. As a consequence, we prove optimality of the lifts described above up to a constant factor, provided the deterministic transport part of the dynamics and the noise are adequately balanced. Our results show for example that an integrated Ornstein-Uhlenbeck process on a locally convex domain with diameter $d$ achieves a relaxation time of the order $d$, whereas, in general, the Poincar\'e constant of the domain is of the order $d^2$.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Hypocoercivity meets lifts

    math.PR 2024-12 conditional novelty 6.0 of 10

    A unified hypocoercivity framework shows adaptive Langevin dynamics is a near-optimal lift, and proves the generalized Langevin equation cannot beat square-root speedup.

Reference graph

Works this paper leans on

59 extracted references · 30 canonical work pages · cited by 1 Pith paper

  1. [25]

    Eberle and F

    A. Eberle and F. L¨ orler. ‘Non-reversible lifts of reve rsible diffusion processes and re- laxation times’. In:Probab. Theory Relat. Fields (Aug. 2024). doi: 10.1007/s00440-024-01308-x

    work page doi:10.1007/s00440-024-01308-x 2024

  2. [1]

    Abraham and J

    R. Abraham and J. E. Marsden. Foundations of Mechanics. 2nd ed. Redwood City, CA: Addison-Wesley Publishing Company, Inc., Oct. 1987. isbn: 0-8053-0102-X

    work page 1987

  3. [2]

    H. M. Afshar and J. Domke. ‘Reflection, Refraction, and Ha miltonian Monte Carlo’. In: Advances in Neural Information Processing Systems . Ed. by C. Cortes, N. Lawrence, D. Lee, M. Sugiyama and R. Garnett. Vol. 28. Curr an Associates, Inc., 2015

    work page 2015

  4. [3]

    Ahn and S

    K. Ahn and S. Chewi. ‘Efficient constrained sampling via th e mirror-Langevin algorithm’. In: Adv. Neural Inf. Process. Syst. 34 (2021), pp. 28405–28418

    work page 2021

  5. [4]

    Albritton, S

    D. Albritton, S. Armstrong, J.-C. Mourrat and M. Novack. ‘Variational meth- ods for the kinetic Fokker-Planck equation’. In: Anal. & PDE (July 2024). doi: 10.2140/apde.2024.17.1953

    doi:10.2140/apde.2024.17.1953 2024

  6. [5]

    Alexander

    S. Alexander. ‘Locally Convex Hypersurfaces of Negativ ely Curved Spaces’. In: Proc. Amer. Math. Soc. 64.2 (1977), pp. 321–325. doi: 10.2307/2041451

    doi:10.2307/2041451 1977

  7. [6]

    Amrouche, P

    C. Amrouche, P. G. Ciarlet and C. Mardare. ‘On a lemma of Ja cques-Louis Lions and its relation to other fundamental results’. In: J. Math. Pures Appl. 104.2 (Aug. 2015), pp. 207–226. doi: 10.1016/j.matpur.2014.11.007

    doi:10.1016/j.matpur.2014.11.007 2015

  8. [7]

    Bakry and M

    D. Bakry and M. ´Emery. ‘Diffusions hypercontractives’. In: S´ eminaire de Probab- ilit´ es XIX 1983/84. Ed. by J. Az´ ema and M. Yor. Springer, 1985, pp. 177–206. doi: 10.1007/BFb0075847

    doi:10.1007/bfb0075847 1983

Show all 59 references
  1. [8]

    Bakry, I

    D. Bakry, I. Gentil and M. Ledoux. Analysis and Geometry of Markov Diffu- sion Operators. Vol. 348. Grundlehren der mathematischen Wissenschaften . Cham: Springer, 2014. isbn: 978-3-319-00227-9. doi: 10.1007/978-3-319-00227-9

    2014 doi

  2. [9]

    Bianchi, B

    D. Bianchi, B. G¨ uneysu and A. G. Setti. Neumann cut-offs and essential self- adjointness on complete Riemannian manifolds with boundar y. 2024. arXiv: 2406.11120 [math.DG]

    2024 arXiv

  3. [10]

    R. L. Bishop. ‘Infinitesimal Convexity Implies Local Co nvexity’. In: Indiana Uni- versity Mathematics Journal 24.2 (1974), pp. 169–172

    1974

  4. [11]

    R. L. Bishop and R. J. Crittenden. Geometry of Manifolds . Vol. 344. AMS Chelsea Publishing Series. American Mathematical Society, 1964. isbn: 978-1-4704-1133-6. doi: 10.1090/chel/344

    1964 doi

  5. [12]

    Bossy and J.-F

    M. Bossy and J.-F. Jabir. ‘On confined McKean Langevin pr ocesses satisfying the mean no-permeability boundary condition’. In: Stochastic Process. Appl. 121.12 (2011), pp. 2751–2775. issn: 0304-4149. doi: https://doi.org/10.1016/j.spa.2011.07.006

    2011 doi

  6. [13]

    Bossy and J.-F

    M. Bossy and J.-F. Jabir. ‘Lagrangian stochastic model s with specular boundary condition’. In:J. Funct. Anal. 268.6 (2015), pp. 1309–1381. doi: 10.1016/j.jfa.2014.11.016

    2015 doi

  7. [14]

    Bou-Rabee and J

    N. Bou-Rabee and J. M. Sanz-Serna. ‘Randomized Hamilto nian Monte Carlo’. In: Ann. Appl. Probab. 27.4 (2017), pp. 2159–2194. doi: 10.1214/16-AAP1255

    2017 doi

  8. [15]

    G. Brigati. Time averages for kinetic Fokker-Planck equations . Nov. 2022. arXiv: 2106.12801 [math.AP] . 37

    2022 arXiv

  9. [16]

    Brigati and G

    G. Brigati and G. Stoltz. How to construct decay rates for kinetic Fokker–Planck equations? Dec. 2023. arXiv: 2302.14506 [math.AP]

    2023

  10. [17]

    Byrne and M

    S. Byrne and M. Girolami. ‘Geodesic Monte Carlo on Embed ded Manifolds’. In: Scand. J. Stat. 40.4 (2013), pp. 825–845. doi: 10.1111/sjos.12036

    2013 doi

  11. [18]

    Y. Cao, J. Lu and L. Wang. ‘On explicit L2-convergence rate estimate for under- damped Langevin dynamics’. In: Arch. Ration. Mech. Anal. 247.90 (Aug. 2023). doi: 10.1007/s00205-023-01922-4

    2023 doi

  12. [19]

    P. R. Chernoff. ‘Essential self-adjointness of powers of generators of hyperbolic equations’. In:J. Funct. Anal. 12.4 (1973), pp. 401–414. doi: 10.1016/0022-1236(73)90003-7

    1973 doi

  13. [20]

    Chevallier, S

    A. Chevallier, S. Pion and F. Cazals. Hamiltonian Monte Carlo with boundary reflections, and application to polytope volume calculatio ns. 2018

    2018

  14. [21]

    Y. Chikuse. Statistics on special manifolds . Vol. 174. Lecture Notes in Statistics. New York, USA: Springer, 2012. isbn: 978-0-387-21540-2. doi: 10.1007/978-0-387-21540-2

    2012 doi

  15. [22]

    Colding and W

    T. Colding and W. Minicozzi. A course in minimal surfaces . Vol. 121. Graduate Studies in Mathematics. American Mathematical Society, 20 11. isbn: 978-1-4704- 7640-3

  16. [23]

    Diaconis, S

    P. Diaconis, S. Holmes and M. Shahshahani. ‘Sampling fr om a manifold’. In: Ad- vances in modern statistical theory and applications: a Fes tschrift in honor of Morris L. Eaton . Vol. 10. Institute of Mathematical Statistics Collection s. Insti- tute of Mathematical Statistics...

    doi

  17. [24]

    Eberle, A

    A. Eberle, A. Guillin, L. Hahn, F. L¨ orler and M. Michel. ‘Convergence of non- reversible Markov processes via lifting and flow Poincar´ e i nequality’. In prepara- tion. 2025

    2025

  18. [26]

    Engel and R

    K.-J. Engel and R. Nagel. One-Parameter Semigroups for Linear Evolution Equa- tions. Vol. 194. Graduate Texts in Mathematics. New York: Springe r, Oct. 1999. isbn: 978-0-387-98463-6. doi: 10.1007/b97696

    1999 doi

  19. [27]

    Gallot, D

    S. Gallot, D. Hulin and J. Lafontaine. Riemannian Geometry. 3rd ed. Universitext. Springer, 2004. isbn: 978-3-642-18855-8. doi: 10.1007/978-3-642-18855-8

    2004 doi

  20. [28]

    Gatmiry, J

    K. Gatmiry, J. Kelner and S. S. Vempala. ‘Sampling Polyt opes with Riemannian HMC: Faster Mixing via the Lewis Weights Barrier’. In: Proceedings of Thirty Seventh Conference on Learning Theory . Ed. by S. Agrawal and A. Roth. Vol. 247. Proceedings of Machine Learning Research....

    2024

  21. [29]

    Girolami and B

    M. Girolami and B. Calderhead. ‘Riemann Manifold Lange vin and Hamiltonian Monte Carlo Methods’. In: J. R. Statist. Soc. 73.2 (Mar. 2011), pp. 123–214. issn: 1369-7412. doi: 10.1111/j.1467-9868.2010.00765.x. 38

    2011

  22. [30]

    Grothaus and M

    M. Grothaus and M. Mertin. ‘Hypocoercivity of Langevin -type dynamics on ab- stract smooth manifolds’. In: Stochastic Process. Appl. 146 (2022), pp. 22–59. issn: 0304-4149. doi: 10.1016/j.spa.2021.12.007

    2022 doi

  23. [31]

    Hongjie Dong, Y

    H. Hongjie Dong, Y. Yan Guo and T. Yastrzhembskiy. ‘Kine tic Fokker-Planck and Landau equations with specular reflection boundary conditi on’. In: Kinet. Relat. Models 15.3 (2022), pp. 467–516. doi: 10.3934/krm.2022003

    2022 doi

  24. [32]

    Ikeda and S

    N. Ikeda and S. Watanabe. Stochastic differential equations and diffusion processes . 2nd ed. Vol. 24. North Holland Mathematical Library. Amster dam: North-Holland Publ. Co., 1989. isbn: 978-0-444-87378-1

    1989

  25. [33]

    J. Jost. Riemannian Geometry and Geometric Analysis . Universitext. Springer,

  26. [34]

    Kolesnikov and E

    A. Kolesnikov and E. Milman. ‘Brascamp–Lieb-type ineq ualities on weighted Rieman- nian manifolds with boundary’. In: J. Geom. Anal. 27 (Sept. 2017), pp. 1680–1702. doi: 10.1007/s12220-016-9736-5

    2017 doi

  27. [35]

    A. V. Kolesnikov and E. Milman. ‘Riemannian metrics on c onvex sets with applic- ations to Poincar´ e and log-Sobolev inequalities’. In: Calc. Var. 55.77 (June 2016). doi: 10.1007/s00526-016-1018-3

    2016 doi

  28. [36]

    A. V. Kolesnikov and E. Milman. ‘Poincar´ e and Brunn-Mi nkowski inequalities on the boundary of weighted Riemannian manifolds’. In: Am. J. Math. 140.5 (Oct. 2018), pp. 1147–1185. doi: 10.1353/ajm.2018.0027

    2018

  29. [37]

    Y. Kook, Y. T. Lee, R. Shen and S. Vempala. ‘Condition-nu mber-independent Convergence Rate of Riemannian Hamiltonian Monte Carlo wit h Numerical In- tegrators’. In: Proceedings of Thirty Sixth Conference on Learning Theory . Ed. by G. Neu and L. Rosasco. Vol. 195. Proceedings...

    2023

  30. [38]

    Lan and L

    S. Lan and L. Kang. ‘Sampling constrained continuous pr obability distributions: A review’. In:WIREs Computational Statistics 15.6 (2023). doi: 10.1002/wics.1608

    2023 doi

  31. [39]

    J. Lee. Introduction to Smooth Manifolds . Vol. 218. Graduate Texts in Mathemat- ics. New York: Springer, 2013. isbn: 978-1-4419-9982-5. doi: 10.1007/978-1-4419-9982-5

    2013 doi

  32. [40]

    J. Lee. Introduction to Riemannian Manifolds . Graduate Texts in Mathematics. Springer International Publishing, 2018. isbn: 978-3-319-91755-9. doi: 10.1007/978-3-319-91755-9

    2018 doi

  33. [41]

    Y. T. Lee and S. S. Vempala. ‘Convergence rate of Riemann ian Hamiltonian Monte Carlo and faster polytope volume computation’. In: Proceedings of the 50th Annual ACM SIGACT Symposium on Theory of Computing . STOC 2018. New York, USA: Association for Computing Machinery, 2018, ...

    2018

  34. [42]

    Y. T. Lee and S. S. Vempala. ‘The Manifold Joys of Samplin g’. In: 49th Inter- national Colloquium on Automata, Languages, and Programmin g (ICALP 2022) . Ed. by M. Boja´ nczyk, E. Merelli and D. P. Woodruff. Vol. 229. Le ibniz Interna- tional Proceedings in Informatics (LIPIcs...

    2022 doi

  35. [43]

    J. Lehec. Convergence in total variation for the kinetic Langevin algo rithm. 2024. arXiv: 2407.09301 [math.PR]

    2024 arXiv

  36. [44]

    Lelievre, M

    T. Lelievre, M. Rousset and G. Stoltz. ‘Langevin dynami cs with constraints and computation of free energy differences’. In: Mathematics of computation 81.280 (2012), pp. 2071–2125

    2012

  37. [45]

    Leli` evre, M

    T. Leli` evre, M. Rousset and G. Stoltz. Free Energy Computations. A Mathematical Perspective. London: Imperial College Press, 2010. isbn: 978-1-908978-75-2. doi: 10.1142/p579

    2010 doi

  38. [46]

    P. Li. Geometric Analysis . Cambridge Studies in Advanced Mathematics. Cam- bridge University Press, 2012

    2012

  39. [47]

    H. C. Lie, D. Rudolf, B. Sprungk and T. J. Sullivan. ‘Dime nsion-independent Markov chain Monte Carlo on the sphere’. In: Scand. J. Stat. 50.4 (2023), pp. 1818–

    2023

  40. [48]

    Lu and L

    J. Lu and L. Wang. ‘On explicit L2-convergence rate estimate for piecewise de- terministic Markov processes in MCMC algorithms’. In: Ann. Appl. Probab. 32.2 (Apr. 2022), pp. 1333–1361. doi: 10.1214/21-AAP1710

    2022 doi

  41. [49]

    Ma and S.-H

    L. Ma and S.-H. Du. ‘Extension of Reilly formula with app lications to eigenvalue estimates for drifting Laplacians’. In: Compt. Rendus Math. 348.21-22 (2010), pp. 1203–1206. doi: 10.1016/j.crma.2010.10.003

    2010 doi

  42. [50]

    R. Reilly. ‘Applications of the Hessian Operator in a Ri emannian Manifold’. In: In- diana Univ. Math. J. 26.3 (1977), pp. 459–472. doi: 10.1512/iumj.1977.26.26036

    1977 doi

  43. [51]

    T. Sakai. Riemannian Geometry . Vol. 149. Translations of Mathematical Mono- graphs. American Mathematical Society, 1997. isbn: 0-8218-0284-4

    1997

  44. [52]

    S. Sasaki. ‘On the differential geometry of tangent bundl es of Riemannian mani- folds’. In:Tohoku Math. J. 10.3 (1958), pp. 338–354. doi: 10.2748/tmj/1178244668

    1958

  45. [53]

    Tabachnikov

    S. Tabachnikov. Geometry and Billiards . Vol. 30. Student Mathematical Library. American Mathematical Soc., 2005. doi: 10.1090/stml/030

    2005 doi

  46. [54]

    M. E. Taylor. Partial Differential Equations II. Qualitative Studies of Line ar Equa- tions. Applied Mathematical Sciences. Springer Cham, 2023. isbn: 978-3-031-33700-

    2023

  47. [55]

    C. Villani. Hypocoercivity. Vol. 202. Memoirs of the American Mathematical Soci- ety 950. Providence, RI: American Mathematical Society, 20 09. isbn: 978-0-8218- 4498-4. doi: 10.1090/S0065-9266-09-00567-5 . 40

    doi

  48. [56]

    F.-Y. Wang. Analysis for Diffusion Processes on Riemannian Manifolds . 1st ed. Vol. 18. Advanced Series on Statistical Science and Applied Probability. Singapore: World Scientific, 2014. isbn: 978-981-4452-64-9. 41

    2014

  49. [57]

    doi: 10.1007/978-3-031-33700-0

    doi

  50. [1858]

    doi: 10.1111/sjos.12653

    doi

  51. [2017]

    doi: 10.1007/978-3-319-61860-9

    isbn: 978-3-319-61860-9. doi: 10.1007/978-3-319-61860-9

    doi

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.