International Association
for Cryptologic Research

Blockchain protocols based on variations of the longest-chain rule—whether following the proof-of-work paradigm or one of its alternatives—suffer from a fundamental latency barrier. This arises from the need to collect a sufficient number of blocks on top of a transaction-bearing block to guarantee the transaction’s stability while limiting the rate at which blocks can be created in order to prevent security-threatening forks. Our main result is a black-box security-amplifying combiner based on parallel composition of m blockchains that achieves \Theta(m)-fold security amplification for conflict-free transactions or, equivalently, \Theta(m)-fold reduction in latency. Our construction breaks the latency barrier to achieve, for the first time, a ledger based purely on Nakamoto longest-chain consensus guaranteeing worst-case constant-time settlement for conflict-free transactions: settlement can be accelerated to a constant multiple of block propagation time with negligible error. Operationally, our construction shows how to view any family of blockchains as a unified, virtual ledger without requiring any coordination among the chains or any new protocol metadata. Users of the system have the option to inject a transaction into a single constituent blockchain or---if they desire accelerated settlement---all of the constituent blockchains. Our presentation and proofs introduce a new formalism for reasoning about blockchains, the dynamic ledger, and articulate our constructions as transformations of dynamic ledgers that amplify security. We also illustrate the versatility of this formalism by presenting robust-combiner constructions for blockchains that can protect against complete adversarial control of a minority of a family of blockchains.

Secure function evaluation (SFE) allows a set of players to compute an arbitrary agreed function of their private inputs, even if an adversary may corrupt some of the players. Secure multi-party computation (MPC) is a generalization allowing to perform an arbitrary on-going (also called reactive or stateful) computation during which players can receive outputs and provide new inputs at intermediate stages. At Crypto~2006, Ishai \emph{et al.} considered mixed threshold adversaries that either passively corrupt some fixed number of players, or, alternatively, actively corrupt some (smaller) fixed number of players, and showed that for certain thresholds, cryptographic SFE is possible, whereas cryptographic MPC is not. However, this separation does not occur when one considers \emph{perfect} security. Actually, past work suggests that no such separation exists, as all known general protocols for perfectly secure SFE can also be used for MPC. Also, such a separation does not show up with \emph{general adversaries}, characterized by a collection of corruptible subsets of the players, when considering passive and active corruption. In this paper, we study the most general corruption model where the adversary is characterized by a collection of adversary classes, each specifying the subset of players that can be actively, passively, or fail-corrupted, respectively, and show that in this model, perfectly secure MPC separates from perfectly secure SFE. Furthermore, we derive the exact conditions on the adversary structure for the existence of perfectly secure SFE resp.~MPC, and provide efficient protocols for both cases.

We show that if a set of players hold shares of a value $a\in Z_p$ for some prime $p$ (where the set of shares is written $[a]_p$), it is possible to compute, in constant round and with unconditional security, sharings of the bits of $a$, i.e.~compute sharings $[a_0]_p, \ldots, [a_{l-1}]_p$ such that $l = \lceil \log_2(p) \rceil$, $a_0, \ldots, a_{l-1} \in \{0,1\}$ and $a = \sum_{i=0}^{l-1} a_i 2^i$. Our protocol is secure against active adversaries and works for any linear secret sharing scheme with a multiplication protocol. This result immediately implies solutions to other long-standing open problems, such as constant-round and unconditionally secure protocols for comparing shared numbers and deciding whether a shared number is zero. The complexity of our protocol is $O(l \log(l))$ invocations of the multiplication protocol for the underlying secret sharing scheme, carried out in $O(1)$.

A broadcast protocol allows a sender to distribute a value among a set of players such that it is guaranteed that all players receive the same value (consistency), and if the sender is honest, then all players receive the sender's value (validity). Classical broadcast protocols for $n$ players provide security with respect to a fixed threshold $t<n/3$, where both consistency and validity are guaranteed as long as at most $t$ players are corrupted, and no security at all is guaranteed as soon as $t+1$ players are corrupted. Depending on the environment, validity or consistency may be the more important property. We generalize the notion of broadcast by introducing an additional threshold $t^+\ge t$. In a {\em broadcast protocol with extended validity}, both consistency and validity are achieved when no more than $t$ players are corrupted, and validity is achieved even when up to $t^+$ players are corrupted. Similarly, we define {\em broadcast with extended consistency}. We prove that broadcast with extended validity as well as broadcast with extended consistency is achievable if and only if $t+2t^+<n$ (or $t=0$). For example, six players can achieve broadcast when at most one player is corrupted (this result was known to be optimal), but they can even achieve consistency (or validity) when two players are corrupted. Furthermore, our protocols achieve {\em detection} in case of failure, i.e., if at most $t$ players are corrupted then broadcast is achieved, and if at most $t^+$ players are corrupted then broadcast is achieved or every player learns that the protocol failed. This protocol can be employed in the precomputation of a secure multi-party computation protocol, resulting in {\em detectable multi-party computation}, where up to $t$ corruptions can be tolerated and up to $t^+$ corruptions can either be tolerated or detected in the precomputation, for any $t,t^+$ with $t+2t^+<n$.

In the {\em strong consensus} problem, $n$ players attempt to reach agreement on a value initially held by {\em one of the good players}, despite the (malicious) behavior of up to $t$ of them. Although the problem is closely related to the standard consensus problem (aka Byzantine agreement), the only known solution with the optimal number of players requires exponential computation and communication in the unconditional setting. In this paper we study this problem, and present {\em efficient} protocols and tight lower bounds for several standard distributed computation models --- unconditional, computational, synchronous, and asynchronous.