Geodesic switches and exceptional times in dynamical Brownian last passage percolation

Basic Information

• Paper ID: 2510.27589
• Title: Geodesic switches and exceptional times in dynamical Brownian last passage percolation
• Author: Manan Bhatia (MIT)
• Classification: math.PR (Probability Theory)
• Submission Date: October 31, 2025
• Paper Link: https://arxiv.org/abs/2510.27589

Abstract

This paper studies Brownian last passage percolation (BLPP) evolving dynamically through discrete resampling processes. The author proves that the expected total number of coarse-grained "switches" accumulated by geodesics from (0,0) to (n,n) during the time interval s,t is at most $n^{5/3+o(1)}(t-s)$, and conjectures that the exponent 5/3 is tight. Using this estimate, the author proves that the Hausdorff dimension of the exceptional time set $\mathscr{T}$ where nontrivial bi-infinite geodesics exist is at most 1/2. Furthermore, for any fixed direction $\theta$, the Hausdorff dimension of the exceptional time set $\mathscr{T}^\theta$ where nontrivial bi-infinite geodesics in direction $\theta$ exist is almost surely 0.

Research Background and Motivation

Problem Background

1. Static vs. Dynamic LPP: The static model of exponential last passage percolation (LPP) has been extensively studied, but understanding of its dynamic evolution version remains limited. In the dynamic version, each vertex has an independent clock attached, and when the clock rings, the weight at that vertex is resampled.
2. Noise Sensitivity: Geodesics are expected to be noise-sensitive, meaning microscopic noise causes macroscopic changes in geodesic structure. Previous research (e.g., GH24) shows that geodesic overlap undergoes a phase transition at the critical time scale $n^{-1/3}$.
3. Bi-infinite Geodesics (Bigeodesics): In static LPP models, nontrivial bigeodesics are conjectured to almost surely not exist, which has been verified for exponential LPP and BLPP. However, in dynamic models, bigeodesics may exist at certain "exceptional times."

Research Motivation

1. Quantitative Analysis of Geodesic Switches: Existing research primarily compares geodesic overlaps at fixed times. This paper is the first to systematically analyze the sample path behavior of geodesics during time evolution, quantifying the number of rapid oscillations.
2. Dimension Bounds on Exceptional Times: Accompanying paper Bha25 proves "near-existence" of exceptional times in exponential LPP. This paper approaches from the opposite direction, providing dimension upper bounds on the exceptional time set in BLPP, demonstrating its "rarity."
3. Technical Challenges: The local behavior of routed weight profiles in exponential LPP is insufficiently understood. Therefore, this paper focuses on BLPP, leveraging its Brownian Gibbs property to enable comparison with Brownian motion.

Core Contributions

1. Quantitative Estimate of Geodesic Switches (Theorem 1): Proves that the expected number of coarse-grained switches accumulated by geodesics from (0,0) to (n,n) in the time interval s,t, away from endpoints, is at most $n^{5/3+\epsilon}(t-s)$, where the exponent 5/3 is believed to be optimal.
2. Dimension Upper Bound on Exceptional Times (Theorem 3): Establishes that the Hausdorff dimension of the exceptional time set $\mathscr{T}$ where nontrivial bigeodesics exist is at most 1/2.
3. Exceptional Times for Directional Bigeodesics (Theorem 4): For any fixed direction $\theta$, the Hausdorff dimension of the exceptional time set $\mathscr{T}^\theta$ where nontrivial bigeodesics in direction $\theta$ exist is almost surely 0.
4. Hitset Estimate (Theorem 2): Proves that the expected cardinality of the coarse-grained point set swept by all geodesics from $L_{-n}$ to $L_n$ in the time interval s,t is at most $n^{1+\epsilon} + n^{5/3+\epsilon}(t-s)$.
5. Technical Innovations:
   • Develops a strategy to access all on-scale geodesics between regions via Poisson point sampling
   • Establishes Brownian regularity estimates for BLPP line ensembles (adapted from Dau24)
   • Proves twin peaks estimates for routed weight profiles

Methodology Details

Task Definition

Dynamic BLPP Model:

• Starts from an i.i.d. family of standard Brownian motions $\{W_n\}_{n\in\mathbb{Z}}$
• Defines process $\{X_{i,n}^t\}_{i,n\in\mathbb{Z},t\in\mathbb{R}}$ where each $(i,n)$ has an independent exponential clock
• When a clock rings, the path $X_{i,n}^t$ is independently resampled
• Defines the dynamic Brownian motion family via $W_n^t(x) - W_n^t(\lfloor x\rfloor) = X_{\lfloor x\rfloor,n}^t(x-\lfloor x\rfloor)$
• Corresponding last passage times and geodesics are denoted $T^t$ and $\Gamma_{p}^{q,t}$

Core Quantity Definitions:

1. Geodesic Switches: For points $p\leq q$ and set $K\subseteq\mathbb{R}^2$, $\text{Switch}_p^{q,[s,t]}(K) = \sum_{r\in T_p^{q,[s,t]}} |\text{Coarse}(K\cap\Gamma_p^{q,r}) \setminus \text{Coarse}(K\cap\Gamma_p^{q,r^-})|$ where $\text{Coarse}(K)$ is the coarse-grained approximation of $K$.
2. Hitset: $\text{HitSet}_{K_1}^{K_2,[s,t]}(K) = \bigcup_{p\in K_1\cap\mathbb{Z}\mathbb{R}, q\in K_2\cap\mathbb{Z}\mathbb{R}, r\in[s,t]} \text{Coarse}(\Gamma_p^{q,r}\cap K)$

Core Strategy

1. Proof Strategy for Geodesic Switch Estimates

Intuitive Motivation (Section 3.1):

• Consider a single resampling: if point $p$ is at transverse distance $k$ from geodesic $\Gamma_0^n$, then geodesic change requires a "twin peaks" event to occur
• Twin peaks probability is estimated via Brownian motion comparison as $\Theta(k^{-3/2})$
• If change occurs, by KPZ 1:2:3 scaling, the change magnitude is $\Theta(k^{3/2})$
• Thus the expected change from a single update is $\Theta(k^{-3/2}\times k^{3/2}) = O(1)$
• In time $(t-s)$, there are expected $O((t-s)n^{5/3})$ updates (since there are $n^{5/3}$ points in the $n^{2/3}$ neighborhood)
• This yields total expected switches $\leq n^{5/3+o(1)}(t-s)$

Technical Implementation:

• Introduces "excursion" concept: paths deviating from the geodesic
• Defines $\text{Exc}_\ell^\delta(m)$ event: existence of near-geodesic excursion at scale $\ell$, position $m$
• Uses twin peaks estimate (Proposition 15) to prove $P(\text{Exc}_\ell^\delta(m))\leq C\ell^{-1/3+2\delta}$
• Excludes pathological geometries via "thin excursion" estimate (Proposition 22, from GH23)
• Controls the union of all qualified excursions (Lemma 25): $E[|\text{Pivot}_\ell^\delta(m)|^2]\leq C'\ell^{5/3}n^{500\delta}$
• The critical 5/3 exponent emerges from this estimate

2. Lifting from Point-to-Point to On-Scale Regions

Challenge: Theorem 1 provides point-to-point estimates, but Theorem 2 requires estimates between segments $L_{-n}$ and $L_n$.

Strategy (Section 3.3):

• Poisson Point Sampling: Independent of dynamic BLPP, samples a Poisson point process $Q_{n,\nu}$ with intensity $n^{-10/3+2\nu}$ on $(\mathbb{Z}\mathbb{R})^2$
• Basin of Attraction Estimate (Proposition 28): For geodesic $\Gamma_p^q$, defines its "basin of attraction" $\text{Basin}_n^\delta(\Gamma_p^q) = \{(p',q'): \Gamma_{p'}^{q'}\cap[-(1-\gamma)n,(1-\gamma)n]\mathbb{R} = \Gamma_p^q\cap[-(1-\gamma)n,(1-\gamma)n]\mathbb{R}\}$ Proves $|\text{Basin}_n^\delta(\Gamma_p^q)|_{\text{hor}} \geq n^{10/3-2\delta}$ with superpolynomial high probability
• Coverage Lemma (Proposition 30): With stretched exponential high probability, all geodesics between on-scale regions can be accessed by geodesics between Poisson points
• Since $|Q_{n,\nu}| = O(n^{5\delta})$, the problem reduces to $O(n^{5\delta})$ point-to-point estimates

3. From Hitset to Exceptional Time Dimension

Key Observation (Section 3.4):

• By Proposition 13, any nontrivial bigeodesic $\Gamma^t$ stays within distance $n^{2/3+\epsilon}$ of its direction line within $[−n,n]\mathbb{R}$
• Defines on-scale parallelogram $B_n = B_{n^{2/3}}(L_{-n}^n)\cap[[−n/2,n/2]]\mathbb{R}$, with $|B_n| = \Theta(n^{5/3})$
• If $0\in\Gamma^t$ for some $t\in[s,s+n^{-2/3}]$, then $P(0\in\text{Coarse}(\Gamma_u^{v,r})\text{ for some }r\in[s,s+n^{-2/3}], u\in L_{-n}, v\in L_n)$$\lesssim \frac{1}{|B_n|}E[|\text{HitSet}_{L_{-n}}^{L_n,[s,s+n^{-2/3}]}|] \lesssim n^{-5/3}(n^{1+\delta} + n^{5/3+\delta}\cdot n^{-2/3}) = O(n^{-2/3+\delta})$
• Covers $[0,1]$ with $n^{2/3}$ intervals of length $n^{-2/3}$, expecting $O(n^\delta)$ intervals containing exceptional times
• Yields $\dim\mathscr{T}^\theta_0 = 0$

Technical Innovations

1. Scale and Position Decomposition of Geodesic Switches: Introduces $\text{Switch}_0^{n,[s,t]}(\ell,m)$ to quantify contributions at scale $\ell$ and position $m$, recovering total switches via logarithmic scale summation.
2. Connection Between Excursions and Near-Geodesics: Lemma 18 establishes precise relationship between $\Gamma_0^{n,r}\setminus\Gamma_0^{n,r^-}$ and excursions, providing structure for analysis.
3. Moment Estimates for Pivot Sets: Through peak counting (Proposition 14) and twin peaks estimates (Proposition 15), precisely controls $E[|\text{Pivot}_\ell^\delta(m)|^2]$, yielding the critical 5/3 exponent.
4. BLPP Version of Brownian Regularity: Adapts Dau24's Airy line ensemble results to BLPP (Propositions 58, 59), overcoming technical difficulties from BLPP's finite domain.
5. Lower Bound on Basin Volume: Uses volume accumulation ideas from BB23 (Proposition 29), combined with transverse fluctuation estimates, to obtain stretched exponential probability lower bounds on basin volume.

Experimental Setup

This is a pure theoretical mathematics paper with no numerical experiments. All results are rigorous mathematical proofs.

Theoretical Tools

1. BLPP Foundational Estimates:
   • Moderate deviation estimates (Propositions 5, from LR10, DV21a)
   • Transverse fluctuation estimates (Propositions 8, 9, from GH23, BBBK25)
   • Brownian Gibbs property of line ensembles
2. Connection Between Dynamic and Static BLPP (Lemma 10): Conditioned on resampling times, the BLPP distribution at each time is that of static BLPP.
3. Directionality of Infinite Geodesics (Propositions 12, 13): Adapts Newman-Howard-Newman argument to dynamic BLPP, proving semi-infinite geodesics are $\theta$-directed and non-axial.
4. Peak Estimates for Routed Weight Profiles (Proposition 14): Uses Brownian comparison to control $|\text{Peak}(n^\delta)\cap\{m\}\mathbb{R}|\leq n^{200\delta}$.

Experimental Results

Main Theorem Statements

Theorem 1 (Geodesic Switches): Fix $\beta\in(0,1/2)$ and $\epsilon>0$. For all sufficiently large $n$ and all $[s,t]\subseteq\mathbb{R}$, $E[\text{Switch}_0^{n,[s,t]}([[\beta n,(1-\beta)n]]\mathbb{R})] \leq n^{5/3+\epsilon}(t-s)$

Theorem 2 (Hitset Estimate): Fix $\gamma\in(0,1)$. For any $\epsilon>0$ and all sufficiently large $n$, $E[|\text{HitSet}_{L_{-n}}^{L_n,[s,t]}([[−(1−\gamma)n,(1−\gamma)n]]\mathbb{R})|] \leq n^{1+\epsilon} + n^{5/3+\epsilon}(t-s)$

Theorem 3 (Exceptional Time Dimension): For dynamic BLPP, almost surely $\dim\mathscr{T}\leq 1/2$.

Theorem 4 (Directional Exceptional Times): For dynamic BLPP and fixed direction $\theta\in(0,\infty)$, almost surely $\dim\mathscr{T}^\theta = 0$.

Key Intermediate Results

1. Twin Peaks Estimate (Proposition 15): For $\beta'\in(0,1/2)$, $\delta\in(0,1/6)$, all $\ell\leq n$ and $m\in[[\beta'n,(1-\beta')n]]$, $P(\text{TP}_{\ell,m}) \leq C\ell^{-1/3+2\delta}$ where $\text{TP}_{\ell,m}$ is the event that the routed profile has $\leq\ell^\delta$ suboptimal peaks at distances $\geq\ell^{2/3-\delta}$.
2. Basin Volume (Proposition 28): There exist constants $c,C,\delta_0$ such that for $\delta<\delta_0$, $P(|\text{Basin}_n^\delta(\Gamma_p^q)|_{\text{hor}} \leq n^{10/3-2\delta}) \leq Ce^{-cn^{3\delta/11}}$
3. Coverage Event (Proposition 30): The Poisson point process $Q_{n,\nu}$ accesses all geodesic basins with probability $\geq 1-Ce^{-cn^{3\delta/11}}-Ce^{-cn^{2\nu-2\delta}}$.

Proof Technique Highlights

1. Origin of the 5/3 Exponent:
   • Expected change from single update $O(1)$ (via twin peaks)
   • $n^{5/3}$ points in $n^{2/3}$ neighborhood
   • Expected $O((t-s)n^{5/3})$ updates in time $(t-s)$
2. Origin of 1/2 Dimension Upper Bound:
   • Hitset expectation $\sim n^{1+\delta} + n^{5/3+\delta}(t-s)$
   • Taking $t-s = n^{-2/3}$ yields probability $O(n^{-2/3+\delta})$
   • Covering $[0,1]$ with $n^{2/3}$ intervals, expected coverage $O(n^\delta)$
   • Interval length $n^{-2/3} = (n^{-2/3})^{-1/2-21\delta/2}$
   • Yields dimension $\leq 1/2 + 21\delta/2$, letting $\delta\to 0$ gives $\leq 1/2$
3. Origin of 0 Dimension: For directional bigeodesics, directly uses hitset estimate without special time scaling, obtaining stronger dimension 0 conclusion.

Related Work

Geodesic Structure in Static LPP

1. Non-existence of Bigeodesics:
   • BHS22, BBS20: Nontrivial bigeodesics almost surely don't exist in exponential LPP
   • RS24: Corresponding result for BLPP
   • LN96: Fixed-direction bigeodesics don't exist in first passage percolation
2. Directionality of Geodesics:
   • New95, HN01: Classical arguments by Newman and Howard-Newman
   • FP05: Implementation in exponential LPP
   • This paper, Propositions 12, 13: Adaptation to dynamic BLPP

Noise Sensitivity in Dynamic LPP

1. Phase Transition in Geodesic Overlap:
   • Cha14: General theory of noise sensitivity
   • GH24: Phase transition in BLPP geodesic overlap at time scale $n^{-1/3}$
   • ADS24: Related stability-to-chaos research
2. Complementarity with Bha25:
   • Bha25: "Near-existence" result for exceptional times in exponential LPP
   • This paper: Dimension upper bounds on exceptional time sets in BLPP
   • Both replace earlier preprint arXiv:2504.12293v1

Brownian Properties of BLPP

1. Brownian Gibbs Property of Line Ensembles:
   • CH14: Brownian Gibbs property of Airy line ensemble
   • Ham22, CHH23: Brownianity of BLPP weight profiles
   • Dau24: Wiener density of Airy line ensemble
   • This paper, Appendix 8: Adaptation of Dau24 to BLPP
2. Twin Peaks Estimates:
   • GH23: Twin peaks for BLPP routed profiles
   • SSZ24: Corresponding result for exponential LPP
   • This paper, Proposition 15: Twin peaks in more general setting

Geodesic Merging and Volume Accumulation

1. Basin Volume:
   • BB23: Volume accumulation for semi-infinite and finite geodesics in exponential LPP
   • This paper, Proposition 29: Adaptation to BLPP
   • Key for Poisson point sampling strategy
2. Transverse Fluctuation Estimates:
   • GH23: Macro and meso transverse fluctuations for BLPP
   • BBBK25: Refined version of meso transverse fluctuations
   • This paper, Lemma 50: Meso transverse fluctuations in dynamic setting

Conclusions and Discussion

Main Conclusions

1. Quantitative Noise Sensitivity: The $n^{5/3}(t-s)$ upper bound on geodesic switches provides the first quantitative characterization of geodesic noise sensitivity, with the 5/3 exponent believed to be optimal.
2. Rarity of Exceptional Times:
   • Exceptional time set where nontrivial bigeodesics exist has dimension $\leq 1/2$
   • For fixed-direction bigeodesics, exceptional time dimension $= 0$
   • Contrasts with "near-existence" result in accompanying paper Bha25
3. Methodological Contributions:
   • Poisson point sampling strategy for accessing on-scale geodesics has universal applicability
   • Establishment of BLPP Brownian regularity estimates
   • Framework for scale-decomposition analysis of geodesic switches

Limitations

1. Model Restrictions:
   • Main results only proven for BLPP, not covering exponential LPP
   • Reason: Local behavior of routed weight profiles in exponential LPP insufficiently understood
   • Accompanying paper Bha25 proves complementary "near-existence" result for exponential LPP
2. Non-tightness of Dimension Upper Bound:
   • Author conjectures $\dim\mathscr{T} = 0$ (rather than 1/2)
   • Theorem 4 only proves dimension 0 for fixed direction, not for union over all directions
   • Bha25, Conjecture 4: Conjectures $\mathscr{T}$ should be empty or a singleton
3. Non-tightness of 5/3 Exponent:
   • While 5/3 is expected to be tight, no lower bound is provided
   • Heuristic arguments support 5/3, but lack rigorous lower bound proof
4. Sub-optimality of Technical Details:
   • Factor $n^{500\delta}$ in Proposition 17 not carefully optimized
   • 1/2 dimension upper bound in Theorem 3 may not be tight

Future Directions

1. Generalization to Exponential LPP:
   • Develop local theory of routed weight profiles in exponential LPP
   • Prove corresponding geodesic switch estimates
   • Unify treatment of BLPP and exponential LPP
2. Exact Dimension Values:
   • Prove $\dim\mathscr{T} = 0$ (or provide counterexample)
   • Prove dimension 0 for union $\bigcup_\theta\mathscr{T}^\theta$ over all directions
   • Improve 1/2 upper bound to 0
3. Lower Bounds on 5/3 Exponent:
   • Construct configurations showing $n^{5/3}(t-s)$ lower bound
   • Understand deep connection between 5/3 and KPZ 1:2:3 scaling
4. Generalization to Other KPZ Models:
   • Corresponding problems in directed landscape
   • Other integrable models (e.g., log-gamma polymer)
   • Non-integrable first passage percolation
5. Fine Structure of Dynamics:
   • Spatial distribution of geodesic switches
   • Fine topological structure of exceptional times
   • Quantitative characterization of geodesic chaos

In-Depth Evaluation

Strengths

1. Pioneering Problem:
   • First systematic study of sample path behavior of geodesics in dynamic LPP
   • Quantitative estimate of geodesic switch count fills important gap
   • Provides new perspective on understanding KPZ universality class dynamics
2. Technical Innovation:
   • Poisson Point Sampling Strategy: Elegantly lifts point-to-point estimates to on-scale estimates, with universal application potential
   • Scale Decomposition: Introduction of $\text{Switch}_0^{n,[s,t]}(\ell,m)$ provides refined analytical framework
   • Brownian Regularity: Adaptation of Dau24 to BLPP overcomes technical difficulties, laying foundation for future work
3. Proof Rigor:
   • 60-page detailed proof including 5 appendices handling technical details
   • Clear distinction between heuristic arguments (Section 3) and rigorous proofs (Sections 4-6)
   • Complete treatment of foundational properties of dynamic BLPP (e.g., geodesic uniqueness, directionality)
4. Result Depth:
   • Appearance of 5/3 exponent reveals deep connection between geodesic dynamics and KPZ scaling
   • Dimension estimates introduce topological methods to dynamic LPP research
   • Complementarity with Bha25 demonstrates problem richness
5. Writing Quality:
   • Section 3 proof overview is clear and intuitive, with beautiful illustrations
   • Technical details well-organized, appendices independently readable
   • Consistent notation system facilitates tracking complex arguments

Weaknesses

1. Model Limitations:
   • BLPP Only: Main results don't cover more physically meaningful exponential LPP
   • Discrete Dynamics: Uses discrete resampling rather than continuous Ornstein-Uhlenbeck dynamics
   • Planar Restriction: Only considers $\mathbb{Z}^2$ model, not generalized to higher dimensions
2. Non-optimality of Results:
   • Dimension Upper Bound: 1/2 expected improvable to 0, but technically not achieved
   • Theorem 3 vs. Theorem 4: For union over all directions only get 1/2, for single direction get 0, creating gap
   • Sub-optimal Constants: Factors like $n^{500\delta}$ clearly not tight
3. Missing Lower Bounds:
   • 5/3 Exponent: While expected optimal, no matching lower bound provided
   • Dimension Lower Bounds: No discussion of whether $\dim\mathscr{T}$ could be $>0$
   • Exceptional Time Structure: No characterization of fine properties of $\mathscr{T}$ (e.g., whether countable)
4. Technical Dependencies:
   • Brownian Gibbs Property: Key results depend on BLPP's special structure, difficult to generalize
   • Reliance on Existing Results: Heavy use of deep results from GH23, Dau24, BB23, limited independence
   • Appendix Complexity: Adaptation of Dau24 arguments occupies substantial space, increasing reading difficulty
5. Insufficient Physical Intuition:
   • Origin of 5/3 Exponent: While heuristic argument provided, deep connection to KPZ scaling insufficiently clarified
   • Bigeodesic Emergence Mechanism: Lacks intuitive explanation of mechanism for bigeodesic appearance
   • Connection to Other KPZ Quantities: No discussion of relationship to fluctuation exponents, correlation lengths, etc.

Impact

1. Contribution to Field:
   • New Direction: Opens new direction of sample path analysis of geodesics in dynamic LPP
   • Methodology: Poisson point sampling strategy potentially applicable to other point-to-region lifting problems
   • Foundational Tools: BLPP Brownian regularity estimates (Appendix 8) will be cited in future work
2. Practical Value:
   • Theoretical Significance: Primarily pure mathematics results, limited direct practical value
   • Concept Clarification: Provides precise quantitative characterization of "noise sensitivity"
   • Conjecture Support: Provides partial evidence for $\dim\mathscr{T}=0$ conjecture
3. Reproducibility:
   • Pure Theory: No numerical experiments, reproducibility manifested in proof verifiability
   • Detail Level: 60-page detailed proof and appendices support independent verification
   • Dependencies: All cited results explicitly listed, facilitating traceability
4. Follow-up Research:
   • Direct Generalizations: Corresponding results for exponential LPP, similar problems in other KPZ models
   • Refinements: Improved dimension bounds, proof of 5/3 exponent tightness
   • New Problems: Fine structure of exceptional times, spatial distribution of geodesic switches

Applicable Scenarios

1. Probability Theory Research:
   • Geodesic research in random geometry
   • Quantitative analysis of noise sensitivity
   • Sample path behavior of dynamic random systems
2. Mathematical Physics:
   • Dynamics of KPZ universality class
   • Time evolution of growth models
   • Dynamic properties of integrable systems
3. Method Borrowing:
   • Poisson point sampling strategy potentially applicable to:
     • Other problems requiring point-to-region lifting
     • Scenarios using "basin of attraction" ideas
     • Research on geodesic merging phenomena
   • Scale decomposition analysis potentially applicable to:
     • Other problems involving multi-scale behavior
     • Scenarios requiring fine control of different scale contributions
4. Inapplicable Scenarios:
   • Non-integrable or weakly integrable models (lacking Brownian Gibbs property)
   • High-dimensional models (methods highly dependent on planar geometry)
   • Applications requiring exact constants (results only give asymptotic orders)

References

Key Citations

1. GH24 Ganguly & Hammond (2024): "Stability and chaos in dynamical last passage percolation" - Phase transition in geodesic overlap
2. Dau24 Dauvergne (2024): "Wiener densities for the Airy line ensemble" - Brownian regularity of Airy line ensemble
3. BB23 Basu & Bhatia (2023): "A Peano curve from mated geodesic trees in the directed landscape" - Volume accumulation estimates
4. GH23 Ganguly & Hammond (2023): "The geometry of near ground states in Gaussian polymer models" - Twin peaks estimates and transverse fluctuations
5. Bha25 Bhatia (2025): "Near-existence of bigeodesics in dynamical exponential last passage percolation" - Accompanying paper, complementary result for exponential LPP
6. BHS22 Basu, Hoffman & Sly (2022): "Nonexistence of Bigeodesics in Planar Exponential Last Passage Percolation" - Non-existence of bigeodesics in static exponential LPP
7. RS24 Rassoul-Agha & Sweeney (2024): "Shocks and instability in Brownian last-passage percolation" - Non-existence of bigeodesics in static BLPP

Overall Assessment: This is an excellent theoretical paper with deep technical content and novel results. The author is the first to systematically study sample path behavior of geodesics in dynamic LPP, establishing the pioneering result that geodesic switch count has upper bound $n^{5/3}(t-s)$, with the 5/3 exponent believed optimal. The appearance of the 5/3 exponent reveals deep connection between geodesic dynamics and KPZ scaling. Technical innovations like Poisson point sampling have universal application potential. The paper's main limitations are restriction to BLPP (not covering exponential LPP) and non-tightness of dimension bounds. Nevertheless, this paper opens new directions for studying dynamic KPZ systems and is expected to have significant follow-up impact.