The Bitcoin Backbone Protocol:

Analysis and Applications∗

Juan A. Garay Aggelos Kiayias†‡
Yahoo Research University of Edinburgh, IOHK
[email protected] [email protected]
Nikos Leonardos†§
National and Kapodistrian University of Athens.
[email protected]

June 23, 2017

Abstract
Bitcoin is the first and most popular decentralized cryptocurrency to date. In this work, we
extract and analyze the core of the Bitcoin protocol, which we term the Bitcoin backbone, and
prove two of its fundamental properties which we call common prefix and chain quality in the
static setting where the number of players remains fixed. Our proofs hinge on appropriate and
novel assumptions on the “hashing power” of the adversary relative to network synchronicity;
we show our results to be tight under high synchronization.
Next, we propose and analyze applications that can be built “on top” of the backbone pro-
tocol, specifically focusing on Byzantine agreement (BA) and on the notion of a public trans-
action ledger. Regarding BA, we observe that Nakamoto’s suggestion falls short of solving it,
and present a simple alternative which works assuming that the adversary’s hashing power is
bounded by 1/3. The public transaction ledger captures the essence of Bitcoin’s operation as
a cryptocurrency, in the sense that it guarantees the liveness and persistence of committed
transactions. Based on this notion we describe and analyze the Bitcoin system as well as a
more elaborate BA protocol, proving them secure assuming high network synchronicity and
that the adversary’s hashing power is strictly less than 1/2, while the adversarial bound needed
for security decreases as the network desynchronizes.
Finally, we show that our analysis of the Bitcoin backbone protocol for synchronous networks
extends with relative ease to the recently considered “partially synchronous” model, where there
is an upper bound in the delay of messages that is unknown to the honest parties.

1 Introduction
Bitcoin, introduced in [Nak08a], is a decentralized payment system that is based on maintaining
a public transaction ledger in a distributed manner. The ledger is maintained by anonymous par-
ticipants (“players”) called miners, executing a protocol that maintains and extends a distributed
data structure called the blockchain. The protocol requires from miners to solve a “proof of work”
∗
An abridged version of this paper appears in Proc. Eurocrypt 2015.
†
Research partly supported by ERC project CODAMODA, # 259152.
‡
Work partly done while at the National and Kapodistrian University of Athens
§
Work partly done while at LIAFA, Université Paris Diderot–Paris 7.

1
(POW, aka “cryptographic puzzle” — see, e.g., [DN92, RSW96, Bac97, JB99]), which essentially
amounts to brute-forcing a hash inequality based on SHA-256, in order to generate new blocks
for the blockchain. The blocks that comprise the blockchain contain sets of transactions that are
generated at will by owners of bitcoins, who issue transactions that credit any entity of their choice
who accepts payments in bitcoin. Payers broadcast transactions and miners include the transactions
they receive into the blocks they generate. Miners are rewarded for maintaining the blockchain by
receiving bitcoins; it is in this manner bitcoins are created and distributed among the miners who
are the first recipients of newly minted bitcoins.
An important concern in Bitcoin (or any e-payment system for that matter) is the prevention of
double-spending attacks. Specifically, in the context of Bitcoin, a double-spending attack can occur
when the attacker initially credits an account, receives service or goods by the account holder, but
then manages to reorganize the transaction ledger so that the transaction that credits the account
holder is reverted. In this way, the attacker keeps her bitcoin while receiving services and thus she
is able to spend it again somewhere else.
In [Nak08a], Nakamoto provides an initial set of arguments of why the Bitcoin system will
prevent double-spending attacks. Specifically, he argues that if a payee waits for the transaction
that gives her credit to advance into the blockchain a number of k blocks, then the probability
that an attacker can build an alternative blockchain that “reorganizes” the public blockchain (which
contains the credit transaction) drops exponentially with k. Nakamoto argues this by modeling the
attacker and the set of honest players as two competing actors performing a random walk moving
toward a single direction with probabilistic steps. He demonstrates that the k blocks the payee
waits are enough to ensure a negligible (in k) probability of the attacker catching up with the
honest players.
Nevertheless, the above analysis can be easily seen to be oversimplified: in particular, it does not
account for the fact that in Bitcoin’s decentralized setting the attacker may attempt to introduce
disagreement between the honest miners, thus splitting their hashing power on different POW
instances. Nakamoto himself appeared to recognize the relevance of agreement in the context of
Bitcoin, arguing in a forum post [Nak08b] that actually “Bitcoin’s basic concept” of building and
exchanging a blockchain is capable of solving Byzantine agreement (BA) [PSL80, LSP82] in the
presence of an actively malicious adversary.1 However a thorough analysis establishing the exact
security properties of the Bitcoin system has yet to appear.
Our results. In this paper we extract, formally describe, and analyze the core of the Bitcoin
protocol. We call this protocol the Bitcoin backbone, as we describe it in a way that is versatile and
extensible and can be used to solve other problems as well — not just the problem of maintaining
a public transaction ledger. The Bitcoin backbone protocol is executed by players that build a
blockchain following the Bitcoin source code [Nak09] and allows a set of players to maintain a
blockchain in a distributed fashion. The protocol is parameterized by three external functions
V (·), I(·), R(·) which we call the content validation predicate, the input contribution function, and
the chain reading function, respectively. At a high level, V (·) determines the proper structure of the
information that is stored into the blockchain, I(·) specifies how the contents of the blocks are formed
by the players, and R(·) determines how a blockchain is supposed to be interpreted in the context
of the application. Note that the structure, contents, and interpretation of the blockchain are not
1
In [Nak08b], Nakamoto refers to the problem as “Byzantine Generals,” which is often used to refer to the single-
source version of the problem. Note that since more than one general may propose a time to attack this in fact is
the case where every party has an input value, i.e., Byzantine agreement. In fact, in an anonymous setting such as
Bitcoin’s, the single-source version is nonsensical. Note that in the traditional cryptographic setting, with trusted
setup, the two problems are not equivalent in terms of the number of tolerated misbehaving parties t (t < n vs.
t < n/2, respectively).

2
important for the description of the backbone protocol and are left to be specified by the three
external functions above, which are application-specific (we provide examples of these functions in
Section 5).
We analyze the Bitcoin backbone protocol in a static setting when the participants operate in
a synchronous communication network (more details below and in Section 2; see also Section 7 for
an extension of the analysis to the “partially synchronous” model [DLS88, PSS16]) in the presence
of an adversary that controls a subset of the players. We assume that the protocol is executed by a
fixed number n of players; note, however, that this number is not necessarily known to the protocol
participants. The players themselves cannot authenticate each other and therefore there is no way
to know the source of a message; we capture this by allowing the adversary to “spoof” the source
address of any message that is delivered. We assume that messages are eventually delivered and all
parties in the network are able to synchronize in the course of a “round.” The notion of round is
not important for the description of the backbone protocol (which can also be executed in a loose
and asynchronous fashion in the same way that Bitcoin works), however, it is important in terms
of Bitcoin’s inherent computational assumption regarding the players’ ability to produce POWs.
Specifically, we assume that in a single round, all parties involved are allowed the same number
of queries to a cryptographic hash function, as well as to communicate with the other participants.
The hash function is modeled as a random oracle [BR93]. For simplicity we assume a “flat model,”
where all parties have the same quota of hashing queries per round, say q; the non-flat model where
parties have differing hashing power capabilities can be easily captured by clustering the flat-model
parties into larger virtual entities that are comprised by more than one flat-model player. In fact
“mining pools” in Bitcoin can be thought of such aggregations of flat-model players. The adversary
itself represents such pool as it controls t < n players; for this reason, the adversary’s quota per
round is t · q hashing queries. Note that in this setting, the fact t < n/2 directly corresponds
to the adversary controlling strictly less than half of the system’s total “hashing power” that all
players collectively harness, thus, we will use terms such as “honest majority” and “(1/2)-bounded
adversary” interchangeably.
In our analysis of the Bitcoin backbone protocol we formalize and prove two fundamental prop-
erties it possesses. The properties are quantified by three parameters γ, β and f ; γ and β roughly
correspond to the collective hashing power per round of the honest players and the adversary, re-
spectively, while f represents the expected number of POWs that may be found per round by the
Bitcoin network participants as a whole.
The common-prefix property. We prove that if n−t t
is suitably bounded below 1, then the
blockchains maintained by the honest players will possess a large common prefix. More specif-
ically, if an honest party “prunes” (i.e., cuts off) k blocks from the end of its local chain, the
probability that the resulting pruned chain will not be a prefix of another honest party’s chain
drops exponentially in the security parameter (see Definition 3 for the precise formulation of
the property and Theorem 15 for the precise statement). The parameter f plays a critical role
in the analysis of the property and should be small in order to approximate honest majority,
i.e., allowing t to be close to n/2. A small choice for f , suggests also that the network synchro-
nizes significantly faster than the rate of finding POW’s. On the other hand, when the network
“desynchronizes” and f gets closer to 1, achieving provably a common prefix requires almost all
parties to be honest.
The chain-quality property. We prove that the ratio of blocks in the chain of any honest player
that are contributed by malicious players is bounded by n−t t
. Again observe that in an honest
majority setting, i.e., allowing t is bounded below n/2, we obtain that the blockchain maintained
by honest players is guaranteed to have few, but still some, blocks contributed by honest players;

3
Figure 1: An overview of the backbone protocol’s applications: Nakamoto’s BA protocol Πnak
BA , our
1/3 1/2
BA protocols ΠBA and ΠBA , and the public ledger protocol ΠPL . All properties must be satisfied
with overwhelming probability. In each box we state the name of the property as well as the
maximum ratio of the adversarial hashing power that we can prove the protocol withstands (based
on the corresponding backbone property). The value  stands for a negligible quantity.

a higher ratio would be necessary to guarantee bigger percentages of blocks contributed by

honest players in the blockchain. We also observe that this result is basically tight, i.e., that the
adversary is capable of following a strategy (that deviates from the strategy of honest players)
that enables the introduction of that many blocks in the blockchain, under a favorable (for the
adversary) assumption on the propagation of adversarial blocks in the network.
While the above two security properties may seem rather abstract since they refer to properties
of the data structure that is maintained distributedly by the parties, we demonstrate that they
are in fact quite powerful and show that the Bitcoin backbone protocol armed with the above
properties can be used as a basis for solving other problems, including the problem of distributively
maintaining a “robust” public transaction ledger. In Figure 1 we show how the two properties relate
to the properties of the applications that are explained below and is provided in order to assist the
reader in conceptualizing the logic behind the security proofs of the applications.
In fact, a third property that is convenient to consider explicitly in conjunction with the above
two and was introduced in follow-up work [KP15] is chain growth, which, at a high level, quantifies
the number of blocks that are added to the blockchain during any given number of rounds. In our
original exposition [GKL15], this property was not identified explicitly but rather proven and used
directly in the form of a lemma. Nevertheless, identifying it as a separate property allows making
more modular our proof arguments and for this reason we make it explicit in this version of the
paper (cf. Section 3.2).
Byzantine agreement for (1/3)-bounded adversaries. As a first application, we show how a ran-
domized BA protocol can be built on top of the Bitcoin backbone protocol more or less directly,
and based solely on the POW assumption. We instantiate the V (·), I(·), R(·) functions so that
parties form blockchains and act according to the following rules: each party i attempts to insert
its own input vi ∈ {0, 1} into the blockchain; a blockchain is valid only if blocks contain elements
in {0, 1}; the protocol terminates when the blockchain has reached a sufficient length; and, the
blockchain is read by the honest parties by pruning k elements from its end and returning the ma-
jority bit appearing in the resulting blockchain’s prefix. We show how the common-prefix property
and the chain-quality property of the backbone protocol ensure Agreement and Validity (BA’s basic

4
properties; see Section 2) with high probability, thus turning the Bitcoin backbone protocol into a
probabilistic BA protocol.
Observe that for the above protocol to work the chain-quality property should ensure that a
majority of blocks in the blockchain originate from the honest players (otherwise Validity is lost).
Our chain quality property enables this with overwhelming probability assuming the adversarial
power is bounded by 1/3. This approach is different from Nakamoto’s proposal [Nak08b] for BA,
which, as we also show, only guarantees Validity with overwhelming probability if the adversary
has a negligible amount of hashing power. On the positive side, we stress that Nakamoto’s protocol
fails gracefully when the adversarial power gets close to 50% as Validity can be shown with constant
probability (but not overwhelming).
Public transaction ledgers and BA for honest majority. Next, we focus on how a “robust public
transaction ledger” can be built on top of the Bitcoin backbone. We instantiate the V (·), I(·), R(·)
functions so that parties form blockchains and act according to the following rules: each party
(which in this context is called a “miner”) receives a set S of transactions on its input tape and
attempts to insert those in its blockchain, omitting any transactions in S that are already included
in it. (A Bitcoin transaction is, for example, a statement of the type “account A credits account
B a z number of bitcoins,” which is signed using the secret key that corresponds to account A’s
Bitcoin address; each account has a unique Bitcoin address.) Reading a blockchain, on the other
hand, amounts to returning the total sequence of transactions that is contained in the blockchain
of the miner (and note that miners may disagree about the chain they report).
We show how the common-prefix property and the chain-quality property ensure two properties
needed by the ledger, which we call Persistence and Liveness, assuming an honest majority and
arbitrary adversarial behavior. Persistence states that once a transaction goes more than k blocks
“deep” into the blockchain of one honest player, then it will be included in every honest player’s
blockchain with overwhelming probability, and it will be assigned a permanent position in the ledger.
On the other hand, Liveness says that all transactions originating from honest account holders will
eventually end up at a depth more than k blocks in an honest player’s blockchain, and hence the
adversary cannot perform a selective denial of service attack against honest account holders. For
both properties to hold we require an honest majority (i.e., that the adversary’s hashing power
is strictly less than 50%) assuming high network synchronicity (i.e., that the expected number of
POW solutions per round is close to 0). If this is violated, Persistence requires stricter bounds
on adversarial hashing power in order to be preserved following the bounds of the common-prefix
property.
In the context of Bitcoin, our analysis implies that the Bitcoin backbone provides an operational
transaction ledger under the assumptions: (i) the adversary controls less than half of the total
hashing power, and (ii) the network synchronizes much faster relative to the POW solution rate,
(iii) digital signatures cannot be forged. On the other hand, when the network desynchronizes our
results cannot support that the ledger is maintained by assuming an honest majority. This negative
result is consistent with the experimental analysis provided by Decker and Wattenhoffer [DW13],
who predicted a drop below 50% in the required adversarial bound for any setting when information
propagation is problematic. Our result also provides some justification for the “slow” rate of 10-
minute increments used in Bitcoin block generation. Specifically, information propagation in the
Bitcoin network is on the order of seconds2 so the ratio (essentially f ) of this time window over
the average 10-minute period is reasonably close to “small” and thus transaction persistence can
be shown for roughly an honest majority. On the other hand, cryptocurrencies including Litecoin,
Primecoin and others, reacting to the demand to offer faster transaction processing, opted for a
2
See, for example, http://bitcoinstats.com/network/propagation/.

5
faster response rate (some as small as 1 minute), which results in more precarious situations, e.g.,
f > 0.1, which is far from being sufficiently close to 0 and thus cannot support our analysis that a
common prefix would be guaranteed by merely assuming an honest majority. We finally note that
the Persistence and Liveness properties we put forth and prove should not be interpreted as proofs
that all Bitcoin’s objectives are met. In particular, they do not guarantee that miners are properly
incentivized to carry out the backbone protocol, and they can only offer guarantees in a setting of
an honest majority amongst a fixed number of players as opposed to a setting where there is an
ever changing population of parties acting rationally; see related work below as well as Section 8
for further discussion.
Next, we present a BA protocol assuming an honest majority, by suitably exploiting the prop-
erties of the robust transaction ledger above. The protocol substitutes Bitcoin’s transactions with
a type of transactions that are themselves based on POWs, and hence uses POWs in two distinct
ways: for the maintenance of the ledger and for the generation of the transactions. We show that
the ledger’s Persistence implies Agreement, and that Liveness implies Validity, because assuming
the ledger is maintained for long enough, a majority of transactions originating from the honest
parties will be included (despite the fact that honest parties may control a minority of blocks in the
blockchain). The protocol requires special care in the way it employs POWs since the adversary
should be incapable of “shifting” work between the two POW tasks that it faces in each round. To
solve this problem, we introduce a special strategy for POW-based protocol composition which we
call “2-for-1 POWs.”
The Bitcoin backbone in the bounded-delay model. In this version of the paper, we show that
our analysis for the synchronous model extends with relative ease to the “partially synchronous”
model [DLS88, PSS16], where, in a nutshell, instead of messages sent in a round being guaranteed
to arrive by the end of the round, there is an upper bound of ∆ rounds for each message delivery
event that is unknown to the honest parties. This new analysis is presented in Section 7.
Related work. Realizing a digital currency with a centralized entity but while achieving strong
privacy was proposed early on by Chaum in [Cha82]. A number of other works improved various
aspects of this concept, however the approach remained centralized. Nakamoto [Nak08a] proposed
the first decentralized currency system based on POWs while relaxing the anonymity property of
the payment system to mere pseudonymity. This work was followed by a multitude of other related
proposals including Litecoin3 , Primecoin [Kin13], and Zerocash [BSCG+ 14], to mention a few. Our
analysis of the Bitcoin backbone covers all these works as well, since they are based on exactly the
same protocol.
It is interesting to juxtapose our positive results to the results of Eyal and Sirer [ES14], who
introduce an attack strategy called “selfish mining” that shows how the number of blocks contributed
to the blockchain by an adversary can exceed the percentage of the hashing power the adversary
possesses. Their results are consistent and complementary to ours. The crux of the issue is (in
our terminology) in terms of the chain-quality property, as its formulation is quite permissive: in
particular we show that if the adversary controls a suitably bounded amount of hashing power,
then it is also suitably bounded in terms of the number of blocks it has managed to insert in the
blockchain that honest players maintain. Specifically, recall that we prove that the adversary may
control at most a n−tt
percentage of the blocks in the chain. For instance, if the adversary controls
up to 1/3 of the hashing power, then it will provably control less than 50% of the blocks in the
honest players’ blockchain. As it can be easily seen, this does not guarantee that the rate of a
party’s hashing power translates to an equal rate of rewards (recall that in Bitcoin the rewards are
3
http://www.litecoin.com.

6
linearly proportional to the number of blocks that a party contributes in the chain). We define as
ideal chain quality the property that for any coalition of parties (following any mining strategy) the
percentage of blocks in the blockchain is exactly proportional to their collective hashing power, i.e.
n . The chain quality property that we prove is not ideal and the results of [ES14] show that in fact
t

there is a strategy that magnifies the percentage of a malicious coalition. Still, their mining attack
does much worse than our bound. To close the gap, we sketch (cf. Remark 3) a simple selfish
mining strategy that matches our upper bound and hence our chain quality result is tight in our
model4 assuming the number of honest parties is large.
Byzantine agreement (BA, aka “consensus”) [PSL80, LSP82] considers a set of n parties connected
by reliable and authenticated pair-wise communication links and with possible conflicting initial
inputs that wish to agree on a common output in the presence of the disruptive (even malicious)
behavior of some of them. The problem has received a considerable amount of attention under
various models. In this paper we are interested in randomized solutions to the problem (e.g.,
[BO83, Rab83, BG93, FM97, FG03, KK09])5 as in the particular setting we are in, deterministic
BA algorithms are not possible. In more detail, we consider BA in the anonymous synchronous
setting, i.e., when processors do not have identifiers and cannot correlate messages to their sources,
even across rounds, and, further, there is no trusted setup nor a limit to the number of messages
delivered per round. Okun, motivated by earlier work [BSV+ 96], considered a slightly weaker
model called “anonymous model without port awareness,” which matches the above except for
imposing a limit on the number of messages per round and proved the aforementioned impossibility
result, that deterministic algorithms are impossible for even a single failure [Oku05b, Oku05a]. In
addition, Okun showed that probabilistic BA is feasible in his setting by suitably adapting Ben-Or’s
protocol [BO83] for the standard, non-anonymous setting (cf. [Oku05b])6 ; the protocol, however,
takes exponentially many rounds. It turns out that by additionally assuming that the parties are
“port-aware” (i.e., they can correlate messages to sources across rounds), deterministic protocols are
possible and some more efficient solutions were proposed in [OB08].
The anonymous synchronous setting for BA without limits on messages per round and hence
prone to Sybil attacks [Dou02], was considered by Aspnes et al. [AJK05] who pointed to the potential
usefulness of proofs of work (e.g., [DN92, RSW96, Bac97, JB99]) as an identity assignment tool, in
such a way that the number of identities assigned to the honest and adversarial parties can be made
proportional to their aggregate computational power, respectively. For example, by assuming that
the adversary’s computational power is less than 50%, one of the algorithms in [AJK05] results in
a number of adversarial identities less than half of that obtained by the honest parties. By running
this procedure in a pre-processing stage, it is then suggested that a standard authenticated BA
protocol could be run. Such protocols, however, would require the establishment of a consistent
PKI (as well as of digital signatures), details of which are not laid out in [AJK05].
In contrast, and as mentioned above, building on our analysis of the Bitcoin backbone protocol,
we propose two BA protocols solely based on POWs that operate in O(k) rounds with error prob-
ability e−Ω(k) . The protocols solve BA with overwhelming probability under the assumption that
the adversary controls less than 1/3 and 1/2 of the computational power, respectively.
The connection between Bitcoin and probabilistic BA was also considered by Miller and LaViola
4
Our model allows the unfavorable event of adversarial messages winning all head-to-head races in terms of delivery
with honestly generated messages in any given round.
5
We remark that, in contrast to the approach used in typical randomized solutions to the problem, where achieving
BA is reduced to (the construction of) a shared random coin, the probabilistic aspect here stems from the parties’ like-
lihood of being able to provide proofs of work. In addition, as our analysis relies on the random oracle model [BR93],
we are interested in computational/cryptographic solutions to the problem.
6
Hence, BA in this setting shares a similar profile with BA in the asynchronous setting [FLP85].

7
in [ML14] where they take a different approach compared to ours, by not formalizing how Bitcoin
works, but rather only focusing on Nakamoto’s suggestion for BA [Nak08b] as a standalone protocol.
As we observe here, and also recognized in [ML14], Nakamoto’s protocol does not quite solve BA
since it does not satisfy Validity with overwhelming probability. The exact repercussions of this
fact are left open in [ML14], while with our analysis, we provide explicit answers regarding the
transaction ledger’s actual properties and the level of security that the backbone realization can
offer.
Finally, related to the anonymous setting, the feasibility of secure computation without authenti-
cated links was considered by Barak et al. in [BCL+ 11] in a more extreme model where all messages
sent by the parties are controlled by the adversary and can be tampered with and modified (i.e., not
only source addresses can be “spoofed,” but also messages’ contents can be altered and messages
may not be delivered). It is shown in [BCL+ 11] that it is possible to limit the adversary so that all
he can do is to partition the network into disjoint sets, where in each set the computation is secure,
and also independent of the computation in the other sets. Evidently, in such system, one cannot
hope to build a global ledger.
Summary of differences with [GKL15]. The most important difference in the current ver-
sion of the paper compared to the original one [GKL15] is the complete rewrite of all our proofs,
which now follow a much more streamlined and easier presentation. In particular, this incorpo-
rates the concept of a typical execution that we have in fact introduced in [GKL16] (analysis of the
backbone protocol with chains of variable difficulty) and which greatly simplifies our probabilistic
analysis, concentrating it on a single theorem (specifically, Theorem 10, showing that most execu-
tions are typical). Second, we present an analysis of the Bitcoin backbone protocol in the partially
synchronous model [DLS88], where an unknown upper bound is imposed on message delivery (cf.
Section 7). This model was considered in recent work by Pass et al. [PSS16]; nevertheless, here we
show how our model can easily extend to the bounded-delay setting and our proof strategy using
typical executions can provide a simple security proof. This affirms that the concept of typical
executions of [GKL16] and the associated analysis approach are useful tools with the potential of
wider applicability in the analysis of blockchain protocols.
Organization of the paper. The rest of the paper is organized as follows. In Section 2 we present
our model within which we formally express the Bitcoin backbone protocol and prove its basic
properties. The backbone protocol builds “blockchains” based on a cryptographic hash function;
we introduce notation for this data structure as well as the backbone protocol itself in Section 3,
followed by its analysis in Section 4. Sections 5 and 6 are dedicated to the applications built on top
of the backbone protocol — (simple) BA protocols and a robust transaction ledger, respectively.
Specifically, Section 5 covers Nakamoto’s (insufficient) suggestion for BA as well as our solution
for 1/3 adversarial power, while in Section 6 we present our treatment of a robust public ledger
(Section 6.1) formalizing the properties of Persistence and Liveness and how they apply to Bitcoin
(Section 6.2). We also include in this section (Section 6.3) our BA protocol for 1/2 adversarial
power. The extension of our analysis to the partially synchronous model is presented in Section 7,
while some directions for future research are offered in Section 8.

2 Model and Definitions

In this section we define our notion of protocol execution and provide a definition of Byzantine agree-
ment in our model. We will describe and analyze our protocols in a multiparty setting that employs
elements from previous formulations of secure multiparty computation (specifically, Canetti’s for-
mulation of “real world” execution as in [Can00a] and [Can00b, Can01]). We adopt the notation and

8
definitions of [Can00b, Can01] while we also employ ideas regarding the formulation of synchronous,
proceeding in rounds, multiparty computation from [KMTZ13].
Programs involved in a protocol execution. The execution of a protocol Π is driven by an
“environment” program Z that may spawn multiple instances running the protocol Π. The programs
in question can be thought of as “interactive Turing machines” (ITM) that have communication,
input and output tapes. An instance of an ITM running a certain program will be referred to as
an interactive Turing machine instance or ITI. The spawning of new ITI’s by an existing ITI as
well as the interaction between them is at the discretion of a control program which is also an ITM
and is denoted by C. The pair (Z, C) is called a system of ITM’s, cf. [Can00b]. As in this latter
paper we will be restricting our exposition to “locally polynomial-bounded” systems of ITM’s which
ensures a polynomial-time execution overall [Can00b, Proposition 3]. Moreover, we will be using a
more stringent control program C that will be forcing the environment to perform a “round-robin”
participant execution sequence for a fixed set of parties.
Specifically, the execution driven by Z is defined with respect to a protocol Π, an adversary A
(also an ITM) and a set of parties P1 , . . . , Pn ; these are hardcoded in the control program C. The
protocol Π is defined in a “hybrid” setting and has access to two “ideal functionalities,” which are
two other ITM’s to be defined below, called the random oracle and the diffusion channel. They are
used as subroutines by the programs involved in the execution (the ITI’s of Π and A) and they are
accessible by all parties once they are spawned.
Initially, the environment Z is restricted by C to spawn the adversary A. Each time the
adversary is activated, it may send one or more messages of the form (Corrupt, Pi ) to C. The control
program C will register party Pi as corrupted, only provided that the environment has previously
given an input of the form (Corrupt, Pi ) to A and that the number of corrupted parties is less or
equal t, a bound that is also hardcoded in C. The first ITI party to be spawned running protocol
Π is restricted by C to be party P1 . After a party Pi is activated, the environment is restricted to
activate party Pi+1 , except when Pn is activated in which case the next ITI to be activated is always
the adversary A. Note that when a corrupted party Pi is activated the adversary A is activated
instead.
Communication and “hashing power.” We describe next the two functionalities that are ac-
cessible to the parties. These functionalities will reflect the parties’ ability (i) to communicate with
each other and (ii) to calculate values of a hash function H(·) : {0, 1}∗ → {0, 1}κ concurrently. We
note that they share a state and thus they can be viewed as a single functionality, nevertheless it is
convenient to describe them as separate entities.

• The random oracle (RO) functionality. When queried by honest party Pi with a value
x marked for “calculation” for the function H(·), assuming x has not been queried before, it
returns a value y which is selected at random from {0, 1}κ ; furthermore, it stores the pair
(x, y) in the table of H(·). Each honest party Pi is allowed to ask q queries in each round as
determined by the “diffuse” functionality (see below). On the other hand, each honest party is
given unlimited queries for “verification” for the function H(·). In a similar vein, the adversary
A is given t0 · q queries in each round as determined by the diffuse functionality where t0 is
the number of corrupted parties. No verification queries are provided to A. Note that q is a
function of κ, the security parameter. We note that the functionality may maintain tables for
functions other than H(·) as well (for instance, in our protocol descriptions, we will utilise a
function G(·)), but, by convention the functionality will impose query quotas to function H(·)
only.

9
 • The diffuse functionality. Initially, the functionality sets a variable round to be 1. It also
maintains a Receive() string defined for each party Pi . A party is allowed at any moment to
fetch the contents of its personal Receive() string. Moreover, when the functionality receives
an instruction to diffuse a message m from party Pi it marks the party as complete for the
current round; note that m is allowed to be empty. At any moment, the adversary A is
allowed to receive the contents of all messages for the round and specify the contents of the
Receive() string for each party Pi . The adversary has to specify when it is complete for the
current round. When all parties are complete for the current round, the functionality inspects
the contents of all Receive() strings and includes any messages m that were diffused by the
parties in the current round but not contributed by the adversary to the Receive() tapes.
The variable round is then incremented.

We note that by adopting the resource bounded computation modeling of systems of ITM’s by
[Can00b, Can01] we obviate the need of imposing a strict upper bound on the number of messages
that may be transmitted by the adversary in each activation. In our setting, honest parties, at
the discretion of the environment, are given sufficient time to process all messages delivered via the
diffuse functionality including all messages that are injected by the adversary. This is also facilitated
by the fact that the q bound that is imposed on queries to H(·) is not imposed for hash verification
(with foresight, the q-bound will be only imposed for hash computations during the proof of work
stage of the protocol).
Note that the above formulation also reflects the fact that the communication graph is not fully
connected and messages are delivered through “diffusion”, a communication means that reflects
Bitcoin’s peer-to-peer structure. As evidenced by the above, our adversarial model in the network
is “adaptive,” meaning that the adversary is allowed to take control of parties on the fly, and
“rushing,” meaning that in any given round the adversary gets to see all honest players’ messages
before deciding his strategy, and, furthermore, there is no definite source information that can be
guaranteed for each delivered message. Note that the adversary cannot change the contents of the
messages sent by honest parties nor prevent them from being delivered as restricted by the diffuse
functionality. Effectively, this parallels communication over TCP/IP in the Internet where messages
between parties are delivered reliably, but nevertheless malicious parties may “spoof” the source of a
message they transmit and make it appear as originating from an arbitrary party (including another
honest party) in the view of the receiver. Note that the adversary is permitted to abuse the diffusion
mechanism and attempt to confuse honest parties by sending and delivering inconsistent messages
to them (thus diffuse does not constitute a reliable broadcast).7
The parties’ inputs are provided by the environment Z which also receives the parties’ outputs.
Parties that receive no input from the environment remain inactive, in the sense that they will
not act when their turn comes in each round. The environment activates parties in each round by
writing to their input tape. Note that C forces the environment to give all parties an activation in
round-robin fashion. In our exposition we will denote by Input() the input tape of each party.
The q-bounded synchronous setting. Based on the above, we can now use the notation
{exect,n
Π,A,Z (z)}z∈{0,1}∗ to denote the random variable ensemble that determines the output of the
environment Z on input z for a protocol Π that uses the two functionalities of random oracle
and diffuse (we will only be concerned with such protocols). Moreover, we will use the notation
7
In the conference version of this paper [GKL15] we used the term Broadcast instead of Diffuse to mean the
same thing. Given that this leads to some misunderstanding we changed the terminology to employ the term “Diffuse”
instead of “Broadcast.” As in the conference version, note that “Diffuse” remains an atomic operation and hence the
corruption of a party may not happen while the operation is taking place (cf. [HZ10, GKKZ11]).

10
{viewP,t,n
Π,A,Z (z)}z∈{0,1}∗ to denote the random variable ensemble describing the view of party P af-
ter the completion of an execution with environment Z, running protocol Π, and adversary A, on
auxiliary input z ∈ {0, 1}∗ .
In our exposition we are concerned with a “stand-alone” execution of Π and thus we will consider
z to be fixed to 1κ for κ ∈ N. For this reason we will simply refer to the ensemble by viewP,t,n
Π,A,Z .
If n parties P1 , . . . , Pn execute Π, the concatenation of the view of all parties hviewPΠ,A,Z
i ,t,n
ii=1,...,n
t,n
is denoted by viewΠ,A,Z . With foresight, we note that, in contrast to the standard setting where
parties are aware of the number of parties executing the protocol, we are interested in protocols
Π that do not make explicit use of the number of parties n or their identities. Further, note
that because of the unauthenticated nature of the communication model the parties may never be
certain about the number of participants in a protocol execution. Nonetheless note that the number
of parties is fixed during the course of the protocol execution, as this is hardcoded in the control
program C.
The parties’ limited ability to produce POWs is reflected in the limit imposed to all parties in
their access of the function H(·). Parties are allowed to perform a number of queries q per round. We
remark that this is a “flat-model” interpretation of the parties’ computation power, where all parties
are assumed equal. In the real world, different honest parties may have different “hashing power;”
nevertheless, our flat-model does not sacrifice generality since one can imagine that real honest
parties are simply clusters of some arbitrary number of honest flat-model parties. The adversary
A is allowed to perform t0 · q queries per round, where t0 ≤ t is the number of corrupted parties.
The environment Z, on the other hand, is not permitted any queries to H(·). The rationale for
this, is that we would like to bound the “CPU power” [Nak08a] of the adversary to be proportionate
to the number of parties it controls while making it infeasible for them to be aided by external
sources or by transferring the hashing power potentially invested in concurrent or previous protocol
executions. This underscores the fact that in our analysis is the standalone setting, where a single
protocol instance is executed in isolation.
We will refer to all the above restrictions on the environment, the parties and the adversary as
the q-bounded synchronous setting.
Properties of protocols. In our theorems we will be concerned with properties of protocols
Π in the q-bounded synchronous setting. Such properties will be defined as predicates over the
random variable viewt,n Π,A,Z by quantifying over all possible adversaries A and environments Z
that are polynomially bounded. Note that all our protocols will only satisfy properties with a
small probability of error in κ as well as in potentially other parameters. The probability space is
determined by the random choices of the random oracle functionality as well as the private coins of
all ITI’s.

Definition 1. Given a predicate Q and a bound q, t, n ∈ N with t < n, we say that the protocol
Π satisfies property Q in the q-bounded setting for n parties assuming the number of corruptions
is bounded by t, provided that for all polynomial-time Z, A, the probability that Q(viewt,n
Π,A,Z ) is
false is negligible in κ.

Note that we will only consider properties that are polynomial-time computable predicates.
Byzantine agreement. As a simple illustration of the formulation above we define the properties
of a Byzantine agreement (BA) protocol.

Definition 2. A protocol Π solves BA in the q-bounded synchronous setting provided it satisfies

the following two properties:

11
 Agreement. There is a round after which all honest parties return the same output if queried
by the environment.
Validity. The output returned by an honest party P equals the input of some party P 0 that is
honest at the round P ’s output is produced.
We note that in our protocols, the participants are capable of detecting agreement and further-
more they can also detect whether other parties detect agreement, thus termination can be easily
achieved by all honest parties. In the traditional cryptographic setting with no trusted setup, it
is known that the problem does not have a solution if t ≥ n3 [Bor96]. Interestingly, one of our
POW-based BA protocols works for t < n2 , assuming only a simultaneous start without a PKI, the
same bound that is achievable when a PKI is available.
The formulation of Validity above is intended to capture security/correctness against adaptive
adversaries. The notion (specifically, the requirement that the output value be one of the honest
parties’ inputs) has also been called “Strong Validity” [Nei94], but the distinction is only important
in the case of non-binary inputs. In either case, it is known that in the synchronous cryptographic
setting with trusted setup the problem has a solution if and only if n > |V |t, where V is the
input/decision domain [FG03]. Our POW-based protocol also achieves this bound.
Remark 1. One may consider a model where a certain percentage of the honest parties is not
always able to receive all messages broadcast on the network. We point out that such a situation is
subsumed by our adversarial model: simply we let the adversary control these players and simulate
them honestly while dropping messages from their incoming tape arbitrarily. Of course, to apply
the theorems we prove, one should adjust the total power of the adversary accordingly and add
these parties to the adversarial ones.

3 The Bitcoin Backbone Protocol

We start by introducing blockchain notation. Let G(·), H(·) be cryptographic hash functions with
output in {0, 1}κ . A block is any triple of the form B = hs, x, ctri where s ∈ {0, 1}κ , x ∈ {0, 1}∗ , ctr ∈
N are such that satisfy predicate validblockTq (B) defined as
(H(ctr, G(s, x)) < T ) ∧ (ctr ≤ q).
The parameter T ∈ N is also called the block’s difficulty level. The parameter q ∈ N is a bound
that in the Bitcoin implementation determines the size of the register ctr; in our treatment we allow
this to be arbitrary, and use it to denote the maximum allowed number of hash queries in a round.
We do this for convenience and our analysis applies in a straightforward manner to the case that
ctr is restricted to the range 0 ≤ ctr < 232 and q is independent of ctr.
A blockchain, or simply a chain is a sequence of blocks. The rightmost block is the head of
the chain, denoted head(C). Note that the empty string ε is also a chain; by convention we set
head(ε) = ε. A chain C with head(C) = hs0 , x0 , ctr0 i can be extended to a longer chain by appending
a valid block B = hs, x, ctri that satisfies s = H(ctr0 , G(s0 , x0 )). In case C = ε, by convention any
valid block of the form hs, x, ctri may extend it. In either case we have an extended chain Cnew = CB
that satisfies head(Cnew ) = B.
The length of a chain len(C) is its number of blocks. Given a chain C that has length len(C) =
n > 0 we can define a vector xC = hx1 , . . . , xn i that contains all the x-values that are stored in the
chain such that xi is the value of the i-th block.
Consider a chain C of length m and any nonnegative integer k. We denote by C dk the chain
resulting from the “pruning” the k rightmost blocks. Note that for k ≥ len(C), C dk = ε. If C1 is a
prefix of C2 we write C1  C2 .

12
 We note that Bitcoin uses chains of variable difficulty, i.e., the value T may change across
different blocks within the same chain according to some rule that is determined by the x values
stored in the chain8 . This is done to account for the fact that the number of parties (and hence
the total hashing power of the system) is variable from round to round (as opposed to the unknown
but fixed number of parties n we assume). See Section 8 for further discussion. We are now ready
to describe the protocol.

3.1 The backbone protocol

The Bitcoin backbone protocol is executed by an arbitrary number of parties over an unauthenti-
cated network. For concreteness, we assume that the number of parties running the protocol is n;
however, parties need not be aware of this number when they execute the protocol. As mentioned
in Section 2, communication over the network is achieved by utilizing a send-to-all Broadcast
functionality that is available to all parties (and maybe abused by the adversary in the sense of
delivering different messages to different parties). Each party maintains a blockchain, as defined
above, starting from the empty chain and mining a block that contains the value s = 0 (by con-
vention this is the “genesis block”)9 . Each party’s chain may be different, but, as we will prove,
under certain well-defined conditions, the chains of honest parties will share a large common prefix.
(Figure 2 depicts the local view of each party as well as the shared portion of their chains.)
In the protocol description we intentionally avoid specifying the type of values that parties try
to insert in the chain, the type of chain validation they perform (beyond checking for its structural
properties with respect to the hash functions G(·), H(·)), and the way they interpret the chain. In our
description, these actions are abstracted by the external functions V (·), I(·), R(·) which are specified
by the application that runs “on top” of the backbone protocol. We will purposely leave these
functions undetermined in our description assuming they conform to the following specifications.
We will provide explicit instantiations of them in Section 5. Briefly, they are described as follows:

• Content validation predicate V (·). The content validation predicate receives as input the
content of a chain C, denoted by xC , and will return 1 if and only if the contents are consistent
with the intended application implemented on top of the chain. In its simplest form, V (·) can
ensure that the elements of xC are of the proper type.

• Input contribution function I(·). It receives as input a tuple, (st, C, round, Input(), Receive()),
that stands respectively for state data st, current chain C, current round round, contents of
input tape Input() and contents of network tape Receive(). Given these, it will produce
an updated state st0 as well as an input x that should be the next input to be inserted in a
block. For instance, I(·) can be as simple as copying the contents of the input tape into x
and keeping st = , or performing a more complex operation that involves parsing C or even
maintaining old input values that have not yet been processed as part of the state st.

• Chain reading function R(·). It receives as input a chain C and provides an interpretation of
it. In the simplest case it can be just returning xC and leaving it to the callee to process the
contents of the chain.
8
In Bitcoin every 2016 blocks the difficulty is recalibrated according to the time-stamps stored in the blocks so
that the block generation rate remains at approximately 10 minutes per block.
9
Alternatively, s can point to an actual block that contains some trusted setup information (in the case of Bitcoin
the genesis block contains the string “The Times 03/Jan/2009 Chancellor on brink of second bailout for banks”). Our
analysis however is in the standalone setting and thus we choose the simplest possible genesis block.

13
Figure 2: Overview of the basic operation of the Bitcoin backbone protocol. Miner M1 receives
from the environment a Read instruction that results in the application of the R(·) function on
the contents of its chain which are equal to the vector hx1 , x2 , x3 , x4 , x5 i. Miner M2 receives from
the environment an Insert instruction and uses the function I(·) to determine the value y5 that it
subsequently successfully inserts in its local block chain by solving a proof of work; this results in
a broadcast of the newly extended chain. Finally miner M3 receives the newly extended chain and
validates it both structurally as well as using the content validation predicate V (·). M3 will adopt
this chain if M3 deems it better than its local chain as specified by the backbone protocol. Note
that the joint view of M1 , M2 , M3 is inconsistent but there is agreement on the prefix hx1 , x2 , x3 i.

In general our treatment will be independent of the exact operation of V, I, R apart from requir-
ing the following minimal set of conditions.

1. Input Validity. The input contribution function should produce values that are deemed ac-
ceptable by the content validation predicate. Formally, for any chain C with xC = hx1 , . . . , xn i,
the value x produced by an invocation of I(·, C, ·, ·) should satisfy V (hx1 , . . . , xn , xi) = 1. By
convention, V (ε) = 1.

2. Input Entropy. The probability of the event that two independent invocations of I(st, C, v, w)
where st, C, v, w are arbitrary values consistent with the input of I(·), result in the same value
x is negligible in κ.

The Bitcoin backbone protocol is specified as Algorithm 4. Before describing it in detail we first
introduce the protocol’s three supporting algorithms.
Chain validation. The first algorithm, called validate performs a validation of the structural
properties of a given chain C, cf. Algorithm 1. It is given as input the values q and T , as well as a
hash function H(·). It is parameterized by the content validation predicate V (·). For each block of
the chain, the algorithm checks that the proof of work is properly solved, that the counter ctr does

14
Algorithm 1 The chain validation predicate, parameterized by q, T , the hash functions G(·), H(·),
and the content validation predicate V (·). The input is C.
1: function validate(C)
2: b ← V (xC )
3: if b ∧ (C 6= ε) then . The chain is non-empty and meaningful w.r.t. V (·)
4: hs, x, ctri ← head(C)
5: s0 ← H(ctr, G(s, x))
6: repeat
7: hs, x, ctri ← head(C)
8: if validblockTq (hs, x, ctri) ∧ (H(ctr, G(s, x)) = s0 ) then
9: s0 ← s . Retain hash value
10: C ← C d1 . Remove the head from C
11: else
12: b ← False
13: end if
14: until (C = ε) ∨ (b = False)
15: end if
16: return (b)
17: end function

not exceed q and that the hash of the previous block is properly included in the block. It further
collects all the inputs from the chain’s blocks and assembles them into a vector xC . If all blocks
verify and V (xC ) is true then the chain is valid; otherwise it is rejected. As mentioned we purposely
leave the predicate V (·) undetermined.
Chain comparison. The objective of the second algorithm, called maxvalid, is to find the “best
possible” chain when given a set of chains, cf. Algorithm 2. The algorithm is straightforward
and is parameterized by a max(·) function that applies some ordering in the space of chains. The
most important aspect is the chains’ length, in which case max(C1 , C2 ) will return the longest of
the two. In case len(C1 ) = len(C2 ), some other characteristic can be used to break the tie. In
our case, max(·, ·) will always return the first operand10 ; alternatively, other options exist, such as
lexicographic order or picking a chain at random. The analysis we will perform will essentially be
independent of the tie-breaking rule11 .
Proof of work. The third algorithm, called pow, is the main “workhorse” of the backbone protocol,
cf. Algorithm 3. It takes as input a chain and attempts to extend it via solving a proof of work. This
algorithm is parameterized by two hash functions H(·), G(·) (which in our analysis will be modeled
as random oracles),12 as well as two positive integers q, T ; q represents the number of times the
10
Note that the way we deploy maxvalid, amounts to parties always giving preference to their local chain as opposed
to any incoming chain. This is consistent with current Bitcoin operation; however, some debate about alternate tie-
breaking rules has ensued in Bitcoin forums, e.g., see [Cun13].
11
It is worth to point out that the behavior of maxvalid(·) is associated with some stability aspects of the backbone
protocol and currently there are proposals to modify it (e.g., by randomizing it — cf. [ES14]). It is an interesting
question whether any improvement in our results can be achieved by randomizing the maxvalid operation.
12
In reality the same hash function (SHA-256) instantiates both G and H; however, it is notationally more conve-
nient to consider them as distinct.

15
Algorithm 2 The function that finds the “best” chain, parameterized by function max(·). The
input is {C1 , . . . , Ck }.
1: function maxvalid(C1 , . . . , Ck )
2: temp ← ε
3: for i = 1 to k do
4: if validate(Ci ) then
5: temp ← max(Ci , temp)
6: end if
7: end for
8: return temp
9: end function

algorithm is going to attempt to brute-force the hash function inequality that determines the POW
instance, and T determines the “difficulty” of the POW. The algorithm works as follows. Given a
chain C and a value x to be inserted in the chain, it hashes these values to obtain h and initializes
a counter ctr. Subsequently, it increments ctr and checks to see whether H(ctr, h) < T ; this is the
only invocation of H(·) that is subject to the bound q. If a suitable ctr is found then the algorithm
succeeds in solving the POW and extends chain C by one block inserting x as well as ctr (which
serves as the POW). If no suitable ctr is found, the algorithm simply returns the chain unaltered.
(See Algorithm 3.)
The backbone protocol. Given the three algorithms above, we are now ready to describe the
Bitcoin backbone protocol, cf. Algorithm 4. This is the protocol that is executed by the miners
and which is assumed to run “indefinitely” (our security analysis will apply when the total running
time is polynomial in κ). It is parameterized by two functions, the input contribution function I(·)
and the chain reading function R(·), which is applied to the values stored in the chain.
Each miner starts a round with a local chain C (we say that the miner has chain C at this round)
and checks its communication tape Receive() to see whether a “better” chain has been received
and in such case it adopts it resulting in chain C˜ (we say that the miner adopts chain C˜ at this
round). Choosing the chain C˜ is done using the maxvalid function; note that it could be that C = C. ˜
Then, the miner attempts to extend C˜ by running the POW algorithm pow described above.
The value that the miner attempts to insert in the chain is determined by function I(·). The
input to I(·) is the state st, the current chain C, the contents of the miner’s input tape Input() (recall
that they can be written by the environment Z at the beginning of any round) and communication
tape Receive(), as well as the current round number round. The protocol expects two types of
entries in the input tape, Read and (Insert, value); other inputs are ignored.
As mentioned, we purposely leave the functions I(·), R(·) undetermined in the description of
the backbone protocol, as their specifics will vary according to the application. When the input x
is determined by I(·), the protocol attempts to insert it into the chain C by invoking pow. In case
the local chain C is modified during the above steps, the protocol transmits (“broadcasts”) the new
chain to the other parties. Finally, in case a Read symbol is present in the communication tape,
the protocol applies function R(·) to its current chain and writes the result onto the output tape
Output(). The round ends when the algorithm diffuses a message (⊥ in case no message is to be
diffused).

16
Algorithm 3 The proof of work function, parameterized by q, T and hash functions H(·), G(·).
The input is (x, C).
1: function pow(x, C)
2: if C = ε then . Determine proof of work instance
3: s←0
4: else
5: hs0 , x0 , ctr0 i ← head(C)
6: s ← H(ctr0 , G(s0 , x0 ))
7: end if
8: ctr ← 1
9: B←ε
10: h ← G(s, x)
11: while (ctr ≤ q) do
12: if (H(ctr, h) < T ) then . This H(·) invocation subject to the q-bound
13: B ← hs, x, ctri
14: break
15: end if
16: ctr ← ctr + 1
17: end while
18: C ← CB . Extend chain
19: return C
20: end function

3.2 (Desired) Properties of the backbone protocol

We next define the two main properties of the backbone protocol that we will prove. The first
property is called the common-prefix property and is parameterized by a value k ∈ N. It considers
an arbitrary environment and adversary in the q-bounded setting, and it holds as long as removing
k blocks from an honest party’s chain results to a prefix of another honest party’s chain.

Definition 3 (Common-Prefix Property). The common-prefix property Qcp with parameter k ∈ N

states that for any pair of honest players P1 , P2 adopting the chains C1 , C2 at rounds r1 ≤ r2 in
dk
viewt,n
Π,A,Z respectively, it holds that C1  C2 .

We note that this is a strong version of the common-prefix property compared what was originally
considered in [GKL15], where the property was stated for r1 = r2 . The stronger formulation enables
a more modular proof of the persistence property in Section 6.1 and for this reason we opt for it
here. This was observed in [PSS16].
The second property, which we call the chain quality property, aims at expressing the number
of honest-player contributions that are contained in a sufficiently long and continuous part of an
honest player’s chain. Specifically, for parameters ` ∈ N and µ ∈ (0, 1), the rate of adversarial block
contributions in a continuous part of an honest party’s chain of length at least ` is bounded by µ.
This is intended to capture that at any moment that an honest player looks at a sufficiently long
part of its blockchain, that part will be of sufficient “quality,” i.e., the number of adversarial blocks
present in that portion of the chain will be suitably bounded.

17
Algorithm 4 The Bitcoin backbone protocol, parameterized by the input contribution function I(·)
and the chain reading function R(·).
1: C←ε
2: st ← ε
3: round ← 1
4: while True do
5: C˜ ← maxvalid(C, any chain C 0 found in Receive())
6: if Input() contains Read then
7: write R(C)˜ to Output() . Produce necessary output before the POW stage.
8: end if
9: ˜ round, Input(), Receive())
hst, xi ← I(st, C, . Determine the x-value.
10: Cnew ← pow(x, C) ˜
11: if C 6= Cnew then
12: C ← Cnew
13: Diffuse(C) . Broadcast the chain in case of adoption/extension.
14: else
15: Diffuse(⊥) . Signals the end of the round to the diffuse functionality.
16: end if
17: round ← round + 1
18: end while

Definition 4 (Chain Quality Property). The chain quality property Qcq with parameters µ ∈ R
and ` ∈ N states that for any honest party P with chain C in viewt,n Π,A,Z , it holds that for any `
consecutive blocks of C the ratio of adversarial blocks is at most µ.

It is easy to see that, in the absence of an adversary, any set of, say, h honest parties, obtain
as many blocks as their proportion of the total hashing power, i.e., h/n. We say that a protocol
Π satisfies ideal chain quality if this is the case for adversarial parties as well, i.e., µ = t/n with
respect to those parties. The ideal chain quality is not achieved by the Bitcoin backbone protocol,
cf. Remark 3.
A third property that is convenient to consider in conjunction to the above two and was intro-
duced in follow up work, [KP15], to our original exposition is chain growth.

Definition 5 (Chain-Growth Property). The chain growth property Qcg with parameters τ ∈ R
and s ∈ N states that for any honest party P with chain C in viewt,n
Π,A,Z , it holds that for any s
rounds there are at least τ · s blocks added to the chain of P .

In our original exposition, [GKL15], this property was not identified explicitly but rather proven
and used directly in the form of a lemma. Nevertheless, as observed in [KP15], identifying it as a
separate property, allows making more modular our proof arguments and for this reason we adopt
it here as well.

4 Analysis of the Bitcoin Backbone

We now proceed to the analysis of the protocol presented in the previous section.

18
 κ: length of the hash function output.
n: number of parties mining; t out of which are controlled by the adversary.
T : the target hash value used by parties for solving POW’s.
t: number of parties controlled by the adversary.
δ: advantage of honest parties, (t/(n − t) ≤ 1 − δ)
f : probability at least one honest party succeeds in finding a POW in a round.
η: parameter determining block to round translation.
: quality of concentration of random variables in typical executions, cf. Definition 9.
k: number of blocks for the common-prefix property.
`: number of blocks for the chain-quality property.
s: number of blocks for the chain-growth property.
L: the total run-time of the system.

Table 1: The parameters in our analysis. Positive integers n, t, L, s, `, T, k, κ where log T is linearly
related to κ; positive reals f, , δ, η, where f, , δ ∈ (0, 1).

4.1 Definitions and preliminary lemmas

Let {0, 1}κ be the range of H(·). Recall that n is the number of parties, t of which can be corrupted
by the adversary. In the analysis we are going to assume that the honest parties are sufficiently
greater in number than the parties the adversary controls. In particular, we are going to require
the following (cf. Table 1 for the definition all of the relevant parameters).

Honest Majority Assumption. A number of t out of n parties are corrupted such

that t ≤ (1 − δ)(n − t) where δ ≥ 2f + 2.

We will call a query of a party successful if it submits a pair (ctr, h) such that H(ctr, h) ≤ T .
For each round i, j ∈ [q], and k ∈ [t], we define Boolean random variables Xi , Yi and Zijk as follows.
If at round i an honest party obtains a POW, then Xi = 1, otherwise Xi = 0. If at round i exactly
one honest party obtains a POW, then Yi = 1, otherwise Yi = 0. Regarding the adversary, if at
round i, the j-th P
query Pof the k-th corrupted party is successful, thenP
Zijk = 1, otherwise Zijk = 0.
Define also Zi = tk=1 qj=1 Zijk . For a set of rounds S, let X(S) = r∈S Xr and similarly define
Y (S) and Z(S). Further, if Xi = 1, we call i a successful round and if Yi = 1, a uniquely successful
round. Uniquely successful rounds have the following important property.
Lemma 6. Suppose the k-th block B of a chain C was computed by an honest party in a uniquely
successful round. Then the k-th block a chain C 0 either is B or has been computed by the adversary.
Proof. Suppose B 0 is the k-th block of a chain C 0 that has been computed by an honest party and
B 0 is not B. Since B was computed in a uniquely successful round, B and B 0 cannot have been
computed in the same round. Let r be the earliest round on which B or B 0 was computed. Since it
was computed by an honest party, every other honest party will receive it in the next round and so
will adopt a chain of length at least k. It follows that every block computed after round r will be
extending a chain of length more than k.

An important parameter of the protocol is the probability that at least one honest party com-
putes a solution at given round. We denote this parameter by f and we have
pq(n − t)
f = 1 − (1 − p)q(n−t) ≥ ,
1 + pq(n − t)

19
where p = T /2κ is the probability of success of a single query. For the inequality we used 1 + x ≤ ex
(valid for all x) twice on the second term. With respect to the expectations of these random
variables, we note the following bounds.13

E[Xi ] = f and E[Yi ] = (n − t)pq(1 − p)q(n−t−1) ≥ f (1 − f ),

where the last inequality follows from Bernoulli’s inequality (which states that (1 + x)α ≥ 1 + αx
holds for x ≥ −1 and α ≥ 1). With respect to the expected number of blocks the adversary can
compute in a single round we note the bound
t t f
E[Zi ] = pqt = · pq(n − t) ≤ · .
n−t n−t 1−f
The inequality follows from the lower bound on f above. In our analysis we are going to treat all
these expectations as constants. (One could consider a very small number of adversarial parties, so
that E[Zi ] = o(κ). Our assumption however seems suitable for the questions we are interested in.)
We next prove the Chain-Growth Lemma. It states that, at any round, the length of any honest
party’s chain will be at least as large as the number of successful rounds. As a consequence, the
chain of honest parties will grow at least at the rate of successful rounds, no matter what strategy
the adversary employs.
Lemma 7 (Chain-Growth Lemma). Suppose that at round r an honest party has a chain P of length
`. Then, by round s ≥ r, every honest party has adopted a chain of length at least ` + s−1
i=r Xi .

Proof. By induction on s − r ≥ 0. For the basis (s = r), observe that if at round r an honest party
has14 a chain C of length `, then that party broadcast C at a round earlier than r. It follows that
every honest party will receive C by round r.
For the inductive step, note that
P by the inductive hypothesis every honest party has received a
chain of length at least `0 = ` + s−2 i=r Xi by round s − 1. When Xs−1 = 0 the statement follows
directly, so assume Xs−1 = 1. Observe that every honest party queried the oracle with a chain of
length at least `0 at round s − 1. If follows that all honest
P parties successful at round s − 1 broadcast
a chain of length at least `0 + 1. Since `0 + 1 = ` + s−1 i=r Xi , this completes the proof.

We now define our typical set of executions. Informally, this set consists of those executions
with polynomially many rounds and with the property that the sum of the X, Y, Z variables over
any Ω(κ) consecutive rounds does not deviate too much from its expectation.
First, we define a few bad events relevant to the properties of the hash function. In the random
oracle model (as long as the execution is polynomially bounded in κ), these events occur with
probability exponentially small in κ.
Definition 8. An insertion occurs when, given a chain C with two consecutive blocks B and B 0 ,
a block B ∗ created after B 0 is such that B, B ∗ , B 0 form three consecutive blocks of a valid chain.
A copy occurs if the same block exists in two different positions. A prediction occurs when a block
extends one which was computed at a later round.
13
In previous versions of this paper, including [GKL15], we expressed the analysis using parameters α = pq(n −
t), β = pqt, γ = α − α2 , f = pqn. Observe that using this notation and for the random variables Xi , Yi , Zi defined
above, it holds that E[Xi ] ≤ α, E[Yi ] ≥ γ, E[Zi ] ≤ β. Also, the previous choice of f is an upper-bound on E[Xi + Zi ].
Instead, now, we opt to define f as E[Xi ] and drop the notation α, β, γ. We believe the new choice of f is more
versatile and highlights the role of the parameter better while dropping α, β, γ simplifies the conditions of the security
theorems. Moreover, it unifies our exposition with our follow up work, [GKL16].
14
Recall that we say a party “has” a chain C ∗ at round r, if the value of variable C at line 5 of Algorithm 4 is C ∗ ,
while it “adopts” a chain C ∗ if the value of variable C˜ is equal to C ∗ .

20
Definition 9 (Typical execution). An execution is (, η)-typical if, for any set S of consecutive
rounds with |S| ≥ ηκ, the following hold.

(a) (1 − )E[X(S)] < X(S) < (1 + )E[X(S)] and (1 − )E[Y (S)] < Y (S).
(b) Z(S) < (1 + )E[Z(S)].
(c) No insertions, no copies, and no predictions occurred.

Theorem 10. An execution is typical with probability 1 − e−Ω(κ) .

Proof. First we note that the input-entropy requirement on the input contribution function I(·),
allows us to condition our probability space on the event that no two honest parties in a polynomially
bounded execution queried the H(·) oracle with the same input. In this space, the variables Xi , Yi ,
Zijk , are independent Bernoulli trials. Since the probability of this event is at least 1 − eΩ(κ) we
may proceed using the standard Chernoff bound. The stated bounds follow since E[X(S)], E[Y (S)],
and E[Z(S)] are all Ω(|S|). (Recall that we are treating E[Xi ], E[Yi ], and E[Zi ] as constants.)
For part (c) and i ∈ {0, 1, 2, 3}, let Bi = hsti , xi , ctri i and gi = G(sti , xi ). If a block extends two
distinct blocks, then a collision has occurred. To see this, suppose block B3 extents two distinct
blocks B1 and B2 . Then st3 = H(ctr1 , g1 ) = H(ctr2 , g2 ); implying a collision either in H or in G,
since B1 and B2 are distinct. It is not hard to see that an insertion or a copy imply the existence
of a block that extends two distinct blocks. If the total running time of the system of ITM’s is L
then it holds that there are

at most L queries posed to G and H. It follows that the probability of
a collision occurring is L2 2−κ+1 = e−Ω(κ) . Finally, note that, for polynomially many rounds in κ,
the probability that a prediction occurs is exponentially small in κ.

The next two Lemmas are simple, but will make convenient the appeal to the properties of a
typical execution.

Lemma 11. The following hold for any set S of at least ηκ consecutive rounds in a typical execution.

(a) (1 − )f |S| < X(S) < (1 + )f |S| and (1 − )f (1 − f )|S| < Y (S).
t f
(b) Z(S) < (1 + ) · n−t · 1−f · |S| ≤ (1 + )(1 − δ)f |S|/(1 − f ).

(c) Z(S) < (1 + δ/2) · t · X(S) < (1 − δ/2)X(S) and Z(S) < Y (S).
n−t

Proof. Parts (a) and (b) follow easily from the bounds shown in the beginning of this subsection.
Part (c), using the bounds of parts (a) and (b), boils down to verifying that (1 + )(1 − δ) <
(1 − )(1 − f )2 , which follows from 2f + 2 ≤ δ.

Remark 2. From the above conditions, one can see the importance of the parameter f and the way
that the moderate hardness of proof of work is relevant to the analysis. In particular f should be
large enough so that E[X(S)] becomes sufficiently bigger than 0 to be useful. If f is too small, (i.e.,
producing PoW’s is too hard) the honest participants will produce too few PoW’s for the system to
make progress (with foresight, chain-growth will be hurt, which in turn will hurt the liveness of the
transaction ledger). On the other hand, f cannot be too large, because then the lower bound on Y
will be too small (as it depends multiplicatively on (1 − f ) as well as f ). This means that uniquely
successful rounds will not produce sufficiently many PoW’s to overcome the PoW’s produced by the
adversary. In practice, this underscores the importance of calibrating the difficulty of the proof of
work to maintain a suitable value of f within the range (0, 1). Such calibration takes place in the
Bitcoin system every 2016 blocks and attempts to keep f somewhere between 2 − 3% (assuming a
full communication round takes place every up to 20 seconds).

21
Lemma 12. Any k ≥ 2ηκf consecutive blocks of a chain in a typical execution have been computed
in at least ηκ consecutive rounds.

Proof. Assume there is a set of consecutive rounds S 0 with |S 0 | < k in which the k blocks where
computed (i.e., if the first of the k blocks was computed at round r and the last at round r0 , let
S 0 = {r, . . . , r0 }). Then, there is a set S of rounds with |S| = ηκ such that X(S) + Z(S) ≥ k (adding
rounds to S 0 to reach size ηκ favors the inequality). However,
 1−δ
X(S) + Z(S) < (1 + )f |S| 1 + < 2ηκf ≤ k,
1−f

where the first inequality uses Lemma 11(a) and (b), and the second uses 2f + 2 ≤ δ.

A corollary of these two lemmas and Chain-Growth Lemma (Lemma 7) is the following theorem
concerning the chain-growth property.

Theorem 13 (Chain-Growth). In a typical execution the chain-growth property holds with param-
eters τ = (1 − )f and s ≥ ηκ.

4.2 Common-Prefix Property

Lemma 14 (Common-Prefix Lemma). Assume a typical execution and consider two chains C1 and
C2 such that len(C2 ) ≥ len(C1 ). If C1 is adopted by an honest party at round r, and C2 is either
dk dk
adopted by an honest party or diffused at round r, then C1  C2 and C2  C1 , for k ≥ 2ηκf .

Proof. Consider the last block on the common prefix of C1 and C2 that was computed by an honest
party and let r∗ be the round on which it was computed (if no such block exists let r∗ = 0). Define
the set of rounds S = {i : r∗ < i < r}. We claim that

Z(S) ≥ Y (S).

We show this by pairing each uniquely successful round in S with an adversarial block computed in
S. Note, that if the block computed at round r∗ was at position `, then each of these blocks will be
extending a chain of length at least `. For a uniquely successful round u ∈ S, let ju be the position
of the corresponding block. Consider the set

J = {ju : u is a uniquely successful round in S}.

Note that len(C1 ) ≥ max J, because the honest party that computed the chain which corresponds to
max J has broadcasted it and thus since C1 is adopted at round r > max J it should be at least as
long. Since len(C2 ) ≥ len(C1 ), it follows that for every j ∈ J there is a block in position j of either
chain. We now argue that for every j ∈ J there is an adversarial block in the j-th position either
in C1 or in C2 . This is obvious if j lies on the common prefix of C1 and C2 (they are adversarial by
the definition of r∗ ) and follows by Lemma 6 for the rest.
To finish the proof, note that by Lemma 12 the properties of a typical execution apply for the
set of rounds S. But then, Z(S) ≥ Y (S) contradicts Lemma 11.

Theorem 15 (Common-Prefix). In a typical execution the common-prefix property holds with pa-
rameter k ≥ 2ηκf .

22
Figure 3: Chains from the proof of Theorem 15. Round r is the first round, where an honest party
dk
adopts a chain C20 with C1 6 C20
.

Proof. Consider chains C1 and C2 of parties P1 , P2 , and corresponding rounds r1 and r2 , in violation
dk
of the common-prefix property, i.e., C1 6 C2 . Now let r be the smallest round r1 ≤ r ≤ r2 such
dk
that an honest party (not necessarily P1 , P2 ) adopts a chain C20 such that C1 6 C20 . Observe that
by our assumptions such r is well defined (since e.g., r2 is such a round, albeit not necessarily the
smallest one); refer to Figure 3 for an illustration.
We consider two cases. First, in case r = r1 , we deduce that at round r, two honest parties have
dk
adopted chains C1 , C20 for which it holds C1 6 C20 . Given that C20 is adopted at round r by an honest
party and C1 is the chain of an honest party in the same round, it holds that that len(C1 ) ≤ len(C20 ).
By the Common Prefix lemma (Lemma 14) we obtain a contradiction.
Second, consider the case r1 < r. For the honest party that adopted C20 , let C10 be the chain it
adopted at round r − 1. We claim that
     
0dk dk 0dk dk dk 0dk
C2  C10 ∧ C1  C10 ∧ len(C2 ) ≥ len(C1 ) =⇒ C1  C2 .

dk
The conclusion implies directly that C1  C20 and contradicts the definition of r. Thus, to finish
the proof, it suffices to argue the truth of each of the hypotheses and the truth of the implication.
The first hypothesis follows from the Common-Prefix Lemma (Lemma 14). This is because C20
was diffused at round r − 1 and len(C20 ) > len(C10 ) (since C20 was preferred to C10 by an honest party at
round r). The second is immediate from the definition of r. The third follows from len(C20 ) ≥ len(C1 ),
which holds because C20 was preferred by an honest party at round r∗ to C1 (note that the honest
party had received C1 at some point, since C1 belongs to an honest party at round r1 < r). Finally,
0dk dk
to see the implication, note that both C2 and C1 are prefixes of C10 and so the smallest has to be
contained in the longest.

4.3 Chain-Quality Property

Theorem 16 (Chain-Quality). In a typical execution the chain-quality property holds with param-
t
eters µ = (1 + δ/2) · n−t < 1 − δ/2 and ` ≥ 2ηκf .

Proof. Let us denote by Bi the i-th block of the chain C of an honest party P at some round r so
that C = B1 . . . Blen(C) and consider some ` consecutive blocks Bu , . . . , Bv . Define L as the least
number of consecutive blocks Bu0 , . . . , Bv0 that include the ` given ones (i.e., u0 ≤ u and v ≤ v 0 )
and have the properties (1) that the block Bu0 was computed by an honest party or is B1 in case

23
such block does not exist, and (2) that there exists a round at which an honest party was trying to
extend the chain ending at block Bv0 . Observe that number L is well defined since Blen(C) is at the
head of a chain that an honest party is trying to extend. Define also r1 as the round that Bu0 was
created (r1 = 0 if Bu0 is the genesis block), r2 as the first round that an honest party attempts to
extend Bv0 , and let S = {r : r1 ≤ r < r2 }.
Now let x denote the number of blocks from honest parties that are included in the ` blocks
and, towards a contradiction, assume that
h  δ t i h  δ t i
x≤ 1− 1+ · ·`≤ 1− 1+ · · L.
2 n−t 2 n−t
Suppose first that all the L blocks {Bj : u0 ≤ j ≤ v 0 } have been computed during the rounds in
the set S. Then
 δ t  δ t
Z(S) ≥ L − x ≥ 1 + · ·L≥ 1+ · · X(S).
2 n−t 2 n−t
The first inequality comes from the fact that the adversary computed L − x of the L blocks. The
second one comes from the postulated relation between x and L. The last inequality follows from
Lemma 7 and the properties (1) and (2) of the L blocks. To show this, we assume X(S) > L and
contradict property (2). Note that at round r1 an honest party has produced block Bu0 and has a
chain of length u0 . By Lemma 7, every honest party at round r2 will have a chain of length at least
u0 + X(S) > u0 + L = v 0 + 1 > v 0 .
On the other hand, by Lemma 12 the properties of a typical execution apply for the set of rounds
S. Thus, Z(S) ≥ (1 + δ/2) · n−t
t
· X(S) contradicts Lemma 11.
To finish the proof we need to consider the case in which these L blocks contain blocks that the
adversary computed in rounds outside S. It is not hard to see that this implies an insertion, a copy,
or a prediction, and cannot occur in a typical execution.

Remark 3. We are able to argue that Theorem 16 is tight under the simplification that ties between
blockchains of equal length always favor the adversary. In particular, we assume that the function
maxvalid at line 5 of Algorithm 4, in case of chains of equal length, will always return the suggestion
of the adversary if there is one. This simplification is made without loss of generality in our model
since the adversary is rushing and hence in case two chains are transmitted in a single round the
adversary can always arrange it so that its own solution arrives first15 . Furthermore, if the number
of honest parties is large, when an honest party discovers a solution in a round, all other honest
parties will prefer the one transmitted by the adversary and thus the effect of a single honest party
opting for its own block will be negligible.
The attack below is a type of “selfish mining” attack (it is a variation of the one in [ES14] and
appears to be folklore in bitcoin circles) that accomplishes the stated bound. The attack is as
follows. Initially, the adversary works on the same chain as every honest party. However, whenever
it finds a solution it keeps it private and keeps on extending a private chain. Whenever an honest
party finds a solution, the (rushing) adversary releases one block from the private chain; if the
private chain is depleted the adversary returns to the public chain. We now argue (informally)
that this strategy exploits the conditions stated above and maximizes the adversarial blocks in the
blockchain up to the upper bound of Theorem 16.
Consider s rounds of the protocol. With high probability, the adversary will obtain more than
(1 − )pqts solutions for some small  > 0. With each one of them it will block one block broadcast
15
In fact, this rushing capability was argued to be realistic in [ES14] through the dispersion of sybil nodes in the
Bitcoin peer-to-peer network that echo the adversary’s messages.

24
by an honest party. At the end of the s rounds, there may be a few “unused” blocks but these
will be, with high probability, at most pqts. This is because during the rounds that the adversary
acquired the blocks that it did not broadcast, none of the honest players obtained a solution; this
is a low probability event. Now, the honest parties will have—with high probability—at most
(1 + )pq(n − t)s successful rounds. It follows that µ ≥ (1 − 3) · n−t
t
. Note that the Chernoff bound
can be used to make the argument more formal and replace the expression “with high probability”
with 1 − e−Ω(s) . From this it follows that in order to obtain better chain quality one should consider
mechanisms that result in more favorable (for the honest parties) behavior in the function maxvalid.

5 Simple POW-based Byzantine Agreement Protocols

We now turn to applications of the Bitcoin backbone protocol, showing how it can be used as a
basis to solve other problems. We start in this section by analyzing Nakamoto’s suggestion for
solving BA, observing that it falls short of satisfying Definition 2; we then present our simple
instantiation which solves BA. This protocol, however, only tolerates an adversarial hashing power
less than 1/3, which takes us to the next section, where we present Bitcoin’s essential task, namely,
distributively maintaining a public transaction ledger, as well as a more elaborate BA protocol
tolerating an adversarial power strictly less than 1/2. An overview of our applications and the way
their properties depend on those of the backbone protocol was already presented in Figure 1.

5.1 Nakamoto’s suggestion for Byzantine agreement

As our first illustration of how the Bitcoin backbone can be used we present Nakamoto’s suggestion
for solving BA, as presented in a forum post [Nak08b].16 We describe his solution (call it ΠnakBA ) via
the backbone protocol by specifying the functions V (·), I(·), R(·) in a suitable way (see Figure 4).
The content validation predicate V (·) will be defined to require that all valid chains contain the
same input value together with a nonce. The chain reading function R(·) simply returns this value
(ignoring the nonce) in case the chain has length at least k (which is the security parameter);
otherwise it is undefined. The input contribution function I(·) examines the contents of the current
chain C and the contents of the input tape Input(). In case C = ε the input contribution for the
next block is taken verbatim from the input tape; otherwise, the input contribution is determined
as the (unique) value that is already present in C (and in this case the local input is ignored). Note
that we will only consider environments Z that provide an input symbol to all parties. Note that
the nonce is added to ensure “work independence”: the parties need to introduce a fresh random
κ-bit nonce at each block (cf. the beginning of Sec. 4).
It follows that initially the protocol builds various chains all containing the same value. The
intuition is that Agreement will follow from the fact that the honest players will eventually agree on
a single chain, as long as the majority of the hashing power lies with the honest parties. While this
is true, as we will demonstrate, the second necessary property does not hold: this protocol cannot
provide Validity (with high probability).
As we now show, Agreement follows easily from the common-prefix property. Indeed, as long as
there is a common prefix (irrespective of its length), it is ensured that when R(·) becomes defined
and all honest parties will produce the same output.

16
Note that Nakamoto’s description is quite informal. We make the most plausible interpretation of it in our formal
framework.

25
 Content validation pred- V (hx1 , . . . , xn i) is true if and only if it holds that v1 = . . . = vn ∈
icate V (·) {0, 1}, ρ1 , . . . , ρn ∈ {0, 1}κ where xi = hvi , ρi i, or n = 0.
Chain reading function If V (xC ) = True and len(C) ≥ k, the value of R(C) is the (unique)
R(·) (parameterized by value v that is present in each block of C, while it is undefined if
k) V (xC ) = False or len(C) < k.
Input contribution func- If C = ∅ and (Insert, v) is in the input tape then
tion I(·) I(st, C, round, Input()) is equal to hv, ρi where ρ ∈ {0, 1}κ is a ran-
dom value; otherwise (i.e., the case C 6= ∅), it is equal to hv, ρi where
v is the unique v ∈ {0, 1} value that is present in C and ρ ∈ {0, 1}κ is
a random value. The state st always remains .

Figure 4: Expressing Nakamoto’s BA protocol Πnak

BA over the Bitcoin backbone protocol via the
specification of V (·), R(·), I(·).

Lemma 17 (Agreement). Under the Honest Majority Assumption, it holds that ΠnakBA from Fig. 4
parameterized with k = 2ηκf running for a total number of rounds L ≥ 4ηκ, satisfies Agreement
(cf. Definition 2) with probability at least 1 − e−Ω(κ) .

Proof. Observe that chains contain unique values (ignoring the nonces), therefore a disagreement
between honest parties implies that two parties have disjoint chains (essentially, this is equivalent to
a fork that happens at the onset). It follows from the common-prefix property (Theorem 15) that
two chains of length at least k that are completely disjoint do not exist in a typical execution.

On the other hand, it is easy to see that Validity cannot be guaranteed with overwhelming
probability unless the hashing power of the adversary is negligible compared to the honest players,
i.e., t/n is negligible. This is because in case the adversary finds a solution first, then every honest
player will extend the adversary’s solution and switch to the adversarial input hence abandoning
the original input. While one can still show that Validity can be ensured with non-zero probability
(and thus the protocol fails gracefully assuming honest majority), Πnak BA falls short from providing
a solution to BA. Interestingly, by appropriately modifying the way the backbone protocol is used,
we show in the next section how a solution can be derived.

5.2 A Byzantine agreement protocol for (1/3)-bounded adversaries

We now show that the Bitcoin backbone can be directly used to satisfy BA’s properties with an
error that decreases exponentially in the length of the chain, assuming however that the adversary’s
hashing power is less than 1/3. There are two important differences with respect to the approach
in the previous section: (i) parties never abandon their original input but instead they do insist
in inserting it into the blockchain, and (ii) after round L they output the majority of their local
length-k prefix (note that here we consider binary BA). The protocol (i.e., the specification of the
functions V (·), I(·), R(·)) is presented in Figure 5.
1/3
Lemma 18 (Agreement). Under the Honest Majority Assumption, it holds that ΠBA from Fig. 5
parameterized with k = 2ηκf running for a total number of rounds L ≥ 8ηκ, satisfies Agreement
with probability at least 1 − e−Ω(k) .

Proof. In order for agreement to be satisfied, it suffices to argue that the chains of all honest parties
begin with the same k blocks. Note that in a typical execution, by Lemma 11 and the Chain-
Growth Lemma, upon termination (i.e., after L rounds) the chain of every honest party has more

26
 Content validation pred- V (hx1 , . . . , xn i) is true if and only if v1 , . . . , vn ∈ {0, 1}, ρ1 , . . . , ρn ∈
icate V (·) {0, 1}κ where vi , ρi are the values from the pair xi = hvi , ρi i, or n = 0.
Chain reading function If V (hx1 , . . . , xn i) = True and n ≥ 2k, the value R(C) is the ma-
R(·) (parameterized by jority bit of v1 , . . . , vk where xi = hvi , ρi i; otherwise (i.e., the case
k) V (hx1 , . . . , xn i) = False or n < 2k) the output value is undefined.
Input contribution func- I(st, C, round, Input()) is equal to hv, ρi if the input tape contains
tion I(·) (Insert, v); ρ is a random κ-bit string. The state st remains always
.
1/3
Figure 5: Protocol ΠBA over the Bitcoin backbone via the specification of V (·), R(·), I(·).

than (1 − )f L ≥ 4ηκf = 2k blocks (for the inequality note that  < δ/2 ≤ 1/2). Therefore,
dk
disagreement in the first k blocks among two chains C1 and C2 implies C1 6 C2 , in violation of the
common-prefix property. The statement follows by Theorem 15, assuming a typical execution.

We now turn to the Validity property. In order to prove it we need to show that, upon termi-
nation of the protocol, the chain of any honest party will contain among the first k inputs more
inputs from honest players than provided by the adversary. As we will see, this is a consequence of
the chain-quality property.

Lemma 19 (Validity). Under the Honest Majority Assumption, and the additional assumption
t 1−δ 1/3
n−t ≤ 2 it holds that ΠBA from Fig. 5 parameterized with k = 2ηκf running for a total number
of rounds L ≥ 8ηκ, satisfies Validity (cf. Definition 2) with probability at least 1 − e−Ω(κ) .

Proof. For the property to be satisfied we only need to ensure that in C dk , where C is the chain of an
honest party upon termination, the majority of the inputs was computed by the honest parties. First
note that, as in the proof of Agreement (Lemma 18), C has at least k blocks. Assuming a typical
execution, Theorem 16 applies and the fraction of adversarial blocks is less than 1−δ 2 · (1 + /2) <
δ

2 − 4 < /2.
1 δ 1

1/3
Note that ΠBA solves BA only in case the adversary’s hashing power is bounded by 1/3. In
case adversarial blocks win all head-to-head races within a round (as it is the case with a rushing
adversary), the result is tight, as argued in Remark 3. In the next section we show a more elaborate
construction based on a transaction ledger which tolerates can tolerate an adversary with hashing
power bounded by 1/2.
Remark 4. As mentioned in Section 2, “Strong Validity” refers to the requirement that the output
value be one of the honest parties’ inputs, and the distinction is relevant in the case of non-binary
inputs, i.e., coming from an arbitrary set V , |V | > 2. It is easy to modify the above algorithm to
also satisfy this property by making the chain reading function the element with highest plurality in
the chain (ties broken favoring the lexicographically smallest element in V ), as opposed to majority,
and by imposing a more stringent bound on the adversary, namely, that γ ≥ |V |(1 + δ)β. This
ensures that the expected number of blocks in the blockchain that are controlled by the adversary
is less than |V1 | , and maintains validity even in the worst case that the honest parties’ inputs are
equally split among all possible values but one (i.e., there are |V | − 1 inputs equally proportioned
among the honest parties). Agreement is ensured in the same way as before via the common-prefix
property. The bound is in-line with the known bounds for the computational setting with trusted
setup, n > |V |t, cf. [FG03].

27
6 Public Transaction Ledgers
We now come to the application which the Bitcoin backbone was designed to solve: maintaining a
public transaction ledger. We first formally introduce this object — a “book” where transactions
are recorded — and its properties, and then we show how it can be used to implement the Bitcoin
ledger and BA in the honest majority setting by properly instantiating the notion of a transaction.

6.1 Robust public transaction ledgers

A public transaction ledger is defined with respect to a set of valid ledgers L and a set of valid
transactions T , each one possessing an efficient membership test. A ledger x ∈ L is a vector of
sequences of transactions tx ∈ T . Each transaction tx may be associated with one or more accounts,
denoted a1 , a2 , . . . etc.
The backbone protocol parties, called miners in the context of this section, process sequences of
transactions of the form x = tx1 . . . txe that are supposed to be incorporated into their local chain
C. The input inserted at each block of the chain C is the sequence x of transactions. Thus, a ledger
is a vector of transaction sequences hx1 , . . . , xm i, and a chain C of length m contains the ledger
xC = hx1 , . . . , xm i if the input of the j-th block in C is xj . The position of transaction txj in the
ledger xC is the pair (i, j) where xi = tx1 . . . txe .
The description and properties of the ledger protocol will be expressed relative to an oracle Txgen
which will control a set of accounts by creating them and issuing transactions on their behalf. In
an execution of the backbone protocol, the environment Z as well as the miners will have access
to Txgen. Specifically, Txgen is a stateful oracle that responds to two types of queries (which we
purposely only describe at a high level):
GenAccount(1κ ): It generates an account a.
˜ It returns a transaction tx provided that tx
IssueTrans(1κ , tx): ˜ is some suitably formed string,
or ⊥.
We also consider a symmetric relation on T , denoted by C(·, ·), which indicates when two trans-
actions tx1 , tx2 are conflicting. Valid ledgers x ∈ L can never contain two conflicting transactions.
We call oracle Txgen unambiguous if it holds that for all PPT A, the probability that ATxgen
produces a transaction tx0 such that C(tx0 , tx) = 1, for tx issued by Txgen, is negligible in κ.
Finally, a transaction tx is called neutral if C(tx, tx0 ) = 0 for any other transaction tx0 . The
presence of neutral transactions in the ledger can be helpful for a variety of purposes, as we will see
next and in the BA protocol that we build on top of the ledger. For convenience we will assume that
a single random nonce ρ ∈ {0, 1}κ is also a valid transaction. Nonces will be neutral transactions
and may be included in the ledger for the sole purpose of ensuring independence between the POW
instances solved by the honest parties.
Next, we determine the three functions V (·), I(·), R(·) that will turn the backbone protocol into
ΠPL , a protocol realizing a public transaction ledger. See Figure 6.
We now introduce two essential properties for a protocol maintaning a public transaction ledger:
(i) Persistence and (ii) Liveness. In a nutshell, Persistence states that once an honest player reports
a transaction “deep enough” in the ledger, then all other honest players will report it indefinitely
whenever they are asked, and at exactly the same position in the ledger (essentially, this means
that all honest players agree on all the transactions that took place and in what order). In a more
concrete Bitcoin-like setting, Persistence is essential to ensure that credits are final and that they
happened at a certain “time” in the system’s timeline (which is implicitly defined by the ledger
itself).

28
 Content validation pred- V (hx1 , . . . , xm i) is true if and only if the vector hx1 , . . . , xm i is a valid
icate V (·) ledger, i.e., hx1 , . . . , xm i ∈ L.
Chain reading function If V (hx1 , . . . , xm i) = True, the value R(C) is equal to hx1 , . . . , xm i;
R(·) undefined otherwise.
Input contribution func- I(st, C, round, Input()) operates as follows: if the input tape contains
tion I(·) (Insert, v), it parses v as a sequence of transactions and retains the
largest subsequence x0  v that is valid with respect to xC (and whose
transactions are not already included in xC ). Finally, x = tx0 x0 where
tx0 is a neutral random nonce transaction.

Figure 6: The public transaction ledger protocol ΠPL , built on the Bitcoin backbone.

Note that Persistence is useful but not enough to ensure that the ledger makes progress, i.e.,
that transactions are eventually inserted in a chain. This is captured by the Liveness property,
which states that as long as a transaction comes from an honest account holder and is provided
by the environment to all honest players, then it will be inserted into the honest players’ ledgers,
assuming the environment keeps providing it as an input for a sufficient number of rounds.17
We define the two properties below.18

Definition 20. A protocol Π implements a robust public transaction ledger in the q-bounded syn-
chronous setting if it organizes the ledger as a hashchain of blocks of transactions and it satisfies
the following two properties:
Persistence: Parameterized by k ∈ N (the “depth” parameter), if in a certain round an honest
player reports a ledger that contains a transaction tx in a block at least k blocks away from
the end of the ledger (such transaction will be called “stable”), then whenever tx is reported as
stable by any honest player it will be in the same position in the ledger.
Liveness: Parameterized by u, k ∈ N (the “wait time” and “depth” parameters, resp.), provided
that a transaction either (i) issued by Txgen, or (ii) is neutral, is given as input to all honest
players continuously for u consecutive rounds, then all honest parties will report this transaction
at least k blocks from the end of the ledger, i.e., all report it as stable.

Remark 5. Since in our model we abstract the transaction transmission to happen within the
environment program Z, we require that all honest parties should be given as input (from Z) each
transaction so that the liveness property will guarantee its inclusion to the ledger. An alternate
formulation of the liveness property that would require transaction exchange between parties is
possible, in which case a single honest party would suffice to receive it as input.
We prove the two properties separately, starting with Persistence. We note first that it is essential
to require that the stability of the transaction is reported from the “next round on” from the time
that an honest party reports it as stable. Indeed, it is not guaranteed that parties simultaneously
report a transaction as stable: the adversary may advance the chain of a certain player at a specific
17
Observe that here we take the view that new transactions are available to all honest players and the way they
are propagated is handled by the environment that feeds the backbone protocol. While this makes sense in the
honest/malicious cryptographic model, it has been challenged in a model where all players are rational [BDOZ12].
Analysis of the backbone protocol in a setting where transaction propagation is governed by rational players is beyond
the scope of our paper. Still, it is straightforward to use our results to argue about liveness even when some players
do not receive all transactions by applying the same reasoning as in Remark 1.
18
We note that we provide a slightly different formulation for persistence and liveness compared to our original
formulation in [GKL15]. Even though the conjunction of the two properties remains equivalent to the original
formulation we believe the current formulation is simpler to present.

29
round and thus make the transaction appear as stable when the environment checks it; nevertheless
at that round other honest parties may still have chains that have not advanced sufficiently enough
and thus report the transaction as not stable. This is akin to the lack of simultaneous termination
in early stopping consensus protocols, cf. [DRS90].
The proof is essentially based on the common-prefix property of the backbone protocol (recall
Definition 3 and Theorem 15); in fact it relies on the stronger formulation of common prefix as
expressed in Lemma 14.
Lemma 21 (Persistence). Under the Honest Majority Assumption, it holds that ΠPL from Fig. 4
parameterized with k = 2ηκf satisfies Persistence (cf. Definition 20) with probability at least 1 −
e−Ω(κ) .
Proof. Let C1 be the chain of some honest player P1 at round r1 . We show that if a transaction tx
dk
is included in C1 at round r1 , i.e., it is stable, then this transaction will be always included in every
honest player’s chain in a typical execution. Since their chain will be at least as long from the next
round on, thus the transaction tx will be also stable for them. But if this is not true for the chain
dk
C2 of a party P2 at a round r2 , then C1 6 C2 and common-prefix property is clearly violated.

We next prove Liveness, which is based on the chain-quality property (recall Definition 4 and
Theorem 16) and the fact that the chain of honest parties grows at least as fast as the number of
blocks they produce proven in Lemma 7.
Lemma 22 (Liveness). Under the Honest Majority Assumption, it holds that ΠPL from Fig. 4
parameterized with u = 4η(1 − )−1 κ rounds and k = 2ηκf satisfies Liveness (cf. Definition 20)
with probability at least 1 − e−Ω(κ) .
Proof. We prove that assuming all honest players receive as input the transaction tx for at least u
rounds, then in a typical execution there exists an honest party with chain C such that tx is included
in C dk . Indeed, in a typical execution, after u rounds the honest parties have at least 4ηκf = 2k
successful rounds. Invoking Chain-Growth Lemma (Lemma 7), we infer that the chain’s length of
any honest party has increased by at least 2k blocks.
Finally, the chain-quality property (Theorem 16) implies that at least one of the blocks in the
length-k suffix of C dk was computed by an honest party. Such a block would include tx since it
is infeasible for adversarial Z, A to produce a conflicting transaction tx0 (which would be the only
event making an honest player drop tx from the sequence of transactions x that it attempts to insert
in the blockchain). Thus, the lemma follows.

6.2 Bitcoin-like transactions and ledger

Next, we show how to instantiate the public transaction ledger for Bitcoin, by defining the sets of
transactions and valid ledgers.
Transactions and accounts are defined with respect to a digital signature scheme that is com-
prised of three algorithms hKeyGen, Sign, Verifyi. An account will be a pair a = (vk, G(vk)) where
G(·) is a hash function and G(vk) is the “address” corresponding to the account.
A transaction tx is of the form “{a1 , a2 , . . . , ai } → (σ, {(a01 , b01 ), . . . , (a0o , b0o )}),” where a1 , . . . , ai
are the accounts to be debited, a01 , . . . , a0o are the addresses of the accounts19 to be credited with
funds b01 , . . . , b0o , respectively, and σ is a vector h(vk1 , σ1 ), . . . , (vki , σi )i of verification keys and
19
In bitcoin terminology every account has an address that is used to uniquely identify it. Payments directed to an
account require only this “bitcoin address.” The actual verification key corresponding to the account will be revealed
only when the account makes a payment.

30
digital signatures issued under them, on the same message {(a01 , b01 ), . . . , (a0o , b0o )}. (We note that
Bitcoin transactions can be more expressive but the above description is sufficient for the purpose
of our analysis).
Next, we specify the Txgen oracle:
GenAccount(1κ ): It generates an account a by running KeyGen and computing the hash G(·) on
the verification key. The account is the pair (vk, G(vk)), where G(vk) is the account’s address.
The corresponding secret key, sk, is kept in the state of Txgen.
˜ It returns a transaction tx provided that tx
IssueTrans(1κ , tx): ˜ is a transaction that is only miss-
ing the signatures by accounts that are maintained by Txgen. (Recall the format of transactions
above.) Each account is only allowed a single transaction.
Note that the above restriction on IssueTrans is without loss of generality, as in Bitcoin, entities
typically maintain a number of accounts and are allowed (although not forced) to move their balances
forward to a new account as they make new transactions. The conflict relation C(·, ·) over T satisfies
that C(tx1 , tx2 ) = 1 if and only if tx1 6= tx2 and tx1 , tx2 have an input account in common20 . Thus,
we can easily prove the unambiguity of the Txgen oracle based on the unforgeability of the underlying
digital signature.

Lemma 23. Assume that hKeyGen, Sign, Verifyi is an existentially unforgeable signature scheme.
Then oracle Txgen is unambiguous.

In order to define the set of valid Bitcoin ledgers we first need to determine in what sense a
transaction may be valid with respect to a ledger. Then we will define the set of valid ledgers
recursively as the maximal set of vectors of sequences of transactions that satisfy this condition. So
here it goes.
A transaction tx is valid with Pi respect P to a Bitcoin ledger x = hx1 , . . . , xm i provided that all
o
digital signatures verify and j=1 bj ≥ j=1 bj , where bj is the balance that was credited to
0

account aj in the latest transaction involving aj in x. In case e = ij=1 bj − oj=1 b0j > 0, then e
P P
is a transaction fee that may be claimed separately in a special transaction of the form “∅ → . . .,”
called a coinbase transaction. In more detail, a coinbase transaction has no inputs and its purpose
is to enable miners to be rewarded P for maintaining the legder. The transaction is of the form
“∅ → {(a1 , b1 ), . . . , (ao , bo )},” and oj=1 bj is determined based on the other transactions that are
“bundled” in the block as well as a flat reward fee, as explain below.
A sequence of transactions x = h∅ → {(a1 , b1 ), . . . , (ao , bo )}, tx1 , . . . , txl i is said to be valid with
respect to a ledger x = hx1 , . . . , xm i, if each transaction txj is valid with respect to the ledger x
extended by the transactions tx1 , . . . , txj−1 . I.e., for all j = 1, . . . , l the transaction txj should be
valid with respect to ledger
hx1 , . . . , xm , tx1 . . . txj−1 i,
Po
and furthermore, the total P fee e = j=1 bj collected in the transaction ∅ → {(a1 , b1 ), . . . , (ao , bo )}
does not exceed rm + m j=1 ej , which includes all the individual fees corresponding to transactions
tx1 , . . . , txe , plus a value rm that is the flat reward given for extending a ledger of length m to a
ledger of length m + 1.21
The set of valid ledgers L with respect to a reward progression {rj }j∈N contains ε (the empty
ledger), and any ledger x which extends a ledger in L by a valid sequence of transactions. Note
20
The conflict relation is more permissive in the actual Bitcoin ledger. We adopt the more simplified version given
above as it does not change the gist of the analysis.
21
Originally, the flat reward for extending the Bitcoin chain was 50 BTC. The sequence r0 , r1 , . . . for Bitcoin follows
a geometric progression with large constant intervals.

31
that the first transaction sequence of any Poledger x ∈ L contains a single transaction of the form
∅ → {(a1 , b1 ), . . . , (ao , bo )} that satisfies j=1 bj = r0 , where r0 is the initial flat reward. This first
transaction “distributes an initial amount of money” to the ledger’s initiator(s).22 It is easy to see
that L has an efficient membership test.
Given the existence of coinbase transactions in this application we can do away with random
nonces as standalone transactions and the description of the input contribution function I in Fig. 6,
is modified to include their generation each time an input sequence of transactions is determined
to be inserted P in the ledger. Specifically, I(·) will form a coinbase transaction ∅ → {(a, b)}, where
b = rlen(C) + m j=1 ej and ej is the fee corresponding to x’s j-th transaction. Account a is a
freshly created account that is obtained via running KeyGen. I(·) will append account a and the
corresponding (vk, sk) to its private state st.
We will refer to the modified ΠPL protocol by the moniker ΠBTC . ΠBTC inherits from ΠPL the
properties of Persistence and Liveness which will ensure the following with overwhelming probability
in k.
Apart from its latest k blocks, the transaction ledger is fixed and immutable for all honest
miners.
If a majority of miners23 receive an honest transaction and attempt to insert it following the
protocol for a sufficient number of rounds (equal to parameter u, the “wait time”), it will become
a permanent entry in the ledger (no matter the adversarial strategy of the remaining miners).

6.3 Byzantine agreement for honest majority

We now use the public transaction ledger formulation to achieve POW-based BA for an honest
majority by properly instantiating the notion of a transaction, thus improving on the simple BA
protocol tolerating a (1/3)-bounded adversary presented in Section 5.
Here we consider a set of valid ledgers L that contain sequences of transactions of the form
hnonce, v, ctri, and satisfy the predicate:
(H1 (ctr, G(nonce, v)) < T ) ∧ (ctr ≤ q), (1)
where H1 (·), G(·) are two hash functions as in the definition of the backbone protocol, and v ∈ {0, 1}
is a party’s input. (Recall that T is the difficulty level and q determines how many calls to H1 (·) a
party is allowed to make per round.) To distinguish the oracles, in this section we will use H0 (·) to
refer to the oracle used in the backbone protocol.
For the ledger we consider in this section, there will be no accounts and all transactions will be
neutral — i.e., the conflict predicate C(·, ·) will be false for all pairs of transactions.
We first provide a high level description of the BA protocol assuming parties have q queries per
round to each oracle H0 (·), H1 (·). We then show how to use a single oracle H(·) to achieve the
combined functionality of both of them while only using q queries per round.
1/2
At a high level, the protocol, ΠBA , works as follows:
Operation: In each round, parties run two protocols in parallel. The first protocol is protocol ΠPL
(Fig. 6), which maintains the transaction ledger and requires q queries to the oracle H0 (·). The
second process is a “transaction production” protocol Πtx (Fig. 7), which continuously generates
transactions satisfying predicate (1). The protocol makes q queries to the H1 (·) oracle.
22
In the case of Bitcoin, it was supposedly Nakamoto himself who collected this first reward of 50 BTC.
23
Recall that we assume a flat model w.r.t. hashing power; a majority of miners corresponds to a set of parties
controlling the majority of the hashing power.

32
Algorithm 5 The POW-based transaction production protocol Πtx , parameterized by q, T and
hash functions H1 (·), G(·).
1: v ← Input()
2: ctr ← 1
3: tx ← ε
4: h ← G(nonce, v) . nonce is a random κ-bit string
5: while (ctr ≤ q) do
6: if (H1 (ctr, h) < T ) then . Proof of work succeeded
7: tx ← hnonce, v, ctri
8: break
9: end if
10: ctr ← ctr + 1
11: end while
12: Broadcast(tx)

Figure 7: The transaction production protocol Πtx .

Termination: After (a predetermined) round L, a party collects all the unique POW transac-
tions that are present in the first (1−)δ
4k
+ k blocks and returns the majority bit from the bits
occurring in these transactions (note that uniqueness takes also the nonce of each transaction
into account).
1/2
As described, protocol ΠBA does not conform to the q-bounded setting since parties require
q queries to oracle H0 (·) and q queries to oracle H1 (·) to perform the computation of a single
round (the setting imposes a bound of q queries to a single oracle for all parties). Note that a
naïve simulation of H0 (·), H1 (·) by a single oracle H(·) in the (2q)-bounded setting (e.g., by setting
Hb (x) = H(b, x)) would violate the restriction imposed on each oracle individually, since nothing
would prevent the adversary, for example, from querying H0 (·) 2q times. Next, we show how we can
combine the two protocols into a single protocol that utilizes at most q queries to a single random
oracle in a way that the adversary will remain q-bounded for each oracle. This transformation,
1/2
explained below, completes the description of ΠBA .
2-for-1 POWs. We now tackle the problem of how to turn a protocol operation that uses two
separate POW subprocedures involving two distinct and independent oracles H0 (·), H1 (·) into a
protocol that utilizes a single oracle H(·) for a total number of q queries per round. Our transfor-
mation is general and works for any pair of protocols that utilize H0 (·), H1 (·), provided that certain
1/2
conditions are met (which are satisfied by protocol ΠBA above). In more detail, we consider two
protocols Π0 , Π1 that utilize a POW step as shown in Algorithm 6 in Figure 8.
In order to achieve composition of the two protocols Π0 , Π1 in the q-bounded setting with access
to a single oracle H(·), we will substitute steps 2-11 in both protocols with a call to a new function,
double-pow, defined below. First, observe that in Πb , b ∈ {0, 1}, the POW steps 2-11 operate with
input wb and produce output in Bb if the POW succeeds. The probability of obtaining a solution
is T · 2−κ .
The modification consists in changing the structure of the POWs from pairs of the form (w, ctr)
to triples of the form (w, ctr, label), where label is a κ-bit string that is neutral from the point of

33
 Algorithm 6 POW-based protocol fragment of Algorithm 7 The double proof of work func-
Πb , b ∈ {0, 1} parameterized by q, T and hash tion, parameterized by q, T and hash func-
functions Hb (·), G(·), b ∈ {0, 1}. The value wb is tions H(·), G(·) that substitutes steps 2-11 of two
determined from the protocol’s context. POW-based protocols.
1: . . . . Value wb is determined
1: function double-pow(w0 , w1 )
2: ctr ← 1
2: B0 , B1 ← ε
3: B ← ε
3: ctr ← 1
4: hb ← G(wb )
4: h ← hG(w0 ), G(w1 )i
5: while (ctr ≤ q) do
5: while (ctr ≤ q) do
6: if (H(ctr, hb ) < T ) then
6: u ← H(ctr, h)
7: Bb ← hwb , ctri
7: if (u < T ) ∧ (B0 = ε) then
8: break
8: B0 ← hw0 , ctr, G(w1 )i
9: end if
9: end if
10: ctr ← ctr + 1
10: if ([u]R < T ) ∧ (B1 = ε) then
11: end while
11: B1 ← hw1 , ctr, G(w0 )i
12: . . . . The POW B is exploited here
12: end if
13: ctr ← ctr + 1
14: end while
15: return hB0 , B1 i
16: end function

Figure 8: The 2-for-1 POW transformation.

view of the proof. This will further require the modification of the verification step for POWs in
both protocols Π0 , Π1 in the following manner.
Any verification step in Π0 of a POW hw0 , ctri which is of the form H(ctr, G(w0 )) < T , will
now operate with a POW of the form hw0 , ctr, labeli and will verify the relation
H(ctr, hG(w0 ), labeli) < T.

Any verification step in Π1 of a POW hw1 , ctri which is of the form H(ctr, G(w1 )) < T , will
now operate with a POW of the form hw1 , ctr, labeli and will verify the relation
[H(ctr, hlabel, G(w1 )i)]R < T,

where [a]R denotes the reverse of the bitstring a.

This parallel composition strategy in the form of function double-pow is shown in Algorithm 7.
Either or both the solutions it returns, B0 , B1 , may be empty if no solution is found.
1/2
Protocol ΠBA will employ double-pow, which will substitute the individual POW operation of
the two underlying protocols Π0 , Π1 as defined in lines 2-11 of Algorithm 6. The correctness of the
above composition strategy follows from the following simple observation.

Lemma 24. Consider a uniform random variable U over the integers in [0, 2κ ) and an integer T
such that T = 2t for some positive integer t < κ/2. Then, the events (U < T ) and ([U ]R < T ) are
independent and they both occur with probability T · 2−κ .

34
Proof. It is easy to see that each event happens with probability T · 2−κ . The conjunction of the
two events involves the choice of an integer U which satisfies U < T and [U ]R < T . Observe that
because T = 2t , it follows that the conditioning on U < T leaves the t least significant bits of U
uniformly random while fixing the remaining κ − t bits. It follows that the t most significant bits
of [U ]R are uniformly random in the conditional space U < T . The event [U ]R < T has probability
(T /2κ−t )/T = T · 2−κ and thus the two events are independent.

Theorem 25. Under the Honest Majority Assumption, strengthened so that δ ≥ 2f + 4, it holds
1/2
that protocol ΠBA parameterized with k = 2ηκf running for a total number of rounds L ≥ 4ηκ(1 +
2 −Ω(κ) .
δ )/(1 − ), satisfies Agreement and Validity (cf. Definition 2) with probability at least 1 − e

Proof. First observe that due to Lemma 24 the success probability for all parties to solve a proof of
work of either kind in each round is independent. In the following, we assume a typical execution.
To verify Agreement, note that by chain-growth (Theorem 13) the chain of every honest party
is of length at least (1 + 2δ )4ηκf = 8k/δ + 2k ≥ (1−)δ
4k
+ 2k at the end of the L rounds. Thus, the
parties prune at least k blocks and Agreement follows directly from the common-prefix property.
We next focus on validity. Consider the prefix C of length (1−)δ
4k
+ k of an honest party’s chain
upon termination. As in Lemma 12, it can be shown that the first (1−)δ4k
blocks of C were computed
4ηκ
after round (1−)δ . By the chain-quality property, the last k blocks of C do contain an honest block.
4ηκ
Thus, for S the set of all the rounds up to the round that this block was computed, |S| ≥ (1−)δ and
there are at least X(S) honest inputs in C. On the other hand, the adversarial blocks in C are less
than Z(S) + k. The following sequence of inequalities shows that the majority of inputs are honest.

f
Z(S) + k (1 + )(1 − δ) 1−f |S| + 2ηκf (1 + )(1 − δ)  2ηκ 
< = 1+
X(S) (1 − )f |S| (1 − )(1 − f ) (1 − )|S|
 δ  2ηκ   δ  δ
< 1− 1+ ≤ 1− 1+ < 1.
2 (1 − )|S| 2 2

(The first inequality uses Lemma 11, the second the assumption δ ≥ 2f + 4, and the third the
lower bound on |S|.) Validity follows.

Remark 6. Regarding strong validity in the multivalued BA setting, i.e., where the input domain is
V and has a constant cardinality strictly larger than 2 we can adapt the above protocol to return the
plurality from the values stored in the transactions that are found in the ledger. In order to ensure
strong validity by this modification we restrict the hashing power of the adversary to (1 − δ)/|V |
since this will ensure that the adversary’s number of transactions cannot overturn the plurality
value as defined by the honest parties’ inputs (even if those are evenly distributed amongst them).

7 The Bitcoin Backbone in the Bounded-Delay Model

We now show how, with relative ease, we can extend our analysis to the bounded delay model. The
bounded delay model is identical to the model of Section 2 with the following modifications: (1) the
parameter q is fixed to be 1, (2) the diffuse functionality allows for ∆-delays, i.e., when it inspects
the contents of all Receive() strings, it will only include messages that were omitted if they are
∆ rounds old. Note that ∆ is a parameter of the execution that is unknown to the honest protocol
participants (in contrast to q). Observe that “rounds” still exist in the model, but now these are
not synchronization rounds where messages are supposed to be delivered to honest parties.

35
 We now describe how to extend our analysis in the bounded delay setting. The most crucial
observation is in the notion of uniquely successful round: specifically, while in the synchronous
setting we were counting uniquely successful rounds, now we need to count ∆-isolated uniquely
successful rounds, where isolated means that there is a window of ∆ rounds before and after where
no other successful round took place for the honest parties. (This notion is analogous to the
definition of convergence opportunity in [PSS16].)
In more details, we will replace the Yi random variable by Yi0 defined as follows. It takes values
in {0, 1} and Yi0 = 1 when round i was uniquely successful (i.e., Yi = 1) and Xj = 0 for each j 6= i
such that |j − i| < ∆. When Yi0 = 1, we call i a ∆-isolated uniquely successful round (we may drop
∆ if it is clear from the context). We have

E[Yi0 ] ≥ f (1 − f )2∆−1 ≥ f [1 − (2∆ − 1)f ].

We remark that this definition is adopted with the goal that the (easy but crucial) observation of
Lemma 6 still holds. This is the main difference with the q-bounded model; in what follows we
will argue that the same lemmas and theorems still hold, with small modifications of proofs and
parameters.
To obtain a Chain-Growth Lemma, cf. Lemma 7, although we could use the isolated uniquely
successful rounds, it is more natural and gives slightly better results if we use a new Boolean variable
Xi0 . For a round i, we define Xi0 = 1, if Xi = 1 and Xj = 0 for each j ∈ {i − ∆ + 1, . . . , i − 1}.
When Xi0 = 1, we call i an isolated successful round. We have

E[Xi0 ] = f (1 − f )∆−1 ≥ f [1 − (∆ − 1)f ].

Lemma 26 (Chain-Growth Lemma in the bounded-delay model). Suppose that at round r an

honest party has a chain of length
Ps−∆ `. Then, by round s ≥ r + ∆ − 1, every honest party has adopted
a chain of length at least ` + i=r Xi0 .

Proof. By induction on s − r − ∆ + 1 ≥ 0. For the basis (s = r + ∆ − 1), observe that if at round

r an honest party has a chain C of length `, then that party broadcast C at a round earlier than r.
It follows that every honest party will receive C by round r + ∆ − 1 = s.
For the inductive step, we consider two cases. First, we consider Xs−∆0 = 0. By the inductive
hypothesis every honest party has received a chain of length at least ` = ` + s−∆−1
0 Xi0 by round
P
i=r
s − 1. Since Xs−∆
0 = 0, we have ` + s−∆ i=r Xi = ` and the statement follows. For the second case,
0 0
P
Xs−∆ = 1. Observe that this implies Xi = 0 for all i ∈ {s − 2∆ + 1, . . . , s − ∆ − 1}. It follows
0 0

that s−∆−1 Xi0 = s−2∆ i . By the inductive hypothesis every honest party has adopted a chain
0
P P
i=r i=r X
of length at least `0 = ` + s−2∆i=r Xi by round s − ∆. It follows that every honest party queried
0
P
the oracle with a chain of length at least `0 at round s − ∆. Hence, all honest parties successful at
round s − ∆ broadcast a chain of length at least `0 + 1. This
P chain0 will be received by every honest
party by round s. Since Xs−∆
0 = 1, we have `0 + 1 = ` + s−∆
i=r Xi and this completes the case and
the proof.

Definition 27 (Typical execution in the bounded-delay model). An execution is (, η, ∆)-typical

if, for any set S of consecutive rounds with |S| ≥ ηκ, the following hold.

(a) (1 − )E[X 0 (S)] < X 0 (S), X(S) < (1 + )E[X(S)], and (1 − )E[Y 0 (S)] < Y 0 (S).
(b) Z(S) < (1 + )E[Z(S)].
(c) No insertions, no copies, and no predictions occurred.

36
Theorem 28. An execution is typical with probability 1 − e−Ω(κ) .

Proof. Note that Yi0 and Yj0 are not independent anymore when |i − j| < 2∆ and the standard
Chernoff bound does not apply. (Similarly for Xi0 and Xj0 .) However, it is not hard to verify that
Y 0 (S) is 2-Lipschitz with respect to the independent variables {Xi : i ∈ S}. This is because Xi
affects Yj0 if |i − j| < ∆ and there can be at most two Yj0 ’s equal to 1 in an interval of length 2∆ − 1.
The statement then follows by Theorem 36.

The honest majority assumption needs to be strengthened to accommodate the decreased value
of E[Yi0 ].

Honest Majority Assumption (Partial Synchrony). A number of t out of n

parties are corrupted such that t ≤ (1 − δ)(n − t), with δ such that δ ≥ 3∆f + 5 + 3∆
ηκ .

Lemma 29. The following hold for any set S of at least ηκ consecutive rounds in a typical execution.

(a) (1 − )f (1 − f )∆−1 |S| < X 0 (S) and (1 − )f (1 − f )2∆−1 |S| < Y 0 (S).
t f
(b) Z(S) < (1 + ) · n−t · 1−f · |S| ≤ (1 + )(1 − δ)f |S|/(1 − f ).
(c) Let S 0 = {r, . . . , r0 } with |S 0 | ≥ ηκ. For S = {r, . . . , r0 + ∆}, Z(S) < (1 + δ/2) · t
n−t · X 0 (S 0 ).
(d) Let S 0 = {r, . . . , r0 } with |S 0 | ≥ ηκ. For S = {r − ∆, . . . , r0 + ∆}, Z(S) < Y 0 (S 0 ).

Proof. Part (a) uses the bounds on the E[Xi0 ] and E[Yi0 ] mentioned in the beginning of this section.
Part (b) is the same as in the q-bounded model. Parts (c) and (d), using parts (a) and (b), both
follow from δ ≥ 2∆f + 4 + 2ηκ
5∆
.

Note that Lemma 12 holds as is, since its proof relies only on the bounds for X(S) and Z(S).
Combining with the Chain-Growth Lemma (Lemma 26) for this model, we obtain the following
theorem with respect to the chain-growth property.

Theorem 30. In a typical execution in the bounded-delay model, the chain-growth property holds
with parameters τ = (1 − )f (1 − f )∆−1 and s ≥ ηκ.

7.1 Common Prefix and Chain Quality in the Bounded-Delay Model

Lemma 14 holds with the depth parameter slightly increased.

Lemma 31. Assume a typical execution and consider two chains C1 and C2 at round r, such that
dk dk
len(C2 ) ≥ len(C1 ). For the same conditions of Lemma 14, it holds C1  C2 and C2  C1 , for
k ≥ 2f ηκ + 2∆.

Proof. The proof is basically the same. The difference is that with respect to the honest parties we
consider the set of rounds S 0 = {i : r∗ + ∆ < i < r − ∆}. One then argues that

Z(S) ≥ Y 0 (S 0 )

as in Lemma 14, pairing each uniquely successful round in S 0 with an adversarial block computed
in S. The contradiction is obtained due to Lemma 29(d).

Then, the common-prefix property for a typical execution follows with exactly the same proof
as that of Theorem 15 stated as follows:

37
Theorem 32 (Common-Prefix). In a typical execution the common-prefix property holds with pa-
rameter k ≥ 2ηκf + 2∆.

Finally, chain quality is also easy to prove.

Theorem 33 (Chain-Quality). In a typical execution the chain-quality property holds with param-
t
eters µ = (1 + δ/2) · n−t and ` ≥ 2ηκf + ∆.

Proof. Let S 0 = {r : r1 ≤ r < r2 − ∆}. It can be argued as in Theorem 16 that

 δ t  δ t
Z(S) ≥ L − x ≥ 1 + · ·L≥ 1+ · · X 0 (S 0 ),
2 n−t 2 n−t
contradicting Lemma 29(c).

Remark 7. Given the above results on the three basic properties, chain growth, chain quality and
common prefix, it follows in a straightforward manner that all our applications from Sections 5 and
6 can be easily ported to the bounded delay setting.

8 Summary and Directions for Future Work

In this paper we presented a formal treatment of the Bitcoin backbone, the protocol used at the core
of Bitcoin’s transaction ledger. We identified and proved basic properties of the backbone protocol —
“common prefix,” “chain quality,” “chain growth” — and showed how they can be used as foundations
for designing Byzantine agreement and robust public transaction ledger protocols. Our results show
that an honest majority among the (equally equipped) participants suffices, assuming the network
synchronizes much faster than the proof of work rate (f is relatively small in our notation), the
proper inputs (e.g., transactions) are available to the honest majority24 , while the bound on the
adversary for honest parties to reach agreement degenerates as f gets larger.
While these are encouraging results, we have demonstrated deviations that are of concern for
the proper operation of Bitcoin. Importantly, we show that as the network ceases to synchronize
fast enough compared to the proof-of-work rate (i.e., the worst-case time that takes honest players
to “hear” each other becomes substantial compared to the time it takes to solve a proof of work),
the honest majority property ceases to hold and the bound offered by our analysis that is required
to obtain a robust transaction ledger approaches 0 as f approaches 1. Note that the effects of bad
synchronization is in the maintenance of the common-prefix property, which is the critical property
for showing agreement.
A second important concern is regarding the chain quality property, where our results show that
if an adversary controls a hashing power ratio corresponding to β then the ratio of the blocks it can
contribute to the blockchain is bounded but can be strictly bigger than β.
The above caveats in the two basic properties of the backbone have repercussions on the Per-
sistence and Liveness properties of the Bitcoin ledger. Firstly, they illustrate that fast information
propagation amongst honest players (in relation to proof of work) is essential for transaction persis-
tence. Secondly, they show that transaction liveness becomes more fragile as the relative adversarial
power n−tt
, gets close to 1. Note that we achieve Liveness for any relative power less than 1 but we
do not assume any upper bound on the number of transactions that may be inserted in a block25 ; it
24
Our formalization is a way to formally express what perhaps was Nakamoto’s intuition when he wrote about
Bitcoin that “it takes advantage of the nature of information being easy to spread but hard to stifle” [Nak09].
25
In the current Bitcoin implementation there is an upper bound of 1MB for blocks, hence the number transactions
per block is limited.

38
is obvious that the fewer blocks the honest miners get into the blockchain the harder may be for a
transaction to get through. Furthermore, the fact that chain quality demonstrably fails to preserve a
one-to-one correspondence between a party’s hashing power and the ratio of its contributions to the
ledger point to the fact that Bitcoin’s rewarding mechanism is not incentive compatible (cf. [ES14]).
In fact we show that the relative hashing power n−t t
is the essential upper bound on chain quality
for the Bitcoin backbone — a result we show also to be tight in our rushing adversary model. In
this way our results flesh out the incentive compatibility problems of the Bitcoin backbone, but (on
a more positive note) they also point to the fact that honest hashing-power majority is sufficient to
maintain the public ledger (under favorable network conditions), and hence suggest that the Bitcoin
protocol can work as long as the majority of the miners want it to work (without taking into account
the rationality of their decision).
The above observations apply to the setting where the number of participants is fixed. In the
dynamic setting (where the number of parties running the protocol may change from round to
round), given the flat model that we consider, the difficulty of the blockchain, determined by the
target T , may be calibrated according to the number of players n that are active in the system.
If T is set by an omniscient trusted party then the analysis carries in a straightforward way but
otherwise, if T is somehow calculated by the parties themselves, the adversary can try to exploit
its calculation. Note that in this case the maxvalid function would need to take the difficulty’s
variability into account and thus choose the “most difficult” chain (as opposed to the longest).
Comparing chains based on difficulty is simply done by computing the length of a chain by counting
blocks proportionally to how difficult they are (for example, a block whose difficulty is two times
larger than a given difficulty value would contribute twice as much in “length”). The analysis of
the Bitcoin backbone protocol with chains of variable of difficulty is the subject of our follow-up
work [GKL16].
Other interesting questions include the security analysis of the Bitcoin backbone protocol in a
rational setting as opposed to honest/malicious, and in a concurrent/universal composition setting
as opposed to standalone. Furthermore, the substitution of the random oracle assumption with
a suitable computational assumption, as well as the development of backbone modifications that
improve its characteristics in terms of common prefix and chain quality. In terms of the ledger
application, transaction processing times (i.e., reducing the wait time parameter u in the Liveness
property) is also an interesting question with implications to practice (since real-world payment
systems benefit greatly from fast transaction confirmation and verification — see, e.g., [CDE+ 16]
for scalability issues and suggestions). In all these cases, our work offers a formal foundation
that allows analyzing the security properties of “tweaks” on the backbone protocol (such as the
randomization rule of [ES14] or the “GHOST” rule in [SZ13] used in Ethereum26 ) towards meeting
the above goals.
We remark that follow-up work to the present paper has examined additional backbone protocol
properties, protocols and model extensions. For instance, we have already mentioned the chain
growth property, explicitly introduced in [KP15] and now also stated in the current version of this
paper, which enables one to abstract the blockchain feature of being able to grow unhindered by
the adversary. While this is a quite simple property to prove for the Bitcoin backbone, it becomes
far more complex in alternative blockchain protocols such as those using the GHOST rule [SZ13];
see [KP16] for an analysis of such protocols. In [PSS16], Pass et al. put forth a property called
self-consistency, which refers to the inability of the adversary to make honest parties disagree with
themselves as the protocol advances. Chain growth and self-consistency are useful if one wants to
do a black-box reduction of Persistence and Liveness of the ledger to the underlying properties of
26
https://www.ethereum.org/

39
the blockchain, an approach also fulfilled in the current version of this paper, where the common
prefix property in Section 3.2 captures self-consistency and chain growth is explicitly treated. Fur-
ther, [PSS16] also studied the robustness of transaction ledger in the partially synchronous model,
where messages may not be delivered at the end of a round, but there is still a certain bound within
which all messages are eventually delivered, see [DLS88]. This version of the paper also shows how
the original analysis for the synchronous model extends to that setting, cf. Section 7.
Another set of interesting directions include the development of other applications that may be
built on top of the Bitcoin backbone protocol. A notable example is secure multiparty computation
(MPC) [Yao82, GMW87] with properties such as fairness and guaranteed output delivery (current
works in this direction, e.g., [ADMM14, BK14a, BK14b, KZZ16, KB16, KVV16] assume an idealized
version of the Bitcoin system).

Acknowledgements
The authors are grateful to Giorgos Panagiotakos, Sergio Rajsbaum and Hong-Sheng Zhou for
helpful comments and discussions.

References
[ADMM14] Marcin Andrychowicz, Stefan Dziembowski, Daniel Malinowski, and Łukasz Mazurek. Secure
multiparty computations on bitcoin. IEEE Security and Privacy, 2014.
[AJK05] James Aspnes, Collin Jackson, and Arvind Krishnamurthy. Exposing computationally-
challenged Byzantine impostors. Technical Report YALEU/DCS/TR-1332, Yale University
Department of Computer Science, July 2005.
[Bac97] Adam Back. Hashcash. http://www.cypherspace.org/hashcash, 1997.
[BCL+ 11] Boaz Barak, Ran Canetti, Yehuda Lindell, Rafael Pass, and Tal Rabin. Secure computation
without authentication. J. Cryptology, 24(4):720–760, 2011.
[BDOZ12] Moshe Babaioff, Shahar Dobzinski, Sigal Oren, and Aviv Zohar. On bitcoin and red balloons. In
Boi Faltings, Kevin Leyton-Brown, and Panos Ipeirotis, editors, EC, pages 56–73. ACM, 2012.
[BG93] Piotr Berman and Juan A. Garay. Randomized distributed agreement revisited. In Digest
of Papers: FTCS-23, The Twenty-Third Annual International Symposium on Fault-Tolerant
Computing, Toulouse, France, June 22-24, 1993, pages 412–419. IEEE Computer Society, 1993.
[BK14a] Iddo Bentov and Ranjit Kumaresan. How to use bitcoin to design fair protocols. In Advances in
Cryptology - CRYPTO 2014 - 34th Annual Cryptology Conference, Santa Barbara, CA, USA,
August 17-21, 2014, Proceedings, Part II, pages 421–439, 2014.
[BK14b] Iddo Bentov and Ranjit Kumaresan. How to use bitcoin to incentivize correct computations.
ACM CCS 2014, 2014.
[BO83] Michael Ben-Or. Another advantage of free choice: Completely asynchronous agreement proto-
cols (extended abstract). In Robert L. Probert, Nancy A. Lynch, and Nicola Santoro, editors,
PODC, pages 27–30. ACM, 1983.
[Bor96] Malte Borderding. Levels of authentication in distributed agreement. In Özalp Babaoglu and
Keith Marzullo, editors, Distributed Algorithms, 10th International Workshop, WDAG ’96,
Bologna, Italy, October 9-11, 1996, Proceedings, volume 1151 of Lecture Notes in Computer
Science, pages 40–55. Springer, 1996.
[BR93] Mihir Bellare and Phillip Rogaway. Random oracles are practical: A paradigm for designing
efficient protocols. In CCS ’93, Proceedings of the 1st ACM Conference on Computer and
Communications Security, Fairfax, Virginia, USA, November 3-5, 1993., pages 62–73, 1993.

40
[BSCG+ 14] Eli Ben-Sasson, Alessandro Chiesa, Christina Garman, Matthew Green, Ian Miers, Eran Tromer,
and Madars Virza. Zerocash: Decentralized anonymous payments from bitcoin. IACR Cryptol-
ogy ePrint Archive, 2014:349, 2014.
[BSV+ 96] Paolo Boldi, Shella Shammah, Sebastiano Vigna, Bruno Codenotti, Peter Gemmell, and Janos
Simon. Symmetry breaking in anonymous networks: Characterizations. In Fourth Israel Sympo-
sium on Theory of Computing and Systems, ISTCS 1996, Jerusalem, Israel, June 10-12, 1996,
Proceedings, pages 16–26. IEEE Computer Society, 1996.
[Can00a] Ran Canetti. Security and composition of multiparty cryptographic protocols. J. Cryptology,
13(1):143–202, 2000.
[Can00b] Ran Canetti. Universally composable security: A new paradigm for cryptographic protocols.
Cryptology ePrint Archive, Report 2000/067, 2000. http://eprint.iacr.org/2000/067.
[Can01] Ran Canetti. Universally composable security: A new paradigm for cryptographic protocols.
In 42nd Annual Symposium on Foundations of Computer Science, FOCS 2001, 14-17 October
2001, Las Vegas, Nevada, USA, pages 136–145. IEEE Computer Society, 2001.
[CDE+ 16] Kyle Croman, Christian Decker, Ittay Eyal, Adem Efe Gencer, Ari Juels, Ahmed E. Kosba, An-
drew Miller, Prateek Saxena, Elaine Shi, Emin Gün Sirer, Dawn Song, and Roger Wattenhofer.
On scaling decentralized blockchains - (A position paper). In Jeremy Clark, Sarah Meikle-
john, Peter Y. A. Ryan, Dan S. Wallach, Michael Brenner, and Kurt Rohloff, editors, Financial
Cryptography and Data Security - FC 2016 International Workshops, BITCOIN, VOTING, and
WAHC, Christ Church, Barbados, February 26, 2016, Revised Selected Papers, volume 9604 of
Lecture Notes in Computer Science, pages 106–125. Springer, 2016.
[Cha82] David Chaum. Blind signatures for untraceable payments. pages 199–203, 1982.
[Cun13] Cunicula. Why doesn’t bitcoin use a tiebreaking rule when comparing chains of equal length?
https://bitcointalk.org/index.php?topic=355644.0. 2013.
[DLS88] Cynthia Dwork, Nancy A. Lynch, and Larry J. Stockmeyer. Consensus in the presence of partial
synchrony. J. ACM, 35(2):288–323, 1988.
[DN92] Cynthia Dwork and Moni Naor. Pricing via processing or combatting junk mail. In Ernest F.
Brickell, editor, CRYPTO, volume 740 of Lecture Notes in Computer Science, pages 139–147.
Springer, 1992.
[Dou02] John R. Douceur. The sybil attack. In Peter Druschel, M. Frans Kaashoek, and Antony
I. T. Rowstron, editors, Peer-to-Peer Systems, First International Workshop, IPTPS 2002,
Cambridge, MA, USA, March 7-8, 2002, Revised Papers, volume 2429 of Lecture Notes in
Computer Science, pages 251–260. Springer, 2002.
[DP12] Devdatt P. Dubhashi and Alessandro Panconesi. Concentration of Measure for the Analysis of
Randomized Algorithms. Cambridge University Press, New York, NY, USA, 2012.
[DRS90] Danny Dolev, Rüdiger Reischuk, and H. Raymond Strong. Early stopping in byzantine agree-
ment. J. ACM, 37(4):720–741, 1990.
[DW13] Christian Decker and Roger Wattenhofer. Information propagation in the bitcoin network. In
P2P, pages 1–10. IEEE, 2013.
[ES14] Ittay Eyal and Emin Gun Sirer. Majority is not enough: Bitcoin mining is vulnerable. In
Financial Cryptography, 2014.
[FG03] Matthias Fitzi and Juan A. Garay. Efficient player-optimal protocols for strong and differential
consensus. In Elizabeth Borowsky and Sergio Rajsbaum, editors, PODC, pages 211–220. ACM,
2003.
[FLP85] Michael J. Fischer, Nancy A. Lynch, and Mike Paterson. Impossibility of distributed consensus
with one faulty process. J. ACM, 32(2):374–382, 1985.

41
[FM97] Pesech Feldman and Silvio Micali. An optimal probabilistic protocol for synchronous byzantine
agreement. SIAM J. Comput., 26(4):873–933, 1997.
[GKKZ11] Juan A. Garay, Jonathan Katz, Ranjit Kumaresan, and Hong-Sheng Zhou. Adaptively secure
broadcast, revisited. In Cyril Gavoille and Pierre Fraigniaud, editors, Proceedings of the 30th
Annual ACM Symposium on Principles of Distributed Computing, PODC 2011, San Jose, CA,
USA, June 6-8, 2011, pages 179–186. ACM, 2011.
[GKL15] Juan A. Garay, Aggelos Kiayias, and Nikos Leonardos. The bitcoin backbone protocol: Analysis
and applications. In Advances in Cryptology - EUROCRYPT 2015 - 34th Annual International
Conference on the Theory and Applications of Cryptographic Techniques, Sofia, Bulgaria, April
26-30, 2015, Proceedings, Part II, pages 281–310, 2015.
[GKL16] Juan A. Garay, Aggelos Kiayias, and Nikos Leonardos. The bitcoin backbone protocol with
chains of variable difficulty. IACR Cryptology ePrint Archive, 2016:1048, 2016.
[GMW87] Oded Goldreich, Silvio Micali, and Avi Wigderson. How to play any mental game or a com-
pleteness theorem for protocols with honest majority. In STOC, pages 218–229. ACM, 1987.
[HZ10] Martin Hirt and Vassilis Zikas. Adaptively secure broadcast. In Henri Gilbert, editor, Advances
in Cryptology - EUROCRYPT 2010, 29th Annual International Conference on the Theory and
Applications of Cryptographic Techniques, French Riviera, May 30 - June 3, 2010. Proceedings,
volume 6110 of Lecture Notes in Computer Science, pages 466–485. Springer, 2010.
[JB99] Ari Juels and John G. Brainard. Client puzzles: A cryptographic countermeasure against
connection depletion attacks. In NDSS. The Internet Society, 1999.
[KB16] Ranjit Kumaresan and Iddo Bentov. Amortizing secure computation with penalties. In Weippl
et al. [WKK+ 16], pages 418–429.
[Kin13] Sunny King. Primecoin: Cryptocurrency with prime number proof-of-work.
http://primecoin.io/bin/primecoin-paper.pdf, July 2013.
[KK09] Jonathan Katz and Chiu-Yuen Koo. On expected constant-round protocols for byzantine agree-
ment. Journal of Computer and System Sciences, 75(2):91 – 112, 2009.
[KMTZ13] Jonathan Katz, Ueli Maurer, Björn Tackmann, and Vassilis Zikas. Universally composable
synchronous computation. In TCC, pages 477–498, 2013.
[KP15] Aggelos Kiayias and Giorgos Panagiotakos. Speed-security tradeoffs in blockchain protocols.
IACR Cryptology ePrint Archive, 2015:1019, 2015.
[KP16] Aggelos Kiayias and Giorgos Panagiotakos. On trees, chains and fast transactions in the
blockchain. IACR Cryptology ePrint Archive, 2016:545, 2016.
[KVV16] Ranjit Kumaresan, Vinod Vaikuntanathan, and Prashant Nalini Vasudevan. Improvements to
secure computation with penalties. In Weippl et al. [WKK+ 16], pages 406–417.
[KZZ16] Aggelos Kiayias, Hong-Sheng Zhou, and Vassilis Zikas. Fair and robust multi-party computation
using a global transaction ledger. In Marc Fischlin and Jean-Sébastien Coron, editors, Advances
in Cryptology - EUROCRYPT 2016 - 35th Annual International Conference on the Theory and
Applications of Cryptographic Techniques, Vienna, Austria, May 8-12, 2016, Proceedings, Part
II, volume 9666 of Lecture Notes in Computer Science, pages 705–734. Springer, 2016.
[LSP82] Leslie Lamport, Robert E. Shostak, and Marshall C. Pease. The byzantine generals problem.
ACM Trans. Program. Lang. Syst., 4(3):382–401, 1982.
[ML14] Andrew Miller and Joseph J. LaViola. Anonymous byzantine consensus from moderately-hard
puzzles: A model for bitcoin. University of Central Florida. Tech Report, CS-TR-14-01, April
2014.
[Nak08a] Satoshi Nakamoto. Bitcoin: A peer-to-peer electronic cash system.
http://bitcoin.org/bitcoin.pdf, 2008.

42
[Nak08b] Satoshi Nakamoto. “the proof-of-work chain is a solution to the byzan-
tine generals’ problem”. The Cryptography Mailing List, https://www.mail-
archive.com/[email protected]/msg09997.html, November 2008.
[Nak09] Satoshi Nakamoto. Bitcoin open source implementation of p2p currency.
http://p2pfoundation.ning.com/forum/topics/bitcoin-open-source, February 2009.
[Nei94] Gil Neiger. Distributed consensus revisited. Inf. Process. Lett., 49(4):195–201, 1994.
[OB08] Michael Okun and Amnon Barak. Efficient algorithms for anonymous byzantine agreement.
Theor. Comp. Sys., 42(2):222–238, January 2008.
[Oku05a] Michael Okun. Agreement among unacquainted byzantine generals. In Pierre Fraigniaud, editor,
DISC, volume 3724 of Lecture Notes in Computer Science, pages 499–500. Springer, 2005.
[Oku05b] Michael Okun. Distributed computing among unacquainted processors in the presence of byzan-
tine distributed computing among unacquainted processors in the presence of byzantine failures.
Ph.D. Thesis Hebrew University of Jerusalem, 2005.
[PSL80] Marshall C. Pease, Robert E. Shostak, and Leslie Lamport. Reaching agreement in the presence
of faults. J. ACM, 27(2):228–234, 1980.
[PSS16] Rafael Pass, Lior Seeman, and Abhi Shelat. Analysis of the blockchain protocol in asynchronous
networks. IACR Cryptology ePrint Archive, 2016:454, 2016.
[Rab83] Michael O. Rabin. Randomized byzantine generals. In FOCS, pages 403–409. IEEE Computer
Society, 1983.
[RSW96] R. L. Rivest, A. Shamir, and D. A. Wagner. Time-lock puzzles and timed-release crypto.
Technical report, Cambridge, MA, USA, 1996.
[SZ13] Yonatan Sompolinsky and Aviv Zohar. Accelerating bitcoin’s transaction processing. fast money
grows on trees, not chains. IACR Cryptology ePrint Archive, 2013:881, 2013.
[WKK+ 16] Edgar R. Weippl, Stefan Katzenbeisser, Christopher Kruegel, Andrew C. Myers, and Shai
Halevi, editors. Proceedings of the 2016 ACM SIGSAC Conference on Computer and Com-
munications Security, Vienna, Austria, October 24-28, 2016. ACM, 2016.
[Yao82] Andrew Chi-Chih Yao. Protocols for secure computations (extended abstract). In FOCS, pages
160–164. IEEE, 1982.

A Useful Inequalities
We will require the following inequalities.

Fact 1 (Bernoulli’s inequality). For q ≥ 1 and 0 ≤ p ≤ 1, (1 − p)q ≥ 1 − pq.

α2
Fact 2. For any real α > 0, 1 − α < e−α < 1 − α + 2 .

We provide here the form of the Chernoff bound we use in the proofs.

Theorem 34 (Chernoff bounds). Suppose {Xi : i ∈ [n]} are P

mutually independent Boolean random
variables, with Pr[Xi = 1] = p, for all i ∈ [n]. Let X = ni=1 Xi and µ = pn. Then, for any
δ ∈ (0, 1],
2 2
Pr[X ≤ (1 − δ)µ] ≤ e−δ µ/2 and Pr[X ≥ (1 + δ)µ] ≤ e−δ µ/3 .

In Section 7 we need more general concentration bounds. We provide here the relevant definition
and bounds. (See [DP12, Corollary 5.2])

43
Definition 35. A function f (x1 , . . . , xn ) is k-Lipschitz if |f (x) − f (x0 )| ≤ k, whenever x and x0
differ in at most one coordinate.

Theorem 36. If f is k-Lipschitz and X1 , . . . , Xn are independent random variables, then

 2t2   2t2 
Pr[f > Ef + t] ≤ exp − 2 and Pr[f < Ef − t] ≤ exp − 2 .
nk nk

44