Research
A Trilemma in AMM Mechanism Design
AMMs let users trade crypto-assets directly against a smart contract, but users are routinely forced into bad prices by front-running and back-running. Prior work by Chan, Wu, and Shi showed that…

- arXiv
- 2605.27602
- Published
- 2026-05-26
- Authors
- Yuhao Li, Elaine Shi, Mengqian Zhang
AI summary
Overview
Research area: Algorithmic game theory applied to decentralized finance — specifically, incentive-compatible mechanism design for Automated Market Makers (AMMs) and the mitigation of Miner Extractable Value (MEV).
Technical level: Advanced. The paper is written for readers comfortable with formal mechanism-design definitions (incentive compatibility, Pareto optimality, individual rationality), utility orderings, and impossibility proofs. The prose is accessible, but the results are theorem-and-proof style.
Scope (one sentence): The paper maps the design space of two-asset AMM mechanisms and proves that incentive compatibility, weak local efficiency, and uniform pricing cannot all hold at once.
What This Paper Is About
AMMs let users trade crypto-assets directly against a smart contract, but users are routinely forced into bad prices by front-running and back-running. Prior work by Chan, Wu, and Shi showed that, assuming the blockchain gives weak sequencing fairness, one can design an AMM mechanism that is fully incentive compatible (IC) — meaning no user or miner gains by lying, reordering, or inserting orders. This paper asks what that strong IC guarantee costs: can an IC mechanism also be efficient, and can it also price every order in a batch at the same exchange rate?
Key Contributions
-
A model and vocabulary for comparing AMM mechanisms. The paper formalizes orders of the form (t, r, q, α), a potential-function-based pool, two adversary models (the plain model and the weak fair-sequencing model), and utility rankings — including a total ordering for Sell(X) orders and a partial ordering for Buy(X) orders that compares outcomes under both extreme interpretations of a user's value for units beyond their stated demand q. On top of this it defines Pareto optimality (PO), local efficiency (LE), weak local efficiency (wLE), and uniform pricing (UP).
-
IC is incompatible with strong efficiency. Theorem 1 / Theorem 3 shows that no AMM mechanism can be simultaneously IC and LE (and therefore not IC and PO either, since PO implies LE), and this holds even under weak fair sequencing. The proof builds a scenario with two Buy(X) orders and one Sell(X) order to show that LE forces prices that hand some subset a risk-free gain in both assets, contradicting the arbitrage resilience that IC is known to imply.
-
A trilemma among IC, wLE, and UP. Theorem 2 states that no mechanism achieves IC, weak local efficiency, and uniform pricing simultaneously — again even under weak fair sequencing — while any two are attainable: wLE + UP in the plain model, and IC + wLE or IC + UP under weak fair sequencing.
-
LE is too strong in a second, independent way. The paper proves that UP and LE are also mutually impossible, which is presented as further justification for relaxing LE to wLE, given that UP + wLE is achievable.
Main Findings
-
IC precludes LE, PO, and hence arbitrage resilience through efficiency. Under LE, all orders in the constructed scenario must be fully executed, which forces the average buy price strictly below the average sell price; for sufficiently small ε, one Sell(X) plus one Buy(X) order then yields strictly positive net gain in both X and Y. Since arbitrage resilience forbids exactly that, and IC implies arbitrage resilience (Fact 2.1 of Chan et al.), IC and LE cannot coexist.
-
This is a mathematical justification for prior work's weaker efficiency guarantee. The authors note that their impossibility result explains why the Chan–Wu–Shi mechanism only satisfies a notion slightly weaker than LE, which they name weak local efficiency.
-
The trilemma is exhaustive in the sense that any two properties are achievable, but all three are not. wLE + UP is feasible in the plain model; IC + wLE and IC + UP are feasible in the weak fair-sequencing model. The impossibility side holds in both models, including the weak fair-sequencing one.
-
Uniform pricing is not a substitute for incentive compatibility. Although UP eliminates internal arbitrage and sandwich attacks, that is a weaker guarantee: it only stops players with zero intrinsic demand from risk-free profit. The paper points to a natural mechanism described in its Section 4.3 that is UP + wLE but not IC, and to miner profit extraction from batch auctions studied by Zhang et al.
-
LE is incompatible with UP as well. A separate impossibility shows that LE is stringent from a second angle, since it cannot be combined with UP either; relaxing to wLE sidesteps this because UP + wLE is feasible.
-
Weak local efficiency restricts efficiency demands to "eligible" orders only. A Buy(X) or Sell(Y) order is eligible if its rate is no less than the initial market price; a Sell(X) or Buy(Y) order is eligible if its rate is no higher. wLE applies the LE conditions only to these orders, which is exactly the gap that lets IC exist.
-
The paper reports no empirical evaluation. There are no datasets, benchmarks, simulations, or measured performance numbers; the results are formal feasibility and impossibility theorems. The only figures cited are contextual, drawn from prior work: $15 billion in assets held by the top six AMMs as of March 2021, and more than 85% of blocks built by the top two block producers in a cited measurement study.
-
One central question is left explicitly open. Whether IC + wLE can be achieved in the plain model, without the weak fair-sequencing assumption, is not resolved here.
Methodology in Plain English
The authors proceed by defining rather than by experimenting. They first write down precisely what an order is, what a trade against a pool is allowed to do, and what makes an outcome legal — no order may be over- or under-filled, no order may gain one asset without losing the other, no order may trade worse than its stated rate, and the pool must respect its pricing curve before and after the whole batch.
They then pin down the adversary. In the plain model, a strategic user or miner can post fake or misreported orders, censor honest orders, rewrite the α field that encodes arrival-time and position, and choose its move after seeing everyone else's orders. In the weak fair-sequencing model, the adversary can still post multiple orders, misreport values and quantities, delay its own submission (α′ ≥ α), and act after observing honest orders — so front-running is still possible — but it can no longer censor others or forge their α values.
With the model fixed, the strategy is to find small, carefully chosen order batches where a desirable property forces prices that contradict IC's prohibition on risk-free profit. Where a property turns out to be achievable, the paper shows which combinations survive in which model, yielding the any-two-of-three structure. The utility definitions are the delicate part: for Buy(X) orders the authors deliberately use a partial ordering, because the model does not specify how a user values units beyond their stated demand q, so they require an outcome to be at least as good under both the "excess units worth r" reading and the "excess units worth 0" reading.
Why This Matters
Impact on research. The paper converts a loose intuition — "strong incentive guarantees must cost something" — into named impossibility theorems and a clean combinatorial structure (any two of IC, wLE, UP). It also sharpens the boundary between consensus-level and application-level MEV mitigation: consensus-layer impossibility results exist, and this work shows the application layer has its own hard limits rather than being a clean escape hatch. It additionally gives a formal reason why earlier IC constructions settled for less efficiency than one might want, and it flags the plain-model question as the next target.
Real-world applications:
- Constant-product AMMs (e.g., Uniswap-style pools) — designers choosing between a pricing rule that is fully manipulation-proof and one that clears more of users' demand now know the trade-off is not free.
- Batch auctions and uniform-clearing venues — the result warns that pricing an entire batch at one rate does not by itself deliver the strong incentive guarantee these designs are often marketed as providing.
- Blockchains with decentralized sequencers — the weak fair-sequencing model is meant to capture protocols like these, and the paper identifies exactly which property combinations they can and cannot support.
- Transaction fee mechanism design — since TFM work is application-agnostic and cannot capture sequence-dependent utility, these AMM-level results complement it by treating the application semantics explicitly.
Industry relevance. MEV extraction has pushed block building toward concentration, with the paper citing a measurement that more than 85% of blocks come from the top two block producers, and private deals for favorable in-block positions. Protocol teams and auditors can use the trilemma as a checklist: no deployed design should claim IC, wLE, and UP together.
Future Directions
-
IC + wLE in the plain model. The authors identify this as the main open question left by their trilemma and by the Chan–Wu–Shi mechanism, and promise further discussion in their Section 5.
-
Characterizing the remaining design space. The trilemma covers three properties; whether other natural desiderata (variants of efficiency, alternative fairness or pricing notions) compose with IC in the same any-two-of-three pattern is not settled here.
-
Quantifying the price of IC. The paper establishes impossibility, not magnitude. How much social efficiency must be given up for IC in practice — as a function of the pricing curve and order mix — is a natural follow-up that this work does not measure.
-
Extending beyond the two-asset, potential-function setting. The model assumes a pool of two assets whose state follows an increasing, differentiable, concave potential function, and requires the curve to hold only at the end of a batch. Whether the same trilemma survives multi-asset pools or per-order curve enforcement (the stricter Ferreira–Parkes regime) is left open.
Target Audience
Researchers in algorithmic game theory and mechanism design; blockchain and DeFi protocol designers and auditors; economists working on market microstructure and batch auctions; and graduate students who already know the basics of AMMs and incentive compatibility and want a rigorous account of what can and cannot be simultaneously guaranteed at the smart-contract layer. Readers looking for empirical results, benchmarks, or implementable protocol specifications will not find them here.
Authors’ abstract
Blockchains have popularized the Automated Market Makers (AMMs), where users trade crypto-assets directly with a smart contract, governed by a pricing function embedded in the contract's code. Today, users of AMMs are often forced to accept unfavorable prices due to widespread front-running and back-running attacks, commonly known as Miner Extractable Value (MEV). Several earlier works show impossibility results suggesting that completely removing MEV at the consensus layer…
The full abstract on arXiv