Research
Tight Characterizations of AMM Mechanism Design under Weak Efficiency
Tight bounds on AMM mechanisms under fair sequencing tell designers which MEV-mitigation guarantees are actually achievable, not merely aspirational.

- arXiv
- 2610.36635
- Published
- 2026-10-27
- Authors
- Chengkai Zhang, Lirong Xia, Elaine Shi, Mengqian Zhang, Qishen Han, Yuhao Li
AI summary
Tight Characterizations of AMM Mechanism Design under Weak EfficiencyOverview
Research area: Blockchain mechanism design and algorithmic game theory, specifically automated market maker (AMM) design and miner extractable value (MEV) mitigation on consensus-layer-aided architectures. The paper appears on arXiv (2610.36635v1 [cs.GT], 27 Oct 2026) under CC BY 4.0.
Technical level: Advanced. The paper is a formal game-theory and mechanism-design paper: it defines a model of two-asset batch AMMs with a concave pricing curve, formalizes utility over sets of identities, and proves impossibility and feasibility theorems. It assumes comfort with incentive compatibility, dominant-strategy truthfulness, and proof-by-contradiction arguments.
Scope: A single-sentence scope: the paper determines exactly when incentive compatibility and weak local efficiency can coexist in a two-asset batch AMM, showing they cannot under plain consensus (even with censorship resilience and unlimited block space) and that removing any one of three strategic capabilities restores feasibility.
What This Paper Is About
Prior work showed that a batch AMM can achieve incentive compatibility (IC) and weak local efficiency (wLE) if the consensus layer provides fair sequencing, which orders transactions by arrival time. The open question was whether these same guarantees are achievable under plain consensus — the single-builder model used by mainstream blockchains such as current Ethereum — where the builder chooses both inclusion and ordering.
The paper answers that question negatively and then pins down exactly why: the impossibility requires three strategic capabilities at once (Sybil identities, budget over-reporting, and value misreporting), and each of the three, if removed, restores feasibility.
Key Contributions
-
An impossibility theorem for plain consensus. The paper proves that no deterministic, anonymous two-asset batch AMM satisfying individual rationality and pricing curve conformity can simultaneously satisfy IC and wLE. The result survives even when the consensus layer guarantees censorship resilience and block space is unlimited (stated as Theorem 1 in the introduction and restated as Theorem 4 in Section 3).
-
The first separation between censorship resilience and fair sequencing in decentralized mechanism design. Prior impossibility results either held only under plain consensus (and disappeared with censorship resilience or uncongested block space) or held even in permissioned settings. This paper's impossibility persists under censorship resilience yet disappears under fair sequencing, which the authors describe as the first such separation "in a precise sense."
-
A tightness characterization via three feasibility constructions. The proof of impossibility uses Sybil identities, aggregate budget over-reporting, and value misreporting together. Restricting any one of them restores IC and wLE (plus other natural properties) under censorship resilience, and the paper supplies a mechanism for each restricted setting: an equal-share hand-in-hand mechanism (no Sybil identities), a proportional hand-in-hand mechanism (no aggregate budget over-reporting), and a layered mechanism processing orders in increasing order of their asks (public values). The specific theorem numbers for these constructions are not visible in the provided text.
-
A reusable decomposition of batch-AMM design. All three constructions share a two-sided matching component: eligible orders on the side with the smaller aggregate reported budget (measured in common units at the initial exchange rate) trade their full budgets at that price, and the remaining orders are handled by a one-sided mechanism on an augmented compensation curve that folds in the matched trades before following the pool's curve. The authors expect initial-price matching to become a standard starting point for batch-AMM mechanism design.
Main Findings
-
The main impossibility. IC plus wLE is unattainable for deterministic, anonymous two-asset batch AMMs under plain consensus, and the obstruction is not censoring: it holds under censorship resilience and with unbounded block space. Since a mechanism that rejects every order is trivially IC, wLE is imposed to make the statement meaningful.
-
Minority-side full-budget lemma. A key lemma shows that eligible orders on the minority side — the side with the smaller aggregate reported budget measured in common units at the initial exchange rate — must trade their full budgets at the initial price. The authors describe this as "seemingly intuitive" but "highly nontrivial" to prove, requiring a chain of intermediate claims because wLE is a weak assumption.
-
Three capabilities are jointly necessary. The impossibility proof combines Sybil identity creation, over-reporting of budgets, and misreporting of values. The paper states explicitly that "we have to use all of them simultaneously."
-
Tightness, setting 1 — no Sybil identities. If each user submits at most one order but may still misreport value and budget (including over-reporting budget) and choose either trading direction, an equal-share hand-in-hand mechanism achieves all desired properties.
-
Tightness, setting 2 — no aggregate budget over-reporting. If users may create Sybil identities and misreport values but the aggregate declared budget in their true trading direction cannot exceed their true budget, a proportional hand-in-hand mechanism achieves all desired properties.
-
Tightness, setting 3 — public values. If a user's identities in its true trading direction must share its public value, while budgets remain private and may be over-reported, a layered mechanism processing orders in increasing order of asks achieves all desired properties. The latter two settings still permit fake bids in both directions with unrestricted asks and declared budgets for opposite-direction orders; the public-value result does not assume budgets are public or that opposite-direction asks are fixed by the public value.
-
IC implies no free lunch (Fact 2). Under reported-budget constraints, individual rationality, and pricing curve conformity, IC forces that any order receiving a positive amount of one token must pay a positive amount of the other: for sell-Y orders, x_i > 0 implies y_i > 0, and symmetrically for sell-X orders. The proof uses an unbounded-value deviator of true type (u, 0).
-
Anonymity plus builder IC implies economic symmetry (Fact 3). Identical reports in a profile must receive equal utilities evaluated at their common reported type. The proof is a position-swap argument available to a builder-as-user. This is used to bound a strategic user's share of the trading loss when it splits an inflated budget across Sybil bids.
-
Exact settlement is the binding constraint. The paper positions its result against Otter (Shi, Zhang, Chung, and Li), which relaxes exact settlement by allowing surplus redistribution — some pool output may go to the broader community rather than to orders in the batch — and achieves IC plus social-welfare maximization under censorship-resistant consensus and uncongested block space. The authors conclude that any mechanism seeking both MEV-resilience and reasonable efficiency in this setting must depart from exact settlement; this is presented as a general insight, not a property of Otter specifically.
-
Model generality. The mechanism class covers any continuous, strictly increasing, concave pricing curve F on (-ŷ, +∞) with F(0) = 0, differentiability at 0 with 0 < F′(0) < ∞, and the lower-endpoint behavior lim_{p↓-ŷ} F(p) = -∞. Constant product (Uniswap-style, with k = x₀y₀, σ₀ = x₀/y₀, ρ₀ = y₀/x₀) is shown to be a special case.
-
Relationship to prior work. Chan, Wu, and Shi achieve IC and wLE under fair sequencing; Li, Shi, and Zhang show that stronger efficiency notions — local efficiency for all orders and Pareto optimality — are incompatible with IC even under fair sequencing; a more recent work obtains an IC mechanism with weak Pareto optimality for eligible bids via adaptive clinching, which the present authors call substantially more sophisticated than their own simpler feasibility constructions; Ferreira and Parkes show that under sequential execution, every sequencing rule admits risk-free builder profit even for constant-product AMMs, a weaker property than IC.
-
No empirical results. The paper is purely theoretical. It reports no datasets, no benchmark instances, no simulations, and no measured numbers; all quantitative content consists of model parameters and curve constants.
Methodology in Plain English
The authors build a formal model of a two-asset batch AMM. Users submit sell-X or sell-Y orders with a private value (reservation price) and a private budget. The mechanism processes the whole block jointly, subject to pricing curve conformity: after the entire batch, the pool's reserves must lie on the prescribed curve, all tokens paid out must be distributed among the batch's orders, and each order's spend is capped by its reported budget. Mechanisms must also be individually rational and anonymous — outcomes depend only on reported values, budgets, and relative ordering, not on identity.
The strategy space is deliberately hostile. A strategic user may misreport value and budget, create any finite number of pseudonymous identities, and submit bids in either direction regardless of its true type. A strategic builder may reorder arbitrarily and inject Sybil bids, and a builder with intrinsic trading demand may also act as a user (a builder-as-user).
To make the proof tractable, the authors rescale units so the initial marginal price is 1 (F′(0) = 1, with X-payout bounded by K := x₀/σ₀ > 0, and F(z) ≤ z by concavity). In this normalization an order is strictly eligible exactly when its reported value is below 1, and wLE forces zero-value orders to spend their full reported budget.
The impossibility argument proceeds by contradiction. First, a lemma pins down that eligible orders on the minority side trade their full budgets at the initial price, fixing those outcomes; because wLE is a weak assumption, this requires incremental intermediate claims rather than a direct argument. Then, assuming (without loss) that sell-X is the minority side, a strategic sell-Y user over-reports its budget and simultaneously submits an opposite-direction order to buy back the excess Y at the initial price, keeping its net sale within its true budget. Splitting the inflated budget evenly between two sell-Y bids that each match another user's zero-value report produces three identical sell-Y orders; economic symmetry then bounds the strategic user's share of the trading loss, where trading loss means the shortfall in X received relative to trading the same amount of Y at the initial price. The final contradiction uses a profile with two identical sell-Y orders and a small sell-X order: the loss bound plus value and budget truthfulness yields a large utility guarantee for either sell-Y user via a deviation reporting value zero, IC forces that guarantee to hold for both under truthful reporting, and their sum exceeds the total utility available from the opposite-side order and the pool.
For the three feasibility results, the authors isolate the tractable part of the design problem. A two-sided matching step identifies the side with the smaller aggregate reported budget (converted to common units at the initial exchange rate) and fills those eligible orders at the initial price, which fixes their outcomes. The rest is a one-sided allocation problem on an augmented compensation curve that accounts for the matched trades before following the pool's pricing curve. A lemma (number not visible in the provided text) establishes the settlement guarantees; incentives still require analyzing the complete two-sided mechanism.
Why This Matters
Impact on research. The paper closes the main open question left by Chan, Wu, and Shi and by Li, Shi, and Zhang, and it does so in the strongest form: the obstruction is not censorship or congestion, the two usual suspects. It also reframes the design space. Rather than asking whether AMMs can be MEV-resilient in general, the field can now ask which single relaxation — restricting Sybil identities, restricting aggregate budget over-reporting, making values public, permitting surplus redistribution, or upgrading to fair sequencing — is cheapest to obtain. The tightness results make each of these options a precise, comparable design choice, and the "minority side trades at the initial price, remainder handled one-sidedly" decomposition gives later work a standard skeleton to build on.
Real-world applications:
- Sandwich-attack resistance for Uniswap-style pools. Because IC in this literature rules out risk-free profit from inserting front-running and back-running orders, the impossibility says that constant-product AMMs run on a single-builder block cannot be made simultaneously sandwich-proof and weakly locally efficient while keeping exact settlement.
- Consensus-layer roadmap decisions. Ethereum's planned adoption of FOCIL and other censorship-resistance efforts motivate the question of whether protecting inclusion suffices. The paper's answer is no: censorship resilience alone does not reconcile IC with wLE, whereas fair sequencing does, which bears directly on whether to invest in fair-ordering protocols and decentralized sequencers.
- Batch-auction and uniform-price clearing design. The analysis clarifies what uniform pricing buys: it removes within-batch internal arbitrage, but IC still fails when the clearing price depends on the batch, so builders can profit by selectively excluding orders.
- MEV mitigation tooling. Encrypted mempools (keeping transaction contents private until finalization), private submission services such as Flashbots Protect, order-flow auctions such as MEV-Share, and proposer-builder separation via MEV-Boost all address MEV under specific information and trust assumptions; this paper delineates what an application-layer mechanism can and cannot promise given the consensus guarantees underneath it.
Industry relevance. For AMM and DEX teams, the takeaway is that exact settlement is the expensive constraint: a design that redirects some pool output to a broader community (as Otter does) can be IC and welfare-maximizing under censorship resilience, while one that insists every output token goes to the current batch cannot. For L2 sequencer and fair-ordering teams, the separation result is an argument that ordering guarantees are a distinct and non-substitutable product from inclusion guarantees.
Future Directions
-
Stronger efficiency under plain consensus. The paper's positive results target wLE, a deliberately weak notion; Li, Shi, and Zhang show local efficiency for all orders and Pareto optimality are incompatible with IC even under fair sequencing, and a recent work achieves weak Pareto optimality for eligible bids when aggregate budget over-reporting is banned. Whether any stronger notion survives under plain consensus with a different restriction is open.
-
Relaxing anonymity or determinism. The impossibility is stated for deterministic, anonymous mechanisms. Whether randomization, identity-aware rules, or reputation systems can restore IC plus wLE under censorship resilience is not settled by this paper, and the anonymity assumption is noted as both desirable and satisfied by real-world AMMs, making its role worth probing.
-
Economically compelling designs from the three restricted settings. The authors describe their three mechanisms as "substantially simpler" and "intended primarily to establish feasibility rather than to offer an economically compelling design." Turning the equal-share, proportional, and layered constructions into designs with good welfare or liquidity properties is a natural next step.
-
Beyond two assets and beyond exact settlement. The model is two-asset with a single concave curve. Extending the tightness characterization to multi-asset pools, or combining surplus redistribution with the three strategy restrictions to see whether the two relaxations compose or overlap, is left open.
Target Audience
Mechanism-design and algorithmic game-theory researchers working on blockchains; consensus-layer and protocol researchers deciding between fair sequencing, censorship resilience, and related guarantees; AMM and DEX designers who need to know which settlement and pricing assumptions are load-bearing; and graduate students in computer science or economics looking for a worked example of a tight impossibility-plus-feasibility characterization. Readers without a game-theory background will find the model and theorem statements accessible but the proofs demanding.
Authors’ abstract
Miner extractable value (MEV) poses a persistent challenge for blockchain ecosystems. It arises from the fact that a block builder can exploit its unilateral control over transaction inclusion and ordering to extract profit, potentially at users' expense. Recent work has explored consensus-aided mechanism design to mitigate MEV. This approach combines generic consensus-layer guarantees with application-layer mechanisms that use these guarantees to address…
The full abstract on arXiv