Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

A Bayesian sequential soft classification problem for a Brownian motion's drift

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper solves the soft-classification version of Bayesian sequential testing for the drift of a Brownian motion: the optimal policy is to stop at the first time the posterior exits an interval $(A^*,B^*)$ when the signal is strong…

desk verdict Interesting new soft-classification sequential testing problem with a largely self-contained free-boundary analysis, but the main verification theorem is delegated to references and should be filled in before the result is treated as fully established. read the letter →

arxiv 2501.11314 v1 pith:A22MT6HC submitted 2025-01-20 math.PR math.STstat.TH

Steven Campbell , Yuchong Zhang This is my paper · ORCID
classification math.PRmath.STstat.TH MSC 60G3560G4062L1062L15
keywords softclassificationsequentialtestingoptimalstoppingfilteringfree-boundaryproblemsBrownianmotiondriftinformationratio
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

In the classic Bayesian sequential test, an observer pays a running cost to watch a noisy Brownian signal and must eventually issue a hard yes/no verdict about the drift. This paper relaxes that verdict: the loss is a soft score $g(\pi)$ charged according to the posterior probability at the stopping time, with zero penalty only when the observer is certain. The paper establishes that the optimal policy is still simple: keep watching while the posterior $\Pi_t$ remains between two constants $A^*$ and $B^*$, stop at the first exit, and if the squared signal-to-noise ratio $K=(\alpha/\sigma)^2$ is too small, stop immediately with $V^*\equiv g$. The two thresholds are the unique solution of a free-boundary problem, and the value function is shown to be the largest convex minorant of a transformed penalty. A sympathetic reader cares because it gives the soft-classification analogue of the classical hard-classification solution with the same semi-explicit structure, plus a phase transition in the information ratio that the hard problem does not have.

What carries the argument

The central object is the free-boundary problem (7)–(13) together with the transformed penalty $H(\pi)=g(\pi)-2K^{-1}\Psi(\pi)$, where $\Psi(\pi)=(1-2\pi)\log(\pi/(1-\pi))$. Equation (7) fixes the curvature of the candidate value function inside the waiting region, equations (8)–(11) impose the value and smooth-fit conditions at the boundaries, and conditions (12)–(13) force the candidate to lie below the penalty inside the waiting region and equal it outside. The existence and uniqueness of the boundary pair is carried by a common-tangent construction: restricting $H$ to the two convex pieces on either side of its concave middle, equations (17)–(18) ask for a single line tangent to both pieces, and the classical theorem that two strictly separated convex bodies admit exactly two common tangents supplies the unique solution. The same structure yields the convex-envelope representation of the value function.

What would settle it

A concrete numerical check: for the L1 penalty $g(\pi)=2\pi(1-\pi)$ with running cost $c=1$ and squared signal-to-noise ratio $K=7$, the theorem says $V^*\equiv g$ and immediate stopping is optimal; a simulation-based approximation of (4) that finds any bounded stopping rule with expected cost below $g(\pi_0)$ at some starting prior would refute the verification theorem. At $K=9$ the same simulation should reproduce the expected cost of first exit from $(A^*,B^*)$ obtained from (17)–(18).

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that the value function $V^*(\pi)=\inf_\tau \mathbb{E}_\pi[c\tau+g(\Pi_\tau)]$ is the unique non-trivial $C^2((0,1)\setminus\{A^*,B^*\})\cap C^1(0,1)$ solution of the free-boundary problem (7)–(13) whenever the curvature condition $Ag(\pi_0)<-K^{-1}$ holds. The continuation region is exactly $(A^*,B^*)$, with $(A^*,B^*)$ the unique solution pair of the consistency equations (17)–(18) subject to $A^*\le \pi_*<\pi^*\le B^*$, and the smallest optimal stopping time is the first exit from that interval. When $Ag(\pi_0)\ge -K^{-1}$, no observation is worthwhile: $V^*\equiv g$. The solution also satisfies $V^*(\pi)-2K^{-1}\Psi(\pi)=\inf_\tau\mathbb{E}_\pi[H(\Pi_\tau)]$, i.e. it is the largest convex minorant of $H=g-2K^{-1}\Psi$ with $\Psi(\pi)=(1-2\pi)\log(\pi/(1-\pi))$.

Load-bearing premise

The load-bearing premise is the single-well curvature condition (G2): the weighted second derivative $Ag(\pi)=\frac12\pi^2(1-\pi)^2g''(\pi)$ must be strictly decreasing then strictly increasing with one minimum, because this is what forces the waiting region to be a single interval and makes the two-boundary solution unique.

Editorial extensions

If this is right

  • The optimal procedure is a two-threshold rule on the posterior: while $\Pi_t$ lies in $(A^*,B^*)$ the observer keeps sampling, and the smallest optimal stopping time is the first exit from that interval.
  • There is a sharp phase transition: when the squared signal-to-noise ratio $K$ is at or below the threshold $\beta^{-1}$, immediate stopping is optimal and no observation is worthwhile; for the L1 and cross-entropy penalties this threshold is $K=8$, whereas the hard-classification problem has a non-empty waiting region for every $K>0$.
  • As $K\to\infty$, the boundaries move to the endpoints: $A^*(K)\downarrow0$ and $B^*(K)\uparrow1$, with the rate bounds $A^*(K)\le1/(1+CK^{1-\epsilon})$ and $B^*(K)\ge CK^{1-\epsilon}/(1+CK^{1-\epsilon})$; as $K$ falls to $\beta^{-1}$, both boundaries collapse to the minimizer $\pi_0$.
  • The value function can be written as $2K^{-1}\Psi$ plus the convex envelope of $H$, so the whole optimal-stopping solution is encoded in a single convex-minorant computation; in symmetric penalties this reduces further to $B^*=1-A^*$.

Reading between the lines

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

  • An ordering left unproved by the paper is that the soft-classification boundaries cross the hard-classification boundaries in $K$: at high information ratios a soft classifier should observe longer than a hard classifier, and at low ratios it should stop sooner; the paper demonstrates this for the L1 and cross-entropy examples without proving a general ordering result.
  • The convex-envelope characterization suggests a computational route valid beyond the two-boundary case: compute the largest convex minorant of $H$ by a one-dimensional convex hull, then read off the value function and the contact set as the stopping region even if (G2) fails and the region disconnects.
  • If the information ratio is made endogenous or time-dependent, the constant thresholds would become moving boundaries and the phase transition at $K=\beta^{-1}$ would become a separating surface in state-time space; the free-boundary formulation in this paper is the natural starting point for such an extension.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper studies the Bayesian sequential soft-classification problem for the drift of a Brownian motion. The observer pays a linear observation cost and a terminal loss g(Πτ) depending on the posterior probability, where g is induced by a soft-classification loss and satisfies concavity and a unimodality condition (G1)-(G2). The main results are: (i) a free-boundary problem (7)-(13) whose nontrivial solution is unique when Ag(π0) < -K^{-1}, with boundaries A*, B* characterized by the two transcendental equations (17)-(18); (ii) a verification theorem (Theorem 3.1) identifying the value function with this solution; (iii) a convex-envelope representation V* - 2K^{-1}Ψ = convex envelope of H; and (iv) asymptotic and monotonicity results for the boundaries as the signal-to-noise ratio K varies. The paper also contains numerical illustrations comparing the soft-classification boundaries with the classical hard-classification boundaries.

Significance. If the verification step were fully supplied, this would be a clean and useful extension of the classical Wiener sequential testing problem to soft-classification losses. The free-boundary construction, the use of Bisztriczky's theorem to prove existence and uniqueness of the common tangent, the explicit equations for the boundaries, and the asymptotic analysis are genuine contributions. The paper has no fitted parameters, and the structural results are derived rather than calibrated. However, the central claim that the value function equals the solution of the free-boundary problem is currently not established in the manuscript, and all subsequent results depend on that identification.

major comments (3)
  1. [§3, Theorem 3.1] The proof of Theorem 3.1 is not a proof of the stated result: it consists of two citations and one cited estimate. [16, Theorem 21.1] is formulated for the hard-classification loss g(π)=a1π∧a2(1−π), which is not C² and has a kink; [13, Proposition 3.1] treats linear-cost problems under different structural assumptions. Neither establishes that the candidate V in (19) is superharmonic, that the stopping time τ_{A*,B*} is optimal among all F^Π-stopping times, or that the relevant stochastic integrals are true martingales for general g satisfying (G1)-(G2). Since Corollary 3.2, Theorem 3.3, and all of Section 4 depend on this identification, this is load-bearing. The authors should provide a self-contained verification under (G1)-(G2) or a rigorous reduction that verifies all hypotheses of the cited results.
  2. [§3, Theorem 3.3] The proof of the convex-envelope representation uses a reduction to 'regular' stopping times delegated to [13, Proposition 2.1] and applies Dynkin's formula to H, g, and Ψ without stating the integrability and boundedness conditions needed for the stochastic integral to be a true martingale. In particular, the estimate |π(1−π)g'(π)|≤M is merely quoted from [2, Remark 2.3]; for general concave g with g(0)=g(1)=0 this is not proved here. The claim that V*−2K^{-1}Ψ is convex also relies on the identification in Theorem 3.1. Please include the missing estimates and make the justification self-contained.
  3. [§4, Proposition 4.2(i)] The proof of Proposition 4.2(i) says that the derivative of A*(K) (resp. B*(K)) 'must be positive (resp. negative)', but the displayed formulas and the stated conclusion (A* decreasing, B* increasing) require the opposite signs: dA*/dK≤0 and dB*/dK≥0. The numerator signs cited appear consistent with the stated monotonicity, so this is likely a typo, but it should be corrected.
minor comments (5)
  1. [§2, Lemma 2.2] The proof of Lemma 2.2 contains an unexplained constant C1 and C0 and a sign change in the double integral; as written, the displayed equality after integrating the bound is not derived. The conclusion is correct, but the argument should be rewritten for clarity.
  2. [§2, Theorem 2.1] In the statement of Theorem 2.1, the notation 'π∗ < π∗' should presumably be 'π_* < π^*', and the constraints should read A* ≤ π_* < π^* ≤ B*. The current typography makes the statement difficult to parse.
  3. [§2, Proposition 2.6] The notation [π, π∗] and [π∗, π] is overloaded: the underline and overline on π are easy to confuse with the asterisks for π_* and π^*. Please use a clearer notation, such as π_l, π_r or π_- and π_+.
  4. [§5] The statements that the soft-classification boundaries are contained in the classical boundaries for small K and not for large K are based on numerical comparison; they should be labelled as numerical observations rather than proved results.
  5. [§1] The paper uses the symbol A both for the infinitesimal operator in (6) and for the lower stopping boundary. Although this is common, a different symbol for one of the two would improve readability.

Circularity Check

1 steps flagged · score 2.0 of 10

No significant circularity: the free-boundary analysis is derived from scratch, but Theorem 3.1's verification leans on a minor same-author citation for a technical estimate.

  1. self citation load bearing [Section 3, Theorem 3.1 (Verification), proof paragraph]
    "Up to showing the martingale property of a stochastic integral, all verification arguments are standard. To this end, Remark 2.3 of [2] gives the necessary estimate |π(1 − π)g′(π)| ≤M for some M >0."

    The only self-citation used in the proof is a boundedness estimate imported from the authors' companion paper [2]. It is a technical supporting ingredient for the martingale property, not the value-function identity itself. The free-boundary construction and boundary equations (17)-(18) are derived internally in Section 2, so the central derivation is not equivalent to its inputs by construction. The score of 2 reflects this minor same-author dependency in the verification step, not a fitted parameter or an equation-for-equation reduction.

full rationale

The paper's main free-boundary result, Theorem 2.1, is proved self-contained: Lemma 2.3 establishes the shape of H′, Corollary 2.4 gives the convex/concave regions, Lemma 2.5 uses Bisztriczky's Theorem as an external geometric fact, and Proposition 2.6 yields the unique boundary pair (A∗, B∗) solving (17)-(18). No parameter is fitted to data, and no prediction is the renaming of an input. The derivation of (14)-(18) from the free-boundary problem is by explicit calculation. Theorem 3.3, the convex-envelope characterization, is proved with internal inequalities, Dynkin's formula, and an external reduction to regular stopping times from [13]. The only concern is Theorem 3.1: the identification of the value function with the free-boundary solution is delegated by analogy to [13, Proposition 3.1] and [16, Theorem 21.1], neither of which is shown to cover the present soft-classification losses, and the martingale estimate is cited from the authors' companion work [2]. This is a verification gap and a minor self-citation, but it is not circularity because the cited estimate is not the target claim and the free-boundary analysis has independent content. Honest non-finding on circularity; score 2.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the standard filtering SDE for the posterior, Assumption 1.1 (G1)-(G2) on the penalty function, Bisztriczky's theorem for the common tangent construction, and a boundedness estimate from the authors' companion paper. No parameters are fitted to data; the model parameters α, σ, c are inputs, not free parameters.

assumptions (6)
  • domain assumption The posterior probability process Π follows dΠ_t = (α/σ) Π_t(1-Π_t) dW_t with innovation Brownian motion W̄.
    Standard filtering result (c.f. [16, (21.0.9)-(21.0.10)]) used in Section 1 to reduce the problem to Markovian optimal stopping.
  • domain assumption Assumption 1.1(G1): g ∈ C²(0,1) is concave with g(0)=g(1)=0.
    Restricts the loss functions under consideration. Used throughout to ensure the free boundary problem is well-posed and the convex envelope result holds.
  • domain assumption Assumption 1.1(G2): there exists π0 such that Ag is strictly decreasing on (0,π0) and strictly increasing on (π0,1).
    This is the load-bearing technical condition ensuring the continuation region is a single interval (A,B) and the free boundary problem has a unique non-trivial solution. If it fails, the two-boundary characterization breaks down (Section 6(ii)).
  • standard math Bisztriczky's Theorem in 2D: two strictly separated convex bodies have exactly two common supporting lines with both bodies on the same side.
    Used in Lemma 2.5 to prove existence and uniqueness of a common tangent to the two convex functions f0, f1, which is the core of Proposition 2.6.
  • domain assumption The boundedness estimate |π(1-π)g'(π)| ≤ M for some M > 0, cited from [2, Remark 2.3].
    Used in the proof of Theorem 3.1 to establish the martingale property of a stochastic integral in the verification argument. Not proved in this paper; drawn from the authors' companion paper.
  • standard math It suffices to minimize over regular stopping times bounded by the first exit from compact intervals ([13, Proposition 2.1]).
    Used in Theorem 3.3 to justify replacing the infimum over all stopping times with regular ones.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Bayesian sequential soft classification problem for a Brownian motion's drift." pith.science (2026). https://pith.science/paper/A22MT6HC

@misc{pith2026250111314,
  author       = {Pith},
  title        = {Pith review of: A Bayesian sequential soft classification problem for a Brownian motion's drift},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A22MT6HC}},
  note         = {Machine review of arXiv:2501.11314}
}
read the original abstract

In this note we introduce and solve a soft classification version of the famous Bayesian sequential testing problem for a Brownian motion's drift. We establish that the value function is the unique non-trivial solution to a free boundary problem, and that the continuation region is characterized by two boundaries which may coincide if the observed signal is not strong enough. By exploiting the solution structure we are able to characterize the functional dependence of the stopping boundaries on the signal-to-noise ratio. We illustrate this relationship and compare our stopping boundaries to those derived in the classical setting.

Figures

Figures reproduced from arXiv: 2501.11314 by the authors.

Figure 1
Figure 1. Visualization of the critical points of H′ for the cross-entropy loss (left). Illus￾tration of the arguments of Lemma 2.5 using an example of H which defines f0, f1 (right). Visualized below the graph of H is the unique common tangent. convex hull. These sets correspond to the intersection of the epigraph of fX (resp. fY ) with the half-infinite strip in X (resp. Y ) ending at the horizontal line y = f ∗ (see [PITH… view at source ↗
Figure 2
Figure 2. Penalties g induced by loss functions (left) and the associated optimal stopping boundaries as functions of K (right) [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗

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. Embedding martingale diffusions as binary posteriors in sequential inference

    math.PR 2026-07 accept novelty 7.0 of 10

    Every autonomous win-martingale on [0,1] is the exact posterior of an explicit binary diffusion experiment, and current-state posteriors exist exactly under a Riccati condition on the drift gap.

Reference graph

Works this paper leans on

24 extracted references · 23 canonical work pages · cited by 1 Pith paper

  1. [13]

    Irle and V

    A. Irle and V. Paulsen. Solving problems of optimal stopping with linear costs of observations. Sequential Analysis, 23(3):297–316, 2004

    work page 2004

  2. [16]

    Peskir and A

    G. Peskir and A. N. Shiryaev. Optimal stopping and free-boundary problems. Springer, 2006

    work page 2006

  3. [2]

    Campbell and Y

    S. Campbell and Y. Zhang. A mean field game of sequential testing. arXiv preprint arXiv:2403.18297, 2024

    arXiv 2024

  4. [1]

    Bisztriczky

    T. Bisztriczky. On separated families of convex bodies. Archiv der Mathematik, 54(2):193–199, 1990

    work page 1990

  5. [3]

    S. E. Cappell, J. E. Goodman, J. Pach, R. Pollack, and M. Sharir. Common tangents and common transversals. Advances in Mathematics, 106(2):198–215, 1994

    work page 1994

  6. [4]

    Common tangents to convex bodies

    F. Castillo, J. Doolittle, and J. A. Samper. Common tangents to convex bodies. arXiv preprint arXiv:2108.13569, 2021

    work page Pith review arXiv 2021

  7. [5]

    Chernoff

    H. Chernoff. Sequential design of experiments. The Annals of Mathematical Statistics, 30(3):755–770, 1959

    work page 1959

  8. [6]

    H. F. Dodge and H. G. Romig. A method of sampling inspection. The Bell System Technical Journal, 8(4):613–631, 1929

    work page 1929

Show all 24 references
  1. [7]

    Dvoretzky, J

    A. Dvoretzky, J. Kiefer, and J. Wolfowitz. Sequential decision problems for processes with continuous time parameter. testing hypotheses. The Annals of Mathematical Statistics, 24(2):254–264, 1953

    1953

  2. [8]

    Ekstr¨ om and I

    E. Ekstr¨ om and I. Karatzas. A sequential estimation problem with control and discretionary stopping. Probability, Uncer- tainty and Quantitative Risk, 7(3), 2022

    2022

  3. [9]

    Ekstr¨ om, I

    E. Ekstr¨ om, I. Karatzas, and J. Vaicenavicius. Bayesian sequential least-squares estimation for the drift of a Wiener process. Stochastic Processes and their Applications, 145:335–352, 2022

    2022

  4. [10]

    Ekstr¨ om and J

    E. Ekstr¨ om and J. Vaicenavicius. Bayesian sequential testing of the drift of a brownian motion. ESAIM: Probability and Statistics, 19:626–648, 2015

    2015

  5. [11]

    P. V. Gapeev and G. Peskir. The Wiener sequential testing problem with finite horizon. Stochastics and stochastic reports, 76(1):59–75, 2004

    2004

  6. [12]

    P. V. Gapeev and A. N. Shiryaev. On the sequential testing problem for some diffusion processes. Stochastics An Interna- tional Journal of Probability and Stochastic Processes, 83(4-6):519–535, 2011

    2011

  7. [14]

    Kiefer and L

    J. Kiefer and L. Weiss. Some properties of generalized sequential probability ratio tests. The Annals of Mathematical Statistics, pages 57–74, 1957

    1957

  8. [15]

    Lewis, B

    T. Lewis, B. von Hohenbalken, and V. Klee. Common supports as fixed points. Geometriae Dedicata, 60:277–281, 1996

    1996

  9. [17]

    A. N. Shiryaev. On optimum methods in quickest detection problems. Theory of Probability & Its Applications, 8(1):22–46, 1963

    1963

  10. [18]

    disorder

    A. N. Shiryaev. Some exact formulas in a “disorder” problem. Theory of Probability & Its Applications, 10(2):348–354, 1965

    1965

  11. [19]

    A. N. Shiryaev. Two problems of sequential analysis. Cybernetics, 3(2):63–69, 1967

    1967

  12. [20]

    A. N. Shiryaev. Optimal Stopping Rules. Springer Science & Business Media, 2007

    2007

  13. [21]

    A. Wald. Sequential tests of statistical hypotheses. The Annals of Mathematical Statistics, 16(2):117–186, 1945

    1945

  14. [22]

    Wald and J

    A. Wald and J. Wolfowitz. Optimum character of the sequential probability ratio test. The Annals of Mathematical Statistics, pages 326–339, 1948

    1948

  15. [23]

    Wald and J

    A. Wald and J. Wolfowitz. Bayes solutions of sequential decision problems. The Annals of Mathematical Statistics, pages 82–99, 1950

    1950

  16. [24]

    R. Xu, T. Zariphopoulou, and L. Zhang. Decision making under costly sequential information acquisition: the paradigm of reversible and irreversible decisions. arXiv preprint arXiv:2401.00569, 2023

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.