Non-atomic MEV: Mechanisms & Implications

• Non-atomic MEV is an extraction technique spanning multiple transactions or blocks, introducing execution risks absent in atomic MEV.
• It includes strategies such as CEX–DEX arbitrage, cross-chain arbitrage, and time-bandit attacks that rely on real-time off-chain data and precise fee management.
• Empirical evidence on Ethereum shows non-atomic MEV accounts for around 30% of DEX volume, influencing block construction, market concentration, and security.

Non-atomic maximal extractable value (MEV) denotes extraction strategies whose profit depends on coordinating actions that cannot be bundled atomically into one transaction or one block, and therefore expose the searcher to non-zero risk of partial failure. In Ethereum-centric decentralized finance, the canonical case is non-atomic arbitrage between an on-chain DEX price and an external price on a CEX or on another blockchain, executed as an on-chain swap followed by an off-chain counter-trade (Heimbach et al., 2024). In the broader MEV literature, the category extends beyond CEX–DEX arbitrage to cross-domain and multi-block strategies, including cross-chain arbitrage, time-bandit attacks, and certain cross-layer sandwiches (Mancino et al., 8 Mar 2026). In multiple concurrent proposer (MCP) blockchains, non-atomicity also arises from protocol structure itself: multiple proposers publish data-available blocks before final execution order is resolved, turning MEV into races, auctions, and censoring games that unfold across concurrently published blocks rather than within a single sequential block (Landers et al., 17 Nov 2025).

1. Definition, scope, and distinction from atomic MEV

Atomic MEV consists of extraction opportunities that can be realized all-or-nothing within one block or one transaction. The Ethereum literature identifies cyclic arbitrage between DEX pools on the same chain, sandwich attacks exploiting a victim’s pending swap, and liquidations on lending protocols as the standard atomic cases (Heimbach et al., 2024). The SoK formalizes this as

$$\mathrm{MEV}_{\text{atomic}}(s_0,A)=\max_{\sigma\in\Sigma(A)} \mathrm{profit}(s_0,\sigma),$$

where $\Sigma(A)$ is the set of orderings of subsets of candidate actions whose execution is guaranteed to be all-or-nothing in one block or transaction (Mancino et al., 8 Mar 2026).

Non-atomic MEV, by contrast, is realized through a sequence of actions that may be split across multiple blocks and/or domains and that admit non-zero risk of partial failure. In the single-domain CEX–DEX example, one leg might be “buy on DEX” and the second “sell on CEX,” with neither leg bilked into one atomic on-chain transaction (Mancino et al., 8 Mar 2026). The SoK writes this as

$$\mathrm{MEV}_{\text{nonatomic}}^{\text{single}}(s_0;A_1,A_2) =\max_{\sigma_1\in\Sigma(A_1)} \left[ \mathrm{profit}(s_0,\sigma_1)+ \max_{\sigma_2\in\Sigma(A_2)} \mathrm{profit}(s_1,\sigma_2) \right],$$

subject to $s_1=\mathrm{state\_after}(s_0,\sigma_1)$ (Mancino et al., 8 Mar 2026).

The SoK divides non-atomic MEV along two axes, atomicity and scope. In the non-atomic, single-domain quadrant, it places CEX–DEX arbitrage and time-bandit attacks. In the non-atomic, cross-domain quadrant, it places sequence-independent arbitrage (SIA), sequence-dependent arbitrage (SDA), and cross-layer sandwiches (Mancino et al., 8 Mar 2026). A common misconception is to treat non-atomic MEV as merely a less convenient form of atomic arbitrage. The literature instead treats it as a distinct class because execution cannot be guaranteed by a single inclusion-and-ordering decision, and because risk, latency, private order flow, and cross-domain state reconstruction become first-order considerations (Heimbach et al., 2024).

2. Economic mechanism and formal models

For Ethereum DEXes, the principal formalization in "Non-Atomic Arbitrage in Decentralized Finance" models non-atomic arbitrage over constant-function market makers (CFMMs). For assets $X,Y$ with reserves $(x,y)$ and liquidity parameter $L$, Uniswap-V2 satisfies

$x\,y=L^2,\qquad P=\frac{y}{x}.$

Let $P_{\text{on}}$ denote the DEX marginal price, $\tilde P_{\text{off}}$ the off-chain average price, and $\Sigma(A)$0 (Heimbach et al., 2024).

A searcher trades $\Sigma(A)$1 of $\Sigma(A)$2 on-chain, paying a fee fraction $\Sigma(A)$3 to liquidity providers. The on-chain leg yields

$\Sigma(A)$4

The searcher then sells the acquired $\Sigma(A)$5 off-chain at price $\Sigma(A)$6 minus trading fees $\Sigma(A)$7, receiving

$\Sigma(A)$8

At optimal trade size, the post-trade marginal price on-chain equals $\Sigma(A)$9, and profitable arbitrage requires

$$\mathrm{MEV}_{\text{nonatomic}}^{\text{single}}(s_0;A_1,A_2) =\max_{\sigma_1\in\Sigma(A_1)} \left[ \mathrm{profit}(s_0,\sigma_1)+ \max_{\sigma_2\in\Sigma(A_2)} \mathrm{profit}(s_1,\sigma_2) \right],$$0

The net profit in units of $$\mathrm{MEV}_{\text{nonatomic}}^{\text{single}}(s_0;A_1,A_2) =\max_{\sigma_1\in\Sigma(A_1)} \left[ \mathrm{profit}(s_0,\sigma_1)+ \max_{\sigma_2\in\Sigma(A_2)} \mathrm{profit}(s_1,\sigma_2) \right],$$1 is

$$\mathrm{MEV}_{\text{nonatomic}}^{\text{single}}(s_0;A_1,A_2) =\max_{\sigma_1\in\Sigma(A_1)} \left[ \mathrm{profit}(s_0,\sigma_1)+ \max_{\sigma_2\in\Sigma(A_2)} \mathrm{profit}(s_1,\sigma_2) \right],$$2

The paper reports that a typical plot of $$\mathrm{MEV}_{\text{nonatomic}}^{\text{single}}(s_0;A_1,A_2) =\max_{\sigma_1\in\Sigma(A_1)} \left[ \mathrm{profit}(s_0,\sigma_1)+ \max_{\sigma_2\in\Sigma(A_2)} \mathrm{profit}(s_1,\sigma_2) \right],$$3 shows profit growing rapidly once $$\mathrm{MEV}_{\text{nonatomic}}^{\text{single}}(s_0;A_1,A_2) =\max_{\sigma_1\in\Sigma(A_1)} \left[ \mathrm{profit}(s_0,\sigma_1)+ \max_{\sigma_2\in\Sigma(A_2)} \mathrm{profit}(s_1,\sigma_2) \right],$$4 exceeds the combined fee threshold (Heimbach et al., 2024).

Operationally, the strategy is a two-leg sequence: an on-chain swap on an Ethereum DEX and an off-chain counter-trade at a better quoted price. Because the two legs cannot be bundled atomically on one chain, searchers face execution risk. In PoS Ethereum with PBS, they privately relay the on-chain leg to block builders to avoid public mempool exposure and pay competitive fees, via priority gas fees or direct coinbase transfers, so that the swap is placed at the top of the block (Heimbach et al., 2024). The SoK’s stylized procedure is correspondingly simple: fetch the current DEX and CEX prices, act only when the price gap exceeds a threshold $$\mathrm{MEV}_{\text{nonatomic}}^{\text{single}}(s_0;A_1,A_2) =\max_{\sigma_1\in\Sigma(A_1)} \left[ \mathrm{profit}(s_0,\sigma_1)+ \max_{\sigma_2\in\Sigma(A_2)} \mathrm{profit}(s_1,\sigma_2) \right],$$5, execute the first leg, wait for confirmation, and then execute the second leg; expected profit, ignoring fees, is $$\mathrm{MEV}_{\text{nonatomic}}^{\text{single}}(s_0;A_1,A_2) =\max_{\sigma_1\in\Sigma(A_1)} \left[ \mathrm{profit}(s_0,\sigma_1)+ \max_{\sigma_2\in\Sigma(A_2)} \mathrm{profit}(s_1,\sigma_2) \right],$$6 when $$\mathrm{MEV}_{\text{nonatomic}}^{\text{single}}(s_0;A_1,A_2) =\max_{\sigma_1\in\Sigma(A_1)} \left[ \mathrm{profit}(s_0,\sigma_1)+ \max_{\sigma_2\in\Sigma(A_2)} \mathrm{profit}(s_1,\sigma_2) \right],$$7 (Mancino et al., 8 Mar 2026).

3. Detection, data, and identification methodology

The Ethereum measurement study analyzes every swap on Ethereum’s five largest DEXes—Uniswap V2, Uniswap V3, Curve, Balancer, and Sushiswap—from the Merge, block 15 537 393, until 31 October 2023, for a dataset of 77 019 583 swaps (Heimbach et al., 2024). The on-chain dataset is enriched with MEV labels from Zeromev, execution-layer rewards from blocks, PBS relay data from eleven public relays, Ethnet mempool timestamps from Mempool Guru, and off-chain price data from Binance 1 s candles and CoinMarketCap daily data (Heimbach et al., 2024).

Because only the on-chain leg is visible on Ethereum, the study identifies likely non-atomic arbitrage swaps through five heuristics. The transaction must be a simple swap: exactly one swap in a standard DEX pool, not already labeled as sandwich, cyclic arbitrage, or liquidation, and with gas usage $$\mathrm{MEV}_{\text{nonatomic}}^{\text{single}}(s_0;A_1,A_2) =\max_{\sigma_1\in\Sigma(A_1)} \left[ \mathrm{profit}(s_0,\sigma_1)+ \max_{\sigma_2\in\Sigma(A_2)} \mathrm{profit}(s_1,\sigma_2) \right],$$8. It must be privately submitted, meaning not seen in the public mempool before block propagation. It must show a significant tip, either a coinbase transfer to the fee recipient or a priority fee of at least 1 Gwei. It must be top-of-pool, meaning the first swap in that direction in the pool for that block, or have the same recipient as prior swaps. Finally, it must involve a liquid off-chain market, operationalized by both tokens appearing among top CEX-listed assets such as ETH, BTC, USDC, USDT, and DAI (Heimbach et al., 2024).

Applied conservatively, these heuristics capture 58–88% of swaps by previously known non-atomic arbitrage searchers—beaversearcher, rsyncsearcher, and mantasearcher—while flagging only approximately 2.8% of all swaps. Clustering the flagged swaps by originating address yields eleven major searchers (Heimbach et al., 2024). The SoK places these methodological choices in a wider measurement context: non-atomic MEV estimation requires distinguishing potential from realized MEV, stitching together CEX order-book snapshots, on-chain DEX logs, bridge events, and off-chain API data, and handling partial failures, reorgs, and heterogeneous finality models. It also notes that, as Chi et al. argue in the “Remeasure” discussion, small changes in detection windows or fee-model assumptions can swing estimated realized extractable value by tens of percent (Mancino et al., 8 Mar 2026).

4. Empirical prevalence on Ethereum

The measured scale of non-atomic arbitrage on Ethereum is large. The study identifies \$\mathrm{MEV}_{\text{nonatomic}}^{\text{single}}(s_0;A_1,A_2) =\max_{\sigma_1\in\Sigma(A_1)} \left[ \mathrm{profit}(s_0,\sigma_1)+ \max_{\sigma_2\in\Sigma(A_2)} \mathrm{profit}(s_1,\sigma_2) \right],$9460 billion volume over the observation period (<a href="/papers/2401.01622" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Heimbach et al., 2024</a>). The paper’s abstract characterizes this as “more than a fourth of the volume on Ethereum&#39;s biggest five DEXes,” and the SoK restates the result as approximately 30% of DEX trading from the Merge to October 2023 (<a href="/papers/2603.07716" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Mancino et al., 8 Mar 2026</a>).</p> <p>The activity level is similarly substantial. Average daily non-atomic arbitrage occurrences are 5 216, comparable to 5 219 average daily sandwich attacks, 1 245 cyclic arbitrages, and 14 liquidations. Average daily on-chain fee value is 112 ETH, at least \$200 K (Heimbach et al., 2024). These values place non-atomic extraction within the same order of magnitude as the most studied atomic MEV categories, rather than at the periphery of DEX activity.

Market participation is highly concentrated. Eleven searchers account for 80% of non-atomic volume, and the top two searchers alone account for 49% (Heimbach et al., 2024). Trade-size distribution has median approximately \$s_1=\mathrm{state\_after}(s_0,\sigma_1)$010 M+, with distinct peaks per searcher reflecting strategic sizing. Token-pair concentration is also strong: 85–99% of non-atomic trades involve ETH, BTC, USDC, USDT, and DAI, accounting for 90–94% of volume (Heimbach et al., 2024). This suggests that non-atomic arbitrage is tightly linked to highly liquid benchmark assets rather than broadly dispersed across long-tail pairs.

Temporal variation tracks market stress and trading hours. The study reports spikes during the FTX collapse in November 2022 and the USDC depeg in March 2023. It also finds high correlation with price volatility: ETH volatility versus non-atomic volume has Pearson $s_1=\mathrm{state\_after}(s_0,\sigma_1)$1 with $s_1=\mathrm{state\_after}(s_0,\sigma_1)$2, and BTC volatility versus non-atomic volume has $s_1=\mathrm{state\_after}(s_0,\sigma_1)$3 with $s_1=\mathrm{state\_after}(s_0,\sigma_1)$4. Daily hour-of-day and weekday heatmaps show peaks at the US market open at 14:30 UTC, the Asian open at 00:00 UTC, the European open at 08:00 UTC, and around 18:00 UTC on Wednesdays during FOMC announcements (Heimbach et al., 2024).

5. Block construction, concentration, and security implications

In PoS Ethereum, PBS introduces a structural connection between non-atomic arbitrage and block construction. Validators choose from privately submitted blocks assembled by specialized builders that bundle searcher transactions and public mempool order flow (Heimbach et al., 2024). The study identifies four “HFT builders” with integrated non-atomic arbitrage searchers: beaverbuild with beaversearcher$s_1=\mathrm{state\_after}(s_0,\sigma_1)$5, rsyncbuilder with rsyncsearcher$s_1=\mathrm{state\_after}(s_0,\sigma_1)$6, mantabuilder with mantasearcher, and builder$s_1=\mathrm{state\_after}(s_0,\sigma_1)$7 with builder$s_1=\mathrm{state\_after}(s_0,\sigma_1)$8searcher (Heimbach et al., 2024).

For each builder $s_1=\mathrm{state\_after}(s_0,\sigma_1)$9 and searcher $X,Y$0, the study compares the daily share of blocks won by $X,Y$1 with the daily share of non-atomic volume by $X,Y$2 and finds strong positive correlations, often $X,Y$3 (Heimbach et al., 2024). On days when ETH or BTC volatility exceeds the 99.9th percentile, HFT builders win more than 75% of blocks, and non-atomic swaps consume more than 80% of block value and more than 10% of gas limit. In October 2023, the correlation between HFT builder share and ETH volatility is $X,Y$4 with $X,Y$5; for BTC volatility it is $X,Y$6 with $X,Y$7 (Heimbach et al., 2024). The paper interprets this as evidence that, during high off-chain price volatility and high non-atomic MEV, a small set of builders with integrated searchers come to dominate block production.

The security implications follow directly from block value concentration. The paper highlights consensus instability through time-bandit attacks: large block values, often more than $X,Y$8 the PoS consensus reward of approximately 0.04 ETH, can incentivize attackers to privately fork past blocks to capture MEV themselves (<a href="/papers/2401.01622" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Heimbach et al., 2024</a>). It also identifies increased base fees and congestion, since overfilled blocks during high volatility force higher base fees and priority fees in subsequent blocks; centralization risk, since HFT builders gain disproportionate market share; and collateral harm, since liquidity providers face <a href="https://www.emergentmind.com/topics/latent-visual-reasoning-lvr" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">LVR</a> and regular traders suffer from front-running and slippage (<a href="/papers/2401.01622" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Heimbach et al., 2024</a>). The SoK adds an unresolved welfare dimension: non-atomic arbitrage is price-aligning, and therefore value-creating in one sense, but also spam-generating, and therefore a negative externality in another (<a href="/papers/2603.07716" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Mancino et al., 8 Mar 2026</a>).</p> <h2 class='paper-heading' id='cross-domain-extensions-mcp-variants-and-mitigations'>6. Cross-domain extensions, MCP variants, and mitigations</h2> <p>The SoK situates non-atomic MEV within a broader historical shift from single-domain MEV to cross-domain extraction. It cites empirical examples beyond Ethereum mainnet CEX–DEX arbitrage: Wu et al. identify \$X,Y$9213 million non-atomic arbitrage on Polygon; Gogol et al. identify 500 000 unexploited non-atomic arbitrages across rollups persisting 10–20 blocks; Oz et al. report 242 535 <a href="https://www.emergentmind.com/topics/cross-chain-arbitrages" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">cross-chain arbitrages</a>, \$(x,y)$08.65 million net profit; and Mancino et al. detect only 10 genuine 3- or 4-hop cross-chain arbitrages in 2.4 billion transactions, with 7–12 minute execution times and 60% profitability (Mancino et al., 8 Mar 2026). These findings indicate that “non-atomic MEV” is not a synonym for one Ethereum microstructure pattern, but a family of extraction problems whose incidence depends on market design, domain boundaries, and execution constraints.

A distinct protocol-native version appears in MCP blockchains. There, each tick may see multiple proposers publish blocks to the DA layer, while a deterministic merge rule resolves relative ordering later. Because transaction content becomes public before merge, other proposers can react intra-tick, and MEV becomes non-atomic by construction (Landers et al., 17 Nov 2025). The MCP model introduces same-tick duplicate steals, proposer-to-proposer auctions, and timing races driven by proof-of-availability latency. Its hazard-normalized model defines a delay envelope

$(x,y)$1

with immediate-inclusion cutoff

$(x,y)$2

so that high tips induce immediate inclusion while lower tips sustain interior delay equilibria (Landers et al., 17 Nov 2025).

Mitigations differ by setting. For Ethereum non-atomic arbitrage, the paper sketches separate top-of-block extraction and shorter block time or migration to faster layers as protocol-level directions. The first decouples the PBS auction into top-of-block slots reserved for high-MEV arbitrage and the remaining block body; the second shrinks the time window for price divergence, reducing $(x,y)$3 and therefore non-atomic arbitrage profit $(x,y)$4 (Heimbach et al., 2024). The SoK adds private order-flow auctions and Flashbots RPCs, commit-reveal on DEX orders, shared sequencers or atomic cross-rollup execution, intent-based protocols, ZK-based bridges, failure-cost mechanisms, and Timelock Shield (Mancino et al., 8 Mar 2026). In MCP designs, the proposed mitigations are duplicate-aware tip splitting and deterministic priority-DAG scheduling, supplemented by parameter choices such as higher proof-of-availability co-signature rate, lower required shape, nonzero base fee, and transaction-fee-mechanism calibration around $(x,y)$5 (Landers et al., 17 Nov 2025).

The open research agenda is correspondingly broad. The SoK highlights standardized metrics for cross-domain MEV, public datasets of linked on-chain, off-chain, and bridge events, graph-based or ML detection of $(x,y)$6-hop arbitrages, reliable reconstruction of partial orders across 10 or more domains, models of sequencer incentives and collusion, welfare analysis, bridge-security interactions, and regulatory implications when one leg occurs on a CEX in one jurisdiction and another leg on a DEX in another (Mancino et al., 8 Mar 2026). Fully preventing non-atomic MEV is described as extremely difficult because it relies on real-time off-chain quotes and, more generally, on value differentials that arise across execution environments rather than within a single ordering surface (Heimbach et al., 2024).