Pith. sign in

REVIEW 2 major objections 4 minor 2 cited by

Discrete turn strategies emerge in information-limited navigation

T0 review · 2 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read This paper proves that without directional information, a memoryless navigator climbs a gradient fastest by using a discrete set of turn angles, with reversals giving way to flicks and more angles as information increases.

desk verdict Core idea is right and the scaling laws are solid, but the discreteness theorem is only proven for analytic strategies, which the abstract overstates. read the letter →

arxiv 2602.23324 v2 pith:2IUGYVZA submitted 2026-02-26 physics.bio-ph cond-mat.stat-mechq-bio.QM

Jose M. Betancourt , Matthew P. Leighton , Thierry Emonet , Benjamin B. Machta , Michael C. Abbott This is my paper · ORCID
classification physics.bio-phcond-mat.stat-mechq-bio.QM
keywords navigationchemotaxisinformation-limitedcontroldiscreteturnanglesrun-and-tumblereversalstrategiesFokker-Planckequationcontactfunction
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 asks how a navigator that cannot tell which way is up should climb a chemical gradient when every bit of sensory information costs something. It establishes that smooth steering is always beaten by sudden turns, and that when arbitrary turn angles are allowed the optimal strategy uses a discrete set of angles rather than a continuum. The proof relies on a contact function that can equal its maximum value only at isolated points; a negative second derivative at zero rules out a plateau. As the information rate rises, the optimal discrete set transitions from direction reversals to adding right-angle flicks and then more angles. The authors also see the same discreteness in three dimensions and point to run-reverse-flick behaviors in microorganisms as a qualitative match.

What carries the argument

The contact function Ψ(Δθ) = ∫ dθ p(θ) exp([χ(θ+Δθ)+χ(θ−Δθ)−2χ(θ)]/2γ), defined from the Lagrange multiplier χ enforcing the Fokker-Planck constraint. Where Ψ(Δθ)=1, the optimal strategy can put turn weight; the equations of motion imply Ψ≤1 everywhere and Ψ=1 on the support. Analyticity of Ψ plus the computed second derivative at 0 forces the set of contact points to be finite, which is exactly the discreteness of optimal turn angles.

What would settle it

Numerically optimize the information-constrained objective in a space of non-analytic strategies (e.g., fine-grained piecewise-constant or neural-network parameterized λ(Δθ, θ)) at intermediate γ; if any solution with continuous support achieves a strictly larger objective than the discrete-support optimum, discreteness does not hold outside the analytic class. Alternatively, experimental turn-angle histograms from a navigating microorganism across a range of gradient steepnesses could show whether the predicted discrete, bifurcating set of angles is realized.

Watch

Extended reading notes

Core claim

The central claim is a statement about optimal information-limited navigation. For an agent moving at fixed speed on a circle of headings, with rotational diffusion and a control law that maps heading to turn rate, the objective is mean up-gradient speed minus γ times the mutual information rate between heading and action. Among strategies whose turn-rate kernel is symmetric in Δθ (so left and right turns are treated equally, which corresponds to not knowing the sign of uphill), the optimizer of this objective is a turn-rate kernel λ(Δθ, θ) whose support—the set of turn angles actually used—is finite. The proof introduces a contact function Ψ(Δθ), an overlap of the Lagrange multiplier χ that

Load-bearing premise

The proof that the optimal turn-angle support is discrete requires the turn-rate strategy λ(Δθ, θ) to be analytic; if a non-analytic optimum existed, the contact function could remain at its maximum over a whole interval and the finite-support conclusion would fail.

Editorial extensions

If this is right

  • Without directional information, continuous steering is never optimal: sudden actions such as reversals and tumbles achieve a given up-gradient speed at lower information cost, and in the low-information limit reversals are twice as efficient as tumbles (v/v0 ~ sqrt(i/2Dr) versus sqrt(i/4Dr)).
  • The optimal turn-angle distribution is discrete, not continuous: the set of turn angles with positive rate is finite, and the number of angles grows through successive bifurcations as the information rate increases.
  • The optimal strategy passes through well-defined transitions: reversals at low information, added right-angle flicks at intermediate information, and eventually fully re-orienting tumbles at high information.
  • In three dimensions the qualitative picture repeats: reverse strategies win at low information, tumble and flick strategies overtake at high information, and the optimal turn kernels remain discrete.

Reading between the lines

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

  • The discreteness result is proven only within analytic strategies; if future work removes that assumption, the conclusion that discrete behavioral repertoires are inevitable under information constraints would be substantially stronger.
  • The contact-function argument parallels support-point proofs in optimal channel-input design, so the same analyticity-based discreteness may recur in other information-limited decision problems, not just navigation.
  • A direct experimental test would vary gradient steepness (the effective information rate) for an organism with run-reverse-flick behavior and record turn-angle histograms; the model predicts the appearance of a second pair of angles at a threshold and a discrete, not continuous, distribution.
  • The paper compares directional and non-directional information at equal bit rates, but real sensors often supply both with different noise; modeling a sensor that delivers both kinds of signal simultaneously is a natural extension, and the optimal-strategy phase diagram for that case remains open.
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

2 major / 4 minor

Summary. This paper studies a rate-information-constrained navigation problem: an agent maximizes mean up-gradient speed minus a trade-off penalty on the information rate from heading to action, with no directional (sign) information. In 2D and 3D, the authors compare continuous steering, tumbles, reversals, flicks, and fully general turn-angle distributions. The main analytic results are low-information Pareto scaling laws (v∼√i in 2D; v∼i for unsigned 3D steering), an exact Mathieu-function solution for signed steering, and a theorem in Appendix D claiming that the optimal target turn-angle distribution has finite support, leading to a sequence of optimal-strategy bifurcations from reversal to flick to multi-angle strategies.

Significance. The framework is clean and the analytic results are genuinely useful: the low-information scaling laws quantify a √2 speed advantage of reversal over tumbling, the exact Mathieu solution provides a benchmark for steering, and the connection to rational inattention is apt. If the discreteness result can be made rigorous for the full admissible strategy class, it would be a substantial conceptual explanation for stereotyped discrete behaviors in biology. The paper is also commendably explicit about the distinction between directional and non-directional information and about limitations such as memoryless control, instantaneous turns, and a single objective.

major comments (2)
  1. [Appendix D.4] The discreteness theorem assumes analytic λ(θ): 'For solutions with analytic λ (which we implicitly constrain our optimization to)', but the optimization problem (D2) only imposes λ≥0 and integrability; no analyticity constraint is stated. Without analyticity, the contact function Ψ could equal 1 on an interval, so the support of the optimal target measure need not be finite. The numerical solutions in Fig. 3 do not sample non-analytic strategies, and the ansatz λ_strong(θ) used elsewhere is piecewise constant. Thus the abstract's unconditional claim that discrete turn angles are best is established only for a restricted strategy class. Please either extend the proof to non-analytic strategies or state the theorem conditionally and soften the abstract/main-text claims. A concrete test is to add a non-analytic perturbation supported on an interval and check the first-order conditions and
  2. [Discrete angles from arbitrary turns; Appendix D] The stated biological setup is absence of directional information, which earlier in the paper is defined as evenness in θ (λ(θ)=λ(|θ|), μ(θ)=μ(|θ|)). The constraint actually imposed in the augmented problem is symmetry in the turn angle, λ(Δθ,θ)=λ(−Δθ,θ). This does not imply sign-blindness: a strategy such as λ(Δθ,θ)=δ(Δθ−π)(1+θ) is even in Δθ but uses the sign of θ. If the numerical optimization and the proof of Appendix D do not enforce θ-evenness, the computed 'sign-blind' frontier may include strategies with directional information, and the comparison with unsigned steering in Fig. 1 is not on the claimed footing. Please specify the exact symmetry group enforced and either prove or explicitly impose θ-evenness.
minor comments (4)
  1. [Appendix D.4, Eq. (D29)-(D32); main text Eq. (7)] There is a factor-of-γ inconsistency in the second derivative of the contact function. Eq. (D29) gives Ψ''(0)=γ^{-1}∫pχ'', while Eq. (D32) and the main text report Ψ''(0)=−γ i/D_r. Combining (D29) with the result of (D31) gives Ψ''(0)=−i/D_r. The sign, which is all that is used, is unaffected, but the displayed formula should be corrected.
  2. [Eq. (B11) vs Eq. (B21)] The drift expression in Eq. (B11) contains λ^{(1)}_1−λ^{(1)}_{−1}, whereas the same optimization step in Eq. (B21) uses λ^{(1)}_1+λ^{(1)}_{−1}. The sum is consistent with the even solution λ^{(1)}=−√2 cosθ and with the subsequent derivation; the minus sign in (B11) appears to be a typo.
  3. [Appendix C.1] The sentence 'which is the desired formula, the first term of (3)' should refer to Eq. (4), the steering information rate, not Eq. (3), which is the tumbling information rate.
  4. [Appendix D.6] The notation Q^s_θ := 1/2(Q_θ + Q_{−θ}) is ambiguous: the text says Q_{−θ} is obtained by the transformation Δθ→−Δθ, but the subscript suggests a heading reflection. Please define the reflection operation explicitly, as this bears directly on the symmetry constraint discussed above.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the optimality results are derived from the stated optimization problem, not from fitted data or load-bearing self-citation; the main caveat is an explicit analyticity restriction in the discreteness proof.

full rationale

The paper's central claim—that information-limited, sign-blind navigation is optimized by discrete turn angles—is derived from the stated variational problem (maximize v/v0 − γ i under the Fokker–Planck constraint and symmetry), not by fitting parameters to data or by citing a conclusion already containing the result. The discreteness proof in Appendix D.4 derives contact conditions (D17–D18), then proves Ψ is analytic and nonconstant, hence Ψ(Δθ)=1 only at isolated points. This does not reduce to its inputs by construction: analyticity plus nonconstancy is a genuine argument, not a definitional equivalence. The most important caveat is explicitly stated in the text: “For solutions with analytic λ (which we implicitly constrain our optimization to), this implies that χ is analytic.” This is a real scope restriction—without analyticity, the contact function could plateau at 1 on a continuum and the support need not be finite—so the abstract's global claim is only proven within an implicitly restricted class. But this is a proof gap or overstatement, not circularity: the optimization problem (D2) itself does not enforce analyticity, and the paper does not define discreteness into the objective. Other potential circularity patterns are absent. The low-information optimality of reversal follows from the elementary bound q1 ≥ −1/2π (Appendix B.1), not from an assumed answer. The λ_strong(θ) strategy is clearly labeled as “a reasonable approximation” and is used only for some numerical points, not fitted and then called a prediction. Self-citations ([4], [9], [10], [12]) are contextual or technical parallels; the discreteness argument is reproduced in the appendix rather than imported as an external uniqueness theorem. No cited prior work is load-bearing for the main derivation. Because the derivation is self-contained and the identified issue is a limitation rather than a circular reduction, the appropriate score is low. I assign 1 rather than 0 to flag the explicit analyticity restriction and the presence of self-citations, but no circular step is present.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central results rest on a well-posed optimization problem with a small number of modeling choices. The most important burden is the analyticity assumption in the discreteness proof; the other assumptions (memoryless, symmetric, information-rate cost) are explicit modeling choices, not hidden free parameters.

free parameters (2)
  • γ (trade-off parameter) = swept across frontier
    Lagrange multiplier in objective (1) that traces the speed–information Pareto frontier; not fitted to data but a free parameter of the optimization.
  • λ_strong threshold θ_thresh and λ_max = not specified analytically; chosen for ansatz
    The high-information approximate strategy (5) used for square points in Fig. 1 contains free threshold and rate; these are picked as a reasonable approximation rather than derived from optimality.
assumptions (5)
  • standard math Stochastic dynamics of heading are described by a Fokker–Planck/master equation with rotational diffusion Dr.
    Used throughout to compute steady-state distributions and velocities; standard mathematical background.
  • domain assumption The resource constraint is the mutual information rate between heading and controlled change of heading, as defined in Eqs. (3) and (6).
    This abstraction equates information cost with the rate of information transmission from sensor to behavior; it is the core modeling choice that makes strategies comparable.
  • domain assumption Without directional information, the strategy is symmetric under left–right reversal: λ(Δθ,θ)=λ(-Δθ,θ) (and µ(|θ|) even).
    Imposed to model agents that can only sense the rate of up-gradient motion, not the sign; the discreteness result is derived under this symmetry.
  • ad hoc to paper The optimal strategy λ(Δθ,θ) is assumed analytic in its arguments.
    Appendix D.4: 'For solutions with analytic λ (which we implicitly constrain our optimization to)...' — the discreteness proof uses analyticity to conclude the contact function is nonconstant analytic, so the contact set is finite.
  • domain assumption The agent has no memory and strategies are stationary Markovian in heading θ.
    The model restricts to memoryless strategies; the conclusion notes that real organisms use memory, which is left to future work.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Discrete turn strategies emerge in information-limited navigation." pith.science (2026). https://pith.science/paper/2IUGYVZA

@misc{pith2026260223324,
  author       = {Pith},
  title        = {Pith review of: Discrete turn strategies emerge in information-limited navigation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2IUGYVZA}},
  note         = {Machine review of arXiv:2602.23324}
}
read the original abstract

Navigation up a smooth sensory gradient is one of the simplest behavioural tasks, and some organisms solve it by making continuous adjustments to their course. Bacteria instead employ a variety of discrete strategies, including run and tumble motion, direction reversals, and turns by specific angles. Here we ask what drives the choice of these strategies, framing the problem as maximising up-gradient speed with a given amount of sensory information per unit time. We find that, without directional information on which way to turn, behavioural strategies that take discrete actions perform better than gradual steering. As the amount of information is increased, we see a series of transitions between optimal strategies, including a shift from direction reversals to fully re-orienting tumbles. Among more complex re-orientation strategies, we show that discrete turn angles are best, and observe transitions in the number of angles employed by the optimal strategy. More broadly, such emergent simplicity in behaviour is a tractable example of a widespread phenomenon in which biology chooses a discrete solution, despite the underlying physics being continuous.

Figures

Figures reproduced from arXiv: 2602.23324 by the authors.

Figure 1
Figure 1. Performance of strategies for two-dimensional nav￾igation, on the speed-information plane. The continuous steer￾ing strategy achieves the highest performance, provided the sign of heading θ is visible (red). Among strategies not us￾ing the sign, fully re-orienting instantaneous tumbles (blue) perform well at high information rates, but at lower rates re￾versing (green) needs half as much information for the same spe… view at source ↗
Figure 2
Figure 2. Numerical optimal strategies for run & tumble, steering, and reversing. (A) Tumble rate λ(θ) in blue, and steady￾state √ p(θ) in grey, for a low-information case, and a high-information case using λstrong(θ). (B) The reverse strategy achieves 2 times the speed at similar information rate i/Dr ≈ 0.01, by acting at half the rate of the tumble strategy. But its high￾information case saturates at speed 2/π, when p(θ) is… view at source ↗
Figure 3
Figure 3. Discrete optimal strategies for three information rates. We impose that rate λ(∆θ, θ) is even in ∆θ, which implies that the strategy ignores the sign of θ. At low information rate (left) we recover the reverse strategy, ∆θ = π, but with increasing information it bifurcates to use three angles (centre), and then five (right). Alongside the rate, we plot the contact function Ψ(∆θ) and a mean distribution q¯(∆θ) ∝ λ¯(∆… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Navigation in three dimensions. (A) Performance of tumble and reverse are qualitatively similar to figure 1, with different prefactors in the v ∝ √ i scaling at low information rates. Flick produces similar performance to tumble. The red steering points are a strategy …

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. On the Information Required for Feedback Control

    cond-mat.stat-mech 2026-07 accept novelty 7.0 of 10

    Feedback control that maintains a target steady state requires an information rate at least equal to the passive entropy rate of the uncontrolled dynamics, and a time-reversal protocol attains this rate for state-inde...

  2. A behavior-environment information loop drives sensory navigation

    physics.bio-ph 2026-07 conditional novelty 6.0 of 10

    Navigation performance is predicted by the geometric mean of reactive and active transfer-entropy rates between sensing and action.

Reference graph

Works this paper leans on

34 extracted references · 4 linked inside Pith · cited by 2 Pith papers

  1. [1]

    H. C. Berg,How bacteria swim, Sci. Am.233(1975) 36–44

    1975

  2. [2]

    B. L. Taylor and D. E. Koshland,Reversal of flagellar rotation in monotrichous and peritrichous bacteria: generation of changes in direction, J Bacteriol119 (1974) 640–642

    1974

  3. [3]

    L. Xie, T. Altindal, S. Chattopadhyay and X.-L. Wu, Bacterial flagellum as a propeller and as a rudder for efficient chemotaxis, PNAS108(2011) 2246–2251

    2011

  4. [4]

    H. H. Mattingly, K. Kamino, B. B. Machta and T. Emonet,Escherichia coli chemotaxis is information limited, Nat. Phys.17(2021) 1426–1431 [arXiv:2102.11732]

    arXiv 2021

  5. [5]

    A. J. Waite, N. W. Frankel, Y. S. Dufour, J. F. Johnston, J. Long and T. Emonet,Non-genetic diversity modulates population performance, Molecular Systems Biology12(2016) 895

    2016

  6. [6]

    S. P. Strong, B. Freedman, W. Bialek and R. Koberle, Adaptation and optimal chemotactic strategy for E. coli, Phys. Rev. E57(1998) 4604–4617

    1998

  7. [7]

    J. G. Smith,The information capacity of amplitude- and variance-constrained sclar Gaussian channels, Information and Control18(1971) 203–219

    1971

  8. [8]

    Hillar and S

    C. Hillar and S. Marzen,Neural network coding of natural images with applications to pure mathematics, in Contemporary Mathematics(H. Harrington, M. Omar and M. Wright, eds.), vol. 685, pp. 189–221. American Mathematical Society, Providence, Rhode Island, 2017

    2017

Show all 34 references
  1. [9]

    H. H. Mattingly, M. K. Transtrum, M. C. Abbott and B. B. Machta,Maximizing the information learned from finite data selects a simple model, PNAS115(2018) 1760–1765

    2018

  2. [10]

    M. C. Abbott and B. B. Machta,A scaling law from discrete to continuous solutions of channel capacity problems in the low-noise limit, J. Stat. Phys.176 (2019) 214–227

    2019

  3. [11]

    Barletta, A

    L. Barletta, A. Dytso and S. Shamai,Improved bounds on the number of support points of the capacity-achieving input for amplitude constrained Poisson channels, arXiv:2401.05045

    arXiv

  4. [12]

    J. Jung, J. H. J. Kim, F. Matějka and C. A. Sims, Discrete actions in information-constrained decision problems, Rev. Econ. Stud.86(2019) 2643–2667

    2019

  5. [13]

    C. A. Sims,Rational inattention: Beyond the linear-quadratic case, Am. Econ. Rev.96(2006) 158–163

    2006

  6. [14]

    A. P. Nikitin, N. G. Stocks, R. P. Morse and M. D. McDonnell,Neural population coding is optimized by discrete tuning curves, Phys. Rev. Lett.103(2009) 138101

    2009

  7. [15]

    D. B. Kastner, S. A. Baccus and T. O. Sharpee,Critical and maximally informative encoding between neural populations in the retina, PNAS112(2015) 2533–2538

    2015

  8. [16]

    T. O. Sharpee,Optimizing neural information capacity through discretization, Neuron94(2017) 954–960

    2017

  9. [17]

    S. Shao, M. Meister and J. Gjorgjieva,Efficient population coding of sensory stimuli, Phys. Rev. Research5(2023) 043205

    2023

  10. [18]

    Tkacik, C

    G. Tkacik, C. G. Callan and W. Bialek,Information flow and optimization in transcriptional regulation, PNAS105(2008) 12265–12270 [arXiv:0705.0313]

    2008 arXiv

  11. [19]

    Witteveen, S

    O. Witteveen, S. J. Rosen, R. S. Lach, M. Z. Wilson and M. Bauer,Optimizing information transmission in optogenetic Wnt signaling, Phys Rev Res. (2026) [arXiv:2506.22633]

    2026

  12. [20]

    Grognot and K

    M. Grognot and K. M. Taute,More than propellers: how flagella shape bacterial motility behaviors, Current Opinion in Microbiology61(2021) 73–81

    2021

  13. [21]

    Grognot, A

    M. Grognot, A. Mittal, M. Mah’moud and K. M. Taute, Vibrio cholerae motility in aquatic and mucus-mimicking environments, Appl. Environ. Microbiol.87(2021) e01293–21

    2021

  14. [22]

    Grognot, J

    M. Grognot, J. W. Nam, L. E. Elson and K. M. Taute, Physiological adaptation in flagellar architecture improves Vibrio alginolyticus chemotaxis in complex environments, PNAS120(2023) e2301873120

    2023

  15. [23]

    K. Son, F. Menolascina and R. Stocker,Speed-dependent chemotactic precision in marine bacteria, PNAS113 (2016) 8624–8629

    2016

  16. [24]

    K. S. Chen, A. K. Sharma, J. W. Pillow and A. M. Leifer,Navigation strategies in Caenorhabditis elegans are differentially altered by learning, PLoS Biol23 (2025) e3003005

    2025

  17. [25]

    Demir, N

    M. Demir, N. Kadakia, H. D. Anderson, D. A. Clark and T. Emonet,Walking Drosophila navigate complex plumes using stochastic decisions biased by the timing of odor encounters, eLife9(2020)

    2020

  18. [26]

    D. A. Clark and L. C. Grant,The bacterial chemotactic response reflects a compromise between transient and steady-state behavior, PNAS102(2005) 9150–9155

    2005

  19. [27]

    Celani and M

    A. Celani and M. Vergassola,Bacterial strategies for chemotaxis response, PNAS107(2010) 1391–1396

    2010

  20. [28]

    K. Son, J. S. Guasto and R. Stocker,Bacteria can exploit a flagellar buckling instability to change direction, Nature physics9(2013), no. 8 494–498

    2013

  21. [29]

    Battista, F

    A. Battista, F. Frischknecht and U. S. Schwarz, Geometrical model for malaria parasite migration in structured environments, Phys. Rev. E90(2014) 042720

    2014

  22. [30]

    Iino and K

    Y. Iino and K. Yoshida,Parallel use of two behavioral mechanisms for chemotaxis in Caenorhabditis elegans, J. Neurosci.29(2009) 5370–5380

    2009

  23. [31]

    L. Vo, F. Avgidis, H. H. Mattingly, K. Edmonds, I. Burger, R. Balasubramanian, T. S. Shimizu, B. I. Kazmierczak and T. Emonet,Nongenetic adaptation by collective migration, PNAS122(2025) e2423774122

    2025

  24. [32]

    G. B. Arfken and H. J. Weber,Mathematical methods for physicists 6th ed.Elsevier, 2005. 7

    2005

  25. [33]

    Hradil, J

    Z. Hradil, J. Řeháček, Z. Bouchal, R. Čelechovsk` y and LL. Sánchez-Soto,Minimum uncertainty measurements of angle and angular momentum, Phys. Rev. Lett.97 (2006) 243601 [arXiv:quant-ph/0605137]

    2006 arXiv

  26. [34]

    1D noise

    J. Řeháček, Z. Bouchal, R. Čelechovsk` y, Z. Hradil and LL. Sánchez-Soto,Experimental test of uncertainty relations for quantum mechanics on a circle, Phys. Rev. A77(2008) 032110 [arXiv:0712.0230]. 8 Figure S1:Performance of strategies for two-dimensional navigation, relative ...

    2008 arXiv

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.