{"id":"193bcd90-c07f-43b1-8223-a9fc63d4ecf0","arxiv_id":"2411.10399","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Concentrated liquidity provision by competing, budget-constrained LPs has a unique Nash equilibrium with a waterfilling structure, and real LPs in risky pools deviate from it.","lead":"This paper builds a game-theoretic model of how liquidity providers compete in concentrated-liquidity automated market makers like Uniswap v3, and shows the complex game reduces to a simpler one with a unique equilibrium. Using real Uniswap v3 data, it finds that LPs in risky pools are far from this equilibrium and could improve daily returns by following a history-based strategy.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Uniqueness of the atomic-game equilibrium (Thm 3.3) rests on the constant-fee assumption; the twinship result survives endogenous fees, but uniqueness and waterfilling do not, and §4.1.3 instantiates games with constant fees.","rationale":"The reader's weakest assumption coincides with the main load-bearing concern. I would keep the CONDITIONAL verdict: the central theory is internally sound under its stated assumptions, and the twinship reduction is even robust to endogenous fees, but the uniqueness and waterfilling claims—and the empirical conclusions built on them—require f_m to be exogenous. This is not a fatal internal inconsistency; the paper flags the assumption explicitly in §4.1.3, but it is precisely the place where the argument is least secure relative to the real-world setting. A small computational search for multiple equilibria under liquidity-dependent fees would settle whether the concern is substantive. No change to the reader's verdict is needed.","tokens_in":43837,"tokens_out":12381,"duration_ms":122782,"concrete_test":"For the M=1, N=2 atomic game with α=1, B=(1,1), τ=0.1, and fee function f(κ)=c·κ^β/(1+κ^β), sweep c∈[0.1,2] and β∈[0.5,2]; solve the best-response system by fixed-point iteration from 20 random starting points. If two iterations converge to different fixed points, the uniqueness theorem does not extend to endogenous fees, confirming that the constant-fee assumption is load-bearing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Thm 3.3 (and its proof, Thm C.5 in App C.5) establishes uniqueness of the atomic-game NE only under the constant-fee specification of §C.2. The Hessian computation treats f_m as a constant factor; if f_m depends on total liquidity κ_m, the derivatives ∂²U/∂K_j∂K_n acquire extra f'_m(κ_m) and f''_m(κ_m) terms, and the negative-definiteness bound used to apply Rosen's DSC theorem no longer follows. The twinship result Thm 3.5 is more robust: App C.1 proves it for fee rewards f_m(κ_m) satisfying condition (C.1). However, the uniqueness theorem, the waterfilling Prop 3.6, and budget-dominance Prop 3.7 all depend on the constant-fee atomic game. In the empirical instantiation (§4.1.3), the authors explicitly set f_m from observed fees and hold it fixed while varying liquidity; hence the computed unique equilibrium and the counterfactual $116/0.009% ROI gain are predictions of a game whose main strategic feedback—fee generation responding to liquidity depth—is assumed away. If liquidity supply affects trading volume and fee generation, the true game may have multiple equilibria, so the computed benchmark and waterfilling description of LP behavior are not guaranteed to apply.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper models strategic liquidity provision in concentrated-liquidity market makers (CLMMs) as a non-cooperative game. Each liquidity provider (LP) chooses liquidity amounts across price ranges subject to a dollar budget, earns a share of per-range trading fees, and bears an expected impermanent-loss cost. The first main result (Thm. 3.5) shows that the original game, whose action space is quadratic in the number of ticks, is payoff-equivalent (a 'twin game') to an atomic game in which each LP chooses liquidity only over adjacent-tick ranges, so Nash equilibria of the two games correspond. The second main result (Thm. 3.3) proves that for 0<alpha<=1 the atomic game has a unique Nash equilibrium, characterized by a waterfilling pattern (Prop. 3.6) and budget dominance (Prop. 3.7): low-budget LPs exhaust their budgets while rich LPs spend equal amounts on active ranges. Empirically, the model is instantiated on five Uniswap v3 pools (one stable, four risky) over 182 daily periods. The authors find that stable-pool LPs' actual positions overlap substantially with the computed Nash equilibrium, whereas risky-pool LPs use fewer and wider price ranges than the equilibrium suggests, and that a strategy computed from a 7-day-history 'inert game' (I_NE) improves median daily utility by $116 (a 0.009% increase in median daily ROI) relative to ground-truth actions.","tokens_in":44168,"tokens_out":21241,"duration_ms":179493,"significance":"The theoretical contribution is substantial and, conditional on its stated assumptions, appears sound. The twinship theorem is proved under weak conditions on the fee-sharing rule (App. C.1), and the uniqueness proof via Rosen's diagonal strict concavity (App. C.5) is detailed and, as far as I can verify, correct for the constant-fee atomic game; the waterfilling and budget-dominance characterizations follow from the KKT conditions. The reduction from a quadratic to a linear action space is clean and practically meaningful. The empirical part is transparent by the standards of this literature: the data-cleaning pipeline is described in detail, a public repository is provided, and the main modeling assumptions (constant per-range fees, the price-shift calibration, per-pool expansion factors) are explicitly stated rather than hidden. The empirical findings are falsifiable and informative: with the caveat that the headline gains are conditional on the model assumptions, the paper demonstrates that simple data-driven equilibrium strategies can improve LP returns in risky pools while stable-pool LPs play close to equilibrium.","major_comments":[{"comment":"The uniqueness theorem (Thm. 3.3), the waterfilling characterization (Prop. 3.6), and budget dominance (Prop. 3.7) are proved only for the constant-fee atomic game in which f_m does not depend on the total liquidity kappa_m. If fee rewards respond to liquidity depth, the Hessian in Thm. C.5 acquires terms in f'_m(kappa_m) and f''_m(kappa_m), and the diagonal-strict-concavity argument used to apply Rosen's theorem no longer yields negative definiteness; the twinship result (Thm. 3.5 / Thm. C.1) survives endogenous fees, but the claimed uniqueness of the atomic liquidity histogram, budget usage, and utilities across original-game equilibria rests on the constant-fee assumption. Because Sec. 4.1.3 instantiates the games with fixed observed fees, the computed unique equilibrium and the $116 / 0.009% counterfactual are predictions of a model that assumes fee generation is unaffected by LP liquidity choices. Please state this scope limitation at the point where Thm. 3.3 is invoked for the empirical analysis, and add a robustness check under a simple endogenous-fee specification, e.g., f_m(kappa_m) = f_m^0 (kappa_m / kappa_m^0)^beta for a small set of beta values, to show whether the qualitative findings survive.","section":"Sec. 3.1 (Eq. 3.3), Sec. 4.1.3, App. C.5 (Thm. C.5)"},{"comment":"The headline improvement of $116 in median daily utility (0.009% median daily ROI) is computed using per-pool expansion factors E (Tab. 6) that are chosen by maximizing the I_BR-versus-GT utility and overlap advantage on the same evaluation data used to report the gain (App. E.3, Fig. 12). This is in-sample selection of a free parameter, so the reported numbers are an optimistic fit rather than an out-of-sample prediction. Please either select E on a training window and evaluate on a holdout window, or report the headline statistic across a grid of E values (including E=1 and a single common E for all pools), and adjust the abstract and Finding 4 so that they do not present the $116 / 0.009% figure as model-free.","section":"App. E.3 (Tab. 6), Sec. 4.3 (Finding 4)"},{"comment":"The shifted dollar prices in Eq. (4.2) are an ad hoc calibration introduced to guarantee tau-hat >= 0, and they change the impermanent-loss term in every utility evaluation. Fig. 14 bounds the resulting relative error in token prices (below 1%), but the utility differences compared in Findings 2-4 are of order 0.01% of budget (Tab. 9), so a small distortion in the loss term could be material for the sign or magnitude of the reported gaps. Please report the sensitivity of the headline utility differences to the calibration choice (e.g., replacing the symmetric shift with the zero-capping alternative mentioned in the text, or perturbing the shift by plausible amounts), and quantify how the $116 gap changes.","section":"Sec. 4.1.4, Eq. (4.2)"},{"comment":"Setting each player's budget equal to their observed total investment for the day means the GT strategy always exhausts the budget by construction, so the 'unused budget' bucket in the overlap metric (Eq. 4.3) is empty for GT and the NE-GT overlap partly reflects this normalization rather than genuine budget constraints. Please clarify whether any GT player leaves budget unused, and discuss how the computed NE would change if budgets were instead estimated from the LP's capital available at the start of the day, consistently with the inertial-game budget estimation described in Sec. 4.3.","section":"Sec. 4.1.1, Eq. (4.3)"}],"minor_comments":[{"comment":"In the negative-definiteness computation, the displayed equality preceding '-alpha Sum_k K^alpha_{m,k}[...]' omits the non-negative chi_m terms; the inequality direction is unaffected, but the sign should be '<=' rather than '='.","section":"App. C.5 (Thm. C.5)"},{"comment":"The contribution list in Sec. 1 refers to the waterfilling result as 'Thm. 3.6', while the statement in Sec. 3.4 is labeled Prop. 3.6; the numbering of this result should be aligned between the introduction and the body.","section":"Sec. 1 vs. Sec. 3.4"},{"comment":"The sentence 'when we observe f_m > 0 but f^P_m for some price range m' is missing the condition '= 0', and for the E100 pool 390 of 833 ranges are uncovered, so the non-player liquidity share chi_m cannot be estimated for nearly half of that pool's ranges; please state explicitly how these ranges enter the game instances.","section":"App. E.2"},{"comment":"The headline figures of $116 median and $222 mean daily utility improvement are not directly traceable to the per-pool statistics in Tabs. 8 and 9 (e.g., the E5 median difference is $150 and the E30 mean difference is $813); please specify the aggregation (pooled observations versus average of per-pool means) used to obtain $116 and $222.","section":"Sec. 4.3 (Finding 4), Tabs. 8-9"},{"comment":"The abstract contains a typo: 'Automated marker makers' should read 'Automated market makers'.","section":"Abstract"},{"comment":"The proposition title contains typos: 'Positive liqidity at eqilibrium' should read 'Positive liquidity at equilibrium'.","section":"Prop. 3.8"}],"recommendation":"major_revision","confidential_remarks":"The theoretical core of this paper is, in my reading, correct under its stated assumptions, and the empirical work is transparent about its modeling choices. The main risk is that the abstract and Finding 4 present the in-sample $116 / 0.009% gain as a robust prediction when it depends on the constant-fee assumption, the per-pool expansion factor E chosen on the evaluation data, and the ad hoc price shift of Eq. (4.2). I would send the revision back to referees with DeFi and game-theory expertise. The most valuable addition would be a sensitivity analysis over fee endogeneity and the expansion factor; extending the pool set is less urgent. If the authors cannot provide any out-of-sample or grid-sensitivity validation, the empirical findings should be reworded as an illustrative calibration of the model rather than a prediction of achievable returns."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The first thing you should know: this paper's central theoretical result is real. The twin-game reduction (Thm 3.5) says that in a CLMM, any liquidity profile over general ranges is payoff-equivalent to an atomic-range profile, and Nash equilibria map across. That holds even if fees depend on total liquidity, as long as weights satisfy the invariance condition (C.1). The uniqueness theorem (Thm 3.3) is conditional on constant fee rewards, but within that model the Rosen/KKT argument is careful and I could not find a gap. The waterfilling and budget-dominance properties are clean corollaries.\n\nThe empirical part is more suggestive than definitive. They are honest that §4.1.3 assumes the fee reward per range stays constant when liquidity changes. The stress-test note is right that uniqueness and waterfilling do not survive endogenous fees, but that is a modeling boundary, not an oversight; the paper says so. The bigger issue is the headline $116 / 0.009% ROI improvement. The expansion factor E in the inert game is chosen per pool to maximize the same utility advantage over ground truth used in the headline (App E.3, Fig 12). So part of that gain is in-sample. The ad hoc price shift (Eq 4.2) is a secondary concern, and they bound its induced error at under 1%, which is fine.\n\nThe stable pool result is more robust: there E=1, overlap with NE is around 40%, and the comparison is cleaner. The risky pool finding—real LPs use wider, fewer ranges than NE—is a useful empirical observation even if the counterfactual is not out-of-sample.\n\nOverall I agree with the reader's conditional verdict. The theoretical contribution deserves a serious referee, and the paper is likely citable for the twinship reduction. The empirical section needs work: either report the I_NE gains as in-sample calibration, or split the data into tuning and evaluation. I would send this to peer review rather than desk reject.","headline":"Solid twinship and uniqueness theorems for CLMM liquidity games; the empirical 'Nash beats real LPs' claim is partly in-sample, so treat the $116 gain as a calibrated illustration rather than a prediction.","tokens_in":44697,"tokens_out":2326,"would_cite":true,"duration_ms":24170,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A10","91A80"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that concentrated-liquidity AMM liquidity provision admits a unique waterfilling Nash equilibrium in an equivalent atomic game, and that risky-pool LPs' deviations from it leave about $116/day on the table.","keywords":["automated market maker","blockchains","decentralized cryptocurrency exchanges","game theory","Nash equilibrium","concentrated liquidity","impermanent loss","waterfilling"],"falsifier":"Take one narrow price range in a live CLMM pool and measure its daily fee revenue before and after a large new liquidity position is added there, holding market conditions fixed; if fee revenue moves systematically with the added liquidity, the constant-fee assumption fails and the simulated gains from the inert game are not trustworthy. Alternatively, an exhaustive numerical search for two distinct equilibrium vectors in a small instance with $0<\\alpha\\le 1$ would directly refute the uniqueness theorem.","tokens_in":43647,"feed_emoji":"🪙","tokens_out":7615,"duration_ms":71788,"temperature":0.7,"pith_summary":"The paper tries to prove that the strategic problem of where and how much liquidity to provide in a concentrated-liquidity AMM is tractable despite its huge strategy space: all Nash equilibria of the full game produce the same per-tick liquidity profile, and that profile is the unique equilibrium of a much smaller atomic game when the fee exponent $\\alpha \\le 1$. If true, LP behavior can be predicted and optimized: the unique equilibrium has a waterfilling structure (poor LPs exhaust their budgets, rich LPs do not), and it can be solved in linear complexity. The paper argues the model matches reality in stable pools, where LPs sit near equilibrium, and explains the gap in risky pools as stale information; adopting the inert-game equilibrium would raise median daily returns by about $116.","feed_headline":"A unique game equilibrium governs CLMM liquidity wars","feed_subtitle":"Concentrated-liquidity AMMs reduce to a linear atomic game; risky-pool LPs leave $116/day.","key_machinery":"The load-bearing object is the atomic game and the linear aggregation map $\\theta$ that summarizes any LP's positions over $O(M^2)$ general price ranges into $M$ atomic per-tick liquidities; Theorem 3.5 (twinship) shows the original and atomic games are payoff-equivalent under this map. Uniqueness is carried by the diagonal strict concavity condition of the atomic utility (Rosen's theorem) combined with the KKT conditions of the budget-constrained concave program, and the waterfilling property follows from those KKT conditions via a monotonicity argument on per-range marginal utilities.","core_discovery":"The paper's central claim is that the high-dimensional liquidity-provision game in a CLMM is payoff-equivalent to an atomic game in which each LP directly chooses liquidity on each tick-to-tick interval. For fee-sharing exponent $0<\\alpha\\le 1$, the atomic game has a unique Nash equilibrium; the map from original strategies to atomic liquidity is a twinship, so every equilibrium of the original game gives the same per-tick liquidity, budget use, and utility as this unique equilibrium. The paper further proves that the equilibrium follows waterfilling: LPs below a budget threshold exhaust their budgets at a common per-range level, while richer LPs invest equal amounts and leave budget unused. Empirically, fitting the model to four Uniswap v3 pools shows stable-pool LPs behave near equilibrium, while risky-pool LPs deviate; against static rivals, switching to the inert-game equilibrium raises median daily utility by about $116 (0.009% median daily ROI).","pith_inferences":["If fee rewards are allowed to depend endogenously on total liquidity, uniqueness and waterfilling need not survive; a natural extension is a volume-as-function-of-liquidity model, which could be tested by regressing per-range fees on lagged liquidity.","The structural result transfers beyond CLMMs: any proportional-sharing resource allocation with one budget constraint per agent and concave per-range returns may inherit the same waterfilling equilibrium, so the theorem could be stated in that generality.","The $116/day counterfactual is measured against static rivals; if all LPs update to the inert equilibrium, the gains should dissipate, so the number is an upper bound on the value of strategic updating rather than a persistent arbitrage.","A testable prediction of the paper's story is that in risky pools, LPs with lower information latency should use narrower ranges and update more often; pool-level data after infrastructure changes could confirm this."],"forward_implications":["For fee-sharing exponents $0<\\alpha\\le 1$, equilibrium strategy computation scales linearly in the number of ticks rather than quadratically, making game-theoretic advice practical for real contracts.","At equilibrium, LPs below a budget threshold spend their full budget, while richer LPs spend a common amount and leave budget idle; equal-budget LPs adopt identical strategies.","The twinship result pins down the per-tick liquidity profile, budget usage, and utility of every Nash equilibrium of the original game, resolving the multiplicity of the original game.","In stable pools, the model's equilibrium matches actual LP actions closely, while in risky pools it does not; this supports using the model to detect suboptimal liquidity provision in volatile markets.","An LP who adopts the inert-game equilibrium while rivals keep their current strategies can improve median daily returns by $116 (0.009% median daily ROI) in risky pools."],"supporting_citations":[{"why":"Defines the CLMM mechanism (ticks, positions, fee sharing) that the paper models.","marker":"[3]"},{"why":"Supplies the on-chain pool data used in the empirical comparison.","marker":"[4]"},{"why":"Provides the relaxation algorithm used to compute Nash equilibria in experiments.","marker":"[11]"},{"why":"Prior single-LP model that also treats fees as independent of total liquidity; the paper inherits this assumption.","marker":"[23]"},{"why":"Prior concave pro-rata game framework for legacy AMM uniqueness that this work extends.","marker":"[35]"},{"why":"Gives the diagonal strict concavity theorem used to prove unique equilibrium in the atomic game.","marker":"[47]"}],"fun_headline_variants":["Unique Nash equilibrium governs CLMM liquidity games","CLMM liquidity game reduces to unique Nash, waterfilling","Risky-pool LPs gain $116/day by adopting Nash equilibrium","Waterfilling equilibrium yields $116 daily gain in CLMM pools"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the fee paid in each price range is a fixed external number, unaffected by how much liquidity LPs put there; if deeper liquidity changes trading volume or fees, the model's utility function, uniqueness proof, and counterfactual gains all lose their grip.","fun_headline_variants_meta":{"raw":{"variants":["Unique Nash equilibrium governs CLMM liquidity games","CLMM liquidity game reduces to unique Nash, waterfilling","Risky-pool LPs gain $116/day by adopting Nash equilibrium","Waterfilling equilibrium yields $116 daily gain in CLMM pools"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000884,"raw_usage":{"total_tokens":3885,"prompt_tokens":1081,"completion_tokens":2804,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":697,"completion_tokens_details":{"reasoning_tokens":2733}},"tokens_in":697,"tokens_out":2804,"duration_ms":19718,"temperature":1.0,"reasoning_tokens":2733,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T19:40:00.477567+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take one narrow price range in a live CLMM pool and measure its daily fee revenue before and after a large new liquidity position is added there, holding market conditions fixed; if fee revenue moves systematically with the added liquidity, the constant-fee assumption fails and the simulated gains from the inert game are not trustworthy. Alternatively, an exhaustive numerical search for two distinct equilibrium vectors in a small instance with $0<\\alpha\\le 1$ would directly refute the uniqueness theorem.","supporting_citations":[{"cited_title":"Uniswap v3 core","cited_arxiv_id":null,"evidence_quote":"Defines the CLMM mechanism (ticks, positions, fee sharing) that the paper models."},{"cited_title":"Allium - enterprise blockchain data platform","cited_arxiv_id":null,"evidence_quote":"Supplies the on-chain pool data used in the empirical comparison."},{"cited_title":"Relaxation techniques and asynchronous algorithms for on-line computation of non-cooperative equilibria","cited_arxiv_id":null,"evidence_quote":"Provides the relaxation algorithm used to compute Nash equilibria in experiments."},{"cited_title":"Marmolejo-Cossío, Ben Altschuler, He Sun, Xintong Wang, and David Parkes","cited_arxiv_id":null,"evidence_quote":"Prior single-LP model that also treats fees as independent of total liquidity; the paper inherits this assumption."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Prior concave pro-rata game framework for legacy AMM uniqueness that this work extends."},{"cited_title":"Existence and uniqueness of equilibrium points for concave n-person games","cited_arxiv_id":null,"evidence_quote":"Gives the diagonal strict concavity theorem used to prove unique equilibrium in the atomic game."}],"review_version":1}