Hit-and-Run Mixes as Fast as the Ball Walk

Ruizhe Zhang

REVIEW 2 major objections 3 minor 60 references

Hit-and-Run Mixes as Fast as the Ball Walk

T0 review · 2 major / 3 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read This paper proves that the lazy hit-and-run walk on any isotropic convex body mixes from an M-warm start in O(n^2 ψ_n^{-2} log^3(M/ε)) steps, matching the ball walk up to logarithmic factors.

desk verdict The main theorem is probably repairable, but the paper as written contains a load-bearing inverted log factor in Corollary 7.2, and the proof of Theorem 7.3 does not actually follow from the stated conductance bound. read the letter →

arxiv 2608.13487 v1 pith:BOBCKLKQ submitted 2026-08-13 cs.DS math.PR

Ruizhe Zhang This is my paper · ORCID

classification cs.DSmath.PR MSC 68W2052A4060J05

keywords hit-and-runwalkmixingtimeconvexbodysamplingKLSconstantconductancelocalizationisoperimetricinequalitywarmstart

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

The paper claims that the lazy hit-and-run walk—a Markov chain that, from a point inside a convex body, samples its next point uniformly along a random chord—mixes in O($n^{2}$ $ψ_n^{{-2}}$ $log^{3}$(M/ε)) steps from any M-warm starting distribution on an n-dimensional isotropic convex body, where $ψ_n^{{-1}}$ is the KLS constant. Up to logarithmic factors this matches the best-known warm-start mixing time of the ball walk, and it replaces the polynomial dependence on the warmness M and accuracy ε in the previous hit-and-run bound with a polylogarithmic one. The result matters because hit-and-run needs no step-size parameter, so a parameter-free sampler becomes asymptotically as fast as the state-of-the-art tuned walk. The proof works through conductance: every cut of small stationary mass has a large weighted perimeter when the boundary is weighted by the local radius r_K(x), and the walk's one-step transition laws overlap uniformly in the metric that this same radius defines.

What carries the argument

The carrying mechanism is a one-dimensional concave-weight isoperimetric inequality (Theorem 4.2) lifted to the body by a guided localization that preserves the mass of the cut on each needle. The central objects are the local conductance radius r_K(x), the largest radius r for which the ball B(x,r) is at least 63/64 contained in K, and the relaxed weighted perimeter P_w(E), the minimum weighted total variation of Lipschitz approximations to the indicator of E under weight w. On a single needle, the inequality bounds the weighted perimeter of a set of mass p below by a constant times (m_I/σ_I) p/(1 + log(1/p)), where m_I is the average of the concave weight and σ_I is the conditional standard deviation of the guiding coordinate. Averaging over needles uses the law of total variance, the spectral-gap inequality, and the reverse Hölder bound for concave weights to show that the average of m_I/σ_I is at least a constant times E[r_K]/√CPI(μ), and hence about ψ_n/√n. The same local radius defines the intrinsic metric d_{r_K} in which one-step overlap is uniform, and a weighted coarea inequality converts the weighted perimeter bound into an ergodic-flow, and therefore conductance, bound.

What would settle it

Compute the relaxed weighted perimeter P_{r_K}(A) for a half-space cut in a cube and compare it with the claimed lower bound of order $n^{{-1/2}}$ p/(1 + log(1/p)); a smaller value, or a measured mixing time on a cube that grows faster than $n^{2}$ $log^{3}$(M/ε) for fixed ε and M, would refute the theorem.

Watch

Extended reading notes

Core claim

The central claim is that for every isotropic convex body K, every M-warm start, and every ε<1/2, the lazy hit-and-run walk has τ_mix(ε, μ_init, μ; P^L_HR) ≲ $n^{2}$ $ψ_n^{{-2}}$ (1 + log(4M/ε))^2 log(2M/ε), which simplifies to O($n^{2}$ $ψ_n^{{-2}}$ $log^{3}$(8M/ε)). The paper derives this from an s-conductance lower bound Φ_s(P^L_HR) ≳ ψ_n $n^{{-1}}$(1 + log(2/s)) that holds for every small-set threshold s<1/4. The geometric estimate underneath is a local-radius weighted isoperimetric inequality: for every measurable A with p = min{μ(A), 1-μ(A)} ≤ 1/2, the relaxed weighted perimeter P_{r_K}(A) is at least a constant times ψ_n $n^{{-1/2}}$ p/(1 + log(1/p)). This all-scale perimeter estimate avoids discarding a boundary layer, and the construction keeps the mass of the cut equal on every localization needle, so that both the fixed-core loss and the probabilistic cut-survival loss of the previous approach disappear.

Load-bearing premise

The proof rests on a balance-preserving localization theorem that decomposes the body into one-dimensional pieces on which every measurable cut keeps exactly its global mass; if that theorem is not valid for arbitrary measurable sets, the averaging step that produces the weighted isoperimetric inequality collapses.

Editorial extensions

If this is right

Warm-start hit-and-run now carries the same O~(n^2/ψ_n^2) mixing dependence as the ball walk, so the parameter-free chord sampler loses nothing asymptotically to a step-size-tuned walk.
With the current best KLS estimate ψ_n^{-1}=O(log^{1/4} n), the bound becomes O(n^2 log^{1/2} n log^3(M/ε)) for every isotropic convex body.
The previous restriction n ≳ log(M/ε) and the polynomial (M/ε)^{11} cost disappear; mixing now depends only polylogarithmically on warmness and accuracy.
Because the conductance lower bound holds uniformly over all small-set thresholds s, the result covers the full range of warm-start accuracies, not just large cuts.

Reading between the lines

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

If the guided-localization machinery extends to general log-concave measures, the same local-radius weighting could give polylog-warm mixing bounds for non-uniform targets; this is an extension the paper does not claim.
The logarithmic factors come from cut transitions near the boundary, where r_K is small; a sharper one-dimensional estimate for endpoint transitions would plausibly remove a log, and this could be checked numerically on one-dimensional log-concave weights.
Cold starts are untouched: a point mass is not M-warm for any finite M, so converting this theorem into a full sampling algorithm still requires a separate warm-start construction or a cold-start argument.

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.

Desk Editor's Note

read the letter

Worth your time: this is the first hit-and-run mixing bound with n^2 psi^{-2} and polylog(M/epsilon), matching the ball walk. The proof architecture is genuinely nice: local-radius weighted perimeter, a one-dimensional concave-weight isoperimetric inequality, lifted by Klartag guided localization, then averaged with Poincare and Berwald. Sections 4 and 5 are careful, and I found no local error there. The high-level approach is believable and the literature is engaged honestly.

But the manuscript as written has a load-bearing sign error in Corollary 7.2. The proof derives P_{r_K}(E) >~ (psi_n/sqrt(n)) p/(1+log(1/p)) and then claims this is >= (psi_n/sqrt(n))(1+log(2/s)) p when p >= s/2. That inequality is backwards: p >= s/2 gives 1+log(1/p) <= 1+log(2/s), so the correct lower bound is p/(1+log(2/s)), not p(1+log(2/s)). Consequently the conductance should be Phi_s >~ (psi_n/n)/(1+log(2/s)), not the printed (psi_n/n)(1+log(2/s)).

This matters for two reasons. First, the wrong direction is exactly the step that converts the isoperimetric estimate into the conductance bound. Second, the theorem proof says "substituting (79)" to get (80), but substituting the printed (79) gives roughly n^2 psi^{-2} log(2M/epsilon)/log^2(4M/epsilon), the reciprocal of (80). So there is an internal inconsistency as written. The stress-test note is correct.

The good news: the final theorem bound (80) matches the corrected reciprocal conductance version, and the rest of the argument does not seem to depend on the erroneous direction. This looks eminently repairable: flip the inequality in Corollary 7.2, state the corollary with the reciprocal log, and the proof of Theorem 7.3 works. I have not found a second issue of the same magnitude.

As written, the paper should not be accepted. The main claim is probably right, but the middle of the proof is currently false, and an editor or referee would need the author to confirm the fix. It is significant enough to deserve referee time in a revised form. I would bring it to a reading group to trace the corrected conductance, and I would cite the final theorem once the correction is in place.

Referee Report

2 major / 3 minor

Summary. The paper proves that the lazy hit-and-run walk on an isotropic convex body K mixes from any M-warm start to total-variation distance ε in O(n^2 ψ_n^{-2} log^3(M/ε)) steps, where ψ_n^{-1} is the KLS constant. The proof goes through s-conductance: it establishes a local-radius weighted isoperimetric inequality via Klartag's guided localization and a one-dimensional concave-weight isoperimetric theorem, then converts this into a conductance bound using an intrinsic local-radius metric and Lovász's overlap lemma. Appendices supply proofs of one-dimensional log-concave estimates, properties of the relaxed weighted perimeter, properties of the local conductance radius, and a Euclidean specialization of Klartag's localization theorem. The claimed bound improves on Chen--Eldan's polynomial dependence on M/ε and removes their condition n ≳ log(M/ε), thereby resolving the open question they posed.

Significance. The main theorem, if correct, is a significant advance in geometric random walks: it places hit-and-run on the same asymptotic footing as the ball walk up to logarithmic factors and improves the warm-start mixing time from polynomial to polylogarithmic dependence on M/ε. The paper is careful and detailed: the one-dimensional isoperimetric theorem (Section 4), the intrinsic geometry and overlap argument (Section 6), and the perimeter and localization appendices are all developed at length, and the external black boxes (Klartag's guided localization, the Lovász overlap lemma, Berwald's inequality, Cheeger--Poincaré equivalence) are identified clearly. The central mechanism—guided localization plus weighted Poincaré averaging—is novel in this context. However, there is a genuine reversed-inequality error in Corollary 7.2 that invalidates the corollary as printed; the final theorem appears repairable because its stated bound matches the corrected reciprocal form. The paper should not be accepted before this error is fixed.

major comments (2)

[§7, Corollary 7.2] The displayed inequality in the proof of Corollary 7.2 is reversed. Since p ≥ s/2 implies 1+log(1/p) ≤ 1+log(2/s), we have p/(1+log(1/p)) ≥ p/(1+log(2/s)), not p/(1+log(1/p)) ≥ (1+log(2/s))p. The asserted inequality is false for small s: for example, with s = 10^{-4} and p = 1/2 the left-hand side is about 0.045 while the right-hand side is about 5. The valid conclusion is γ_s = c ψ_n/(√n(1+log(2/s))), and Proposition 7.1 then yields Φ_s(P_HR^L) ≳ ψ_n/(n(1+log(2/s))), which is the reciprocal of the printed (79). This step is load-bearing because it is exactly the conversion of the isoperimetric estimate into the conductance bound used in Theorem 7.3. The printed Corollary 7.2 is therefore false as stated, even though the final theorem's bound (80) is consistent with the corrected reciprocal form.
[§1.2, Eq. (3)] The same reversed inequality appears in the technical overview. Inequality (3) states Φ_s(P_HR^L) ≳ (ψ_n/n)(1+log(2/s)); if this were true it would imply the stronger mixing-time bound ~ n^2 ψ_n^{-2} log(2M/ε)/(1+log(2/s))^2, not the theorem's stated n^2 ψ_n^{-2} log^3(8M/ε). The correct reduction, consistent with Theorem 7.3, is Φ_s ≳ (ψ_n/n)/(1+log(2/s)). The overview and the corollary must be corrected together so that the paper's stated reduction, its intermediate conductance bound, and the final theorem are mutually consistent.

minor comments (3)

[§6, Lemma 6.5] The explicit constants c0 = 1/(64 C_loc) and α0 = 1/1000 are used throughout the proof; stating them in the lemma statement would help the reader and make the subsequent numerical checks in Proposition 7.1 easier to follow.
[Appendix D, proof of Lemma 5.1] The derivation of the conditional balance property (50) relies on the statement that every measurable saturated set has zero f-integral, imported from [Kla17, Lemma 4.6]. Since this is the only point at which the balance property enters, a few more sentences explaining that import would increase confidence in this black-box step.
[Throughout] The provided text contains numerous spacing and line-break artifacts (for example, 'hit-and-run[ Smi84,Lov99,LV06a]; bothextend' and inconsistent spaces in expressions like 'λ(x,r )'). These should be cleaned up in the journal version.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the hit-and-run mixing bound is proved from independent external inputs (KLS, Poincaré, Berwald, guided localization) and does not assume its conclusion.

full rationale

The derivation chain is non-circular. Theorem 7.3's mixing-time bound is the output of a standard Lovász–Simonovits conductance argument (Lemma 2.3), and the intermediate conductance estimate is derived from an isoperimetric inequality for the weighted perimeter, not from any assumed mixing time of hit-and-run. The external inputs—Klartag's guided localization (Lemma 5.1 / Theorem D.1), the Poincaré–Cheeger equivalence (Proposition 2.7), Berwald's inequality (Lemma 2.8), and Lovász's overlap lemma (Lemma 6.4)—are parameter-free theorems whose stated assumptions do not include the target result. The balanced-needle property μ_I(A) = μ(A) is imported from [Kla17] and proved in Appendix D from Klartag's theorem; it carries no hit-and-run content and is not circular. No fitted parameter is renamed as a prediction, no ansatz is smuggled in via self-citation, and no uniqueness theorem from the author's own prior work is used to force a choice. The cited works by other authors (including Chen–Eldan and Lovász–Vempala) are used as external benchmarks or as sources of standard lemmas, and the present author's own prior work is not load-bearing anywhere in the proof. The apparent inequality reversal in Corollary 7.2 flagged by a critical reader is a correctness concern about the printed algebra, not a circularity: even if the lower bound is misstated, the theorem is not assuming its conclusion. Thus the paper does not exhibit any of the enumerated circularity patterns, and the appropriate score is 0.

Assumptions & free parameters 0 free parameters · 7 assumptions · 0 invented entities

The central proof imports several deep external theorems but introduces no new entities and fits no free parameters. The listed axioms are the unproved background results the derivation rests on.

assumptions (7)

standard math Klartag's guided localization theorem, Euclidean specialization (Lemma 5.1 and Theorem D.1)
The main lifting step from one-dimensional inequalities to K uses the existence of a common 1-Lipschitz guiding function u and exact balance μ_I(A)=p on almost every needle. The paper derives this from [Kla17] in Appendix D; if this specialization fails, the proof collapses.
standard math Cheeger-Poincaré equivalence CPI(ν) ≍ ψ_ch(ν)^{-2} for log-concave ν (Proposition 2.7)
Used to convert the KLS constant into a Poincaré constant bound in Corollary 5.3.
standard math Berwald's inequality for concave weights (Lemma 2.8)
Used in Theorem 5.2 to control the second moment of conditional means m_I.
standard math Lovász's one-step overlap lemma (Lemma 6.4)
Converts closeness in the local-radius metric into a constant total-variation overlap of hit-and-run transition kernels.
standard math Log-concave density estimates of Lemma 2.4, from [LV07, CE26]
Inputs to the one-dimensional weighted density lemma.
standard math Properties of the local conductance radius r_K from [LV06a] (Lemma 3.2)
Concavity, Lipschitz bound, and average lower bound E[r_K] ≳ n^{-1/2} are load-bearing in the weighted perimeter and overlap arguments.
standard math Letwin's KLS estimate ψ_n^{-1}=O(log^{1/4} n) [Let26]
Used only for the concrete corollary after the main theorem; the main theorem is stated in terms of ψ_n directly.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Hit-and-Run Mixes as Fast as the Ball Walk." pith.science (2026). https://pith.science/paper/BOBCKLKQ

@misc{pith2026260813487,
  author       = {Pith},
  title        = {Pith review of: Hit-and-Run Mixes as Fast as the Ball Walk},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BOBCKLKQ}},
  note         = {Machine review of arXiv:2608.13487}
}

abstract

Let $K\subset\mathbb{R}^n$ be an isotropic convex body. We prove that the hit-and-run walk, started from any $M$-warm distribution, reaches total-variation distance $\varepsilon$ from the uniform distribution on $K$ in $O\!\left(n^2\psi_n^{-2}\log^3(M/\varepsilon)\right)$ steps, where $\psi_n^{-1}$ is the Kannan-Lov\'asz-Simonovits (KLS) constant. Up to logarithmic factors, this matches the best-known warm-start mixing time for the ball walk. Chen and Eldan [Discrete Comput. Geom. 2026] obtained the same $n^2\psi_n^{-2}$ dependence for hit-and-run, but with polynomial dependence on $M/\varepsilon$. Our result improves that polynomial dependence to a polylogarithmic one, fully resolving their open question about warm-start mixing of hit-and-run in isotropic convex bodies.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 54 canonical work pages

[1]

Ambrosio, Luigi and Fusco, Nicola and Pallara, Diego , title=

work page
[2]

Verallgemeinerung eines Mittelwertsatzes von J

Berwald, Ludwig , journal=. Verallgemeinerung eines Mittelwertsatzes von J. Favard f. 1947 , publisher=

work page 1947
[3]

Discrete & Computational Geometry , volume=

Hit-and-run mixing via localization schemes , author=. Discrete & Computational Geometry , volume=. 2026 , publisher=

work page 2026
[4]

Discrete & Computational Geometry , volume=

Isoperimetric problems for convex bodies and a localization lemma , author=. Discrete & Computational Geometry , volume=. 1995 , publisher=

work page 1995
[5]

Needle Decompositions in Riemannian Geometry , volume=

Klartag, Bo'az , year=. Needle Decompositions in Riemannian Geometry , volume=. Memoirs of the American Mathematical Society , publisher=. doi:10.1090/memo/1180 , number=

doi:10.1090/memo/1180
[6]

Geometriae Dedicata , volume=

Mai, Cong Hung , title=. Geometriae Dedicata , volume=. 2019 , doi=

work page 2019
[7]

Commentarii Mathematici Helvetici , volume=

Mai, Cong Hung and Ohta, Shin-ichi , title=. Commentarii Mathematici Helvetici , volume=. 2021 , doi=

work page 2021
[8]

Stability of Eigenvalues and Observable Diameter in

Bertrand, J. Stability of Eigenvalues and Observable Diameter in. The Journal of Geometric Analysis , volume=. 2022 , doi=

work page 2022

Show all 60 references

[9]

Weighted Berwald's inequality

Dylan Langharst and Eli Putterman. Weighted Berwald's inequality. Indiana Univ. Math. J
[10]

Mathematical programming , volume=

Hit-and-run mixes fast , author=. Mathematical programming , volume=. 1999 , publisher=

1999
[11]

Random structures & algorithms , volume=

Random walks in a convex body and an improved volume algorithm , author=. Random structures & algorithms , volume=. 1993 , publisher=

1993
[12]

Proceedings [1990] 31st annual symposium on foundations of computer science , pages=

The mixing rate of Markov chains, an isoperimetric inequality, and computing the volume , author=. Proceedings [1990] 31st annual symposium on foundations of computer science , pages=. 1990 , organization=

1990
[13]

Hit-and-Run from a Corner , journal =

Lov. Hit-and-Run from a Corner , journal =. 2006 , doi =

2006
[14]

Random Structures & Algorithms , volume=

The geometry of logconcave functions and sampling algorithms , author=. Random Structures & Algorithms , volume=. 2007 , publisher=

2007
[15]

Operations Research , volume=

Efficient Monte Carlo procedures for generating points uniformly distributed over bounded regions , author=. Operations Research , volume=. 1984 , publisher=

1984
[16]

, title=

Vempala, Santosh S. , title=. Combinatorial and Computational Geometry , editor=
[17]

Ars Inveniendi Analytica , year=

Klartag, Bo'az , title=. Ars Inveniendi Analytica , year=
[18]

2026 , eprint=

Letwin, Brayden , title=. 2026 , eprint=

2026
[19]

and Frieze, Alan M

Dyer, Martin E. and Frieze, Alan M. and Kannan, Ravi , title=. Journal of the ACM , volume=. 1991 , doi=

1991
[20]

Random Walks and an

Kannan, Ravi and Lov. Random Walks and an. Random Structures & Algorithms , volume=
[21]

Simulated Annealing in Convex Bodies and an

Lov. Simulated Annealing in Convex Bodies and an. Journal of Computer and System Sciences , volume=. 2006 , doi=

2006
[22]

, title=

Cousins, Ben and Vempala, Santosh S. , title=. SIAM Journal on Computing , volume=. 2018 , doi=

2018
[23]

Geometric and Functional Analysis , volume=

Eldan, Ronen , title=. Geometric and Functional Analysis , volume=. 2013 , doi=

2013
[24]

, title=

Lee, Yin Tat and Vempala, Santosh S. , title=. Annals of Mathematics , volume=. 2024 , doi=

2024
[25]

Geometric and Functional Analysis , volume=

Chen, Yuansi , title=. Geometric and Functional Analysis , volume=. 2021 , doi=

2021
[26]

Geometric and Functional Analysis , volume=

Klartag, Bo'az and Lehec, Joseph , title=. Geometric and Functional Analysis , volume=. 2022 , doi=

2022
[27]

2024 , eprint=

Guan, Qingyang , title=. 2024 , eprint=

2024
[28]

Geometric and Functional Analysis , volume=

Klartag, Bo'az and Lehec, Joseph , title=. Geometric and Functional Analysis , volume=. 2025 , doi=

2025
[29]

Concentration, Functional Inequalities and Isoperimetry , series=

Eldan, Ronen and Klartag, Bo'az , title=. Concentration, Functional Inequalities and Isoperimetry , series=. 2011 , doi=

2011
[30]

2025 , eprint=

Bizeul, Pierre , title=. 2025 , eprint=

2025
[31]

2025 , eprint=

Klartag, Bo'az and Lehec, Joseph , title=. 2025 , eprint=

2025
[32]

2026 , eprint=

Chen, Yuansi and Klartag, Bo'az , title=. 2026 , eprint=

2026
[33]

Bulletin of the Hellenic Mathematical Society , volume=

Giannopoulos, Apostolos and Pafis, Minas and Tziotziou, Natalia , title=. Bulletin of the Hellenic Mathematical Society , volume=
[34]

, title=

Kook, Yunbum and Vempala, Santosh S. , title=. 2025 , eprint=

2025
[35]

2022 IEEE 63rd Annual Symposium on Foundations of Computer Science , pages=

Chen, Yuansi and Eldan, Ronen , title=. 2022 IEEE 63rd Annual Symposium on Foundations of Computer Science , pages=. 2022 , publisher=

2022
[36]

, title=

Jia, He and Laddha, Aditi and Lee, Yin Tat and Vempala, Santosh S. , title=. Journal of the ACM , volume=. 2026 , doi=

2026
[37]

Problems in Analysis , editor=

Cheeger, Jeff , title=. Problems in Analysis , editor=
[38]

Transactions of the American Mathematical Society , volume=

The central limit problem for convex bodies , author=. Transactions of the American Mathematical Society , volume=
[39]

Geometric Aspects of Functional Analysis: Israel Seminar 2001-2002 , pages=

On the central limit property of convex bodies , author=. Geometric Aspects of Functional Analysis: Israel Seminar 2001-2002 , pages=. 2004 , organization=

2001
[40]

, title=

Kook, Yunbum and Zhang, Matthew S. , title=. Proceedings of the 2025 Annual ACM-SIAM Symposium on Discrete Algorithms , pages=. 2025 , publisher=

2025
[41]

Proceedings of the Thirty-Fourth Conference on Learning Theory , series=

Lee, Yin Tat and Shen, Ruoqi and Tian, Kevin , title=. Proceedings of the Thirty-Fourth Conference on Learning Theory , series=. 2021 , publisher=

2021
[42]

Mathematics of Operations Research , volume=

Kannan, Ravindran and Narayanan, Hariharan , title=. Mathematics of Operations Research , volume=. 2012 , doi=

2012
[43]

and Yu, Bin , title=

Chen, Yuansi and Dwivedi, Raaz and Wainwright, Martin J. and Yu, Bin , title=. Journal of Machine Learning Research , volume=. 2018 , url=

2018
[44]

Proceedings of Thirty Seventh Conference on Learning Theory , pages =

Sampling Polytopes with Riemannian HMC: Faster Mixing via the Lewis Weights Barrier , author =. Proceedings of Thirty Seventh Conference on Learning Theory , pages =. 2024 , editor =

2024
[45]

The Thirty Seventh Annual Conference on Learning Theory , pages=

Gaussian cooling and Dikin walks: The interior-point method for logconcave sampling , author=. The Thirty Seventh Annual Conference on Learning Theory , pages=. 2024 , organization=

2024
[46]

Beyond the d^

Yunbum Kook , year=. Beyond the d^. 2607.13943 , archivePrefix=

arXiv
[47]

, title=

Lee, Yin Tat and Vempala, Santosh S. , title=. Proceedings of the 50th Annual ACM SIGACT Symposium on Theory of Computing , pages=. 2018 , publisher=

2018
[48]

, title=

Laddha, Aditi and Lee, Yin Tat and Vempala, Santosh S. , title=. Proceedings of the 52nd Annual ACM SIGACT Symposium on Theory of Computing , pages=. 2020 , publisher=

2020
[49]

Proceedings of the Thirty-Fifth Conference on Learning Theory , series=

Chen, Yongxin and Chewi, Sinho and Salim, Adil and Wibisono, Andre , title=. Proceedings of the Thirty-Fifth Conference on Learning Theory , series=. 2022 , publisher=

2022
[50]

Journal of the ACM (JACM) , volume=

Solving convex programs by random walks , author=. Journal of the ACM (JACM) , volume=. 2004 , publisher=

2004
[51]

Mathematics of Operations Research , volume=

Simulated annealing for convex optimization , author=. Mathematics of Operations Research , volume=. 2006 , publisher=

2006
[52]

Bayesian Analysis , year=

Bayesian auxiliary variable models for binary and multinomial regression , author=. Bayesian Analysis , year=
[53]

Journal of the American statistical Association , volume=

Bayesian analysis of binary and polychotomous response data , author=. Journal of the American statistical Association , volume=. 1993 , publisher=

1993
[54]

, title=

Kook, Yunbum and Vempala, Santosh S. , title=. 2025 IEEE 66th Annual Symposium on Foundations of Computer Science , pages=. 2025 , publisher=

2025
[55]

, title=

Kook, Yunbum and Vempala, Santosh S. , title=. 2026 , eprint=

2026
[56]

and Zhang, Matthew S

Kook, Yunbum and Vempala, Santosh S. and Zhang, Matthew S. , title=. Advances in Neural Information Processing Systems , volume=. 2024 , doi=

2024
[57]

, title=

Kook, Yunbum and Vempala, Santosh S. , title=. Proceedings of the 57th Annual ACM Symposium on Theory of Computing , pages=. 2025 , publisher=

2025
[58]

American Journal of Mathematics , volume=

On high dimensional maximal functions associated to convex bodies , author=. American Journal of Mathematics , volume=. 1986 , publisher=

1986
[59]

Annales Scientifiques de l'\'Ecole Normale Sup\'erieure , volume=

Buser, Peter , title=. Annales Scientifiques de l'\'Ecole Normale Sup\'erieure , volume=. 1982 , doi=

1982
[60]

Surveys in Differential Geometry , volume=

Ledoux, Michel , title=. Surveys in Differential Geometry , volume=. 2004 , doi=

2004

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Ambrosio, Luigi and Fusco, Nicola and Pallara, Diego , title=

work page

[2] [2]

Verallgemeinerung eines Mittelwertsatzes von J

Berwald, Ludwig , journal=. Verallgemeinerung eines Mittelwertsatzes von J. Favard f. 1947 , publisher=

work page 1947

[3] [3]

Discrete & Computational Geometry , volume=

Hit-and-run mixing via localization schemes , author=. Discrete & Computational Geometry , volume=. 2026 , publisher=

work page 2026

[4] [4]

Discrete & Computational Geometry , volume=

Isoperimetric problems for convex bodies and a localization lemma , author=. Discrete & Computational Geometry , volume=. 1995 , publisher=

work page 1995

[5] [5]

Needle Decompositions in Riemannian Geometry , volume=

Klartag, Bo'az , year=. Needle Decompositions in Riemannian Geometry , volume=. Memoirs of the American Mathematical Society , publisher=. doi:10.1090/memo/1180 , number=

doi:10.1090/memo/1180

[6] [6]

Geometriae Dedicata , volume=

Mai, Cong Hung , title=. Geometriae Dedicata , volume=. 2019 , doi=

work page 2019

[7] [7]

Commentarii Mathematici Helvetici , volume=

Mai, Cong Hung and Ohta, Shin-ichi , title=. Commentarii Mathematici Helvetici , volume=. 2021 , doi=

work page 2021

[8] [8]

Stability of Eigenvalues and Observable Diameter in

Bertrand, J. Stability of Eigenvalues and Observable Diameter in. The Journal of Geometric Analysis , volume=. 2022 , doi=

work page 2022