Theoretical Computer Science 942 (2023) 93–106

Contents lists available at ScienceDirect

Theoretical Computer Science

journal homepage: www.elsevier.com/locate/tcs

Decision on block size in blockchain systems by evolutionary

equilibrium analysis ✩,✩✩
Jinmian Chen, Yukun Cheng ∗ , Zhiqi Xu, Yan Cao
Suzhou University of Science and Technology, Suzhou 215009, China

a r t i c l e i n f o a b s t r a c t

Article history: By using the PoW protocol, mining pools compete to successfully mine blocks to pursue
Received 27 March 2022 rewards. Generally, the reward from a mined block includes the fixed block subsidies and
Received in revised form 15 September the time-varying transaction fees. The latter are offered by the senders whose transactions
2022
are packaged into blocks and are increasing with the block size being larger. However, the
Accepted 20 November 2022
Available online 23 November 2022
larger block size brings the longer latency, resulting in a smaller probability of successfully
mining. Therefore, decision on the optimal block size of a block to trade off two factors
Keywords: above mentioned is a complex and crucial problem for the mining pools. In this paper, we
Blockchain model the repeated mining competition dynamics among mining pools as an evolutionary
Block size game, in which each mining pool has two strategies: following the upper bound of block
Mining competition size B̄, or selecting a block size smaller than B̄. Because of the bounded rationality, each
Evolutionary game mining pool pursues its evolutionary stable strategy (ESS) on block size by continuous
learning and adjustments during the whole mining process. A framework is built for the
general evolutionary game, based on which we then explore the existence and stability of
the ESSs for a case of two mining pools. Numerical experiments using the real Bitcoin data
are conducted to demonstrate the theoretical results in this paper.
© 2022 Published by Elsevier B.V.

1. Introduction

Blockchain was initiated from Bitcoin in 2008 [1], a decentralized payment system based on a public transaction ledger.
To reach consensus, the Proof-of-Work (PoW) protocol is used in Bitcoin system to maintain the stability and sustainability
of blockchain. Under the PoW-based consensus protocol, miners are required to provide proofs of solving the PoW puzzle at
the expense of computing power, and propagate the block with the proofs over the whole network to obtain other nodes’
verification. The first block finishing the verification shall be successfully added to the chain. The miner who successfully
appends a block to the chain will be rewarded by the transaction fees, paid by users whose transactions are included in this
block, and the block mining subsidies. The huge success of Bitcoin can be attributed to the economic incentive measure to
motivate miners to mine blocks. In the early stage of Bitcoin, the reward mainly comes from the Bitcoin subsidies. However,
because of a 21,000,000-unit upper bound of Bitcoin supply, the subsidy of a mined block is set to halve every four years
starting from 50 BTC, and will eventually go to zero. Thus, the transaction fees will gradually become the dominant part of
miners’ rewards. Especially, with the rapid increment of the number of transactions in blockchain system, the senders are

✩
This article belongs to Section A: Algorithms, automata, complexity and games, Edited by Paul Spirakis.
✩✩
The preliminary version of this paper appeared in Proceedings of the COCOA2021, pp. 180-194, 2021.
* Corresponding author.
E-mail addresses: [email protected] (J. Chen), [email protected] (Y. Cheng), [email protected] (Z. Xu), [email protected] (Y. Cao).

https://doi.org/10.1016/j.tcs.2022.11.026
0304-3975/© 2022 Published by Elsevier B.V.
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

likely to offer higher fees to compete with others to make sure their transactions are preferentially packaged into a block
with limited space.
At the same time, miners try their best to package as many transactions as possible to earn more rewards, which needs
larger block size to contain them. However, a block with a large size is a heavy burden for the network to propagate it
to all nodes, and thus this block is more likely to be orphaned [2] due to long latency. On the one hand, the rewards of a
block increase with the block size increasing. On the other hand, the probability of successful mining decreases with the
block size being larger. Therefore, there exists a miners’ dilemma, in which miners need to decide a suitable block size to
maximize their expected rewards.
Because of the economic incentive mechanism, miners take active part into the mining game and compete to be the
winner, the first one to solve the puzzle of PoW protocol, for handsome income. Though the expected reward is positive, a
single miner has to wait for a long time with consuming a large amount of computing power to create a block, due to the
constantly adjusted difficulty of block generation. In order to improve the efficiency of mining and get stable revenue, many
miners may join into a mining pool to mine blocks together. In each mining pool, there is a manager who is responsible to
assign the partial targets to miner members, broadcast the mined block to all nodes and allocate the obtained rewards to
miner members according to their partial proof of work. Therefore, the pool managers also face the difficulty on the block
size.
Recently, some literatures focus on the decision problem on the block size to maximize the expected mining rewards.
Rizun [3] introduced the block space supply curve and the mempool demand curve, to show how a rational miner selects
transactions to maximize her profit without a block size limit. With considering the orphaning risk and the revenue from
transaction fees, the block size corresponding to the intersection point of these two curves is proved to be the optimal size
to maximize the miner’s profit. Also, by comparing three market conditions for Bitcoin transaction fees: healthy, unhealthy
and nonexistent, it is found that a rational miner would like to produce a big block if the mempool is full of high fee
density transactions. But the fee is not sufficient to compensate the orphaning risk, and thus the miners may produce
empty blocks, which brings a huge waste. In the evaluation of Bitcoin unlimited security, Zhang and Preneel [4] discussed
two mining games: the EB (excessive block size) Choosing Game and the Block Size Increasing Game. In the EB Choosing
Game, all miners are profitable with any EB value, and a Nash Equilibrium is proved to exist when all miners choose the
same EB. In the Block Size Increasing Game, all miners are assumed to be honest and their maximum profitable block sizes
are public information. Under this setting, it is proved that the miners, who handle the blocks with larger sizes, tend to
form a coalition, and thus increase their block sizes. This result causes other miners to drop out of the mining competition.
Jiang and Wu [5] considered a game in which miners may strategically manipulate their block sizes, that is not following
the default block size, to maximize their expected payoffs. The authors proved that a miner with less computing power
prefers a smaller block size and the Nash equilibria are shown by numerical experiments.
Some work studied the decision problems in blockchain system from the dynamic evolution perspective. Liu et al. [6]
modeled the mining pool selection by an individual miner in the PoW blockchain as an evolutionary game, in which the
computing power of miners and the propagation delay of one block are considered as two major influence factors. The
authors proved that the evolutionary stable solution exists under some conditions for the case of two mining pools. Pan
et al. [7] built a theoretical framework to study how dividend affects user decisions and welfare in a Delegated Proof-of-
Stake (DPoS) blockchain. Especially, they used an evolutionary game analysis to examine the process dynamics of super
node selection. Chen et al. [8] developed an evolutionary game framework based on the temporary fork modeling to reveal
the long-term trends of computing power distributions over mining pools. They proved that the higher uncle block reward
generally results in the lower degree of centralization of computing power distributions in the whole network. In addition,
a longer network delay or a shorter block generation interval lead to a higher probability of temporary fork. These results
finally make the mining power distributions more centralized.
Inspired by [5], this paper adopts the expected payoff function to study the block size determination of mining pools in a
dynamic process. Different with the work in [5], which only considered one-shot block size game by the miners, we focus on
the optimal decision on block size in a repeated mining process, which is more practical. We assume that the mining pools
(each an individual miner is also viewed as a pool containing one miner) are bounded rational and adapt their strategies
according to the received rewards. Accordingly, we model the repeated mining competition as an evolutionary game, where
each pool, controlling a certain amount of computing power, has to decide whether following the upper bound size B̄ or
not, to pursue the maximum expected payoffs. In addition, if the optimal decision is not following B̄ , then an optimal block
size B ∗ (< B̄ ) shall be computed. For the constructed evolutionary game, evolutionary stable strategies (ESS) on block sizes
are considered to be the ideal solutions. We thus propose a theoretical analysis scheme for the general case containing n
mining pools, based on which we continue to explore the necessary conditions for the existences of the ESSs for a case of
two mining pools. To verify the results of our analysis, a series of numerical experiments are conducted, and the sensitivity
analysis of the upper bound B̄ and the unit propagation time is also presented.
The rest of this paper is organized as follows. Section 2 introduces the blockchain system model and the payoff function
of mining pools. In Section 3, we formulate the evolutionary game model to study the block size decision problem, and
propose the theoretical analysis on the necessary conditions for the existence of ESSs under the case of two mining pools.
We also conduct the numerical experiments to verify our theoretical results and present the sensitivity analysis of the upper
bound B̄ and the unit propagation time in Section 4. Finally, the whole work is concluded in the last section.

94
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

2. System model

We consider the PoW-based blockchain system where there are n mining pools, denoted by N = {1, 2, · · · , n}, and each
contains some miners. All mining pools compete to mine blocks by costing an amount of computing power, and thus to
pursue the corresponding rewards. Similar to [9] and [10], we assume the whole system is in a quasi-static state, meaning
no miners join in or leave the system. Under this assumption, each miner keeps its state unchanged, including which mining
pool it is in and how much computing power it has. This leads to the mining pools’ constant scale and their total computing
power.
By the consensus protocol of PoW, a mining pool, who obtains the reward, must satisfy the following two conditions: it
is the first one to solve a proof-of-work puzzle by consuming an amount of computing power and it is also the first one
to make its mined block reach the consensus. The expected average block arriving interval is about of T = 600 seconds, by
automatically adjusting the difficulty of the proof-of-work puzzle. The whole mining process in a blockchain system consists
of a series of one-shot competitions, in each of which one block is mined. A one-shot competition can be viewed as a
non-cooperative game, where all mining pools are the players and they shall make decisions on the mined blocks’ size to
maximize their individual earnings.
Similar to the model in [5], we compute the expected reward of each mining pool i by regarding its block finding time
in one-shot competition as a random variable, denoted by X i , which follows the exponential distribution. Specifically, let B i
be the block size decided by pool i. Denote h i (0 ≤ h i ≤ 1) to be the relative computing power of pool i, that is the ratio
of pool i’s computing power to the total computing power in blockchain system. The propagation time of pool i’s block is
linear with its size B i , i.e., qi = ρ B i . Pool i’s block cannot reach the consensus if the block finding time is less than the
h
propagation time q i . The mining rate of pool i is denoted by λi = Ti , where T = 600 secs is the average block arriving
interval. By the definition of the exponential distribution, the probability density function (PDF) of X i is
⎧
⎨ 0, t < q ;
i
f X i (t ; B i , λi ) = (1)
⎩ λ e −λi (t −qi ) , t ≥ q .
i i

The cumulative distribution function (CDF) of X i is

⎧
⎨ 0, t < q ;
i
F X i (t ; B i , λi ) = P r ( X i ≤ t ) = (2)
⎩ 1 − e −λi (t −qi ) , t ≥ q .
i

So, the probability that block finding time of pool i is larger than t is
⎧
⎨ 1, t < q ;
i
P r ( X i > t ) = 1 − F X i (t ; B i , λi ) = (3)
⎩ e −λi (t −qi ) , t ≥ q .
i

Define X to be the block finding time among all mining pools. Then X = mini ∈ N { X i }, i.e. the first time to find a block,
and hence

P r ( X > t ) = i =1,··· ,n P r ( X i > t ) = i ∈ Acti ve(t ) P r ( X i > t ) = e i ∈ Acti ve (t ) [−λi (t −q i )] , (4)
where Acti ve (t ) = {i |q i ≤ t } is the pool set, each pool i in which has the propagation time less than time t. For convenience,
we call each pool i in Acti ve (t ) an active pool at time t.
From the probability function of (4), the CDF and PDF of random variable X are as the follows

F X (t ; B, λ) = 1 − P r ( X > t ) = 1 − e i ∈ Acti ve (t ) [−λi (t −q i )] ,
 
f X (t ; B, λ) = ( λi )e i ∈ Acti ve (t ) [−λi (t −q i )] ,
i ∈ Acti ve (t )

where B = ( B 1 , · · · , B n ) and λ = (λ1 , · · · , λn ).

The reward of a mined block comes from two aspects: the fixed subsidies R (e.g., 6.25 BTC for one block currently), and
a variable amount of transaction fees. Particularly, the transaction fees are more dependent on the size of a block, since a
block with a larger size contains more transactions. For the sake of simplicity, we assume the total transaction fee is linearly
dependent on the block size, i.e., α B i . This is similar to the suggested pricing standard of transaction fee for users in some
token wallets, such as 0.0005 BTC per KB [11]. So the total reward for a block mined by pool i is R + α B i . In addition, the
probability that a pool i solves the proof-of-work puzzle at time t is the ratio of its computing power to all other active
mining pools at time t. Then the payoff of mining pool i in expectation at time t is
⎧
⎨ 0, t < q i ;
E (re wardi | X = t ) = (5)
⎩  hi
( R + α B i ), t ≥ qi ,
j ∈ Acti ve (t ) h j

95
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

and then its reward in expectation is expressed as

+∞
U i = E [ E (re wardi | X = t )] = E (re wardi | X = t ) · f X (t ; B, λ)dt
−∞ (6)
 

n
e j λ j (q j −ql )
−e j λ j (q j −ql+1 )
= λi ( R + α B i )  ,
l =i j ∈ Acti ve (ql ) λ j

where qn+1 = +∞.

3. Decision on block size by evolutionary equilibrium analysis

In a PoW-based blockchain system, we suppose that there are n independent mining pools, each pool i owning an
amount of relative computing power h i . The whole mining process is a series of one-shot competitions, and all mining pools
compete to mine a block to win the reward in each one-shot. In this paper, we model the mining competition dynamics as
an evolutionary game to study the dynamic interactions among mining pools. In our evolutionary game model, each pool
has two kinds of strategies: to follow the upper bound size B̄ , or to choose the block size less than B̄. For simplicity, the two
strategies are named as “following” strategy and “not following” strategy, respectively. Because of the bounded rationality,
each mining pool pursues its evolutionary stable block size through continuous learning and adjustments.
In this section, we first propose the analysis framework for general case, and then theoretically analyze the existence
and stability of the equilibrium solutions for a case of two mining pools.

3.1. Analysis framework

In our evolutionary game model, there is a crucial problem for each mining pool i that is how to decide the optimal
block size to maximize its payoff in expectation. Note that each pool i has two kinds of strategies: one is to fix the block
size to the upper bound size B̄, and the other is to choose a block size B i < B̄. So, in the k-th shot, let us define two subsets:

N 1 (k) = {i ∈ N | B i (k) < B̄ } and N 2 (k) = N − N 1 (k) = {i ∈ N | B i (k) = B̄ },

and call ( N 1 (k), N 2 (k)) a subset profile. Clearly, subset profile ( N 1 (k), N 2 (k)) is determined after all mining pools making
decisions on their block sizes. There are 2n subset profiles totally in each one-shot, and hence we denote the collection of
subset profiles in the k-th slot by N (k) = {( N 1 (k), N 2 (k))}.
Suppose that a subset profile ( N 1 (k), N 2 (k)) in the k-th shot is given. Each pool i ∈ N 1 (k) selects the “not following”
strategy. In addition, it continues to decide the optimal block size B ∗i < B̄ by maximizing its expected payoff π(iN 1 (k), N 2 (k)) =
max B i < B̄ U i , such that
  

n
e j λ j (q j −ql )
−e j λ j (q j −ql+1 )
∗ 
B i = arg max U i = arg max λi ( R + α B i ) .
B i < B̄ B i < B̄
l =i j ∈ Acti ve (ql ) λ j

Each mining pool i ∈ N 2 (t ) sets its block size as B̄ and has its payoff
 

n
e j λ j (q j −ql )
−e j λ j (q j −ql+1 )
π(iN 1 (k),N 2 (k)) = λi ( R + α B̄ )  . (7)
l =i j ∈ Acti ve (ql ) λ j

In the evolutionary game, the mining pools keep learning to adjust their low-income strategies to a higher-income one
dynamically. Until n mining pools reach a stable strategy profile, at which no one would like to change its strategy, an
equilibrium state of block size ( B ∗1 , B ∗2 , · · · , B n∗ ) is obtained. Though all the mining competitions are carried out during a
series of discrete slots, we can view each block generating slot as a very small interval with respect to the whole mining
process, and hence deal with it as a continuous version. It allows us to apply the standard technique to study the evolution-
ary process for the decisions on block size. Let xi (k), 0 ≤ xi (k) ≤ 1, represent the probability of mining pool i ∈ N to choose
the “not following” strategy at the k-th slot. Correspondingly, the probability of pool i to choose the upper bound size is
1 − xi (k).
If the choice of pool i is not to follow the upper bound size, then its conditional expected payoff is
⎛ ⎞

E i1 (k) = ⎝ xl (k) (1 − xl (k)) · π(iN 1 (k), N 2 (k)) ⎠ ; (8)
1 2
( N (k), N (k))∈N (k), l∈ N 1 (k),l=i l∈ N 2 (k)
i ∈ N 1 (k)

96
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

If mining pool i selects the “following” strategy, then its conditional expected payoff is
⎛ ⎞

E i2 (k) = ⎝ xl (k) (1 − xl (k)) · π(iN 1 (k), N 2 (k)) ⎠ . (9)
( N 1 (k), N 2 (k))∈N (k), l∈ N 1 (k) l∈ N 2 (k),l=i
i ∈ N 2 (k)

Combining (8) and (9), the average payoff of mining pool i is

Ē i (k) = xi (k) E i1 (k) + (1 − xi (k)) E i2 (k). (10)

By [12], the growth rate of a strategy selected by a participant just equals to the difference between the payoff of this
strategy and its average payoff. Then the replicator dynamic equations for all mining pools are:

f i (x) = ẋi (k) = xi (k)( E i1 (k) − Ē i (k)), ∀i ∈ N . (11)

According to the replicator dynamics (11), a mining pool would like to choose a smaller block size, when its conditional
payoff E i1 (k) is larger than the average payoff E i (k). Otherwise, it will set its block size as B̄ . A state is stable if no mining
pool would like to change its strategy over time in the replicator dynamics, and such a stable state is considered to be the
evolutionary equilibrium [13]. The strategies in this state are evolutionary stable, called ESS. Specifically speaking, when the
payoff of “not following” strategy is equal to the average payoff for each pool, all mining pools achieve the ESS and no one
has incentive to change its current strategy. Therefore, the ESS can be obtained by solving ẋi (k) = 0 for all i ∈ N, whose
solution is called the fixed equilibrium point of replicator dynamics.

3.2. A case study of two mining pools

Based on the analysis framework for general case in previous subsection, we continue to study the case of two mining
pools (n = 2) to exemplify the equilibrium analysis for the decision on block size. Wenormalize the whole computing power
2
in system, thus mining pool i’s relative computing power is h i ∈ h = {h1 , h2 } and i =1 h i = 1. The whole mining process
contains a series of one-shot competitions, and pool 1 and pool 2 need to decide their block sizes B 1 and B 2 to pursue the
optimal payoffs in each one-shot competition. Without loss of generality, we concentrate on the case of 0 ≤ h 1 ≤ 12 ≤ h2 ≤ 1.
The analysis for the case of h2 ≤ h1 is symmetric, and thus we omit the discussion. By the result in [5] “a miner with less
mining power prefers a smaller block size in order to optimize his payoff”, we have the case of 0 ≤ h1 ≤ h2 ≤ 1 leads to
B 1 ≤ B 2 and then the propagation time q1 = ρ B 1 ≤ q2 = ρ B 2 .
In the following, we shall explore the expected payoffs of pool 1 and 2 by distinguishing two subcases.

Subcase 1. In a one-shot mining competition, if B 1 < B 2 ≤ B̄, then

 

2
e λ j (q j −ql ) −e λ j (q j −ql+1 )
U 1 = λ1 ( R + α B 1 )  = ( R + α B 1 )[1 − h2 e λ1 ρ ( B 1 − B 2 ) ];
l =1 j ∈ Acti ve (ql ) λ j
  (12)

2
e λ j (q j −ql ) −e λ j (q j −ql+1 )
 λ1 ρ ( B 1 − B 2 )
U 2 = λ2 ( R + α B 2 ) = ( R + α B 2 )h2 e .
l =2 j ∈ Acti ve (ql ) λ j

Subcase 2. In a one-shot mining competition, if B 1 = B 2 ≤ B̄, then

λ1
U1 = ( R + α B 1 ) = h1 ( R + α B 1 );
λ1 + λ2
(13)
λ2
U2 = ( R + α B 2 ) = h2 ( R + α B 2 ).
λ1 + λ2
Since each mining pool has two kinds of strategies, i.e., to follow the upper bound size B̄, and not to follow, in a one-shot
competition, there are four strategy profiles: ( B 1 , B 2 ), ( B 1 , B̄ ), ( B̄ , B 2 ), and ( B̄ , B̄ ). Note that ( B̄ , B 2 ) either does not exist,
or equals to ( B̄ , B̄ ) under the condition of B 1 ≤ B 2 ≤ B̄. So we do not discuss this strategy profile any more. The payoffs of
two pools under different strategy profiles in the following payoff matrix (Table 1).
Specifically, for the strategy profile ( B 1 , B 2 ), meaning that both pools choose the “not following” strategy, we define the
1 2
payoffs of two pools π11 and π11 representing their optimal payoffs subject to the conditions of B 1 < B̄ and B 2 < B̄. The
corresponding optimal block sizes ( B ∗1 , B ∗2 ) can be obtained by solving ∂ U 1 ∂( BB1 , B 2 ) = 0 and ∂ U 2 ∂( BB1 , B 2 ) = 0, simultaneously. For
1 2
1
the strategy profile ( B 1 , B̄ ), showing that pool 1 would not follow the upper bound size and pool 2’s block size is B̄ , π12
2 1
and π12 are denoted to be the payoffs of pool 1 and 2. To be specific, π12 is defined to be the optimal payoff of pool 1
under the condition of B 1 < B̄ and the corresponding optimal block size B ∗1 can be determined by solving dU 1 ( B 1 , B̄ )
dB 1
= 0. For
1 2
the strategy profile of ( B̄ , B̄ ), both of two pools set their block sizes as B̄ , then their payoffs are denoted π
by 22 π
and 22 .

97
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

Table 1
Payoff matrix of the case of two mining pools.

Mining Pool 2

B2 B̄
1 2 1 2
Mining Pool 1 B1 (π11 , π11 ) (π12 , π12 )
B̄ (\, \) 1
(π22 2
, π22 )

Lemma 1. In a one-shot mining competition, if B 1 ≤ B 2 ≤ B̄, then

1. for strategy profile ( B 1 , B 2 ) with 0 ≤ B 1 < B 2 < B̄, the optimal block size of pool 2 is B ∗2 = λ 1ρ − αR , if 0 < λ 1ρ − αR < B̄. Let 
B ∗1
1 1
dU 1 ( B 1 , B ∗2 )
be the solution satisfying dB 1
B ∗1 < B ∗2 , then the optimal block size of pool 1 is 
= 0. If 0 ≤  B ∗1 . Thus, the payoffs are

∗ ∗ ∗ ∗
1
π11 = ( R + α
B ∗1 )[1 − h2 e λ1 ρ ( B 1 − B 2 ) ], π11
2
= ( R + α B ∗2 )h2 e λ1 ρ ( B 1 − B 2 ) . (14)

If 
B ∗ < 0, then the best choice of pool 1 is to set its block size as zero and the payoffs are
1
∗ ∗
1
π11 = R (1 − h2 e −λ1 ρ B 2 ), π11
2
= ( R + α B ∗2 )h2 e −λ1 ρ B 2 . (15)

2. for strategy profile ( B 1 , B̄ ) with 0 < B 1 < B 2 = B̄, the block size of pool 2 is B̄. Let 
B ∗1 be the solution satisfying
dU 1 ( B 1 , B̄ )
dB 1
= 0. If
0≤
B∗ 1< B̄, then the optimal block size of pool 1 is 
B ∗ . Thus, the payoffs are1

∗ λ1 ρ (
B ∗1 − B̄ ) ∗
1
π = ( R + α B 1 )[1 − h2 e
12 ], π 2
12 = ( R + α B̄ )h2 e λ1 ρ ( B 1 − B̄ ) . (16)

If 
B ∗1 < 0, then the best choice of pool 1 is to set its block size as zero and the payoffs are

1
π12 = R (1 − h2 e −λ1 ρ B̄ ), π12
2
= ( R + α B̄ )h2 e −λ1 ρ B̄ . (17)

3. for strategy profile ( B̄ , B̄ ), the block sizes of two pools are both equal to B̄ and the corresponding payoffs are

1 2
π22 = h1 ( R + α B̄ ), π22 = h2 ( R + α B̄ ). (18)

Remark. By Lemma 1-(1) and (2), the optimal block size of pool 1 depends on the block size of pool 2 in strategy profiles
∗
(
B ∗1 , B ∗2 ) and ( = 0, if 0 ≤ 
B ∗1 < B ∗2 and 0 ≤ 
dU 1 ( B 1 , B 2 )
B ∗1 , B̄ ), and can be obtained from equations = 0 and dU 1dB B ∗1 < B̄,
( B 1 , B̄ )
dB 1 1
respectively. For convenience, we denote g ( B 2 ) to be the implicit function, satisfying

dU 1 ( B 1 , B 2 )
= α − [α + λ1 ρ ( R + α g ( B 2 )]h2 e λ1 ρ ( g ( B 2 )− B 2 ) = 0.
dB 1
Therefore,
1 R

B ∗1 = g ( B ∗2 ) = g ( − ) and 
B ∗1 = g ( B̄ ).
λ1 ρ α

dU 1 ( B 1 , B 2 )
Lemma 2. Let B 1 = g ( B 2 ) be the implicit function satisfying dB 1
= 0. Then g ( B 2 ) is monotone increasing with B 2 and g ( B 2 ) <
B 2 for all B 2 ≥ 0.

Proof. Because g ( B 2 ) satisfies

dU 1 ( B 1 , B 2 )
= α − [α + λ1 ρ ( R + α g ( B 2 ))]h2 e λ1 ρ ( g ( B 2 )− B 2 ) = 0,
dB 1
we have
α + λ1 ρ ( R + α g ( B 2 ))
1 > g (B2) = > 0.
2α + λ1 ρ ( R + α g ( B 2 ))
So g ( B 2 ) is monotonically increasing with respect to B 2 . Moreover, when B 2 = 0, U 1 ( B 1 , 0) = ( R + α B 1 )[1 − h2 e λ1 ρ B 1 ], and
for all B 1 ≥ 0,

d 2 U 1 ( B 1 , 0) dU 1 ( B 1 , 0)
< 0, and | B 1 =0 < 0.
dB 21 dB 1

98
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

It means that function U 1 ( B 1 , 0) is strictly concave, and the maximal point g (0) is less than 0. Let us define an auxiliary
function G ( B 2 ) = g ( B 2 ) − B 2 . Since g ( B 2 ) < 1, G ( B 2 ) decreases on interval [0, ∞). Thus G ( B 2 ) ≤ G (0) < 0 for all B 2 ≥ 0,
indicating g ( B 2 ) < B 2 . 

Based on the monotonicity of g ( B 2 ), we have B ∗1 < 

B ∗1 , if B ∗2 = λ 1ρ − αR < B̄. Moreover, the property of g ( B 2 ) ≤ B 2
1
ensures 
B 1 < B 2 = λ ρ − α and 
∗ ∗ 1 R ∗
B 1 < B̄.
1
In the following, we go to analyze the strategy selections of two mining pools by distinguishing three conditions:

(1) λ 1ρ − αR ≥ B̄; (2) 0 < λ 1ρ − αR < B̄; (3) λ 1ρ − αR ≤ 0,

1 1 1

to explore the equilibrium solutions in the evolutionary game in the following.

3.2.1. Discussion under the condition of λ 1ρ − αR ≥ B̄

1

Theorem 3. In a one-shot mining competition, if 0 ≤ B 1 ≤ B 2 < B̄ and λ 1ρ − αR ≥ B̄, then pool 2 decides its block size as B̄, and
1

1. (B ∗1 , B̄ ) is a strict Nash equilibrium, when 0 ≤  B ∗1 < B̄; or

2. (0, B̄ ) is a strict Nash equilibrium, when  B ∗1 < 0; or
3. ( B̄ , B̄ ) is a strict Nash equilibrium, when B ∗1 ≥ B̄.

2
Proof. Since ∂∂ UB 2 | B = 1 − R = 0 and ∂ U22 < 0 for any block size of B 1 , U 2 is strictly concave and achieves its maximality at
2 2 λ ρ
1 α ∂ B2
the boundary point of B̄ subject to the condition of λ 1ρ − αR ≥ B̄. Thus if λ 1ρ − αR ≥ B̄, then B̄ is the dominant strategy of
1 1
pool 2. In addition, given pool 2’s strategy of B̄, the payoff function U 1 only depends on B 1 , denoted by U 1 ( B 1 , B̄ ). Since

B ∗1 is the solution of
dU 1 ( B 1 , B̄ )
B ∗1 if 0 ≤ 
= 0, U 1 reaches its maximality at  B ∗1 < B̄. Moreover,
dB 1

d2 U 1 ( B 1 , B̄ )
= −h2 λ1 ρ [2α + λ1 ρ ( R + α B 1 )]e λ1 ρ ( B 1 − B̄ ) < 0,
dB 21
showing U 1 is strictly concave with respect to B 1 .
Therefore,

1
π12 = U 1 (
B ∗1 , B̄ ) > lim U 1 ( B 1 , B̄ ) = h1 ( R + α B̄ ) = π22
1
,
B 1 → B̄ −

B ∗1 . Hence, (
meaning the best response of pool 1 is  B ∗1 , B̄ ) is a strict Nash equilibrium.
 ∗
If B 1 ≥ B̄, then

1
π22 = h1 ( R + α B̄ ) = lim U 1 ( B 1 , B̄ ) > max B 1 ∈[0, B̄ ) U 1 ( B 1 , B̄ ),
B 1 → B̄ −

by the strict concavity of U 1 . It means the best response of pool 1 is B̄ . So ( B̄ , B̄ ) is a strict Nash equilibrium, if  B ∗1 ≥ B̄.
When  ∗
B 1 < 0, the strict concavity of U 1 ensures that U 1 ( B 1 , B̄ ) decreases when B 1 ∈ [0, B̄ ], and thus U 1 (0, B̄ ) =
max B 1 ∈[0, B̄ ] U 1 ( B 1 , B̄ ). Therefore, the best response of pool 1 is 0 and (0, B̄ ) is a strict Nash equilibrium if 
B ∗1 < 0. 

Theorem 3 illustrates that the dominant strategy of pool 2 is to follow the upper bound size B̄ if λ 1ρ − αR ≥ B̄, while the
1
optimal strategy of pool 1 depends on the value of 
B ∗1 .

3.2.2. Discussion under the condition of 0 < λ 1ρ − αR < B̄

1
Next, we concentrate on the condition of 0 < λ 1ρ − αR < B̄, under which pool 2 may set its block size of B ∗2 = λ 1ρ − αR
1 1
or B̄. Recall that xi (k), i = 1, 2, is the probability of mining pool i to adopt the “not following” strategy in the k-th slot
competition, then 1 − xi (k) is the probability of mining pool i to follow the upper bound size. The expected payoffs of
mining pool 1 in the k-th slot competition, when it chooses “not following” or “following” strategy, are

E 11 (k) = x2 (k)π11
1 1
+ (1 − x2 (k))π12 ; E 12 (k) = (1 − x2 (k))π22
1
.
The average payoff of mining pool 1 is

E 1 (k) = x1 (k) E 11 (k) + (1 − x1 (k)) E 12 (k).

99
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

Similarly, we derive the expected payoffs of mining pool 2 as follows,

E 21 (k) = x1 (k)π11
2
; E 22 (k) = x1 (k)π12
2 2
+ (1 − x1 (k))π22 .
The average payoff of mining pool 2 is

E 2 (k) = x2 (k) E 21 (k) + (1 − x2 (k)) E 22 (k).

Based on the analysis framework (11) for the general case, the replicator dynamic system of pool 1 and 2 for the case of
two mining pools are:
  
f 1 (x) = ẋ1 (k) = x1 (1 − x1 )( E 11 − E 1 ) = x1 (1 − x1 ) (π11
1 1
− π12 1
+ π22 1
)x2 + (π12 1
− π22 ) ;
  (19)
f 2 (x) = ẋ2 (k) = x2 (1 − x2 )( E 21 − E 2 ) = x2 (1 − x2 ) (π11
2 2
− π12 2
+ π22 2
)x1 − π22 .

Note that all the solutions satisfying

f 1 (x) = ẋ1 (k) = 0, and f 2 (x) = ẋ2 (k) = 0

are the fixed equilibrium points of the replicator dynamic system. There exist four fixed equilibrium points of this system:

(0, 0), (1, 0), (1, 1) and (x∗1 , x∗2 ),

2
π22 1
π22 −π12
1
where x∗1 = ∗
2 , x2 =
∗ ∗
1 . To fulfill the condition for probability vector x, x1 and x2 must be in [0, 1].
2
π11 −π12
2
+π22 1
π11 −π12
1
+π22

Theorem 4. For the evolutionary game between two mining pools, if 0 ≤ B 1 ≤ B 2 ≤ B̄ and 0 < λ 1ρ − αR < B̄, then
1

∗
1. the fixed equilibrium point (1, 0) is an ESS, when (1) 0 ≤  B ∗1 <  B ∗1 < B̄ or (2) B ∗1 < 0 < 
R
B ∗1 < B̄ and λ1 ρ ( αR + B̄ )e λ1 ρ ( B 1 − B̄ − α )+1 >
1;
∗
2. the fixed equilibrium point (1, 1) is an ESS, when (1)  B ∗1 <  B ∗1 < 0 or (2)  B ∗1 < 0 < 
R
B ∗1 < B̄ and λ1 ρ ( αR + B̄ )e λ1 ρ ( B 1 − B̄ − α )+1 < 1;
3. the fixed equilibrium points (0, 0) and (x∗1 , x∗2 ) cannot be ESSs.

Proof. To obtain the ESS of the evolutionary game for block size selection, we first compute the Jacobian matrix of the
replicator dynamic system (19):
⎡ ⎤
∂ f 1 (x) ∂ f 1 (x)
J =⎣ ∂ x1 ∂ x2 ⎦,
∂ f 2 (x) ∂ f 2 (x)
∂ x1 ∂ x2

where

∂ f 1 (x) 1 1 1 1 1
= (1 − 2x1 ){[π11 − (π12 − π22 )]x2 + (π12 − π22 )};
∂ x1
∂ f 1 (x) 1 1 1
= x1 (1 − x1 )[π11 − (π12 − π22 )];
∂ x2
∂ f 2 (x) 2 2 2
= x2 (1 − x2 )[(π11 − π12 ) + π22 ];
∂ x1
∂ f 2 (x) 2 2 2 2
= (1 − 2x2 ){[(π11 − π12 ) + π22 )]x1 − π22 }.
∂ x2
By the results in [12], if a fixed equilibrium point (x1 , x2 ) is an ESS, then the Jacobian matrix of the replicator dynamic
system is negative definite at (x1 , x2 ), equivalent to determinant Det ( J (x1 , x2 )) > 0 and trace T r ( J (x1 , x2 )) < 0.
To simplify the discussion, let us denote

1 1 1 2 2 2
K = π12 − π22 , L = π11 , M = π22 , N = π11 − π12 .
Table 2 shows the determinants and the traces of Jacobian matrix at different fixed equilibrium points.
Next we propose the following fact, based on which we can obtain this theorem.

Fact 5. Based on the expressions of i

π11 i
, π12 i
and π22 , i = 1, 2, in eq. (14)-(18), we have K = π12
1
− π22
1
> 0, L = π11
1
> 0, M =
2
π22 > 0, and

100
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

Table 2
The determinants and traces of Jacobian ma-
trix J at fixed equilibrium points.

Det ( J ) T r( J )

(0,0) K · (− M ) K −M

(1,0) (− K ) · N N−K

(1,1) (− L ) · (− N ) −L − N
(x∗1 , x∗2 ) K LM N
( L − K )( M + N ) 0

2 2
N = π11 − π12
⎧ 
⎪ λ1 ρ (
B ∗1 − λ 1ρ + αR ) ∗
⎪
⎪ R + α ( 1
− R
α h2 e
) 1 − ( R + α B̄ )h2 e λ1 ρ ( B 1 − B̄ ) < 0, if 0 ≤ B ∗1 < 
B ∗1 < B̄ ;
⎪
⎪
λ1 ρ

⎪ 1 R
⎨ R + α ( 1 − R ) h2 e λ1 ρ (− λ1 ρ + α ) − ( R + α B̄ )h2 e λ1 ρ (B ∗1 − B̄ ) < 0(> 0),
λ1 ρ α
=
⎪
⎪ if B 1 < 0 < B 1 < B̄ and λ1 ρ ( αR + B̄ )e λ1 ρ (B 1 − B̄ − α )+1 > 1(< 1);
 ∗ ∗ ∗ R
⎪
⎪
⎪
⎪  
⎩ R + α ( 1 − R ) h e λ1 ρ (− λ11ρ + αR ) − ( R + α B̄ )h e −λ1 ρ B̄ > 0, if  B∗ < B ∗ ≤ 0.
λ1 ρ 2
α 2 1 1

Proof. It is not hard to see L > 0 and M > 0. By the concavity of U 1 ( B 1 , B̄ ) shown in the proof of Theorem 3, U 1 ( B 1 , B̄ ) is
monotonically decreasing when B 1 ∈ [ B ∗1 , B̄ ]. Hence, if 0 ≤ 
B ∗1 < B̄, then π12
1
= U 1 (
B ∗1 , B̄ ) > U 1 ( B̄ , B̄ ) = π22
1
; and if 
B ∗1 < 0,
then π12 = U 1 (0, B̄ ) > U 1 ( B̄ , B̄ ) = π22 . So K > 0.
1 1

If 0 ≤ 
B ∗1 < 
B ∗1 < B̄, let us define
∗
F ( B 2 ) = ( R + α B 2 )h2 e λ1 ρ ( B 1 − B 2 ) = ( R + α B 2 )h2 e λ1 ρ ( g ( B 2 )− B 2 ) .
Then
dF ( B 2 )  
= h2 α + λ1 ρ ( R + α B 2 )( g ( B 2 ) − 1) e λ1 ρ ( g ( B 2 )− B 2 )
dB 2
 
2 − λ1 ρ ( B 2 − g ( B 2 ))
= h2 α e λ1 ρ ( g ( B 2 )− B 2 ) > 0,
2α + λ1 ρ ( R + α g ( B 2 ))
2 2 ρ ( B − g ( B )) ρB
where the inequality comes from 0 < h1 T
< T 2 < 1, because the propagation time ρ B 2 is no more than the
2
interval T = 600 secs in each one-shot competition. F ( B 2 ) is monotonically increasing with respect to B 2 . Therefore, π11 =
∗
F ( B 2 ) < F ( B̄ ) = π12 , and N < 0.
2

If B ∗1 < 
B ∗1 ≤ 0, let us define

F ( B 2 ) = ( R + α B 2 )h2 e −λ1 ρ B 2 ,
then
dF ( B 2 )
= h2 e −λ1 ρ B 2 [α − ( R + α B 2 )λ1 ρ ] .
dB 2
Clearly, F ( B 2 ) decreases when B 2 ∈ [ B ∗2 , B̄ ], and thus 2
π11 = F ( B ∗2 ) > F ( B̄ ) = π12
2
. So we have N = π11
2
− π12
2
> 0. 

Because T r ( J (x∗1 , x∗2 )) = 0 and Det ( J (0, 0)) = − K · M < 0, (x∗1 , x∗2 ) and (0, 0) cannot be ESS.
According to the results of Fact 5, we have Det ( J (1, 0)) > 0 and T r ( J (1, 0)) < 0 when (1) 0 ≤  B ∗1 < 
B ∗1 < B̄ or (2)
 ∗
B ∗1 < 0 < 
R
B ∗1 < B̄ and λ1 ρ ( αR + B̄ )e λ1 ρ ( B 1 − B̄ − α )+1 > 1, and then the fixed equilibrium points (1, 0) is an ESS.
∗
When (1)  B ∗1 ≤ 0 or (2) 
B ∗1 <  B ∗1 < 0 < 
R
B ∗1 < B̄ and λ1 ρ ( αR + B̄ )e λ1 ρ ( B 1 − B̄ − α )+1 < 1, Det ( J (1, 1)) > 0 and T r ( J (1, 1)) < 0,
i.e., the fixed equilibrium points (1, 1) is an ESS.
This theorem holds. 

Theorem 4 illustrates the evolutionary stable solutions of mining pools in the evolutionary mining game. Based on the
definition of the probability of each strategy selected by mining pool, we have the strategy selection of each pool at last.

Corollary 6. For the evolutionary game between two mining pools, if 0 ≤ B 1 ≤ B 2 ≤ B̄ and 0 < λ 1ρ − αR < B̄, then
1

 ∗
1. (
B ∗1 , B̄ ) is an evolutionary stable strategy profile, when 0 ≤ 
B ∗1 < 
R
B ∗1 < B̄, or under the condition of λ1 ρ ( αR + B̄ )e λ1 ρ ( B 1 − B̄ − α )+1
∗ 
> 1 when B 1 < 0 < B 1 < B̄;∗

101
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

Table 3
The payoff matrix under the condition of 
B ∗1 ≤ 0.

B2 = 0 B 2 = B̄

B1 = 0 U 1 (0, 0), U 2 (0, 0) U 1 (0, B̄ ), U 2 (0, B̄ )

B 1 = B̄ \, \ U 1 ( B̄ , B̄ ), U 2 ( B̄ , B̄ )

∗
B ∗1 < 
2. (0, λ 1ρ − αR ) is an evolutionary stable strategy profile, when 
R
B ∗1 < 0, or under the condition of λ1 ρ ( αR + B̄ )e λ1 ρ ( B 1 − B̄ − α )+1
1
 ∗  ∗
< 1 when B 1 < 0 < B 1 < B̄.

Proof. If one of conditions in Theorem 4-1 is satisfied, then (1, 0) is the evolutionary stable solution, meaning pool 2
chooses its block size as B̄ at last and pool 1’s block size is smaller than B̄ . In addition, because 
B ∗1 > 0 and
dU 1 ( B 1 , B̄ )
= dB 1
| B 1 =B ∗
1
 ∗ ∗
0, we have the optimal block size of pool 1 is B 1 . Hence, ( B 1 , B̄ ) is the evolutionary stable strategy profile.
If one of conditions in Theorem 4-2 is satisfied, then (1, 1) is the evolutionary stable solution, meaning pool 1 and pool
2 both choose block size smaller than B̄. Since ∂∂ UB 2 | B = 1 − R = 0 and λ 1ρ − αR < B̄, we have B ∗2 = λ 1ρ − αR is the optimal
2 2 1
λ1 ρ α 1
dU 1 ( B 1 , B ∗2 )
block size of pool 2. Furthermore, as 
B ∗1 < 0, where dB 1
| B 1 =B ∗ = 0, the optimal block size of pool 1 must be zero. So
1
(0, λ1 ρ − α ) is the evolutionary stable strategy profile. 
1 R

3.2.3. Discussion under the condition of λ 1ρ − αR ≤ 0

1

Theorem 7. If 0 ≤ B 1 ≤ B 2 < B̄ and λ 1ρ − αR ≤ 0, then

1

1. in a one-shot mining competition,

(a) ( B 1 = 0, B 2 = 0) is a strict Nash equilibrium when  B ∗1 ≤ 0;

(b) ( B̄ , B̄ ) is a strict Nash equilibrium when 
B ∗1 ≥ B̄.

2. for the evolutionary game between two pools when 0 < 

B ∗1 < B̄,

∗
(a) ( B 1 = 0, B 2 = 0) is an evolutionary stable strategy profile, if R
R +α B̄
> e λ1 ρ ( B 1 − B̄ ) ;
(b) (
B ∗1 , B̄ ) is an evolutionary stable strategy profile, if R
<e λ1 ρ (
B ∗1 − B̄ )
.
R +α B̄

2
Proof. Since ∂∂ UB 2 | B = 1 − R = 0 and ∂ U22 < 0 for any block size of B 1 , U 2 is strictly concave and achieves its maximality at
2 2 λ ρ
1 α ∂ B2
B 2 = 0 subject to the condition of λ 1ρ − αR ≤ 0, thus U 2 ( B 1 , 0) > U 2 ( B 1 , B̄ ).
1

dU 1 ( B 1 , 0)
= α − h2 e λ1 ρ B 1 [α + λ1 ρ ( R + α B 1 )] ≤ α [1 − h2 e λ1 ρ B 1 (2 + λ1 ρ B 1 )] < α (1 − 2h2 ) < 0;
dB 1
d 2 U 1 ( B 1 , 0)
= −h2 λ1 ρ e λ1 ρ B 1 [2α + λ1 ρ ( R + α B 1 )] < 0.
dB 21
dU 1 ( B 1 ,0)
For dB 1
, the first inequality is correct under λ 1ρ − αR ≤ 0, and the second one is correct due to 0 < λ1 ρ B 1 < 1, then
1
d2 U 1 ( B 1 ,0)
the last one is correct because of 1/2 < h2 < 1. That < 0 for 0 ≤ B 1 ≤ B̄ indicates U 1 ( B 1 , 0) is strictly concave.
dB 21

Therefore, U 1 ( B 1 , 0) achieves its maximality at point (0, 0), i.e., U 1 (0, 0) > U 1 ( B̄ , 0). By 
B ∗1 obtained from
dU 1 ( B 1 , B̄ )
dB 1
= 0 and
2
d U 1 ( B 1 , B̄ )
< 0, U 1 ( B 1 , B̄ ) is strictly concave. Thus, B 1 = 0 is the maximal point if 
B ∗1 ≤ 0, B 1 = 
B ∗1 is the maximal point if
dB 21
0<
B ∗1 < B̄, and B 1 = B̄ achieves the maximum of U 1 ( B 1 , B̄ ) if 
B ∗1 ≥ B̄. Specifically, when 
B ∗1 ≤ 0, the game is similar to a
Prisoner’s Dilemma (Table 3).
It is obvious that the best response of pool 1 to B 2 = B̄ is B 1 = 0, due to U 1 (0, B̄ ) > U 1 ( B̄ , B̄ ). However, for any B 1 ≥ 0,
the optimal strategy of pool 2 is B 2 = 0. Hence, the strategy profile (0, 0) is a strict NE if  B ∗1 ≤ 0. As 
B ∗1 ≥ B̄, U 1 ( B 1 , B̄ )
reaches its maximum at point B 1 = B̄. Then the corresponding payoff matrix is shown in Table 4.
Since U i (0, 0) < U i ( B̄ , B̄ ), i = 1, 2, the strategy profile ( B̄ , B̄ ) is a strict NE if 
B ∗1 ≥ B̄.
Next, we analyze the case of 0 <  ∗
B 1 < B̄ in an evolutionary game. For strategy profile ( B 1 = 0, B 2 = 0), the payoffs are
1 2
π11 = Rh1 , π11 = Rh2 .

102
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

Table 4
The payoff matrix under the condition of 
B ∗1 ≥ B̄.

B2 = 0 B 2 = B̄

B1 = 0 Rh1 , Rh2 \, \
B 1 = B̄ \, \ ( R + α B̄ )h1 , ( R + α B̄ )h2

For strategy profile (

B ∗1 , B̄ ), the payoffs are
∗ ∗
1
π12 = ( R + α
B ∗1 )[1 − h2 e λ1 ρ ( B 1 − B̄ ) ], π12
2
= ( R + α B̄ )h2 e λ1 ρ ( B 1 − B̄ ) .
For strategy profile ( B̄ , B̄ ), the payoffs are

1 2
π22 = ( R + α B̄ )h1 , π22 = ( R + α B̄ )h2 .
Then L = π11
1
> 0, M = π22
2
> 0, K = π12
1
− π22
1
and N = π11
2
− π12
2
. Furthermore, we define function H (x) = ( R + α x)[1 −
h2 e λ1 ρ (x− B̄ ) ], x ∈ (0, B̄ ), thus K = H (
B ∗1 ) − H ( B̄ ). So

dH (x)
= α − h2 e λ1 ρ (x− B̄ ) [α + λ1 ρ ( R + α x)];
dx
d2 H (x)
= −h2 λ1 ρ e λ1 ρ (x− B̄ ) [2α + λ1 ρ ( R + α x)] < 0.
dx2
d2 H (x)
Note that x = 
B ∗1 satisfies α − h2 eλ1 ρ (x− B̄ ) [α + λ1 ρ ( R + α x)] = 0, i.e., H (
B ∗1 ) = 0. With dx2
< 0, there exists H (
B ∗1 ) >
H ( B̄ ), i.e., K > 0.
⎧
⎨ > 0, if R ∗
> e λ1 ρ ( B 1 − B̄ ) ;
∗
N = Rh2 − ( R + α B̄ )h2 e λ1 ρ ( B 1 − B̄ ) R +α B̄
⎩ < 0, if R ∗
< e λ1 ρ ( B 1 − B̄ ) .
R +α B̄
∗
So, the fixed point (1, 1), corresponding to strategy profile ( B 1 = 0, B 2 = 0), is an ESS if R
R +α B̄
> e λ1 ρ ( B 1 − B̄ ) ; the fixed point
∗
(1, 0), corresponding to strategy profile (
B ∗1 , B̄ ), is an ESS if R
R +α B̄
< e λ1 ρ ( B 1 − B̄ ) . 

4. Numerical experiments

In this section, we would analyze the case of two mining pools on the decision of block size, based on the statistic data
of Bitcoin blockchain on [14], and the case of three mining pools is to be further discussed.
Firstly, we consider the following setting to explore the impact of different upper bound block sizes on the strategy
selection between two mining pools: the unit transaction fee α = 1 × 10−6 BTC/Byte, the propagation speed ρ = 2 × 10−4
s/Byte, the block subsidy R = 6.25 BTC and the average mining time T = 600 seconds, and two different upper bounds of
block size, i.e., B̄ = 2 MB or 3 MB, as well as the relative computing power of each pool is h i ∈ (0, 1) and h1 + h2 = 1.
In a one-shot mining competition, Jiang and Wu [5] proved that a player’s optimal block size is positively related to its
relative mining power, thus given h1 < h2 , B ∗1 < B ∗2 holds. Let us set h1 = 0.4 and h2 = 0.6 here and take 0.2, 0.5 and 0.8
respectively as the initial values of xi , (i = 1, 2).
Under this setting, we get  B ∗1 = 3.48 × 105 , h Tρ − αR = 1.25 × 106 , and 
B ∗1 = 8.40 × 105 , 1.51 × 106 corresponding to
1
B̄ = 2 × 106 , 3 × 106 , respectively. These results satisfy the conditions of 0 <  B ∗1 < h Tρ − αR < B̄ and 0 < 
B ∗1 < B̄. Fig. 1
1
illustrates the evolution processes of two pools with different upper bound block sizes, verifying the result that (1, 0) is an
ESS in Theorem 4. It is clear that the speed of convergence to fixed equilibrium point (1, 0), i.e., the strategy profile ( B ∗1 , B̄ ),
is faster with the increasing of the maximum capacity of a block. This is because that larger upper bound B̄ brings pool 2
more payoff, stimulating the speed of convergence to the strategy of B̄ . At the same time, g ( B 2 ) > 0 shows the optimal
size of B 1 increases with B 2 , and thus the rate of convergence to  B ∗1 is accelerated with the increasement of B̄.
Since the upper bound of block size can be extended to about 1.7 MB by applying Segregated Witness (SegWit) [15] at
present, then in the following we fix the upper bound block size as 2 MB for simplicity.
Then, we set the value of α and ρ , i.e., α = 0.8 × 10−6 BTC/Byte and ρ = 3 × 10−4 s/Byte, then take h1 = 0.24 to satisfy
the case of 0 < B ∗2 = λ 1ρ − αR < B̄ and B ∗1 < 
B ∗1 < 0. The corresponding evolutionary behaviors of the two pools are shown
1
in Fig. 2, in which the ESS is (1, 1), i.e., (0, λ 1ρ − αR ) is the evolutionary stable strategy under this case. From Fig. 2(a), x1
1
converges to 1 in a relatively short time, while x2 takes much longer time to converge to 1 shown in Fig. 2(b). In addition,
when x2 (0) ≤ 0.5, pool 2 has a strong tendency to take B 2 = B̄ initially, considering the small mining rate of pool 1 while
its own large winning probability. However, as time goes by, pool 2 tends to realize the best response is just B ∗2 = λ 1ρ − αR .
1

103
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

Fig. 1. The impact of upper bound size B̄ on pools’ decisions.

Fig. 2. The evolution of two pools when (1,1) is ESS.

This explains the transition “x2 = 0  x2 = 1” when x2 (0) ≤ 0.5. Hence, with  B ∗1 < 
B ∗1 < 0, the best response of pool 1 is
B 1 = 0, meaning a block without any transactions just for faster propagation.
Furthermore, we discuss the evolutionary process of three mining pools, under such a setting: h1 = 0.2, h2 = 0.3, h3 =
0.5, α = 1 × 10−6 BTC/Byte, R = 6.25 BTC, T = 600 secs, B̄ = 2 × 106 MB, and two different values of ρ , i.e., ρ = 1.5 × 10−4
or 1.25 × 10−4 s/Byte. Note that the values of h i ensure the relationship of B i which is B 1 ≤ B 2 ≤ B 3 . Through some simple
calculation, we can get 0 < B ∗3 = ρ (h T+h ) − αR < B̄ when ρ = 1.5 × 10−4 s/Byte, and ρ (h T+h ) − αR > B̄ resulting in the
1 2 1 2
optimal block size of pool 3 B ∗3 = B̄ when ρ = 1.25 × 10−4 s/byte. Combining Fig. 3(a), 3(b) and 3(d), (1,1,1) and (1,0,0) are
ESSs when ρ = 1.5 × 10−4 s/Byte, while (1,0,0) is ESS when ρ = 1.25 × 10−4 s/Byte. And Fig. 3(c) enlarges the left part of
Fig. 3(d).
Fig. 3(a) shows that pool 1 who has the least amount of computing power would always choose not to follow the upper
bound size B̄, i.e., x1 = 1. And with a shorter unit propagation time ρ , x1 converges faster towards 1. This is because that
when ρ = 1.25 × 10−4 s/Byte, pool 3 has to choose its block size as B̄ and pool 1 only needs to consider the decision-
making process of pool 2; while if ρ = 1.5 × 10−4 s/Byte, then both of pool 2 and pool 3 have two alternative strategies. So
pool 1 needs more time to make its decision on block size for the case of ρ = 1.5 × 10−4 s/Byte. In Fig. 3(b), we can see
the blue curves all go to x2 = 1 at first, but reverse gradually towards 0 after x1 approaching 1. Because pool 2’s computing
power is approximate to that of pool 1, their decision behaviors are similar before pool 1 gets the stable strategy. However,
pool 2 begins to realize that following the upper bound size could bring higher expected payoff when pool 1 chooses not
to follow B̄ and pool 3 decides to follow B̄. As for the red curves in Fig. 3(b) and Fig. 3(d), xi (t ), i = 2, 3 converge to 0
with xi (0) = 0.2, 0.5, but they converge to 1 with xi (0) = 0.8. It is easy to get that x3 approaches 0 much faster than x2
when xi (0) = 0.2, 0.5, i = 2, 3. Then, given x1 = 1 and x3 = 0, the best response of pool 2 is x2 = 0. When xi (0) = 0.8, pool
3 spends a long time staying near x3 = 0.7, which means pool 3 is likely to choose “not following” strategy. So pool 2 who
has much fewer computing power must choose “not following” as well, i.e., x2 converges to 1. Hence, if x1 = 1, x2 = 1 and
ρ = 1.5 × 10−4 s/Byte, following B̄ can bring maximum payoff for pool 3.

5. Conclusion

In this paper, the dynamic decision-making process of mining pools on the optimal block size, in a PoW blockchain
system, is discussed. We model this block size determination problem as an evolutionary game with two kinds of strategies,
i.e., “follow” or “not follow” the upper bound of block size B̄. In addition, the “not follow” strategy means that the mining

104
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

Fig. 3. The evolutionary process of three mining pools with different ρ.

pool shall continue to decide its optimal block size B ∗ (< B̄ ). Supposing all mining pools to be bounded rational, each
pool has to switch the low-payoff strategy to a higher one by learning others’ better strategies, until everyone finds its
evolutionary stable strategy (ESS) that the whole network reaches an evolutionary stable state. Based on the framework
built for the general evolutionary game, we theoretically explore the set of convergent stable states under various conditions
for a case of two mining pools. Our results show that one mining pool owning a lower relative computing power would
not like to follow the upper bound block size, while the other one with a higher relative computing power would choose
the upper bound size. With realistic Bitcoin blockchain data, numerical simulations are conducted to verify our theoretical
results, as well as some sensitivity analyses of the upper bound of block size and the unit propagation time.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have
appeared to influence the work reported in this paper.

Data availability

Data will be made available on request.

Acknowledgements

This work is supported by the National Nature Science Foundation of China (No. 11871366), Qinglan Project of Jiangsu
Province of China, and the Postgraduate Research & Practice Innovation Program of Jiangsu Province (No. KYCX21-2998).

References

[1] N. Satoshi, Bitcoin: a peer-to-peer electronic cash system, http://bitcoin.org, 2008.

[2] N. Houy, The bitcoin mining game, Ledger 1 (2016) 53–68, https://doi.org/10.5195/LEDGER.2016.13.
[3] P.R. Rizun, A transaction fee market exists without a block size limit, https://www.bitcoinunlimited.info/resources/feemarket.pdf, 2015.
[4] R. Zhang, B. Preneel, On the necessity of a prescribed block validity consensus: analyzing bitcoin unlimited mining protocol, in: Proc. 13th International
Conference on Emerging Networking EXperiments and Technologies, New York, NY, USA, 2017, pp. 108–119.
[5] S. Jiang, J. Wu, Bitcoin mining with transaction fees: a game on the block size, in: Proc. 2019 IEEE International Conference on Blockchain (Blockchain),
Atlanta, GA, USA, 2019, pp. 107–115.

105
J. Chen, Y. Cheng, Z. Xu et al. Theoretical Computer Science 942 (2023) 93–106

[6] X. Liu, W. Wang, D. Niyato, N. Zhao, P. Wang, Evolutionary game for mining pool selection in blockchain networks, IEEE Wirel. Commun. Lett. 7 (5)
(2018) 760–763, https://doi.org/10.1109/LWC.2018.2820009.
[7] D. Pan, J.L. Zhao, S. Fan, Z. Zhang, Dividend or no dividend in delegated blockchain governance: a game theoretic analysis, J. Syst. Sci. Syst. Eng. 30
(2021) 288–306, https://doi.org/10.1007/s11518-021-5487-3.
[8] C. Chen, X. Chen, J. Yu, W. Wu, D. Wu, Impact of temporary fork on the evolution of mining pools in blockchain networks: an evolutionary game
analysis, IEEE Trans. Netw. Sci. Eng. 8 (1) (2021) 400–418, https://doi.org/10.1109/TNSE.2020.3038943.
[9] J. Garay, A. Kiayias, N. Leonardos, The bitcoin backbone protocol: analysis and applications, in: Proc. Advances in Cryptology - EUROCRYPT 2015, Sofia,
Bulgaria, 2015, pp. 281–310.
[10] A. Gervais, G.O. Karame, K. Wüst, V. Glykantzis, H. Ritzdorf, S. Capkun, On the security and performance of proof of work blockchains, in: Proc. 2016
ACM SIGSAC Conference on Computer and Communications Security, New York, NY, USA, 2016, pp. 3–16.
[11] NervosFans, Bitcoin transaction fee rules, https://zhuanlan.zhihu.com/p/38479785. (Accessed 15 September 2022).
[12] D. Friedman, On economic applications of evolutionary game theory, J. Evol. Econ. 8 (1) (1998) 15–43, https://doi.org/10.1007/s001910050054.
[13] J. Li, G. Kendall, R. John, Computing Nash equilibria and evolutionarily stable states of evolutionary games, IEEE Trans. Evol. Comput. 20 (3) (2016)
460–469, https://doi.org/10.1109/TEVC.2015.2490076.
[14] TOKENVIEW, Bitcoin Browser, https://btc.tokenview.io/cn/charts. (Accessed 15 September 2022).
[15] E. Lombrozo, J. Lau, P. Wuille, Segregated witness (consensus layer), https://github.com/bitcoin/bips/blob/master/bip-0141.mediawiki, 2015.

106