# Splitting for Stability: a Modified Reward Scheme for Transaction-Fee-Only Regimes

Crypto.world

> Reward splitting for fee-only Bitcoin: quantifies how winner share and loser-side subsidy shift one-step fork profitability, giving protocol designers a tunable stability lever.

**arXiv**: 2611.13252

**Published**: 2026-11-09

**Authors**: Eric Brigham, Baigang Chen, Nicholas Hopper

## AI summary

## Overview

**Research area:** Blockchain security and incentive design — specifically algorithmic game theory and mechanism design for proof-of-work (PoW) consensus, in the setting where miner revenue comes only from transaction fees. The paper is filed under Cryptography and Security (cs.CR) and is authored by Eric Brigham, Baigang Chen and Nicholas Hopper of the University of Minnesota.

**Technical level:** Intermediate. The analysis is formal (threshold theorems, comparative statics), but the core idea — splitting each block reward between the winner and other miners — is stated in plain terms and the paper supplies an intuitive reading of each result.

**Scope (one sentence):** The paper proposes and evaluates a block-reward-splitting mechanism, with an optional reward-smoothing extension, intended to weaken the profitability of undercutting and other fork-based deviations in transaction-fee-only PoW systems.

## What This Paper Is About

As Bitcoin's block subsidy halves toward zero, miners will be paid only by transaction fees, and prior work by Carlsten et al. showed that this fee-only regime enables "Petty Compliance" and "Undercutting" — strategies in which miners deliberately leave fees unclaimed or fork blocks to attract other miners onto a lower-fee branch, raising the fork rate and reducing throughput. The authors ask whether changing *how* a block's reward is allocated, rather than how much is paid, can reduce the profitability of those deviations. Their answer is a scheme in which the winning miner keeps an α-portion of the block reward while the remaining β-portion is shared among other "atomic" (sufficiently large) miners, optionally with a δ-portion deferred and released over the next `L` blocks.

## Key Contributions

1. **A novel reward structure ("splitting").** Each canonical block's reward `R` is divided so the winner receives α·R and eligible non-winning atomic miners share β·R, with α + β = 1. Two redistribution rules are defined: **even splitting** (subsidy divided equally among non-winning atomic miners) and **proportional splitting** (subsidy divided in proportion to hash power). A **smoothing extension** adds a deferred δ·R portion, released evenly over the next `L` canonical blocks and then split by the same α/β rule.
2. **A general analytical threshold theorem.** The authors build a one-step deviation model over an arbitrary fee distribution and prove a single local profitability theorem for the whole family of schemes (winner-take-all, deferral-only, splitting-only, splitting with smoothing). The classical 2/3 undercutting threshold of Carlsten et al. is recovered as a special case, and closed-form thresholds are derived for general fee distributions.
3. **Simulation of the mechanism.** The reward structure is implemented in a mining simulator based on the framework of Carlsten et al., evaluating multiple reward-splitting rules (even and proportional) and the smoothing component inspired by Budinský et al., measuring resulting profitability thresholds and learned miner behavior.
4. **A check on selfish mining.** The paper reports that the modified reward scheme does not appear to worsen selfish mining incentives.

## Main Findings

- **Splitting alone does not lower the local undercutting threshold.** Proposition 8 is described as "a deliberate negative result": with δ = 0, the winner share α merely rescales both sides of the comparison while the loser-side subsidy strictly cheapens failed undercuts, so the splitting-only local cutoff is never below the winner-take-all cutoff, with equality only when the loser-side payment d_i = 0.
- **A clean deterrence condition.** Under the paper's one-step payoff proxy, the payoff advantage of a Function-Fork deviation over honest play is A_i − y·B_i, with A_i = α·Δ and B_i = α·η_G + p_G·(s_i^W − d_i^tot). The deviation is weakly profitable exactly when y·B_i ≤ A_i, where `y` is the share of hash power extending the default branch. If B_i > 0, deterrence requires y > A_i/B_i; if B_i ≤ 0, no feasible y deters.
- **The two subsidy channels push in opposite directions.** The loser-side subsidy always weakens deterrence, while the winner-side smoothing payment always strengthens it.
- **The 2/3 threshold is recovered and generalized.** For winner-take-all with i.i.d. continuous positive fees of mean μ, the cutoff is y*_wta = 2(1−ρ)/(2−ρ) with ρ = E[min(Rem,Gap)]/μ, depending on the fee distribution only through ρ. Exponential fees give ρ = 1/2 and y*_wta = 2/3; uniform fees on [0,b] give ρ = 2/3 and y*_wta = 1/2; Pareto fees with tail index a > 1 give ρ = 2(a−1)/(2a−1) and y*_wta = 1/a. Heavy-tailed fee backlogs push the cutoff toward 1; light-tailed fees push it toward 0.
- **Deferral alone lowers the cutoff monotonically.** For the FRSC idealization (β = 0, δ ∈ [0,1), α = 1 − δ), the cutoff is strictly decreasing in δ from Δ/η_G at δ = 0 to 0 as δ → 1; under the exponential model normalized to μ = 1, y*(δ) = 2(1−δ)/(3−δ), so y*(1/2) = 0.4. The paper notes this is an idealization, not a reproduction of Budinský et al.'s simulations, which report the required Default-Compliant share dropping from 2/3 to roughly 0.3 in their main configuration.
- **A subsidy-dominated failure regime exists.** When p_G·d_i ≥ α·η_G (equivalently B_i ≤ 0 with δ = 0), failed undercuts are subsidized rather than punished, the deviation is weakly profitable for every y in [0,1], and additional default-following hash power does not deter — it can strictly encourage the deviation.
- **Simulations find regions where honest behavior dominates.** Simulations show parameter regions in which honest behavior dominates fork-based deviations, and suggest that suitable reward-splitting parameters can reduce the honest participation needed to suppress aggressive forking strategies relative to the winner-take-all structure of prior work.
- **Forking, when it dominates, is the mildest variant.** In the cases where forking becomes dominant, the learned strategy is the least aggressive variant studied — the channel through which splitting helps, since the local bound of Proposition 8 rules out a direct local effect.
- **Selfish mining incentives are not worsened.** The authors state the modified reward scheme does not appear to worsen selfish mining incentives.

## Methodology in Plain English

The authors work at two levels.

First, they model a single decision. A "focal" atomic miner observes the current chain head and the fees available, and compares two options: mine normally on the head and publish immediately, or apply a one-step Function-Fork rule — mine a competing block on the parent of the head, deliberately leaving some fees unclaimed so the next miner prefers to extend that branch. They write down expected payoffs for both options, including the winner's share, the loser-side subsidy, the smoothing payment, and the probability the fork wins, which depends on the share `y` of hash power that follows the default branch. Subtracting gives a threshold: a deviation is profitable only when the default-following share is small enough. They then plug each reward scheme into that formula to get closed-form cutoffs.

Second, they simulate. The reward structure is implemented in a mining simulator built on the Carlsten et al. framework, with even and proportional redistribution rules and an added smoothing of rewards across blocks following Budinský et al. Miners in the simulation learn strategies, so the authors can observe which fork variants get selected and how much honest participation is needed to suppress them. For experiments using empirical mining distributions, a miner counts as atomic if its share of global hash power is at least 1%; for controlled threshold experiments, atomicity is assigned by the setup so that reward parameters can be varied independently.

The paper is explicit about scope: the analytical thresholds are *local* one-step profitability conditions, not statements about total system welfare, long-run stationary revenue, or the full downstream effects of fork instability. Those system-level quantities are left to simulation.

## Why This Matters

**Impact on research.** The paper separates two things that are easily conflated: making undercutting locally unprofitable, and changing which strategies miners actually learn to play. Its negative result — that splitting cannot lower the local threshold — is a useful constraint on future mechanism-design proposals, and its positive simulation result points to a different channel (shifting behavior toward less aggressive fork variants). It also extends the line of work on fee redistribution (Budinský et al.'s fee-redistribution smart contracts, Aumayr et al.'s FairChain/PRS, Brunjes et al.'s stake-pool reward sharing) into explicit modelling of strategic fee selection in the fee-only regime.

**Real-world applications:**

- Designing block-reward rules for PoW chains approaching or past their final halving, where fees are the only miner revenue.
- Evaluating fee-redistribution proposals before deployment, by giving protocol designers closed-form cutoffs to compare against measured fee distributions.
- Informing mining-pool and large-miner policy, since the mechanism explicitly distinguishes atomic from non-atomic participants and conditions subsidy eligibility on hash-power share.
- Providing a baseline for analyzing mixing incentives when combined with existing smoothing schemes such as FRSCs.

**Industry relevance.** For protocol developers and mining operators, the paper quantifies a trade-off: reducing the winner's immediate cut weakens the payoff from winning a fork race, but the simultaneous loser-side subsidy pays miners even when their undercut fails, which can make deterrence impossible once the subsidy is large enough relative to the winner-side term. That is a concrete parameter-safety warning for anyone considering reward redistribution.

## Future Directions

- **Deployment mechanics.** The authors explicitly do not address how a protocol would track deferred rewards and enforce their distribution across future canonical blocks; designing and securing that bookkeeping is an open engineering problem.
- **Determining atomicity in practice.** The paper notes that determining atomicity in a real PoW system is non-trivial, and sidesteps it with a 1% hash-power threshold in empirical experiments and by assignment in controlled ones.
- **Analyzing the channel splitting actually exploits.** Because Proposition 8 closes off the local explanation, the simulation-identified shift toward the least aggressive fork variants needs a theory of its own — presumably one that tracks multi-step and learning dynamics rather than one-step deviations.
- **Richer allocation rules.** The authors describe even and proportional splitting as intentionally simple baselines and suggest more sophisticated reward allocations, in the spirit of prior pool-allocation work, as a direction for future work.

## Target Audience

Researchers and protocol designers working on proof-of-work incentive compatibility, mechanism design for blockchain rewards, and fee-market dynamics — particularly readers already familiar with the Carlsten et al. undercutting results or with fee-redistribution proposals and who want a formal, parameterized treatment of whether redistribution helps. It is also useful to mining-economics analysts who need to reason about how reward-splitting parameters interact with the share of honest hash power needed to keep forks unprofitable.

## Authors’ abstract

As Bitcoin's block subsidy declines over time, miner revenue will come to depend increasingly on transaction fees. In fee-dominated regimes, winner-take-all reward structures can create incentives for undercutting and fork-based mining strategies, leading to instability and reduced transaction throughput. To prevent this instability, we propose a reward-splitting mechanism in which the winning block producer receives an $α$-portion of the block reward, while the rest is…

[The full abstract on arXiv](https://arxiv.org/abs/2611.13252)

[Read the original paper](https://arxiv.org/abs/2611.13252)

---

Canonical: https://crypto.world/paper/splitting-for-stability-a-modified-reward-scheme-for-transaction-fee-2611.13252
