Research
Capital Costs Protect the Network: Mining Gap Dynamics in Proof-of-Work Blockchains Under a Stochastic Mempool
Derives a unique long-run equilibrium hashpower for fee-dependent PoW where capital cost, not difficulty alone, bounds the mining gap and stops rounds stalling entirely.

- arXiv
- 2610.22451
- Published
- 2026-10-15
- Authors
- Hasan Selçuk Kılıç, Nail Akar
AI summary
Overview
Research area: Blockchain security and cryptoeconomics — specifically Proof-of-Work (PoW) incentive design, transaction-fee markets, and mempool dynamics, with tools borrowed from queueing theory and non-atomic game theory.
Technical level: Advanced. The paper uses measure-theoretic non-atomic games, Lindley-type recursions for queueing, stationary distribution derivations, and a stability analysis of Bitcoin's difficulty-adjustment fixed point. The introduction and the policy discussion are readable on their own, but the derivations are not beginner material.
Scope: The paper derives a closed-form stationary distribution for the length of the "mining gap" — the idle period at the start of each round when honest miners wait for enough high-fee transactions to accumulate — under a persistent, stochastic two-tier mempool, and uses it to analyze equilibrium hashpower, the difficulty-adjustment mechanism, and Bitcoin's long-run security under the halving schedule.
What This Paper Is About
As block subsidies halve roughly every four years, miners increasingly depend on transaction fees. Because each new block clears the highest-fee waiting transactions, the start of the next round often offers too little fee revenue to cover miners' operating costs, so honest miners switch off until fees build up again. This "mining gap" halts the growth of the honest chain while adversaries are free to keep mining, and prior analytical models assumed the fee backlog resets to zero every block, so nobody had derived the gap's length. The paper builds a stochastic mempool model that preserves the backlog across rounds, derives the gap's distribution and expectation, and uses it to show how capital costs bound equilibrium hashpower and therefore protect the network.
Key Contributions
-
A myopic activation rule proved weakly dominant. Mining is formalized as long-run profit-rate maximization over an infinite horizon by a non-atomic population. The authors show that activating hardware exactly when the instantaneous reward rate covers the operating cost is a weakly dominant strategy for every miner — despite the rule ignoring future rounds — because each miner is negligible and cannot move the backlog, the round duration, or the difficulty.
-
A closed-form stationary distribution of the mining gap. The high-fee backlog is modeled as a Lindley-type recursion with a floor, where the floor captures that mining does not start until enough Type-1 transactions accumulate. The authors first give the transient round-by-round distribution, then derive the stationary distribution of the gap and its expectation for each winning probability.
-
A characterization of the invariant winning probability and its stability. Following Kim and Seo (2026), the paper identifies the winning probability at which average round length equals the protocol target under the stochastic mempool model, and analyzes when that fixed point is stable under small disturbances.
-
A security and equilibrium analysis linking hashpower, costs and the gap. The paper shows the expected gap is strictly increasing in aggregate hashpower, derives a free-entry equilibrium hashpower, and shows that a positive capital cost keeps that equilibrium finite and prevents the gap from consuming the entire round. It then estimates the model on Bitcoin data from 2026 and follows the halving schedule.
Main Findings
-
The gap grows strictly with hashpower. Under Bitcoin's difficulty-adjustment mechanism, expected gap length is strictly increasing in the hashpower active on the network. As hashpower grows without bound, the difficulty rises so blocks still arrive at target pace, each unit of hashpower earns less per unit time, and the expected gap approaches the full length of a round.
-
Unbounded hashpower means an ever-smaller attacker share. As the gap approaches the whole round, an attacker who mines continuously needs an ever smaller share of aggregate hashpower to outpace the honest chain. Carlsten et al. (2016) reached a similar conclusion in a simplified model with no fee carryover.
-
Capital costs cap equilibrium hashpower. Free entry drives hashpower to a unique long-run equilibrium where revenue covers both operating and capital costs. A positive capital cost bounds this equilibrium, which in turn stops the gap from occupying the entire round — hence the paper's title.
-
Subsidy and cost parameters move the gap in opposite directions. A larger block subsidy shortens the gap and raises equilibrium hashpower. A higher ratio of operating to capital cost — for example from cheaper mining hardware — brings the gap forward, lengthens it, and can make the invariant winning probability unstable.
-
Under free entry with 2026 Bitcoin data, the gap arrives decades away. With estimated costs and free entry, a gap forms only after several more halvings: 2044 or 2056 in the authors' scenarios, or not before 2070 if fees fall in bitcoin. It lasts about two and a half minutes per ten-minute round by 2070, and the invariant winning probability stays stable.
-
Without entry or exit, the picture is much worse. If no miner could enter or leave after 2026, a gap would form after the next one or two halvings unless the bitcoin price rose enough to offset them, and the invariant winning probability would be unstable for many years.
-
Forks are excluded, treated as the favorable case. The analysis assumes no forking and immediate block adoption. The authors argue that since forks waste honest work and create further attack incentives, the problems they identify should persist when forks occur.
Methodology in Plain English
The authors treat miners as a continuum of negligible units rather than as a handful of firms — the standard "non-atomic" setup from Aumann (1964) and Schmeidler (1973), also used in mining mean-field work by Li et al. (2024). Because each unit is infinitesimally small, no single miner can shift the backlog, the round length, or the difficulty. That makes the state the miner faces independent of that miner's own decision, so maximizing a long-run profit rate collapses into maximizing expected profit round by round, and yields the simple switch-on/switch-off rule.
The mempool is reduced to two fee tiers: high-fee ("Type-1") transactions arriving at constant rate λ with fee r₁ per unit, and low-fee ("Type-0") transactions with fee r₀ < r₁ assumed to exist in an infinite reservoir to fill the rest of a block up to capacity s_b. The high-fee backlog carries over from round to round and is tracked with a Lindley-type recursion with a floor — the same structure as an N-policy queue where a server idles until the queue crosses a threshold, except that here the threshold emerges from miner economics rather than being a fixed service-policy parameter. The reward available at elapsed time t is R_b + min((X + λt)r₁ + (s_b − X − λt)r₀, s_b r₁), which is non-decreasing and saturates at R_max = R_b + s_b r₁.
Bitcoin's difficulty rule is modeled exactly as in the protocol: every E = 2016 blocks (roughly two weeks), the winning probability is multiplied by the realized average round length divided by the target B = 10 minutes, with the multiplier clamped to [1/4, 4]. Results are derived under this rule. Because block discovery is random, the backlog feeds back into round lengths, which feed back into difficulty — the paper solves the stationary distribution of this loop in closed form and then checks whether the fixed point is stable. Finally, the parameters are estimated from Bitcoin data of 2026 and the model is marched along the halving schedule (210,000 blocks, roughly four years, containing about 104 epochs of 2016 blocks). The specific estimated parameter values are not reported in the available text.
Why This Matters
Impact on research. Prior analytical gap models — Tsabary and Eyal (2018), Noguchi (2026), Kim and Seo (2026) — either reset the mempool after every block or use a deterministic fee ramp within a round. Only Carlsten et al. (2016) preserved fee carryover, and only via simulation. This paper supplies what security analyses by Doger et al. (2024, 2026) had to assume as an input: a closed-form gap distribution derived from first principles. It also puts the epoch-level profit-per-time objective of Carlsten et al. on an analytical footing and justifies the per-round objective for negligible miners.
Real-world applications:
- Protocol and monetary-policy debates. Arguments over "tail emission" — a permanent nonzero subsidy, as implemented in Monero and Dogecoin and discussed for Bitcoin — gain a quantitative lever: the paper shows how the subsidy level maps to equilibrium hashpower and gap length.
- Security-budget assessments. Analysts estimating whether fees alone will secure the network after future halvings can use the equilibrium-hashpower and gap formulas rather than assuming continuous honest mining.
- Attack-cost modeling. Double-spend, selfish-mining and fee-sniping analyses need the fraction of time honest hashpower is actually online; this paper makes that an endogenous output rather than an assumption.
- Hardware-market and cost forecasting. The result that a higher operating-to-capital cost ratio brings the gap forward and can destabilize difficulty gives a concrete channel through which cheaper or more efficient mining hardware affects network security.
Industry relevance. Mining operators, mining-pool analysts, protocol developers, and crypto-asset investors evaluating long-horizon security all depend on the fee-transition question. The finding that capital costs, not fees alone, are what keep the gap bounded reframes capital expenditure as a security parameter rather than a pure cost of doing business.
Future Directions
- Atomic, large miners. The non-atomic assumption deliberately excludes mining pools and other participants with non-negligible hashpower, who can profit from block withholding or from slowing down to ease difficulty (Eyal and Sirer 2018; Fiat et al. 2019). A model mixing a continuum of negligible miners with a few large ones is explicitly left open.
- Forks and propagation delays. The analysis assumes no forking and immediate block adoption. Carlsten et al. (2016) show that under fee dominance miners may deliberately fork a block to claim its fees, so a gap model with forking is an identified next step.
- Richer fee markets. The two-tier reduction ignores the full distribution of fees and strategic fee bidding (Huberman et al. 2021). A many-tier or continuously-priced mempool would require substantially more complex analysis.
- Protocol interventions. The paper takes the current protocol as given, in contrast to Shi et al. (2020), who design incentives to switch rigs on earlier, and Budinský et al. (2023), who pass part of each block's fees to later blocks via smart contracts. Evaluating such designs against the derived gap distribution is a natural follow-up.
Target Audience
Researchers in blockchain security and cryptoeconomics; graduate students in applied probability, queueing theory, or game theory looking for a well-posed application; protocol designers and economists working on the Bitcoin fee transition and the security budget; and quantitatively trained analysts at mining firms, exchanges, or asset managers who need to reason about honest-hashpower downtime. Readers without comfort in measure theory, stochastic processes, and fixed-point stability arguments will get the most from the introduction, the model description, and the Bitcoin calibration in Section 13.
Authors’ abstract
Proof-of-Work blockchains increasingly rely on transaction fees to keep miners engaged as block subsidies decline. An immediate outcome of this is that miners may sit idle at the start of a round until enough high-fee transactions accumulate in order to make mining worthwhile. This idle period is known as the mining gap. We model the high-fee transaction backlog as a Lindley-type recursion under a two-tier fee market and use its stationary distribution to characterize the…
The full abstract on arXiv