betaFeedback

WorksPhysics and Astronomy

Quantum Keynesian Beauty Contest with Continuous Variables

Abstract

The development of quantum computing has opened up opportunities to re-explore conventional game models by the integration of quantum processes. Through the established framework of quantum game theory, we examine the problem of speculative pricing in the Keynesian Beauty Contest model – the pp-Beauty Contest (pp-BC). In the pp-BC, players are rewarded with the optimal payoff for successfully bidding at some fraction pp of the average of all of the players' bids. Consequently, rational plays in the pp-BC inexorably lead to boom or bust strategies for p>1p>1 or p<1p<1 respectively. We demonstrate a quantum continuous-variable implementation of pp-BC where the boom-and-bust rational behavior is mitigated through a global transformation of the players' bids generated by increasing multimode squeezing. While we focus on a quantum mechanical implementation of the pp-BC, we show that an entirely analogous game can be formulated via classical channels by a global mediator. Thus we identify global mediation as the underlying game-theoretic mechanism for quenching boom and bust phenomena in our model. We analyze the quantum pp-BC for two, three, and many players in a mean-field type implementation. We derive an analytical form for the probability distribution of the payoff under single-value measurements of the quantum processing circuit, and explore how the rational behavior differs from the expected-value measurement setting.

Introduction

The Keynesian beauty contest (KBC) is a simplified analogy for stock market pricing, where investors choose their bids based on their perceptions of the likely strategies of the other investors [1, 2, 3, 4, 5]. The KBC involves a selection of judges tasked with ordering the best-looking contestants from a crowd, where the judge who selects the most popular contestant, as determined by the judges, wins. One quickly realizes selecting contestants based on one’s own genuine opinion of beauty is a losing strategy. Instead, a better strategy is to rank the contestants according to the rankings that the other judges are likely to make.

Inspired by the idea of KBC, numerous studies have investigated the nested rational process in speculative decision-making [6, 7]. Such experimental realizations of KBC have participants guessing some scaled average of all bids submitted by all players, where the bidding bias introduced by the scaling factor is compounded by rational speculative strategies, which results in players bidding towards extreme values. Specifically, a popular setting of this game has been Guess 2/32/3 of the Average. To quantify the pattern of rational decision making arising from the experiments, the pp-Beauty Contest (pp-BC) was then created as a quantitative game model to explicitly examine such interactions of participants, where pp is defined to be the scaling factor. This captured the rationalization mechanism explaining how speculative trading contributes to boom-and-bust behavior [8, 9, 10].

The pp-BC game highlights the nested logical assumptions of rational players in attempting to derive the optimal strategy. Under this symmetric game, rational players may assume other players are able to think of the same rational strategy as themselves. From here, a counter strategy can be derived for the given assumption, applying the assumption again with the newly derived counter strategy, then a new counter strategy is found, and this cycle of assumptions and counter strategy can continue on and on. Consequently, the continuing iterations lead rational players in a game of pp-BC to bid towards the extremes of the allowed strategy space: for p<1p < 1, players bid towards the minimum, and for p>1p > 1 players bid towards the maximum. The consequences of such speculative process are reflective of the form of market bubbles and destabilization observed in the economy [5, 11].

Quantum game theory began by considering an expansion of conventional game models, where quantum processes are integrated into the player strategies [12, 13]. By exploring the rational behaviors between quantum actors, quantum games have demonstrated a quantum mechanical approach in exploring games by embedding the strategy space in a richer quantum state space [14, 15, 16]. While many quantum games utilize quantum operators and entanglement in discrete-variable quantum systems, continuous-variable quantum systems can be a natural and effective choice in expanding classical game models [17, 18, 19, 20, 21]. With the development of quantum computing, quantum games have been used to explore potential economic environments implemented on quantum hardware, sparking interest in designing quantum-based markets through the framework of quantum game theory in recent years [22, 23, 24].

Collective or global structures are also familiar in classical game theory. In aggregative games, individual incentives depend on population-level quantities constructed from the complete strategy profile, while in beauty-contest settings common information can acquire an amplified strategic role because players respond both to the information itself and to the anticipated responses of others [5, 25]. Global effects may also arise through explicit coordination between players; for example, Flitney and Greentree considered coalitions in the Minority Game, comparing classically coordinated strategies based on communication with strategies enabled by shared quantum entanglement [26]. These examples motivate a broader question explored here: whether the strategic consequences associated with a quantum collective process can be understood in terms of an equivalent global operation in a classical game.

Here, we present an NN-player quantum version of the pp-BC, where the player bids are implemented as single-value (SV) or expectation-value (EV) measurements of a quantum state. We first present the resultant rational behavior in the quantum pp-BC under EV measurements to demonstrate the ability of the quantization circuit in mitigating boom-and-bust behaviors. We show that the multimode quantum circuit acts globally across the complete strategy profile, preserving the population mean while modifying each player’s deviation from that mean. This transformation renormalizes the effective beauty-contest parameter, ultimately mitigating the iterative boom-and-bust behavior of the classical game. Through the classical process of the probability-weighted average, the EV quantum pp-BC enables a reparameterization of the quantum payoff function which resembles the classical payoff. This treatment allows the application of established conventional game theoretic analysis, resulting in a clear description for the rational behavior in the quantum game. Importantly, the Gaussian quantum construction admits an exact quantum-inspired classical realization in which a global mediator applies the same mean-dependent transformation directly to all submitted bids. The quantum and classical mechanisms are physically distinct, but occupy the same architectural role: the strategic variables are transformed globally rather than independently player by player. Using the results of the EV game, the SV quantum pp-BC can be explored with greater clarity under the added complexity of probabilistic payoff. We demonstrate that the mitigation of boom-and-bust is preserved in the SV game, while the probabilistic payoff gives rise to complex player interactions.

This paper is organized as follows: Section II introduces the structure and results of the classical pp-BC game with the established game-theoretic analysis. Section III presents the quantization scheme and the EV quantum pp-BC, where the EV quantum payoff is re-parameterized to apply conventional game-theoretic analysis and mean-field type analyses. Section IV will introduce the SV quantum pp-BC, derive a general expression for the random payoff function, and explore the rational behavior under the probabilistic quantum payoff.

Contributions

The main results presented in this article are as follows:

  1. Continuous-variable quantum pp-Beauty Contest to mitigate boom-and-bust behavior

We demonstrate a CV quantization process for the pp-BC which mitigates the existing boom-and-bust behavior of the classical game. The strength of this effect is controlled by the squeezing parameter γ\gamma, and we show that the resulting global transformation of the player strategies admits an exact quantum-inspired classical interpretation.

  1. Reparameterization of quantum payoff to apply conventional game theoretic analysis

Through a reparameterization of EV quantum pp-BC payoff function, we apply the same game theoretic analysis of iterative elimination on the quantum game, providing a clear and comparable result to the classical game.

  1. Mean-field type analysis for the NN-player quantum pp-Beauty Contest

We perform a mean-field type analysis on the EV quantum pp-BC under different settings. Through a simplified NN-player quantum game, we demonstrate the effect of increasing levels of entanglement γ\gamma on the players’ best responses.

  1. Game theoretic analysis with probabilistic quantum payoff

We derive an analytical expression for the pdf of the probabilistic payoff in SV quantum pp-BC. We present the characteristics of the probabilistic SV payoff under varying discretization settings, including a divergence in its continuous probability density. We demonstrate the on-average optimal strategy of the probabilistic SV quantum payoff resembles the non-probabilistic EV quantum payoff.

Classical pp-Beauty Contest

The pp-BC is more commonly known as Guess pp of the Average, where p∈(0,∞)p \in(0,\infty) is some scalar multiple for the average. The NN-player game is encapsulated in its payoff function [27]:

Φi(x)=−(xi−p∑j=1NxjN)2.(1)\Phi_i(\mathbf{x})=-\left(x_i-p\frac{\sum_{j=1}^{N}x_j}{N}\right)^2. \tag*{(1)}

This is the payoff function of player ii, with xix_i their input bid; NN, the total number of players in the game; x=(x1,x2,…,xi,…,xN)\mathbf{x}=(x_1,x_2,\ldots,x_i,\ldots,x_N) the strategy profile of all players. The payoff is an inverted parabola to penalize bids far from pp of the average, with 0 being the maximum payoff.

Iterative Elimination

The concept of kk-rationality, as derived from Iterative Deletion of Strictly Dominated Strategies (IDSDS), is widely used in pp-BC to describe how logical assumptions and speculative thinking leads rational players to bid towards the extremum of the strategy space [7, 8, 10]. The kk represents the number of iterations the player goes through in predicting the behavior of other players.

Here, we demonstrate the iterative process through an example game of (p=2/3)(p=2/3)-BC, with valid bids between 0 and 100. Regardless of strategy, all players will bid within the strategy space xj≠ik=0=[0,100]x^{k=0}_{j\ne i}=[0,100], hence the maximum 2/32/3 average is 2/3⋅100≈66.672/3\cdot100\approx66.67. The set of optimal strategies is then limited to xik=1=[0,66.67]x_i^{k=1}=[0,66.67]. If a rational player assumes all other players know this, then their new optimal strategy is now xik=2=[0,44.44]x_i^{k=2}=[0,44.44] with the new maximum 2/32/3 average (2/3)2⋅100≈44.44(2/3)^2\cdot100\approx44.44. After kk steps, the optimal strategy is then xik=[0,(2/3)k⋅100]x_i^k=[0,(2/3)^k\cdot100]. For k→∞k\to\infty, the optimal strategy is simply the bust xik→∞=0x_i^{k\to\infty}=0, which is also the Nash equilibrium of the (p=2/3)(p=2/3)-BC game [27]. This rational iterative process is visualized in Figure 1.

Visualization of the IDSDS process

Figure 1. A visualization of the IDSDS process for “Guess 2/3 of the Average”, where each player chooses a number from [0,100][0,100]. With increasing level of rationality, rational players are compelled to choose smaller and smaller numbers. The naive player chooses a number from the entire range xk=0=[0,100]x^{k=0}=[0,100]; a 1-rational player picks their number within a smaller range of xk=1=[0,66.67]x^{k=1}=[0,66.67]; a 2-rational player chooses within xk=2=[0,44.44]x^{k=2}=[0,44.44]; and finally, the ∞\infty-rational player only picks 0, which is the Nash equilibrium of this game.

At every iteration, the rational player ii as the kk-rational player assumes the crowd to be (k−1)(k-1)-rational players, then the strictly decreasing region of player ii’s payoff spectrum $i(x1,x2,…,xik,…,xN)>$i(x1,x2,…,xikdominated,…,xN)\$_i(x_1,x_2,\ldots,x_i^k,\ldots,x_N)>\$_i(x_1,x_2,\ldots,x_i^k{}_{\mathrm{dominated}},\ldots,x_N) is found and eliminated to obtain the optimal strategy [27]. A quantitative IDSDS process is shown in Section III D, where the exact classical game-theoretic analysis is applied to the EV quantum pp-BC.

Nash Equilibria

The movements of the players’ bids are driven by the iterative elimination processes described above. Under the limiting assumption of infinitely rational players k→∞k\to\infty, this kk-rational strategy results in players bidding at the respective Nash equilibrium, and game theoretically eliminates any subsequent deviating bids.

For (p<1)(p<1)-BC or Guessing Smaller than the Average, the Nash equilibrium is bust xNash={(0,0,…,0)}\mathbf{x}_{\mathrm{Nash}}=\{(0,0,\ldots,0)\}. For Guessing Larger than the Average (p>1)(p>1)-BC, the relevant Nash equilibrium is boom xNash={(100,100,…,100)}\mathbf{x}_{\mathrm{Nash}}=\{(100,100,\ldots,100)\}. For Guessing the Average (p=1)(p=1)-BC, there are no dominated strategies and hence infinitely many Nash equilibria [28] across the continuous strategy space. Significantly, given neither boom nor bust is the sole Nash equilibrium in p=1p=1, rational strategies no longer result in extremum bids.

The kk-rational strategy derived from single-round game analysis is a major tool in analyzing multi-round experimental pp-BC by categorizing the players as different levels of kk-rational [7, 9, 10]. The pp-BC experiments showed how different kk-rational players learn to bid towards the extremum Nash equilibria. Importantly, as pp deviates further from 1, the players learn to bid towards the extremum faster.

In this paper, we will continue to use p<1p<1 for mathematical clarity, while the same derivations hold equally true for p>1p>1 with mirroring results. In following sections, we present the quantum game theoretic quantization and discuss how the quantum pp-BC mitigates boom-and-bust behaviors.

Continuous-variable Quantization

Now we explore a quantum version of pp-BC, where a quantum circuit introduces entanglement between continuous-variable player bids. Our NN-player quantization scheme is an extension of the 2-player Li-Du-Massar quantization [17] as shown in Figure 2.

N-player continuous-variable quantization scheme

Figure 2. NN-player continuous-variable quantization scheme. The core element in the quantization is this quantum circuit. Here, D^i\hat{D}_i is the single-mode displacement operator encoding player ii’s bid on their corresponding mode, where the displacement parameter xix_i is limited to the position axis, omitting the complex momentum component. The two NN-mode squeezing operators J^\hat{J} & J^†\hat{J}^{\dagger} are as described in Equation (3), which introduce correlations between the modes and hence the strategies of all players. Finally, each mode is measured in its corresponding positional basis X^i\hat{X}_i given D^i(xi)\hat{D}_i(x_i), with the measurement results as qiq_i.

The present CV construction in the quantization process is Gaussian and hence classically simulable. We utilize this property to isolate a classical game-theoretic mechanism from the resultant quantized game, which employs multimode squeezing and intermediate entanglement. This process results in an equivalent quantum-inspired classical game, through which we characterize the strategic consequences of the quantum-derived global transformation.

In this quantum pp-BC, each player inputs their bid via their assigned single-mode positional displacement operators D^i=exp⁡(−ixiP^i)\hat{D}_i=\exp(-\mathrm{i}x_i\hat{P}_i), where P^i=i/2 (a^i†−a^i)\hat{P}_i=\mathrm{i}/\sqrt{2}\,(\hat{a}_i^\dagger-\hat{a}_i) [29]. The imaginary component of the displacement xix_i is excluded, and the displacement operator D^i\hat{D}_i limited to the position basis X^i\hat{X}_i. By performing quadrature measurements of the modes with respect to the positional basis X^i=1/2(a^i†+a^i)\hat{X}_i = 1/\sqrt{2}(\hat{a}_i^\dagger+\hat{a}_i), the quantization circuit produces a set of quantum-processed bids q\mathbf{q}. These measurement outcomes q\mathbf{q} are substituted into (1) to calculate every player’s quantum payoff. From here, we can write the quantum payoff function as:

ΦiQ(x)=Φi[q(x)]=−(qi−p∑j=1NqjN)2.(2)\Phi_i^Q(\mathbf{x}) = \Phi_i[\mathbf{q}(\mathbf{x})] = -\left(q_i-p\frac{\sum_{j=1}^{N}q_j}{N}\right)^2. \tag*{(2)}

There are two NN-mode squeezing operations J^\hat{J} and J^†\hat{J}^{\dagger} which introduce entanglement between the modes and players’ bids. The operations are described by [29, 30]:

J^(γ)=exp⁡(−γ∑i,j=1N(δij−1/N)(a^i†a^j†−a^ia^j))(3)\hat{J}(\gamma)=\exp\left(-\gamma\sum_{i,j=1}^{N}(\delta_{ij}-1/N)(\hat{a}_i^\dagger\hat{a}_j^\dagger-\hat{a}_i\hat{a}_j)\right) \tag*{(3)}

where γ\gamma represents the level of squeezing, and J^†\hat{J}^{\dagger} is obtained by reversing the sign of the exponent.

Importantly, through this bid entangling process, each player’s input xix_i can deviate from their own quantum-processed bid qiq_i depending on all other players’ bids. A rational player in the quantum game must not only consider the other players’ inputs, but additionally account for intrinsic quantum measurement noise introduced by the quantization circuit. While in the classical game, a rational player can devise a strategy to bid a particular number, and their input will simply be their resultant bid.

Expected-Value Quantum pp-Beauty Contest

In the EV quantum pp-BC, each set of player inputs is repeated over a large ensemble of experiments, with the expected quantum-processed bid ⟨q⟩\langle q\rangle determining the outcome of the game. The expected quantum payoff for player ii is thus:

⟨ΦiQ⟩=−(⟨qi⟩(γ)−p∑j=1N⟨qj⟩(γ)N)2(4)\langle\Phi_i^Q\rangle=-\left(\langle q_i\rangle(\gamma)-p\frac{\sum_{j=1}^{N}\langle q_j\rangle(\gamma)}{N}\right)^2 \tag*{(4)}

where we use ⟨ΦiQ⟩\langle\Phi_i^Q\rangle to denote the EV processed quantum payoff. This treatment significantly simplifies the game-theoretic analysis, and facilitates the derivation of the average rational behavior for the quantum pp-BC.

To express ⟨ΦiQ⟩\langle\Phi_i^Q\rangle as a function of players’ inputs x\mathbf{x}, we derive ⟨qi⟩(x)\langle q_i\rangle(\mathbf{x}) by finding the final wave function in the quantization circuit, then the expectation value with respect to the positional basis X^i\hat{X}_i:

∣ψf⟩=J^†∏k=1ND^k(xk)J^∣0⟩⊗N=⨂k=1Nexp⁡(−i[e2γ(xk−xˉ)+xˉ]P^k)∣0⟩k\begin{aligned} |\psi_f\rangle=\hat{J}^{\dagger}\prod_{k=1}^{N}\hat{D}_k(x_k)\hat{J}|0\rangle^{\otimes N} \\ &=\bigotimes_{k=1}^{N}\exp\left(-i\left[e^{2\gamma}(x_k-\bar{x})+\bar{x}\right]\hat{P}_k\right)|0\rangle_k \end{aligned}
⟨qi⟩(x,γ)=Tr⁡(ρ^fX^i)=e2γ(xi−xˉ)+xˉ(5)\langle q_i\rangle(\mathbf{x},\gamma)=\operatorname{Tr}\left(\hat{\rho}_f\hat{X}_i\right)=e^{2\gamma}(x_i-\bar{x})+\bar{x} \tag*{(5)}

where xˉ\bar{x} is the averaged player input xˉ=1N∑i=1Nxi\bar{x}=\frac{1}{N}\sum_{i=1}^{N}x_i; ρ^f=∣ψf⟩⟨ψf∣\hat{\rho}_f=|\psi_f\rangle\langle\psi_f|.

The chosen squeezing operations and the displacement operators result in a separable state as shown in the above form of ∣ψf⟩|\psi_f\rangle. Each mode is a coherent state with a positional displacement determined by the inputs from all players through the preceding multimode entangling transformation. The resulting global dependence of each processed bid on the complete strategy profile is controlled by the squeezing parameter γ\gamma.

Under the EV setting, the classical pp-Beauty Contest model is fully retained as the quantum game under no entanglement ⟨qi⟩γ=0=xi\langle q_i\rangle_{\gamma=0}=x_i. Importantly, the average bid is also preserved, where the average EV quantum-processed bid equals the average player input ⟨qˉ⟩=xˉ\langle\bar{q}\rangle=\bar{x}.

Effect of Squeezing Operations

The influence of γ\gamma on the correlation of bids can be demonstrated through a 2-player example of the EV quantum game. The quantum states across the circuit are presented in Figure 3. In this example, player 1 and 2 provide the inputs x1=5x_1=5, x2=1x_2=1, respectively, and the expected positional measurement X^1,X^2\hat{X}_1,\hat{X}_2 for the quantization circuit output is simply the same as each player’s input: ⟨q1⟩=x1\langle q_1\rangle=x_1, ⟨q2⟩=x2\langle q_2\rangle=x_2. Under zero squeezing (γ=0\gamma=0), the displacement operators D^1(x1)\hat{D}_1(x_1) & D^2(x2)\hat{D}_2(x_2) simply encode the classical game as shown in Figure 3a.

Wigner function plots for the displaced coherent states

(a)

Wigner function plots for the entangled states

(b)

Figure 3. Wigner Function Plots. Given 2 players who have input bids x1=5x_1 = 5 and x2=1x_2 = 1 through the quantization circuit respectively: a) For γ=0\gamma= 0, there is no entanglement, and only the displacement operations D^i\hat{D}_i are in effect. The modes encoding each player’s input are displaced coherent states with positions corresponding to the inputs. With γ=0.5\gamma= 0.5, the states of the modes remain displaced coherent states through the sandwiching gates J^\hat{J} & J^†\hat{J}^{\dagger} under non-zero squeezing, where the entanglement results in the correlation of the expected position ⟨X^i⟩\langle\hat{X}_i\rangle. b) On the other hand, if J^\hat{J} and D^i\hat{D}_i are applied without J^†\hat{J}^{\dagger}, the resultant modes are in an entangled state, where measurement outcomes of the modes are anti-correlated.

With the addition of the first squeezing operator J^\hat{J} for γ>0\gamma>0 to the displacement operators, the quantization circuit creates a two-mode displaced squeezed state, where the positional measurements of each mode become anti-correlated as depicted in Figure 3b. For example, if a larger than average q1>x1q_1 > x_1 measurement is obtained for player 1’s mode, then there is a high probability of obtaining a smaller than average q2<x2q_2 < x_2 for player 2. The effect is amplified by increased level of squeezing γ↑\gamma\uparrow.

By adding the second squeezing operation J^†\hat{J}^{\dagger}, the anti-correlation in the positional measurement now translates to anti-correlation in the positional displacement of the coherent states as shown in Figure 3a. The effect of this entanglement configuration on the bids is best described by Equation (5), where the expected distance between each player’s inputs and the average input is amplified by the squeezing parameter γ\gamma. For example, a player who bids away from the average may find their quantum-processed bids ending up further away.

In the following, we utilize the dimensionless squeezing parameter γ=2\gamma= 2, unless otherwise stated. This corresponds to state-of-the-art squeezing parameters by γdB=10log⁡10(e2γ)≈15dB\gamma_{\mathrm{dB}} = 10\log_{10}(e^{2\gamma}) \approx15dB as reported in recent experimental implementations [31, 32, 33].

Continuing from Equation (2) and Equation (5), the EV quantum payoff ⟨SiQ⟩\langle\mathcal{S}^{Q}_{i}\rangle can be expressed with respect to x\mathbf{x}:

⟨SiQ⟩=−e4γ(xi−pQ(γ)∑j=1NxjN)2(6)\langle\mathcal{S}^{Q}_{i}\rangle= -e^{4\gamma}\left(x_i-p_Q(\gamma)\frac{\sum_{j=1}^{N}x_j}{N}\right)^2 \tag*{(6)}
pQ(γ)=1+e−2γ(p−1).(7)p_Q(\gamma)=1+e^{-2\gamma}(p-1). \tag*{(7)}

The EV quantum payoff is now expressed in a near identical form to the classical payoff in Equation (1). Most importantly, pQp_Q in Equation (6) emerges as the effective pp scalar parameter in the EV quantum game, where the valid values of pQp_Q lie between pp and 1, depending on γ\gamma. Additionally, the scaling introduced by e4γe^{4\gamma} only alters the magnitude of the payoff spectrum, while the Nash equilibria and the trend of rational strategy remain unchanged.

By taking expected values, the EV treatment brings the quantum pp-BC closer to the original classical game, which allows the EV quantum pp-BC to produce a definite rational behavior through an established game theoretic analysis technique. This will provide necessary insight for discussing the more complex SV quantum pp-BC in Section IV.

Global mediation and a quantum-inspired classical game

(5) reveals a global structure in the transformation of the player strategies. Defining λ=e2γ\lambda=e^{2\gamma}, the expected processed bid may be written as:

⟨qi⟩=xˉ+λ(xi−xˉ)=λxi+(1−λ)xˉ.(8)\langle q_i\rangle=\bar{x}+\lambda(x_i-\bar{x}) =\lambda x_i+(1-\lambda)\bar{x}. \tag*{(8)}

This transformation preserves the population mean, ⟨qˉ⟩=xˉ\langle\bar{q}\rangle=\bar{x}, while amplifying each player’s deviation from that mean by the common factor λ\lambda. Equivalently, every player’s bid is subject to the same global contribution (1−λ)xˉ(1-\lambda)\bar{x}, which depends on the complete strategy profile rather than on that player’s strategy alone.

The same transformation can therefore be implemented without a quantum system. A classical mediator can collect the strategy profile x\mathbf{x}, calculate the global mean xˉ\bar{x}, and apply Equation (8) to every bid before evaluating the original pp-BC payoff. This produces the same effective beauty-contest parameter:

pQ=1+λ−1(p−1)=1+e−2γ(p−1),(9)p_Q=1+\lambda^{-1}(p-1) =1+e^{-2\gamma}(p-1), \tag*{(9)}

and therefore leaves all subsequent game-theoretic results unchanged. Increasing the strength of the global mediation drives pQp_Q towards unity, suppressing the iterative tendency towards boom or bust.

A logic flow diagram visualizing the necessary computations to obtain ⟨qi⟩\langle q_i\rangle is shown in Figure 4, where the number of classical operations scales linearly as (N−1)+(N+1)+2N→O(N)(N-1)+(N+1)+2N \to O(N). Comparatively, the quantization circuit shown in Figure 2 also scales linearly as O(N)O(N).

Logic flow diagram of the classical computation

Figure 4. Logic Flow Diagram to visualize the classical computational complexity. This diagram implements the expected processed bid ⟨qi⟩\langle q_i\rangle in Equation (8), where λ∈[1,∞)\lambda\in[1,\infty) is the common scaling factor. Each player’s input bid xix_i passes through the multiplication operations as square boxes and the summing junctions as circled pluses to produce each player’s processed bid as the output.

The quantum and quantum-inspired implementations are physically different but share the same global architecture. In the quantum circuit, multimode squeezing and the associated intermediate entanglement provide a transformation acting collectively across the player modes. In the classical construction, the analogous role is played by the global mediator through the population mean. Thus, entanglement acts as a physical implementation resource in the quantum protocol, while the underlying game-theoretic resource can be understood more generally as global mediation: the players’ strategic variables are transformed collectively rather than independently.

Expected Iterative Elimination

A qualitative description for the rational process of IDSDS was presented in the Section II A. We apply the formal mathematical process of this conventional game theoretic technique to the form of EV quantum payoff in Equation (6).

In general, the players are presented with the strategy set x∈[a,b]x \in[a,b], and the region of dominated strategy ∂⟨ΦiQ⟩/∂xi<0\partial\langle\Phi_i^Q\rangle/\partial x_i<0 can be obtained by the derivative of the EV quantum payoff Φi\Phi_i with respect to the player input xix_i:

∂⟨ΦiQ⟩∂xi=−2e4γ(1−pQN){(1−pQN)xi−pQN∑j=1j≠iNxj}.(10)\frac{\partial\langle\Phi_i^Q\rangle}{\partial x_i}=-2e^{4\gamma}\left(1-\frac{p_Q}{N}\right)\left\{\left(1-\frac{p_Q}{N}\right)x_i-\frac{p_Q}{N}\sum_{\substack{j=1\\j\ne i}}^{N}x_j\right\}. \tag*{(10)}

Now starting with xik=0∈[a,b]x_i^{k=0}\in[a,b], if we iteratively derive and eliminate the ∂⟨ΦiQ⟩/∂xi<0\partial\langle\Phi_i^Q\rangle/\partial x_i<0 region under every (k−1)(k-1)-rational assumption, the expression for the kk-rational strategy can be written as [27]:

xik∈{[a,(Ielim)k⋅b]for p<1[(Ielim)k⋅a,b]for p>1{(x,…,x)∣x∈xValid}for p=1(11)x_i^k\in\begin{cases} \left[a,(\mathcal{I}_{\mathrm{elim}})^k\cdot b\right] & \text{for }p<1\\ \left[(\mathcal{I}_{\mathrm{elim}})^k\cdot a,b\right] & \text{for }p>1\\ \{(x,\ldots,x)\mid x\in x_{\mathrm{Valid}}\} & \text{for }p=1 \end{cases} \tag*{(11)}
Ielim=pQ(N−1)N−pQ→{<1for p<1>1for p>1=1for p=1.(12)\mathcal{I}_{\mathrm{elim}}=\frac{p_Q(N-1)}{N-p_Q}\to\begin{cases} <1 & \text{for }p<1\\ >1 & \text{for }p>1\\ =1 & \text{for }p=1 \end{cases}. \tag*{(12)}

The optimal kk-rational strategies are derived for all (k−1)(k-1)-rational assumption as shown in Equation (11), where the magnitude of the iterative term Ielim\mathcal{I}_{\mathrm{elim}} as defined in Equation (12) ensures the lower bound does not rise above the upper bound. For a sizable player population N≫1N\gg1, the iterative term converges rapidly to pQp_Q. Hence, under a large population assumption, the kk-rational expression in Equation (11) reduces to xik=[a,pQk⋅b]x_i^k=[a,p_Q^k\cdot b] and aligns exactly with the described qualitative analysis of the classical Guess 2/3 of the Average example in Section II A. Hence, the rational behavior in the EV quantum pp-BC can be controlled by changing pQp_Q, which is achieved by adjusting the level of squeezing γ\gamma as shown in Equation (7).

Effective scaling factor as a function of squeezing

Figure 5. Effective scaling factor pQp_Q as a function of γ\gamma: by adjusting γ\gamma the level of squeezing, the effective pp in the quantum game (pQp_Q) can be adjusted towards 11 regardless of the chosen value of pp in the classical game.

Under the limit of large squeezing γ≫1\gamma\gg1, the effective scaling parameter pQp_Q is approximately 11. As a result, the region of dominated strategy becomes negligible, and the strategy space eliminated through the iterative elimination process becomes insignificant. As the kk-rational strategy becomes less relevant, the Nash equilibria approximately spans the entire strategy space xNash={(x,…,x)∣x∈xValid}\mathbf{x}_{\mathrm{Nash}}=\{(x,\ldots,x)\mid x\in x_{\mathrm{Valid}}\}, which means boom and bust are no longer the only Nash equilibria, and the rational behavior is no longer to bid towards those extremum values.

With the rational behavior being largely independent of pp under high level of squeezing, the rational behavior remains the same for a version of the EV quantum pp-BC where each player has a separate individual scaling parameter pip_i in their payoff function.

N-player Quantum p-BC

The game-theoretic analysis above has provided the trend of the rational strategies and the typical game of quantum pp-BC. To see the game from each player’s perspective and to understand how rational players may choose to interact with the rest of the players, we explore the NN-player pp-BC under different settings and examine the payoff spectra.

Two Players

The 2-player pp-BC game can be a special case under a certain setting [34]. Each of the 2 players knows the 2/3 average of any two numbers is always closer to the smaller number, and hence can choose 0 to ensure an equal or higher payoff compared to the other player. However, the goal of our rational player is to maximize their own personal payoff, hence our kk-rational strategy in Equation (12) remains unchanged for the 2-player case.

The 2-player case is the simplest setting of the EV quantum pp-BC. In Figure 6a&Figure 6b), the player 1’s payoff is presented as a spectrum with respect to both players’ choices, and the player 2’s payoff is identical by an exchange of the axis labels under the symmetric game pp-BC. By examining the payoff spectra changes with increasing γ\gamma in Figure 6a, the most notable feature is the shifting gradient of the peak payoff curves, which are the best response curves where player 1 bids exactly on 2/3 of the average. These curves can be derived as:

Figure 6. Two-player quantum pp-BC, player 1’s payoff spectrum with respect to both players’ bids: a) the payoff spectra for γ\gamma set at {0,1,2}\{0,1,2\}, where the peak gradient approaches 1 with increasing level of squeezing; b) since the only significant feature here is the peaks of the payoff spectra, which are incidentally the best-response curves for player 1, the player interactions in the quantum game can then be simplified to observing how the best response changes with respect to γ\gamma. Additionally, the iterative elimination process is visualized by the dashed lines drawn.

x1,best=pQ2−pQx2(13)x_{1,\mathrm{best}} = \frac{p_Q}{2-p_Q}x_2 \tag*{(13)}
x2,best=pQ2−pQx1.(14)x_{2,\mathrm{best}} = \frac{p_Q}{2-p_Q}x_1. \tag*{(14)}

The vertical extent of the best response curve in Figure 6b demonstrates the influence of player 2 on player 1’s payoff, with the bounds indicated by the intercepts for the minimum and maximum bid of player 2. Under no entanglement γ=0\gamma= 0, player 1’s optimal bid range is limited between [0,50][0,50]1. With increasing level of entanglement increases γ↑\gamma\uparrow, player 1’s optimal bid range grows, spanning the entire strategy space under the limit γ≫1\gamma\gg1. With larger squeezing, a player must be more sensitive to the bids of the other.

The best response curves can also provide a visual demonstration of the iterative elimination process. For γ=0\gamma= 0, the 1-rational strategy is again the range of best response xk=11=[0,50]x_{k=1}^{1} = [0,50], with the upper limit marked by the x2=100x_2 = 100. Assuming player 2 follows the 1-rational strategy, the 2-rational strategy reduces to xk=22=[0,25]x_{k=2}^{2} = [0,25], and the process continues till the Nash equilibria bust at k→∞k \to\infty. Additionally, this visualization also shows the invariance of this payoff spectrum and the best response curves under the changing upper bound in the strategy space.

Three Players

By introducing a third player, the expressions of best response curves are now:

x1,best=pQ3−pQ(x2+x3)(15)x_{1,\mathrm{best}} = \frac{p_Q}{3-p_Q}(x_2+x_3) \tag*{(15)}
x2,best=pQ3−pQ(x1+x3)(16)x_{2,\mathrm{best}} = \frac{p_Q}{3-p_Q}(x_1+x_3) \tag*{(16)}
x3,best=pQ3−pQ(x1+x2).(17)x_{3,\mathrm{best}} = \frac{p_Q}{3-p_Q}(x_1+x_2). \tag*{(17)}

Comparing the intercepts in Figure 7 to Figure 6b, player 2 now has less influence on player 1’s payoff as anticipated, with the span of the best response at γ=2\gamma= 2 decreasing from 100 to 50. On the other hand, similar to the 2-player game, there is an increase in responsiveness between the players under higher levels of entanglement.

Three-player quantum p-BC: Player 1’s best response curves for different $\gamma$ settings given the different bids from players 2 and 3

Figure 7. Three player quantum pp-BC: Player 1’s best response curves for different γ\gamma settings given the different bids from player 2 & 3.

Mean-field Type N-player Games

To visualize the EV quantum pp-BC for N>3N > 3, a mean-field type approach is used by creating a group of players who always coordinate to bid the same number. Similar to the 2-player game, the best response curves for a game of 1 vs N−1N - 1 players are found with the addition of weighted average:

x1,best=pQ(N−1)N−pQxN−1(18)x_{1,\mathrm{best}} = \frac{p_Q(N-1)}{N-p_Q}x_{N-1} \tag*{(18)}
xN−1,best=pQN−pQ(N−1)x1.(19)x_{N-1,\mathrm{best}} = \frac{p_Q}{N-p_Q(N-1)}x_1. \tag*{(19)}

The influence of γ\gamma on player 1’s best response curves is very similar to the behavior found in the 2-player game as shown in Figure 8a and Figure 6b. Despite N−1N-1’s significant influence on the average, player 1 can continue to receive the maximum payoff by bidding 2/3 of the N−1N-1 group’s number. The term pQ(N−1)N−pQ\frac{p_Q(N-1)}{N-p_Q} which accounts for player 1’s own bid in the average quickly converges to pQp_Q with increasing player population NN, which agrees with player 1’s best response curves being relatively unchanged for N>5N > 5.

Figure 8. Mean-field type game for 1 vs N−1N-1: a) Player 1’s best response curve under different γ\gamma settings with respect to the N−1N-1 group’s bid; b) The term pQ(N−1)N−pQ\frac{p_Q(N-1)}{N-p_Q} shows how the player should account for their bid’s influence on average in calculating their bid for 2/3 average. When pQN−pQ\frac{p_Q}{N-p_Q} is far from pp, the player’s bid has a large sway on the average bid. When pQN−pQ→p\frac{p_Q}{N-p_Q} \to p, the average is independent of the player’s bid; c) The N−1N-1 group’s best response curve under different γ\gamma settings with respect to the player 1’s bid. Here an unrealistic upper limit of γ=10\gamma=10 is used to demonstrate what is required for the N−1N-1 players to be influenced by the single player.

From the perspective of the N−1N-1 group, player 1’s bid should be insignificant due to its weighting on the average. This aligns with N−1N-1 group’s best response curve being horizontal under γ=0\gamma= 0 for N≫1N \gg1 in Figure 8c. However, in the limit of extreme squeezing γ=10\gamma= 10, the best response of the N−1N-1 crowd is shifted towards following the single player’s bid, in the same way the single player follows the N−1N-1 crowd.

Here, we present an example of the game with large squeezing γ≫1\gamma\gg1. Under this limit, the EV quantum pp-BC resembles a game of Guess the Average with pQ=1p_Q = 1, where the best response curve of the N−1N-1 group in (19) becomes xN−1,best=x1x_{N-1,\mathrm{best}} = x_1. This is independent of the player population size NN. For a quantitative analysis, we assume a game of N=1000N = 1000 players, where player 1 bids x1=20x_1 = 20 and the N−1N-1 group bids xN−1=10x_{N-1} = 10. The N−1N-1 group sees an error 0.01, which is insignificant relative to the error of 9.99 for player 1. Nevertheless, the optimal strategy derived from the mathematical game-theoretic analysis suggests that the N−1N-1 group will always look to improve their own payoff when possible. In this case, this error of 0.01 from the bid xN−1=10x_{N-1} = 10 decreases as xN−1→x1x_{N-1} \to x_1, and hence the best response of the N−1N-1 crowd is to move towards the single player’s bid.

Observing the best response curves of the N−1N-1 population, the magnitude of the gradients decreases as NN increases, reflecting the diminishing impact of the single player’s bid on the average. In the limit where the single player’s contribution becomes negligible, the best response for the N−1N-1 population converges to 0 for p<1p < 1. However, higher levels of squeezing amplify the single player’s influence on the average, which lifts the best response from 0 as shown in Figure 8c.

Expanding to a game of 1 vs N−2N - 2, the best response curves are found as:

x1,best=pQ(N−2)N−pQ(x2N−2+xN−2)(20)x_{1,\mathrm{best}} = \frac{p_Q(N-2)}{N-p_Q}\left(\frac{x_2}{N-2}+x_{N-2}\right) \tag*{(20)}
x2,best=pQ(N−2)N−pQ(x1N−2+xN−2)(21)x_{2,\mathrm{best}} = \frac{p_Q(N-2)}{N-p_Q}\left(\frac{x_1}{N-2}+x_{N-2}\right) \tag*{(21)}
xN−2,best=pQ(x1+x2)N−pQ(N−2).(22)x_{N-2,\mathrm{best}} = \frac{p_Q(x_1+x_2)}{N-p_Q(N-2)}. \tag*{(22)}

For the 2 single players, under the large population limit, due to their increasingly insignificant contribution to the average, the 2 players no longer have any meaningful interactions with each other as demonstrated in Figure 9. Under this setting, both players are effectively only playing the game against the N−2N-2 group, where player behavior in such interactions is reflected in the above 1 vs N−1N-1 analysis.

Mean-field type game for 1 vs 1 vs $N-2$

Figure 9. Mean-field type game for 1 vs 1 vs N−2N-2: Player 1’s best response curve given γ\gamma, player 2’s bid and the N−2N-2 group’s bids.

Through the reparameterization of EV quantum payoff function, the conventional kk-rational game theoretic analysis was applied and demonstrated the ability of the quantum game to disincentivize boom-and-bust behavior from rational players. Additionally, while behaviors in the classical game are shaped by players following pp, players become more responsive to the strategy of other players in the quantum game under higher levels of entanglement.

Single-Value Quantum pp-Beauty Contest

Practical applications may require the design of the quantum game to accommodate a large number of participants under heavy computational constraints. Despite the EV quantum pp-BC’s simple result, its limitation lies in the need for a large number of measurements for every participant.

In this section, we present the SV quantum pp-BC, where each player’s quantum-processed bid is produced by a single quantum measurement.

The introduced quantum probabilistic measurements can substantially alter the game’s physical implementation and rational behavior. Each player’s quantum-processed bid is not only influenced by the bids of other players through the entanglement, but is now also sampled from a random distribution due to the Born rule. Hence, a player may find their strategies more difficult to implement accurately, while the strategies of other players also become less predictable. This is amplified by the addition of probabilistic payoff, where each player’s payoff is now dependent on the independent quantum measurements of every players’ quantum-processed bids. To begin to understand the SV game and how it can shape the behaviors of rational players, we derive the form of probability distribution for the payoff function.

In the following sections, we explore the properties and the resultant rational strategies of the SV setting of the quantized game, drawing comparisons to the EV setting described above.

Payoff Probability Distribution

Using the positional basis, we assume the wave function of the initial vacuum state in the quantization circuit takes the form [35]:

ψ0(q)=⨂iN⟨ai∣ψ0⟩=⨂iN⟨ai∣0⟩i=1π1/4⨂iNexp⁡(−ai2/2).(23)\psi_0(\mathbf{q}) = \bigotimes_{i}^{N}\langle a_i|\psi_0\rangle= \bigotimes_{i}^{N}\langle a_i|0\rangle_i = \frac{1}{\pi^{1/4}}\bigotimes_{i}^{N}\exp(-a_i^2/2). \tag*{(23)}

The final wave function ψf\psi_f of the quantization circuit ends_mid=1 consists of separable coherent states with correlated positional displacement, where probability density function (pdf) of each measured mode fqif_{q_i} can be expressed as:

ψf(q,γ)=⨂i=1N⟨ai∣ψf⟩=1πN/4⨂i=1Nexp⁡(−(ai−⟨qi⟩)2)(24)\psi_f(\mathbf{q},\gamma)=\bigotimes_{i=1}^{N}\langle a_i|\psi_f\rangle = \frac{1}{\pi^{N/4}}\bigotimes_{i=1}^{N}\exp\left(-(a_i-\langle q_i\rangle)^2\right) \tag*{(24)}
fqi(ai)=∣⟨ai∣ψf⟩∣2=1π1/2exp⁡(−(ai−⟨qi⟩)2)(25)f_{q_i}(a_i)=|\langle a_i|\psi_f\rangle|^2 = \frac{1}{\pi^{1/2}}\exp\left(-(a_i-\langle q_i\rangle)^2\right) \tag*{(25)}

where ∣ai⟩|a_i\rangle refers to the positional basis, ai∈(−∞,∞)a_i\in(-\infty,\infty) is the support for the random variable qiq_i, and ⟨qi⟩\langle q_i\rangle is as defined previously in Equation (5).

The measurement outcomes qiq_i follow a continuous Gaussian distribution described by fqi(ai)f_{q_i}(a_i). With the quantum payoff QiQQ_i^{\mathrm{Q}} described in Equation (2), the variance of the measurements is then σqi2=1/2\sigma_{q_i}^2=1/2 under no squeezing γ=0\gamma=0, and the expected value of each mode aligns with the input of the respective player ⟨qi⟩(γ=0)=xi\langle q_i\rangle(\gamma=0)=x_i.

Payoff Distribution by Monte Carlo

In this section, we show two separate numerical approaches to demonstrate a characteristic of the SV quantum payoff. Following the 2-player quantum pp-BC example from Section III B, the two approaches will produce two different payoff spectra, which converge in the continuous limit. Both approaches begin with the Gaussian random distributions of q1q_1 and q2q_2 in Equation (25).

The first approach is Monte Carlo. This process is visualized in Figure 10, where the qiq_i distributions are randomly sampled. For each pair of measurement samples, the quantum payoff QiQQ_i^{\mathrm{Q}} is computed for each player. With 10,000 samples, the probability distributions of each player’s payoff can be constructed, which is presented as raw samples in Figure 10b, and a histogram in Figure 10c.

Figure 10. Payoff Probability Spectrum by Monte Carlo: a) Histograms of 10,000 random samples from q1q_1 and q2q_2 each, discretized with 50 uniform bins, where the player inputs are x1=5x_1=5 and x2=1x_2=1; b) Calculated payoff values for both players from 10,000 pairs of q1q_1 and q2q_2; c-e) Histogram of 10,000 calculated payoff samples for both players Q1QQ_1^{\mathrm{Q}} & Q2QQ_2^{\mathrm{Q}} with 1,000 bins, 200 bins, and 50 bins.

By observing player 2’s payoff spectrum in Figure 10c with increasing number of bins, the closest bin to 0 rapidly increases in height relative to the others, and a diverging pattern emerges at 0 in the continuous limit. In the following, we investigate how this divergence can be introduced by processing samples of two finite probability density functions of q1q_1 and q2q_2 through the payoff function in Equation (2).

Divergence in Payoff PDF

To confirm the divergence is not an artifact, we compute the spectrum through a different approach. Instead

of randomly sampling qiq_i, we begin with discretizing the pdf of q1q_1 and q2q_2, and constructing the payoff matrix of player 2 as visualized in Figure 11a. By computing the probability for obtaining each individual payoff outcome with the pdf of q1q_1, the payoff probability spectrum is created as presented in Figure 11b. As shown in the results of the two different numerical approaches, both have resulted in a concentrated probability near 0 which increases rapidly with increasing number of samples and bins. This suggests that the divergence at 0 may be an inherent feature of the pdf of the SV quantum payoff.

Figure 11. Computing payoff distribution through uniform sampling: a) Here, the probability distributions of q1q_1 and q2q_2 are sampled at 8 distinct outcomes for each mode. The 8 pairs of q1q_1 and q2q_2 samples give 64 possible payoff results, each with probability given by fq1(a1)×fq2(a2)f_{q_1}(a_1) \times f_{q_2}(a_2). The payoff values are presented by a color-coded plot, with their corresponding probability value (scaled by ×10−1\times10^{-1}) at the center of each colored box. b) By gathering the possible payoff results and their associated probabilities, a histogram of the payoff probability distribution is constructed. c) Similarly, the probability distribution of the bid error d2d_2 described in Equation (26) can also be built.

To investigate the origin of the divergence, we examine the component inside the quantum payoff in Equation (2):

di=qi−pN∑jqj(26)d_i = q_i - \frac{p}{N}\sum_j q_j \tag*{(26)}

where $iQ=−di2\$_i^Q = -d_i^2.

The bid error did_i is introduced to demonstrate the effect of squaring a random variable. The bid error represents the distance between a player’s bid and pp of the average. The resultant pdf of did_i through the numerical computation is shown in Figure 11c, where high probability amplitudes are gathered between −2-2 and 00. When the random variable did_i is squared in the quantum payoff function, the pdf support bunches and spreads for ∣di∣<1\lvert d_i\rvert< 1 and ∣di∣>1\lvert d_i\rvert> 1 respectively. This behavior stems from the fact that squaring a small number ∈(0,1)\in(0,1) creates a smaller number, and squaring a large number ∈(1,∞)\in(1,\infty) creates a larger number. The values of did_i close to zero bunch closer and closer. Amplifying the effect, the negative values of did_i also switch to positive in the process of squaring.

In the 2-player example, the consequence of squaring a random variable is most significant in player 2’s payoff due to their expected payoff being much closer to 0 compared to player 1. With the number of uniform samples for q1q_1 and q2q_2 increasing indefinitely in the continuous limit, the number of probability amplitudes around 0 in Figure 11b grows at a much faster rate relative to the rest of the spectrum, which results in the divergence in the SV quantum payoff pdf.

Analytical Expression of Payoff

To work towards a general expression for the SV quantum payoff’s pdf, we return to the pdf of qiq_i described in Equation (25). Each player’s quantum payoff involves subtracting their bid qiq_i by the scaled summation of all player’s random bid and taking the square as in Equation (2). The subtraction and the summation of random variables can be derived through the convolution of their pdf, and the pdf of the bid error did_i is found to be:

fdi(adi)=Aexp⁡[−B(adi−Ci)2](27)f_{d_i}(a_{d_i}) = A\exp\left[-B\left(a_{d_i}-C_i\right)^2\right] \tag*{(27)}

where adi∈(−∞,∞)a_{d_i} \in(-\infty,\infty) is the support for the random variable

able did_i, while AA, BB, and CiC_i are defined as:

A=Nπ(N−2p+p2) ⟶ 12πVar⁡[di](28)A=\sqrt{\frac{N}{\pi(N-2p+p^2)}}\ \longrightarrow\ \frac{1}{\sqrt{2\pi\operatorname{Var}[d_i]}} \tag*{(28)}
B=NN−2p+p2 ⟶ 12Var⁡[di](29)B=\frac{N}{N-2p+p^2}\ \longrightarrow\ \frac{1}{2\operatorname{Var}[d_i]} \tag*{(29)}
Ci=⟨qi⟩(γ)−pN∑jN⟨qj⟩(γ) ⟶ E[di].(30)C_i=\langle q_i\rangle(\gamma)-\frac{p}{N}\sum_{j}^{N}\langle q_j\rangle(\gamma)\ \longrightarrow\ \mathbb{E}[d_i]. \tag*{(30)}

Squaring the did_i involves using multiplicative convolution2^{2}, and hence the pdf of the quantum payoff Q=−di2Q=-d_i^2 is:

fsiQ(asiQ)=A2−asiQ{exp⁡[−B(−asiQ−Ci)2]+exp⁡[−B(−asiQ+Ci)2]}.(31)f_{s_i^Q}\left(a_{s_i}^Q\right)=\frac{A}{2\sqrt{-a_{s_i}^Q}}\left\{\exp\left[-B\left(\sqrt{-a_{s_i}^Q}-C_i\right)^2\right]+\exp\left[-B\left(\sqrt{-a_{s_i}^Q}+C_i\right)^2\right]\right\}. \tag*{(31)}

For a simpler picture of the payoff pdf under different input bids and γ\gamma, we derive the pdf of expected value and variance:

E[SiQ]=∫−∞0asiQ⋅fsiQ dasiQ=−(12B+Ci2)\begin{aligned} \mathbb{E}\left[S_i^Q\right]=\int_{-\infty}^{0}a_{s_i}^Q\cdot f_{s_i^Q}\,da_{s_i}^Q \\ &=-\left(\frac{1}{2B}+C_i^2\right) \end{aligned}
⟶−Var⁡[di]−[E[di]]2(32)\longrightarrow-\operatorname{Var}[d_i]-[\mathbb{E}[d_i]]^2 \tag*{(32)}
Var⁡[SiQ]=∫−∞0(asiQ)2⋅fsiQ dasiQ−[E[SiQ]]2=1B(12B+2Ci2)\begin{aligned} \operatorname{Var}\left[S_i^Q\right]=\int_{-\infty}^{0}\left(a_{s_i}^Q\right)^2\cdot f_{s_i^Q}\,da_{s_i}^Q-\left[\mathbb{E}\left[S_i^Q\right]\right]^2 \\ &=\frac{1}{B}\left(\frac{1}{2B}+2C_i^2\right) \end{aligned}
⟶2[Var⁡[di]]2+4Var⁡[di]⋅[E[di]]2.(33)\longrightarrow2[\operatorname{Var}[d_i]]^2+4\operatorname{Var}[d_i]\cdot[\mathbb{E}[d_i]]^2. \tag*{(33)}

In the EV game, the expected payoff of each player ⟨SiQ⟩\langle S_i^Q\rangle is determined by the expected measurement outcomes ⟨qi⟩\langle q_i\rangle. In the SV game, we compute the payoff for each player using the processed bids qiq_i's obtained in a single measurement iteration, where the expected payoff E[SiQ]\mathbb{E}[S_i^Q] is derived as the average of these resulting payoff values over multiple iterations.

Under increasing level of squeezing, the variance of the payoffs increases while the expected payoff decreases for the same set of input bids. This effect can be seen by observing Equation (32) and (33). The two features are

also shown for the given example in Figure 12b&Figure 12c). The decreasing expected payoff is a result of the squeezing operations amplifying the distance between the player’s bids and the average bid as discussed in Section III B, while under lower expected payoffs, the variance increases as a result of the contribution from (−asiQ)−1/2(-a_{s_i}^Q)^{-1/2} term in Equation (31).

Figure 12. Analytical Solution of SiQS_i^Q's Probability Density: a) for player 1 and player 2 with support as1Qa_{s_1}^Q and as2Qa_{s_2}^Q respectively; b) S1QS_1^Q with γ∈[0,0.8]\gamma\in[0,0.8] (in 0.02 increments); c) S2QS_2^Q for γ∈[0,0.8]\gamma\in[0,0.8], where small values are plotted separately to show the divergence; d) The expected value and variance of the payoffs as a function of γ\gamma.

With an increasing variance and uncertainty under higher levels of squeezing, estimating the strategies of others becomes increasingly difficult. While the transient learning process of rational players can only be clearly understood through a multi-round analysis of the game, the introduction of bid uncertainty may disincentivize speculative behaviors due to the increased difficulty and significantly weakens high-order rationality strategies.

Due to the contribution of (−asiQ)−1/2\left(-a_{s_i^Q}\right)^{-1/2} in the payoff pdf, the divergence near 0 becomes significant for a near 0 expected payoff E[SiQ]≈0\mathbb{E}\left[S_i^Q\right]\approx0. Given the probability is defined by the integral of some continuous interval of the pdf, the narrowness of the peak near 0 from the divergence does not necessarily lead to the players getting better payoff on average.

Probabilistic Payoff & Optimal Strategy

The SV quantum payoff probability spectrum is the integration of its pdf function over an interval of outcomes. Given the measurement outcomes are always of finite resolution in practice, the discretization produced by the Monte Carlo simulation may be close to the observations in an implemented quantum game. Here, we demonstrate the characteristics of the SV payoff through the integration of the payoff pdf in Equation (31) over equal-width intervals.

Despite the additions of probabilistic quantum measurements, the SV expected payoff E[SiQ]\mathbb{E}\left[S_i^Q\right] has retained the linear dependence on −Ci2-C_i^2 from the EV quantum payoff as proven in Equation (32) and visualized in Figure 13a. The dashed line in the figure demonstrates the equivalence between negative squared expected bid error and the EV payoff function ⟨SiQ⟩\left\langle S_i^Q\right\rangle. This is a significant result for the SV game, as it demonstrates that any rational player seeking to maximize the expected payoff in the SV quantum pp-BC will adopt a strategy consistent with the analysis of the EV quantum pp-BC, hence preserving the mitigation of the boom-and-bust strategy.

Figure 13. a) Expected value and variance of the SV payoff E[SiQ]\mathbb{E}\left[S_i^Q\right] & Var⁡[SiQ]\operatorname{Var}\left[S_i^Q\right] as a function of the negative squared expected bid error −Ci2-C_i^2, with dashed lines drawn as references for gradients −1 and 1. b) Resultant payoff probability spectrum given nsamples=100n_{\mathrm{samples}}=100 by adjusting the expected bid error CiC_i and hence expected payoff E[SiQ]\mathbb{E}\left[S_i^Q\right]. c) Resultant payoff probability for nsamples=1000n_{\mathrm{samples}}=1000.

A rational player may also approach the game to optimize their bid for the most probable payoff outcome. For the purpose of this analysis, we define the most probable payoff for a given discretization setting nsamplesn_{\mathrm{samples}}, using the notation [SiQ∣Max⁡(PsiQ)](nsamples)\left[S_i^Q\mid\operatorname{Max}(P_{s_i^Q})\right](n_{\mathrm{samples}}). In any game of pp-BC, each player’s payoff is solely dependent on the inputs of the participating players. Accordingly, under fixed squeezing parameter γ\gamma, scaling parameter pp, number of players NN and the number of discrete samples for pdf nsamplesn_{\mathrm{samples}}, the expected value and the variance of payoff is only a function of the expected bid error CiC_i and the expected measurements ⟨qi⟩\langle q_i\rangle as defined in Equation (31).

For a player who bids exactly on pQpQ of the average with an expected bid error of zero Ci=0C_i=0, their expected payoff becomes less than 0 E[SiQ]=−Var⁡(di)<0\mathbb{E}\left[S_i^Q\right]=-\operatorname{Var}(d_i)<0, as demonstrated in Equation (32). Namely, if player 1 guesses exactly on 2/3 of the average in the 2-player case, their expected payoff is E[SiQ]≈−0.28\mathbb{E}\left[S_i^Q\right]\approx-0.28 in this example instead of 0 as in the EV quantum game. Despite the lowered expected payoff, it is important to note that the rational strategy in game theory depends on the gradient of payoff rather than the absolute value. The more significant change under the SV game is how each individual bid yields a non-zero probability of receiving the maximum payoff, as a result of the probability density diverging at 0.

The peak probability shifts under different discretization settings, which is largely dependent on the bid error did_i. By increasing the magnitude of the expected bid error ∣Ci∣\lvert C_i\rvert from 0, there exists a value where the most probable payoff is non-zero [SiQ∣Max⁡(PsiQ)]≠0\left[S_i^Q\mid\operatorname{Max}(P_{s_i^Q})\right]\ne0 for a given nsamplesn_{\mathrm{samples}}, as demonstrated in Figure 13b & Figure 13c. We define this value as the threshold E[SiQ]=SthresQ\mathbb{E}\left[S_i^Q\right]=S_{\mathrm{thres}}^Q.

Below the threshold, the most probable payoff 3^{3} is 0, where the divergence term (−asiQ)−1/2\left(-a_{s_i^Q}\right)^{-1/2} dominates the Gaussian term of the pdf. Above the threshold, the

most probable outcome approaches the expected payoff 4^{4}, where the Gaussian contribution in the pdf dominates the divergence term in Equation (31). Under the limit of infinite discretization samples nsamples→∞n_{\mathrm{samples}} \to\infty, the threshold diverges to negative infinity $thresQ→−∞\$^{Q}_{\mathrm{thres}} \to-\infty as demonstrated in the right figure of Figure 14a. Hence, there is no value where the [$iQ∣Max⁡(P$iQ)][\$^{Q}_{i}|\operatorname{Max}(P_{\$^{Q}_{i}})] becomes non-zero under the continuous limit.

Figure 14. a) Left: The most probable payoff [$iQ∣Max⁡(P$iQ)][\$^{Q}_{i}|\operatorname{Max}(P_{\$^{Q}_{i}})] as a function of negative squared expected bid error −Ci2-C_{i}^{2} for the discretization setting nsamples∈{100,500,1000}n_{\mathrm{samples}} \in\{100,500,1000\}. Right: The negative squared expected bid error threshold −Ci2-C_{i}^{2} for the most probable payoff [$iQ∣Max⁡(P$iQ)]≠0[\$^{Q}_{i}|\operatorname{Max}(P_{\$^{Q}_{i}})] \ne0 to be non-zero, as a function of nsamplesn_{\mathrm{samples}}; b) Probability of obtaining better than expected payoff $iQ>E[$iQ]\$^{Q}_{i} > \mathop{\mathrm{E}}[\$^{Q}_{i}] as a function of −Ci2-C_{i}^{2}.

Intuitively, one would expect players to benefit from the SV quantization given the non-zero probability of obtaining the maximum payoff when a player bids close to the pp of the average in the quantum game. To confirm this quantitatively, for each given nsamplesn_{\mathrm{samples}}, we again shift CiC_i and calculate the probability for a larger than expected value outcome ($iQ>E[$iQ]\$^{Q}_{i} > \mathop{\mathrm{E}}[\$^{Q}_{i}]) by integration. The result is as shown in Figure 14c. For nsamples≫1n_{\mathrm{samples}} \gg1, bidding right on the pp of average gives E[$iQ]=−0.28\mathop{\mathrm{E}}[\$^{Q}_{i}] = -0.28 with around 70% probability of getting better than −0.28-0.28 as payoff. For worse bids under increased ∣Ci∣|C_i|, the chance of getting better payoff becomes the same as the chance of getting worse – 50/50, and this is where f$iQf_{\$^{Q}_{i}} becomes approximately a Gaussian, and the expected value simply

becomes E[$iQ]≈−Ci2\mathop{\mathrm{E}}[\$^{Q}_{i}] \approx-C_{i}^{2}.

Lastly, the noise introduced by the uncertainty principle is static in magnitude, and its impact on the payoff is dependent on the range and resolution of all possible bids. If the range of the displacement operation can be widened without increasing noise or decreasing bid resolution, then the pdf of the quantum measurement qiq_i and payoff $iQ\$^{Q}_{i} converges to Dirac delta functions in the limit of large strategy range (xi∈[a,b]x_i \in[a,b] where b≫ab \gg a), and hence, SV quantum pp-BC converges to the EV game.

Despite the added layers of complexity, the SV quantum pp-BC approach significantly reduces the requirement on computational resources, while retaining the capability in mitigating boom-and-bust rational strategies on-average, similar to the EV quantum pp-BC. As shown, the SV probabilistic payoff strongly rewards accurate estimations of the average bid, which may contribute to further incentivize non-speculative rational strategies.

Conclusion

We have demonstrated, by game theoretic analysis of the designed quantum pp-BC, that the boom-and-bust rational strategy present in the classical pp-BC can be contained by the quantization process. We showed the quantization scheme can relax the kk-rationality process for the rational players, where this feature is present in both the EV and the SV setting of the quantum pp-BC. With a chosen level of squeezing γ\gamma and the adjustment of the quantum effective scaling parameter ρQ\rho_Q, the resistance against boom-and-bust strategies can be appropriately strengthened for a desirable level. In the last section, we presented some unique features of the quantum pp-BC under the single-measurement setting as the SV quantum pp-BC. We have provided quantitative insights for the SV quantum game under the measurement uncertainty, as well as the differences and similarities compared to the simpler EV game.

The SV quantum pp-BC has the potential to create a game with similar characteristics to the EV setting while requiring significantly fewer resources. On the other hand, the complexity could lead to some unexpected exploitative strategies in the multi-round game that would alter the above described player behavior. One way to explore this can be simulation with neural network agents. This approach will also allow comparison between the learning processes of the agents in the SV and EV game, and potentially finding convergence in rational strategies as well as Nash equilibria in the SV quantum pp-BC.

By incorporating a designed quantization process, there is potential for a quantum market system to be built as an environment with innate resistance to boom-and-bust behavior, allowing genuine pricing of commodities to emerge naturally in a market without external intervention. Given the broad scope of the simple pp-BC model, the quantization process has the potential to accommodate the rich complexity of a functioning economic system, and contribute to designing a sustainable market and economy.

Author Contribution

  • Hiu Ming Lau: Conceptualization, Formal Analysis, Investigation, Methodology, Visualization, Writing – Original Draft Preparation

  • Kareem Raslan: Conceptualization, Investigation, Validation

  • Kieran Hymas: Supervision, Writing – review & editing

  • Qiang Sun: Investigation, Software, Supervision, Writing – review & editing

  • Azhar Iqbal: Conceptualization, Formal Analysis

  • Derek Abbott: Funding Acquisition, Methodology, Project Administration, Resources, Supervision, Validation, Writing – review & editing.

  • James Q. Quach: Conceptualization, Methodology, Project Administration, Supervision, Validation, Writing – review & editing

  • Andrew D. Greentree: Conceptualization, Funding Acquisition, Methodology, Project Administration, Resources, Supervision, Validation, Writing – review & editing

Acknowledgments

We acknowledge the joint financial support of this project from RMIT and CSIRO through the PhD program. Support from the Australian Research Council (DP240103404 and FL240100217) is gratefully acknowledged. We would like to thank Tinula Kariyasasam for supportive discussions on the Mean-Field type analysis and stylistic comments.

References

Appendix

Quantization Circuit Wave function derivation

The final state of the quantization circuit before measurement ∣ψf⟩|\psi_f\rangle as shown in Figure 2 can be written as:

∣ψf⟩=J^†[∏jND^j(xj)]J^∣000...000⟩.(34)|\psi_f\rangle= \hat{J}^{\dagger}\left[\prod_{j}^{N}\hat{D}_j(x_j)\right]\hat{J}|000...000\rangle. \tag*{(34)}

The expectation value of measurement result for each mode with respect to the positional basis is then:

qi=⟨ψf∣X^i∣ψf⟩=⟨000...000∣J^†[∏jND^j†(xj)]J^X^iJ^†[∏jND^j(xj)]J^∣000...000⟩.(35)\begin{aligned} q_i = \langle\psi_f|\hat{X}_i|\psi_f\rangle\\ &= \langle000...000|\hat{J}^{\dagger}\left[\prod_{j}^{N}\hat{D}^{\dagger}_j(x_j)\right]\hat{J}\hat{X}_i\hat{J}^{\dagger}\left[\prod_{j}^{N}\hat{D}_j(x_j)\right]\hat{J}|000...000\rangle. \tag*{(35)} \end{aligned}

The form of the NN-mode squeezing operators [30] J^(γ)\hat{J}(\gamma) & J^†(γ)\hat{J}^{\dagger}(\gamma) is chosen to be:

J^(γ)=e−γj^J^†(γ)=eγj^j^=∑i,j=1N(δij−1/N)(a^i†a^j†−a^ia^j).(36)\begin{aligned} \hat{J}(\gamma) &= e^{-\gamma\hat{j}}\\ \hat{J}^{\dagger}(\gamma) &= e^{\gamma\hat{j}}\\ \hat{j} &= \sum_{i,j=1}^{N}(\delta_{ij}-1/N)\left(\hat{a}^{\dagger}_i\hat{a}^{\dagger}_j-\hat{a}_i\hat{a}_j\right). \tag*{(36)} \end{aligned}

With the displacement operator encoding each player’s strategy:

D^i(xi)=exp⁡(−ixiP^i)(37)\hat{D}_i(x_i)=\exp\left(-\mathrm{i}x_i\hat{P}_i\right) \tag*{(37)}

where P^i=i/2(a^i†−a^i)\hat{P}_i=\mathrm{i}/\sqrt{2}\left(\hat{a}^{\dagger}_i-\hat{a}_i\right).

In the following, we derive an expression of (34) in terms of all players’ inputs xix_i by applying Baker-Campbell-Hausdorff (BCH) to J^†D^kJ^\hat{J}^{\dagger}\hat{D}_k\hat{J} [29, 36]. We begin by applying BCH to the term J^†a^kJ^\hat{J}^{\dagger}\hat{a}_k\hat{J}:

eA^a^ke−A^=∑m=0∞1m!(ad⁡A^)m(a^k)=10!a^k+11![A^,a^k]+12![A^,[A^,a^k]]+13![A^,[A^,[A^,a^k]]]+…(38)\begin{aligned} e^{\hat{A}}\hat{a}_k e^{-\hat{A}}=\sum_{m=0}^{\infty}\frac{1}{m!}(\operatorname{ad}_{\hat{A}})^m(\hat{a}_k) \\ &=\frac{1}{0!}\hat{a}_k+\frac{1}{1!}[\hat{A},\hat{a}_k]+\frac{1}{2!}[\hat{A},[\hat{A},\hat{a}_k]]+\frac{1}{3!}[\hat{A},[\hat{A},[\hat{A},\hat{a}_k]]]+\ldots \tag*{(38)} \end{aligned}

where A^=−γ∑i,j=1N(δij−1/N)(a^i†a^j†−a^ia^j)\hat{A}=-\gamma\sum_{i,j=1}^{N}(\delta_{ij}-1/N)(\hat{a}^{\dagger}_i\hat{a}^{\dagger}_j-\hat{a}_i\hat{a}_j).

By expanding out the multi-layered commutators, we can find a pattern to help simplify the BCH expression:

[A^,a^k]=−2γ(−a^k†+1N∑lNa^l†)(39)[\hat{A},\hat{a}_k] = -2\gamma\left(-\hat{a}_k^\dagger+ \frac{1}{N}\sum_l^N \hat{a}_l^\dagger\right) \tag*{(39)}
[A^,[A^,a^k]]=2γ[A^,a^k†]for m>1:(ad⁡A^)m(a^k)=(2γ)m−1[A^,a^k†](ad⁡A^)m(a^k†)=(2γ)m−1[A^,a^k].(40)\begin{aligned} [\hat{A},[\hat{A},\hat{a}_k]] &= 2\gamma[\hat{A},\hat{a}_k^\dagger] \\ \text{for }m>1:\qquad (\operatorname{ad}_{\hat{A}})^m(\hat{a}_k) &= (2\gamma)^{m-1}[\hat{A},\hat{a}_k^\dagger] \\ (\operatorname{ad}_{\hat{A}})^m(\hat{a}_k^\dagger) &= (2\gamma)^{m-1}[\hat{A},\hat{a}_k]. \tag*{(40)} \end{aligned}

Substitute the commutator pattern into the Equation (38), we obtain:

eA^a^ke−A^=a^k+[A^,a^k]+(2γ)12![A^,a^k†]+13!(2γ)2[A^,a^k]+(2γ)314![A^,a^k†]+(2γ)415![A^,a^k]−⋯=a^k−(2γ)(−a^k†+1N∑lNa^l†)−(2γ)22!(−a^k+1N∑lNa^l)−(2γ)33!(−a^k†+1N∑lNa^l†)−(2γ)44!(−a^k+1N∑lNa^l)−(2γ)55!(−a^k†+1N∑lNa^l†)−⋯=∑m∞(2γ)2m(2m)!a^k+∑m∞(2γ)2m+1(2m+1)!a^k†+(1−∑m∞(2γ)2m(2m)!)1N∑lNa^l−∑m∞(2γ)2m+1(2m+1)!1N∑lNa^l†.(41)\begin{aligned} e^{\hat{A}}\hat{a}_k e^{-\hat{A}} &= \hat{a}_k + [\hat{A},\hat{a}_k] + (2\gamma)\frac{1}{2!}[\hat{A},\hat{a}_k^\dagger] + \frac{1}{3!}(2\gamma)^2[\hat{A},\hat{a}_k] \\ &\quad+ (2\gamma)^3\frac{1}{4!}[\hat{A},\hat{a}_k^\dagger] + (2\gamma)^4\frac{1}{5!}[\hat{A},\hat{a}_k] - \cdots\\ &= \hat{a}_k - (2\gamma)\left(-\hat{a}_k^\dagger+ \frac{1}{N}\sum_l^N\hat{a}_l^\dagger\right) - \frac{(2\gamma)^2}{2!}\left(-\hat{a}_k + \frac{1}{N}\sum_l^N\hat{a}_l\right) \\ &\quad- \frac{(2\gamma)^3}{3!}\left(-\hat{a}_k^\dagger+ \frac{1}{N}\sum_l^N\hat{a}_l^\dagger\right) - \frac{(2\gamma)^4}{4!}\left(-\hat{a}_k + \frac{1}{N}\sum_l^N\hat{a}_l\right) \\ &\quad- \frac{(2\gamma)^5}{5!}\left(-\hat{a}_k^\dagger+ \frac{1}{N}\sum_l^N\hat{a}_l^\dagger\right) - \cdots\\ &= \sum_m^\infty\frac{(2\gamma)^{2m}}{(2m)!}\hat{a}_k + \sum_m^\infty\frac{(2\gamma)^{2m+1}}{(2m+1)!}\hat{a}_k^\dagger\\ &\quad+ \left(1-\sum_m^\infty\frac{(2\gamma)^{2m}}{(2m)!}\right)\frac{1}{N}\sum_l^N\hat{a}_l - \sum_m^\infty\frac{(2\gamma)^{2m+1}}{(2m+1)!}\frac{1}{N}\sum_l^N\hat{a}_l^\dagger. \tag*{(41)} \end{aligned}

The resultant summation terms in Equation (41) are incidentally Taylor series of hyperbolic functions. Hence, the action of the squeezing operators on the annihilation and creation operators gives us:

J^a^kJ^†=cosh⁡(2γ)a^k+sinh⁡(2γ)a^k†+(1−cosh⁡(2γ))1N∑lNa^l−sinh⁡(2γ)1N∑lNa^l†.(42)\begin{aligned} \hat{J}\hat{a}_k\hat{J}^{\dagger} &= \cosh(2\gamma)\hat{a}_k + \sinh(2\gamma)\hat{a}_k^\dagger\\ &\quad+ (1-\cosh(2\gamma))\frac{1}{N}\sum_l^N\hat{a}_l - \sinh(2\gamma)\frac{1}{N}\sum_l^N\hat{a}_l^\dagger. \tag*{(42)} \end{aligned}
J^†a^kJ^=cosh⁡(−2γ)a^k+sinh⁡(−2γ)a^k†+(1−cosh⁡(−2γ))1N∑lNa^l−sinh⁡(−2γ)1N∑lNa^l†.(43)\begin{aligned} \hat{J}^{\dagger}\hat{a}_k\hat{J} &= \cosh(-2\gamma)\hat{a}_k + \sinh(-2\gamma)\hat{a}_k^\dagger\\ &\quad+ (1-\cosh(-2\gamma))\frac{1}{N}\sum_l^N\hat{a}_l - \sinh(-2\gamma)\frac{1}{N}\sum_l^N\hat{a}_l^\dagger. \tag*{(43)} \end{aligned}

With Equation (43), we can then express J^†D^k\hat{J}^{\dagger}\hat{D}_k as:

J^†D^k(xk)J^=J^†exp⁡{xi2(a^k†−a^k)}J^(44)\hat{J}^{\dagger}\hat{D}_k(x_k)\hat{J} = \hat{J}^{\dagger}\exp\left\{\frac{x_i}{\sqrt{2}}(\hat{a}_k^\dagger-\hat{a}_k)\right\}\hat{J} \tag*{(44)}
=exp⁡{−ixi[cosh⁡(−2γ)P^k−sinh⁡(−2γ)P^k†+(1−cosh⁡(−2γ))1N∑lNP^l+sinh⁡(−2γ)1N∑lNP^l†]}.(45)\begin{aligned} &= \exp\left\{-ix_i\left[\cosh(-2\gamma)\hat{P}_k-\sinh(-2\gamma)\hat{P}_k^\dagger\right.\right.\\ &\qquad\left.\left. +(1-\cosh(-2\gamma))\frac{1}{N}\sum_l^N\hat{P}_l+\sinh(-2\gamma)\frac{1}{N}\sum_l^N\hat{P}_l^\dagger\right]\right\}. \tag*{(45)} \end{aligned}

Hence, ∣ψf⟩|\psi_f\rangle can now be written as:

∣ψf⟩=J^†∏kND^k(xk)J^∣0⟩⊗N=⨂kNexp⁡{−i[xk(cosh⁡2γ+sinh⁡2γ)+(1−cosh⁡2γ)1N∑lNxl−sinh⁡2γ1N∑lNxl]P^k}=⨂kNexp⁡{−i[e2γ(xk−xˉ)+xˉ]P^k}∣0⟩k.(46)\begin{aligned} |\psi_f\rangle&= \hat{J}^{\dagger}\prod_{k}^{N}\hat{D}_k(x_k)\hat{J}|0\rangle^{\otimes N} \\ &= \bigotimes_{k}^{N}\exp\left\{-i\left[x_k(\cosh2\gamma+\sinh2\gamma)+(1-\cosh2\gamma)\frac{1}{N}\sum_{l}^{N}x_l-\sinh2\gamma\frac{1}{N}\sum_{l}^{N}x_l\right]\hat{P}_k\right\} \\ &= \bigotimes_{k}^{N}\exp\left\{-i\left[e^{2\gamma}(x_k-\bar{x})+\bar{x}\right]\hat{P}_k\right\}|0\rangle_k . \tag*{(46)} \end{aligned}

For quadrature measurement on the positional basis, the expectation value is:

⟨qk⟩(γ,x)=Tr⁡(X^k∣ψf⟩⟨ψf∣)=⟨ψf∣X^k∣ψf⟩=k⟨e2γ(xk−xˉ)+xˉ∣X^k∣e2γ(xk−xˉ)+xˉ⟩k=e2γ(xk−xˉ)+xˉ.(47)\begin{aligned} \langle q_k\rangle(\gamma,\mathbf{x}) &= \operatorname{Tr}\left(\hat{X}_k|\psi_f\rangle\langle\psi_f|\right)=\langle\psi_f|\hat{X}_k|\psi_f\rangle\\ &= {}_k\langle e^{2\gamma}(x_k-\bar{x})+\bar{x}|\hat{X}_k|e^{2\gamma}(x_k-\bar{x})+\bar{x}\rangle_k \\ &= e^{2\gamma}(x_k-\bar{x})+\bar{x}. \tag*{(47)} \end{aligned}

Finally, we find the expected payoff as a function of all players’ inputs:

⟨ΦiQ⟩(γ,x)=Φi(⟨q⟩)=−(⟨qi⟩−pN∑j=1N⟨qj⟩)2=−e4γ{xi−[1+e−2γ(p−1)]xˉ}2=−e4γ(xi−pQxˉ)2(48)\begin{aligned} \langle\Phi_i^Q\rangle(\gamma,\mathbf{x}) &= \Phi_i(\langle q\rangle) \\ &= -\left(\langle q_i\rangle-\frac{p}{N}\sum_{j=1}^{N}\langle q_j\rangle\right)^2 \\ &= -e^{4\gamma}\left\{x_i-\left[1+e^{-2\gamma}(p-1)\right]\bar{x}\right\}^2 \\ &= -e^{4\gamma}(x_i-p_Q\bar{x})^2 \tag*{(48)} \end{aligned}

where pQ=1+e−2γ(p−1)p_Q=1+e^{-2\gamma}(p-1).

Payoff PDF Derivation

First, we express the wave function of the initial vacuum state with respect to the positional basis as [35]:

ψ0(ai)=⟨ai∣D^i(xi)∣0⟩1=1π1/4exp⁡(−ai22).(49)\psi_0(a_i) = \langle a_i|\hat{D}_i(x_i)|0\rangle_1 = \frac{1}{\pi^{1/4}}\exp\left(-\frac{a_i^2}{2}\right). \tag*{(49)}

where aia_i is the support of mode ii’s distribution under positional basis X^i\hat{X}_i.

Recall (46) and (47), the wave function before measurement is:

∣ψf⟩=exp⁡(−i⟨q1⟩P^1)∣0⟩1⊗exp⁡(−i⟨q2⟩P^2)∣0⟩2⊗…⊗exp⁡(−i⟨qN⟩P^N)∣0⟩N(50)|\psi_f\rangle= \exp\left(-i\langle q_1\rangle\hat{P}_1\right)|0\rangle_1 \otimes\exp\left(-i\langle q_2\rangle\hat{P}_2\right)|0\rangle_2 \otimes\ldots\otimes\exp\left(-i\langle q_N\rangle\hat{P}_N\right)|0\rangle_N \tag*{(50)}

where the γ\gamma dependence in the expectation value ⟨qi⟩(γ,x)\langle q_i\rangle(\gamma,\mathbf{x}) is omitted to avoid clutter.

For some displacement η1=(α1+iβ1)/2\eta_1 = (\alpha_1+i\beta_1)/\sqrt{2}, the action of our displacement operators on the vacuum state is:

⟨a1∣D^1(η1)∣0⟩1=⟨a1∣exp⁡(−iη1P^1)∣0⟩1=⟨a1∣exp⁡(−iα1β1/2)exp⁡(iβ1X^1)exp⁡(−iα1P^1)∣0⟩=exp⁡(−iα1β1/2)exp⁡(iβ1X^1)exp⁡(−iα1P^1)exp⁡(−a12/2)=exp⁡(−iα1β1/2)exp⁡(−(a1−α1)2/2)→β1=0exp⁡(−(a1−α1)2/2).(51)\begin{aligned} \langle a_1|\hat{D}_1(\eta_1)|0\rangle_1 &= \langle a_1|\exp\left(-i\eta_1\hat{P}_1\right)|0\rangle_1 \\ &= \langle a_1|\exp\left(-i\alpha_1\beta_1/2\right)\exp\left(i\beta_1\hat{X}_1\right)\exp\left(-i\alpha_1\hat{P}_1\right)|0\rangle\\ &= \exp\left(-i\alpha_1\beta_1/2\right)\exp\left(i\beta_1\hat{X}_1\right)\exp\left(-i\alpha_1\hat{P}_1\right)\exp\left(-a_1^2/2\right) \\ &= \exp\left(-i\alpha_1\beta_1/2\right)\exp\left(-\left(a_1-\alpha_1\right)^2/2\right) \\ &\xrightarrow{\beta_1=0} \exp\left(-\left(a_1-\alpha_1\right)^2/2\right). \tag*{(51)} \end{aligned}

Hence, we find the final wave function in positional basis ψf\psi_f, as well as the probability density function of each mode fqif_{q_i} at the end of the quantization circuit:

ψf(⟨q⟩)=⟨a1a2…aN∣ψf⟩=1πN/4exp⁡(−(a1−⟨q1⟩)2/2)exp⁡(−(a2−⟨q2⟩)2/2)…exp⁡(−(aN−⟨qN⟩)2/2)(52)\begin{aligned} \psi_f(\langle\mathbf{q}\rangle) = \langle a_1a_2\ldots a_N|\psi_f\rangle &= \frac{1}{\pi^{N/4}}\exp\left(-\left(a_1-\langle q_1\rangle\right)^2/2\right)\exp\left(-\left(a_2-\langle q_2\rangle\right)^2/2\right)\ldots\\ &\quad\exp\left(-\left(a_N-\langle q_N\rangle\right)^2/2\right) \tag*{(52)} \end{aligned}
⟨qi∣ψf⟩=1π1/4exp⁡(−(ai−⟨qi⟩)2/2)(53)\langle q_i|\psi_f\rangle= \frac{1}{\pi^{1/4}}\exp\left(-\left(a_i-\langle q_i\rangle\right)^2/2\right) \tag*{(53)}
fqi(ai)=∣⟨ai∣ψf⟩∣2=1π1/2exp⁡(−(ai−⟨qi⟩)2).(54)f_{q_i}(a_i) = |\langle a_i|\psi_f\rangle|^2 = \frac{1}{\pi^{1/2}}\exp\left(-\left(a_i-\langle q_i\rangle\right)^2\right). \tag*{(54)}

Recall the quantum payoff function:

Φiq(q)=−(qi−p∑j=1NqjN)2=−((1−pN)qi−pN∑j≠iNqj)2.(55)\begin{aligned} \Phi_i^q(\mathbf{q}) &= -\left(q_i-p\frac{\sum_{j=1}^{N}q_j}{N}\right)^2 \\ &= -\left(\left(1-\frac{p}{N}\right)q_i-\frac{p}{N}\sum_{j\ne i}^{N}q_j\right)^2. \tag*{(55)} \end{aligned}

Recall the random variable properties given XX and YY with continuous probability density distributions:

for A=X+Y:fA(a)=∫−∞∞fX(τ)fY(a−τ)dτ(56)\text{for } A=X+Y:\qquad f_A(a)=\int_{-\infty}^{\infty}f_X(\tau)f_Y(a-\tau)d\tau \tag*{(56)}
fc⋅X(a)=1∣c∣fX(ac)(57)f_{c\cdot X}(a)=\frac{1}{|c|}f_X\left(\frac{a}{c}\right) \tag*{(57)}
for A=XY:fA(a)=∫−∞∞fX(x)fY(a/x)1∣x∣ dx.(58)\text{for } A=XY:\qquad f_A(a)=\int_{-\infty}^{\infty}f_X(x)f_Y(a/x)\frac{1}{|x|}\,dx. \tag*{(58)}

The summation term for all player bids except player ii’s in Equation (55) is the convolution of fqj(aj)f_{q_j}(a_j):

f∑qj(a∑qj)=fq1(a1)∗fq2(a2)∗...∗fqN(aN)=(∫fq1(τ)fq2(a1−τ) dτ)∗...fqN=1π12(∫exp⁡(−(τ−⟨q1⟩)2)exp⁡(−(a2−τ−⟨q2⟩)2) dτ)∗...fqN=1ππ2(exp⁡(−12(a1∗2−⟨q1⟩−⟨q2⟩)2))∗fq3∗...∗fqN=1ππ21π2π3(exp⁡(−13(a1∗2∗3−∑j3⟨qj⟩)2))∗fq4∗...∗fqN=1(N−1)πexp⁡(−1N−1(a∑qj−∑j≠iN⟨qj⟩)2).(59)f_{\sum q_j}(a_{\sum q_j}) = f_{q_1}(a_1) * f_{q_2}(a_2) * ... * f_{q_N}(a_N) = \left(\int f_{q_1}(\tau) f_{q_2}(a_1-\tau) \,d\tau\right) * ... f_{q_N} \\ = \frac{1}{\pi^{\frac{1}{2}}}\left(\int\exp\left(-(\tau-\langle q_1\rangle)^2\right)\exp\left(-(a_2-\tau-\langle q_2\rangle)^2\right)\,d\tau\right) * ...f_{q_N} \\ = \frac{1}{\pi}\sqrt{\frac{\pi}{2}}\left(\exp\left(-\frac{1}{2}(a_{1*2}-\langle q_1\rangle-\langle q_2\rangle)^2\right)\right) * f_{q_3} * ... * f_{q_N} \\ = \frac{1}{\pi}\sqrt{\frac{\pi}{2}}\frac{1}{\sqrt{\pi}}\sqrt{\frac{2\pi}{3}}\left(\exp\left(-\frac{1}{3}\left(a_{1*2*3}-\sum_{j}^{3}\langle q_j\rangle\right)^2\right)\right) * f_{q_4} * ... * f_{q_N} \\ = \frac{1}{\sqrt{(N-1)\pi}}\exp\left(-\frac{1}{N-1}\left(a_{\sum q_j}-\sum_{j\ne i}^{N}\langle q_j\rangle\right)^2\right). \tag*{(59)}

Then for the bid error di=(1−p/N)qi−p/N∑qjd_i=(1-p/N)q_i-p/N\sum q_j, its probability distribution is:

fdi(adi)=1∣1−p/N∣fqi(ai/(1−p/N))∗1∣−p/N∣f∑qj(a∑qj/(−p/N))=11−p/N(1πexp⁡(−(ai/(1−p/N)−⟨qi⟩)2))∗1p/N(1(N−1)πexp⁡(−1N−1(a∑qj/(−p/N)−∑j≠iN⟨qj⟩)2))=NN−p(1πexp⁡(−N2(N−p)2(ai−(1−p/N)⟨qi⟩)2))∗1p/N(1(N−1)πexp⁡(−1N−1(a∑qj−(−p/N)∑j≠iN⟨qj⟩)2))f_{d_i}(a_{d_i})=\frac{1}{|1-p/N|}f_{q_i}\left(a_i/(1-p/N)\right)*\frac{1}{|-p/N|}f_{\sum q_j}\left(a_{\sum q_j}/(-p/N)\right) \\ =\frac{1}{1-p/N}\left(\frac{1}{\sqrt{\pi}}\exp\left(-(a_i/(1-p/N)-\langle q_i\rangle)^2\right)\right)* \\ \frac{1}{p/N}\left(\frac{1}{\sqrt{(N-1)\pi}}\exp\left(-\frac{1}{N-1}\left(a_{\sum q_j}/(-p/N)-\sum_{j\ne i}^{N}\langle q_j\rangle\right)^2\right)\right) \\ =\frac{N}{N-p}\left(\frac{1}{\sqrt{\pi}}\exp\left(-\frac{N^2}{(N-p)^2}\left(a_i-(1-p/N)\langle q_i\rangle\right)^2\right)\right)* \\ \frac{1}{p/N}\left(\frac{1}{\sqrt{(N-1)\pi}}\exp\left(-\frac{1}{N-1}\left(a_{\sum q_j}-(-p/N)\sum_{j\ne i}^{N}\langle q_j\rangle\right)^2\right)\right)

Given Gaussian distributions, take the sum of the means, sum of variances

=Nπ(N−2p+p2)⋅exp⁡(−NN−2p+p2(adi−[(1−p/N)⟨qi⟩−p/N∑j≠iN⟨qj⟩])2)=Aexp⁡(−B(adi−Ci)2)(60)=\sqrt{\frac{N}{\pi(N-2p+p^2)}}\cdot\exp\left(-\frac{N}{N-2p+p^2}\left(a_{d_i}-\left[(1-p/N)\langle q_i\rangle-p/N\sum_{j\ne i}^{N}\langle q_j\rangle\right]\right)^2\right) \\ =A\exp\left(-B(a_{d_i}-C_i)^2\right) \tag*{(60)}

where:

A=Nπ(N−2p+p2)=BπA=\sqrt{\frac{N}{\pi(N-2p+p^2)}}=\sqrt{\frac{B}{\pi}}
B=NN−2p+p2B=\frac{N}{N-2p+p^2}
Ci=⟨qi⟩−p∑j⟨qj⟩.C_i=\langle q_i\rangle-p\sum_j\langle q_j\rangle.

Finally, to find the pdf of the random quantum payoff $Q\$Q, we square the random bid error did_i and find its pdf by [37]:

Y=X2FY(y)=P(Y≤y)=P(X2≤y)=P(∣X∣≤y)=P(−y≤X≤y)=FX(y)−FX(−y)(61)\begin{aligned} Y &= X^2 \\ F_Y(y) &= P(Y \le y) = P(X^2 \le y) = P(|X| \le y) \\ &= P(-\sqrt{y} \le X \le\sqrt{y}) = F_X(\sqrt{y}) - F_X(-\sqrt{y}) \tag*{(61)} \end{aligned}
fY(y)=dFY(y)dy=12yfX(y)+12yfX(−y).(62)f_Y(y) = \frac{dF_Y(y)}{dy} = \frac{1}{2\sqrt{y}}f_X(\sqrt{y}) + \frac{1}{2\sqrt{y}}f_X(-\sqrt{y}) . \tag*{(62)}

Hence finally, the analytical expression for the payoff probability distribution in our continuous-variable quantization can be found as:

f−siQ(asiQ)=A2asiQ(exp⁡(−B(asiQ−Ci)2)+exp⁡(−B(asiQ+Ci)2))(63)f_{-s_i^Q}(a_{s_i^Q}) = \frac{A}{2\sqrt{a_{s_i}^Q}}\left(\exp\left(-B\left(\sqrt{a_{s_i}^Q}-C_i\right)^2\right) + \exp\left(-B\left(\sqrt{a_{s_i}^Q}+C_i\right)^2\right)\right) \tag*{(63)}
fsiQ(asiQ)=A2−asiQ(exp⁡(−B(−asiQ−Ci)2)+exp⁡(−B(−asiQ+Ci)2)).(64)f_{s_i^Q}(a_{s_i^Q}) = \frac{A}{2\sqrt{-a_{s_i}^Q}}\left(\exp\left(-B\left(\sqrt{-a_{s_i}^Q}-C_i\right)^2\right) + \exp\left(-B\left(\sqrt{-a_{s_i}^Q}+C_i\right)^2\right)\right) . \tag*{(64)}

The expressions for expected value and variance can be obtained by symbolic computation:

E[siQ]=∫−∞0asiQ⋅fsiQ dasiQ=Aπ2B3/2(−1−2BCi2)for A=Bπ:=−Bππ2B3/22B(12B+Ci2)(65)\begin{aligned} \mathbb{E}\left[s_i^Q\right] &= \int_{-\infty}^{0} a_{s_i^Q}\cdot f_{s_i^Q}\, da_{s_i^Q} \\ &= \frac{A\sqrt{\pi}}{2B^{3/2}}(-1-2BC_i^2) \\ \text{for } A &= \sqrt{\frac{B}{\pi}}: \\ &= -\sqrt{\frac{B}{\pi}}\frac{\sqrt{\pi}}{2B^{3/2}}2B\left(\frac{1}{2B}+C_i^2\right) \tag*{(65)} \end{aligned}
=−(12B+Ci2)(66)= -\left(\frac{1}{2B}+C_i^2\right) \tag*{(66)}
=−Var⁡[di]−[E[di]]2(67)= -\operatorname{Var}[d_i]-\left[\mathbb{E}[d_i]\right]^2 \tag*{(67)}
Var⁡[siQ]=∫−∞0(asiQ)2⋅fsiQ dasiQ−(E[siQ])2=14B3(AB(3+4BCi2(3+BCi2))π−(A+2ABCi2)2π)=A4B3((3Bπ−Aπ)+4B(3Bπ−Aπ)Ci2+4B2(Bπ−Aπ)Ci4)for A=Bπ & Aπ=Bπ:=Bπ14B3(2Bπ+8BBπCi2)(68)\begin{aligned} \operatorname{Var}\left[s_i^Q\right] &= \int_{-\infty}^{0}\left(a_{s_i^Q}\right)^2\cdot f_{s_i^Q}\, da_{s_i^Q}-\left(\mathbb{E}\left[s_i^Q\right]\right)^2 \\ &= \frac{1}{4B^3}\left(A\sqrt{B}\left(3+4BC_i^2(3+BC_i^2)\right)\sqrt{\pi}-\left(A+2ABC_i^2\right)^2\pi\right) \\ &= \frac{A}{4B^3}\left((3\sqrt{B\pi}-A\pi)+4B(3\sqrt{B\pi}-A\pi)C_i^2+4B^2(\sqrt{B\pi}-A\pi)C_i^4\right) \\ \text{for } A &= \sqrt{\frac{B}{\pi}}\ \&\ A\pi=\sqrt{B\pi}: \\ &= \sqrt{\frac{B}{\pi}}\frac{1}{4B^3}\left(2\sqrt{B\pi}+8B\sqrt{B\pi}C_i^2\right) \tag*{(68)} \end{aligned}
=14B34B2(12B+2Ci2)(69)= \frac{1}{4B^3}4B^2\left(\frac{1}{2B}+2C_i^2\right) \tag*{(69)}
=2[Var⁡[di]]2+4Var⁡[di]⋅[E[di]]2.(70)= 2\left[\operatorname{Var}[d_i]\right]^2+4\operatorname{Var}[d_i]\cdot\left[\mathbb{E}[d_i]\right]^2 . \tag*{(70)}

References

  1. [1]J. M. Keynes, The Quarterly Journal of Economics 51, 209 (1937).
  2. [2]A. Lanteri and A. Carabelli, The European Journal of the History of Economic Thought 18, 269 (2011).
  3. [3]F. Allen, S. Morris, and H. S. Shin, Review of Financial Studies 19, 719 (2006).
  4. [4]G. P. Brockway, Journal of Post Keynesian Economics 5, 515 (1983).
  5. [5]S. Morris and H. S. Shin, The American Economic Review 92, 1521 (2002).
  6. [6]A. Bosch-Domènech, J. G. Montalvo, R. Nagel, and A. Satorra, The American Economic Review 92, 1687 (2002).
  7. [7]R. Nagel, in Handbook of Experimental Economics Results, Vol. 1 (Elsevier, 2008) pp. 391–410.
  8. [8]T. Ho, C. Camerer, and K. Weigelt, The American Economic Review 88, 947 (1998).
  9. [9]P. Sbriglia, Experimental Economics 11, 107 (2008).
  10. [10]E. Cartwright, Behavioral Economics, Routledge advanced texts in Economics and Finance No. 16 (Routledge, London ; New York, NY, 2011).
  11. [11]S. Hirota and S. Sunder, SSRN Electronic Journal 10.2139/ssrn.302393 (2002).
  12. [12]D. A. Meyer, Physical Review Letters 82, 1052–1055 (1999).
  13. [13]J. Eisert, M. Wilkens, and M. Lewenstein, Physical Review Letters 83, 3077–3080 (1999).
  14. [14]A. P. Flitney, Review of quantum game theory (Nova Science Publishers, 2009) Chap. 1, pp. 1–40.
  15. [15]F. S. Khan, N. Solmeyer, R. Balu, and T. S. Humble, Quantum Information Processing 17, 309 (2018).
  16. [16]L. Sanz-Martín, G. Rivas, N. Clavijo-Buritica, M. Herrera, and J. Parra-Domínguez, Quantum Information Processing 24, 291 (2025).
  17. [17]H. Li, J. Du, and S. Massar, Physics Letters A 306, 73 (2002).
  18. [18]J. Zhou, L. Ma, and Y. Li, Physics Letters A 339, 10 (2005).
  19. [19]H. Guo, J. Zhang, and G. J. Koehler, Decision Support Systems 46, 318 (2008).
  20. [20]Z. L. Zhou, G. Qin, L. Hu, S. Li, N. Xu, and J. Du, Physics Letters A 372, 3586 (2008).
  21. [21]Z. Yang and Q. Gong, Quantum Information Processing (Print) 17, 10.1007/s11128-018-2079-3 (2018).
  22. [22]K. Hymas, H. Lau, K. Raslan, Q. Sun, A. Iqbal, D. Abbott, A. D. Greentree, and J. Q. Quach, Quenching speculation in quantum markets via entangled neural traders (2026), preprint, arXiv:2602.06367 [quant-ph], arXiv:2602.06367 [quant-ph].
  23. [23]K. Ikeda and S. Aoki, Quantum Information Processing 21, 10.1007/s11128-021-03378-5 (2022).
  24. [24]D. Ding and L. Jiang, Coordinating decisions via quantum telepathy (2025), arXiv:2407.21723 [quant-ph], arXiv:2407.21723 [quant-ph].arxiv.org/abs/2407.21723
  25. [25]D. Acemoglu and M. K. Jensen, Games and Economic Behavior 81, 27 (2013).
  26. [26]A. P. Feynney and A. D. Greentree, Physics Letters A 362, 132 (2007).
  27. [27]M. Yildiz, Economic Applications of Game Theory (2012).
  28. [28]R. Nagel, The American Economic Review 85, 1313 (1995).
  29. [29]C. F. Lo and R. Sollie, Phys. Rev. A 47, 733 (1993).
  30. [30]C. Lo and D. Kiang, Europhysics Letters (EPL) 64, 592 (2003).
  31. [31]H. Vahlbruch, M. Mehmet, K. Danzmann, and R. Schnabel, Phys. Rev. Lett. 117, 110801 (2016).
  32. [32]H. Liu, M. L. Ju, N. Hamdash, and A. S. Helmy, Towards arbitrary time-frequency mode squeezing with self-conjugated mode squeezing in fiber (2025), arXiv:2406.19991 [quant-ph].arxiv.org/abs/2406.19991
  33. [33]S. L. Danilishin and F. Y. Khalili, Living Reviews in Relativity 15, 10.12942/lrr-2012-5 (2012).
  34. [34]B. Grosskopf and R. Nagel, Games and Economic Behavior 62, 93 (2008).
  35. [35]A. I. Lvovsky, Squeezed light, in Photonics (John Wiley & Sons, Ltd, 2015) Chap. 5, pp. 121–163, https://onlinelibrary.wiley.com/doi/pdf/10.1002/9781119009719.ch5.
  36. [36]P. Frackiewicz and J. Sladkowski, Quantum Information Processing 15, 3637 (2016).
  37. [37]J. C. Watkins, Transformations of random variables, Department of Mathematics, The University of Arizona (2009).

Paper details

vlnc:W_hf2hvjfr
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents