Research
Merged Bitcoin: Proof of Work Blockchains with Multiple Hash Types
Bitcoin's security rests entirely on one hash function (SHA-256) and specialised circuits that compute it, which concentrates mining power among a small number of ASIC manufacturers and exposes the…

- arXiv
- 2601.09090
- Published
- 2026-01-14
- Authors
- Christopher Blake, Chen Feng, Xuachao Wang, Qianyu Yu
AI summary
Overview
Research area: Cryptography and security — specifically proof-of-work (PoW) blockchain consensus, adversarial security modelling, and multi-resource consensus design.
Technical level: Advanced. The paper is written for readers comfortable with Poisson arrival processes, fork-choice rules, Borel–Cantelli arguments, and closed-form bounding of stochastic processes. The core protocol idea, however, can be understood without the mathematics.
Scope (one sentence): The paper proposes "Merged Bitcoin," a Bitcoin-style protocol in which blocks may be linked using several different hash functions, and derives closed-form upper and lower bounds on its security region under a Δ-bounded network delay model, alongside a proof that no permissionless protocol can achieve an "AND of 51% attacks" security region.
What This Paper Is About
Bitcoin's security rests entirely on one hash function (SHA-256) and specialised circuits that compute it, which concentrates mining power among a small number of ASIC manufacturers and exposes the network to hardware backdoor risk. The authors ask whether a chain could instead be built from blocks linked by multiple hash types, so that an attacker must dominate several independent mining ecosystems at once. They show that the ideal version of this idea (security failing only when every hash type is majority-adversarial) is impossible for any permissionless protocol, then introduce Merged Bitcoin as a practical alternative and quantify exactly how much it does improve attack cost.
Key Contributions
-
An impossibility theorem for "Big-And" security. The authors prove (Theorem 10) that in any permissionless blockchain protocol satisfying a chain property, an adversary-equivalence requirement, and adversary withhold ability, if a blockrate configuration (a, b) is secure then the swapped configuration (b, a) is not secure. A corollary shows that no permissionless protocol can be secure whenever at least one hash type has honest majority — the "AND of 51% attacks" region is unreachable.
-
The Merged Bitcoin protocol. A minimal modification of Bitcoin in which each block carries a type i with an associated score c_i > 0, the chain score is Σ c_i N_i, and the fork choice rule selects the highest-score chain. Blocks of different hash types link into a single ledger; everything else is unchanged from Bitcoin.
-
Closed-form security bounds. Using Theorem 11 (from the companion paper) — Merged Bitcoin is secure when λ_a < λ_h and insecure when λ_a > λ_h, where λ_a = Σ c_i b_i — the paper derives a lower bound (Lemma 13, via the Delta Interval Deletion Process) and an upper bound (Lemma 15, via the Small Block Increase Process) on the honest fully-delayed score growth rate λ_h for the two-block-type case, and compares both to simulation.
-
A linear cost-per-hash economic model and a difficulty adjustment method. The paper proves that with appropriate choice of constants, Merged Bitcoin maximises cost of attack in the linear cost-per-hash model in the negligible-delay case, and introduces a difficulty adjustment scheme argued to partly remedy asymmetric hashing advantages arising from algorithmic advances, Grover's algorithm, or hardware backdoors.
Main Findings
-
The exponential-error dream fails for permissionless protocols, but partly survives. The authors' information-theoretic motivation was that insecurity requiring an AND of several 51% thresholds would give error probability scaling like p^n ≤ A exp(−cn). Theorem 10 rules this out for any permissionless protocol, not just Merged Bitcoin, because an adversary can simulate honest mining. However, under a limited-mining-power backdoor adversary model — where hacking hash type i is an independent event K_i with probability p < 0.5, and the adversary needs roughly n/2 of the hash types — Merged Bitcoin does achieve P(insecure) ≤ A exp(−cn).
-
Zero-delay security condition. When Δ = 0, λ_h = Σ c_i h_i, giving Corollary 12: Merged Bitcoin is secure if Σ c_i b_i < Σ c_i h_i.
-
Lower bound on honest growth rate (Lemma 13). For two block types, λ_h ≥ c_1h_1/(1+h_1Δ+h_2Δ) + c_2h_2/(1+h_1Δ+h_2Δ), obtained by deleting all blocks arriving within Δ of each "guaranteed" block. Corollary 14 turns this into a sufficient condition for security, comparing that expression against c_1b_1 + c_2b_2.
-
Upper bound on honest growth rate (Lemma 15). When c_2 > c_1 (Big Blocks vs Small Blocks), λ_h < c_1h_1e^{−h_2Δ}/(1+h_1Δ+h_2Δ) + c_2(h_2+(1−e^{−h_2Δ})h_1)/(1+h_1Δ+h_2Δ), obtained by artificially promoting small blocks that are followed by a big block within Δ.
-
Different scores break the classic formula. In plain Bitcoin the fully-delayed honest growth rate is h/(1+hΔ). With heterogeneous block scores this no longer holds: a block arriving within Δ of another block is not necessarily orphaned, because the later miner may see a higher-score subchain. The paper's Figure 1 illustrates this with identical arrival times producing chain ABD in the equal-score case but ACE in the unequal-score case.
-
Simulation comparison. Figure 2 reports upper and lower bounds plus a 1000-second simulation for honest blockrates h_1 = 2 and h_2 = 1 (hashes/second) with score constants c_1 = 1 and c_2 = 2 (points per hash), varying Δ in increments of 0.5 from 0 to 5 seconds. At Δ = 0 the bounds coincide, and the simulated result falls below the lower bound due to random variation.
-
Security definition is stricter than prior work. Unlike Minotaur-style definitions framed around a transaction getting through "with overwhelming probability," the authors define security around an honest block becoming a forever block, bounded by a probability bounding function f(t) → 0. This excludes finite-but-unbounded delay attacks, which the authors argue are practically unacceptable and which would otherwise block any impossibility proof. Lemma 9 shows a secure protocol has infinitely many honest blocks with probability 1, via Borel–Cantelli.
Methodology in Plain English
The authors model the network as a set of hashing resources, each producing valid blocks as a Poisson process. For each hash type i they convert raw hashrate into a blockrate by dividing by the difficulty, giving honest rates h_i and adversarial rates b_i; the pair (h, b) is the "blockrate configuration." Security is then decided by comparing two growth rates: the adversary's score growth rate λ_a = Σ c_i b_i, which is straightforward, and the honest chain's score growth rate λ_h under worst-case (fully delayed) network conditions, which is not.
The technical work is in bounding λ_h. For the lower bound they delete every block that arrives within Δ of a "guaranteed" block, which can only reduce the chain score, producing a tractable rate they solve for in closed form. For the upper bound they inflate the score of small blocks that are followed by a big block within Δ, which can only increase the chain score. Because the true λ_h sits between these, comparing each bound against λ_a yields conservative security and insecurity statements. They also simulate the growth rate directly by generating many Poisson arrival sequences, computing the longest chain under the optimal delay schedule, and averaging the score per second — a randomised estimate that can be pushed to arbitrary precision and is used to sanity-check the bounds.
Why This Matters
For research, the paper's negative result is arguably as valuable as the protocol: it closes off a whole family of proposed designs that hope to convert independent 51% thresholds into exponential security, and it does so with a security definition precise enough to make the impossibility argument go through where prior definitions (which left "overwhelming probability" undefined) did not.
Real-world applications:
- Diversifying mining hardware dependence. Because different hash functions have different computational bottlenecks — the paper cites Ethash's memory-access bandwidth bottleneck as an example — Merged Bitcoin could let ordinary personal computers contribute meaningfully rather than ceding all mining to SHA-256 ASIC specialists.
- Reducing ASIC supply-chain risk. With a limited number of Bitcoin ASIC manufacturers, a hardware backdoor could hijack otherwise honest miners; requiring several independent hardware ecosystems to be compromised at once raises the bar.
- Hedging against algorithm-specific cryptanalysis. A quantum or algorithmic advance against one hash function (the paper names Grover's algorithm) would no longer be a single point of failure, and the proposed difficulty adjustment is argued to partly offset such asymmetric advantages.
- Raising the cost of short-term 51% attacks. The linear cost-per-hash optimality result bears directly on the concern that financial instruments could make a short-term attack profitable, given the scale of value secured and proposals for Bitcoin as a reserve asset.
Industry relevance: the protocol is deliberately a minimal delta on Bitcoin — same block structure, same transaction validity rules, same fork choice style — so it speaks directly to protocol upgrade debates rather than requiring a new consensus paradigm.
Future Directions
- Extending the bounds beyond two block types. The closed-form bounds are derived for the two-block-type case; the general n-type case is left to the simulation-based estimator and to the companion paper's results.
- Formalising the difficulty adjustment. Section XI introduces a difficulty adjustment method and argues informally that it mitigates asymmetric hashing advantages; a full analysis of its behaviour under adversarial hashrate shifts is not reported in the provided content.
- Comparison with merged PoW/PoS designs. The paper states it discusses tradeoffs against Minotaur-style merged protocols in Subsection XII, but the provided content does not include that discussion, leaving the relative merits open for follow-up.
- Tightening the gap between upper and lower bounds. The simulation in Figure 2 shows the bounds bracketing the true growth rate with a visible gap at intermediate Δ, so sharper closed-form bounds — or exact expressions for special score configurations — remain an open problem.
Target Audience
Blockchain protocol researchers and graduate students working on consensus security proofs; cryptographers interested in impossibility results for permissionless systems; and protocol engineers or technical policy reviewers evaluating multi-hash and merged-mining-style designs. The paper assumes fluency in stochastic-process arguments, so beginners will get the most from the Overview and protocol-description sections rather than the bound derivations.
Authors’ abstract
Proof of work blockchain protocols using multiple hash types are considered. It is proven that the security region of such a protocol cannot be the AND of a 51\% attack on all the hash types. Nevertheless, a protocol called Merged Bitcoin is introduced, which is the Bitcoin protocol where links between blocks can be formed using multiple different hash types. Closed form bounds on its security region in the $Δ$-bounded delay network model are proven, and these bounds are…
The full abstract on arXiv