Extended nonlocal games

In this tutorial, we will define the concept of an extended nonlocal game. Extended nonlocal games are a more general abstraction of nonlocal games wherein the referee, who previously only provided questions and answers to the players, now share a state with the players and is able to perform a measurement on that shared state.

Every extended nonlocal game has a value associated to it. Analogously to nonlocal games, this value is a quantity that dictates how well the players can perform a task in the extended nonlocal game model when given access to certain resources. We will be using |toqito⟩ to calculate these quantities.

We will also look at existing results in the literature on these values and be able to replicate them using |toqito⟩. Much of the written content in this tutorial will be directly taken from [1].

Extended nonlocal games have a natural physical interpretation in the setting of tripartite steering [2] and in device-independent quantum scenarios [3]. For more information on extended nonlocal games, please refer to [4] and [1].

The extended nonlocal game model

An extended nonlocal game is similar to a nonlocal game in the sense that it is a cooperative game played between two players Alice and Bob against a referee. The game begins much like a nonlocal game, with the referee selecting and sending a pair of questions \((x,y)\) according to a fixed probability distribution. Once Alice and Bob receive \(x\) and \(y\), they respond with respective answers \(a\) and \(b\). Unlike a nonlocal game, the outcome of an extended nonlocal game is determined by measurements performed by the referee on its share of the state initially provided to it by Alice and Bob.

An extended nonlocal game.

Specifically, Alice and Bob’s winning probability is determined by collections of measurements, \(V(a,b|x,y) \in \text{Pos}(\mathcal{R})\), where \(\mathcal{R} = \mathbb{C}^m\) is a complex Euclidean space with \(m\) denoting the dimension of the referee’s quantum system–so if Alice and Bob’s response \((a,b)\) to the question pair \((x,y)\) leaves the referee’s system in the quantum state

\[\sigma_{a,b}^{x,y} \in \text{D}(\mathcal{R}),\]

then their winning and losing probabilities are given by

\[\left\langle V(a,b|x,y), \sigma_{a,b}^{x,y} \right\rangle \quad \text{and} \quad \left\langle \mathbb{I} - V(a,b|x,y), \sigma_{a,b}^{x,y} \right\rangle.\]

Strategies for extended nonlocal games

An extended nonlocal game \(G\) is defined by a pair \((\pi, V)\), where \(\pi\) is a probability distribution of the form

\[\pi : \Sigma_A \times \Sigma_B \rightarrow [0, 1]\]

on the Cartesian product of two alphabets \(\Sigma_A\) and \(\Sigma_B\), and \(V\) is a function of the form

\[V : \Gamma_A \times \Gamma_B \times \Sigma_A \times \Sigma_B \rightarrow \text{Pos}(\mathcal{R})\]

for \(\Sigma_A\) and \(\Sigma_B\) as above, \(\Gamma_A\) and \(\Gamma_B\) being alphabets, and \(\mathcal{R}\) refers to the referee’s space. Just as in the case for nonlocal games, we shall use the convention that

\[\Sigma = \Sigma_A \times \Sigma_B \quad \text{and} \quad \Gamma = \Gamma_A \times \Gamma_B\]

to denote the respective sets of questions asked to Alice and Bob and the sets of answers sent from Alice and Bob to the referee.

When analyzing a strategy for Alice and Bob, it may be convenient to define a function

\[K : \Gamma_A \times \Gamma_B \times \Sigma_A \times \Sigma_B \rightarrow \text{Pos}(\mathcal{R}).\]

We can represent Alice and Bob’s winning probability for an extended nonlocal game as

\[\sum_{(x,y) \in \Sigma} \pi(x,y) \sum_{(a,b) \in \Gamma} \left\langle V(a,b|x,y), K(a,b|x,y) \right\rangle.\]

Standard quantum strategies for extended nonlocal games

A standard quantum strategy for an extended nonlocal game consists of finite-dimensional complex Euclidean spaces \(\mathcal{U}\) for Alice and \(\mathcal{V}\) for Bob, a quantum state \(\sigma \in \text{D}(\mathcal{U} \otimes \mathcal{R} \otimes \mathcal{V})\), and two collections of measurements

\[\{ A_a^x : a \in \Gamma_A \} \subset \text{Pos}(\mathcal{U}) \quad \text{and} \quad \{ B_b^y : b \in \Gamma_B \} \subset \text{Pos}(\mathcal{V}),\]

for each \(x \in \Sigma_A\) and \(y \in \Sigma_B\) respectively. As usual, the measurement operators satisfy the constraint that

\[\sum_{a \in \Gamma_A} A_a^x = \mathbb{I}_{\mathcal{U}} \quad \text{and} \quad \sum_{b \in \Gamma_B} B_b^y = \mathbb{I}_{\mathcal{V}},\]

for each \(x \in \Sigma_A\) and \(y \in \Sigma_B\).

When the game is played, Alice and Bob present the referee with a quantum system so that the three parties share the state \(\sigma \in \text{D}(\mathcal{U} \otimes \mathcal{R} \otimes \mathcal{V})\). The referee selects questions \((x,y) \in \Sigma\) according to the distribution \(\pi\) that is known to all participants in the game.

The referee then sends \(x\) to Alice and \(y\) to Bob. At this point, Alice and Bob make measurements on their respective portions of the state \(\sigma\) using their measurement operators to yield an outcome to send back to the referee. Specifically, Alice measures her portion of the state \(\sigma\) with respect to her set of measurement operators \(\{A_a^x : a \in \Gamma_A\}\), and sends the result \(a \in \Gamma_A\) of this measurement to the referee. Likewise, Bob measures his portion of the state \(\sigma\) with respect to his measurement operators \(\{B_b^y : b \in \Gamma_B\}\) to yield the outcome \(b \in \Gamma_B\), that is then sent back to the referee.

At the end of the protocol, the referee measures its quantum system with respect to the measurement \(\{V(a,b|x,y), \mathbb{I}-V(a,b|x,y)\}\).

The winning probability for such a strategy in this game \(G = (\pi,V)\) is given by

\[\sum_{(x,y) \in \Sigma} \pi(x,y) \sum_{(a,b) \in \Gamma} \left \langle A_a^x \otimes V(a,b|x,y) \otimes B_b^y, \sigma \right \rangle.\]

For a given extended nonlocal game \(G = (\pi,V)\), we write \(\omega^*(G)\) to denote the standard quantum value of \(G\), which is the supremum value of Alice and Bob’s winning probability over all standard quantum strategies for \(G\).

Unentangled strategies for extended nonlocal games

An unentangled strategy for an extended nonlocal game is simply a standard quantum strategy for which the state \(\sigma \in \text{D}(\mathcal{U} \otimes \mathcal{R} \otimes \mathcal{V})\) initially prepared by Alice and Bob is fully separable.

Any unentangled strategy is equivalent to a strategy where Alice and Bob store only classical information after the referee’s quantum system has been provided to it.

For a given extended nonlocal game \(G = (\pi, V)\) we write \(\omega(G)\) to denote the unentangled value of \(G\), which is the supremum value for Alice and Bob’s winning probability in \(G\) over all unentangled strategies. The unentangled value of any extended nonlocal game, \(G\), may be written as

\[\omega(G) = \max_{f, g} \lVert \sum_{(x,y) \in \Sigma} \pi(x,y) V(f(x), g(y)|x, y) \rVert\]

where the maximum is over all functions \(f : \Sigma_A \rightarrow \Gamma_A\) and \(g : \Sigma_B \rightarrow \Gamma_B\).

Non-signaling strategies for extended nonlocal games

A non-signaling strategy for an extended nonlocal game consists of a function

\[K : \Gamma_A \times \Gamma_B \times \Sigma_A \times \Sigma_B \rightarrow \text{Pos}(\mathcal{R})\]

such that

\[\sum_{a \in \Gamma_A} K(a,b|x,y) = \rho_b^y \quad \text{and} \quad \sum_{b \in \Gamma_B} K(a,b|x,y) = \sigma_a^x,\]

for all \(x \in \Sigma_A\) and \(y \in \Sigma_B\) where \(\{\rho_b^y : y \in \Sigma_B, b \in \Gamma_B\}\) and \(\{\sigma_a^x: x \in \Sigma_A, a \in \Gamma_A\}\) are collections of operators satisfying

\[\sum_{a \in \Gamma_A} \sigma_a^x = \tau = \sum_{b \in \Gamma_B} \rho_b^y,\]

for every choice of \(x \in \Sigma_A\) and \(y \in \Sigma_B\) and where \(\tau \in \text{D}(\mathcal{R})\) is a density operator.

For any extended nonlocal game, \(G = (\pi, V)\), the winning probability for a non-signaling strategy is given by

\[\sum_{(x,y) \in \Sigma} \pi(x,y) \sum_{(a,b) \in \Gamma} \left\langle V(a,b|x,y) K(a,b|x,y) \right\rangle.\]

We denote the non-signaling value of \(G\) as \(\omega_{ns}(G)\) which is the supremum value of the winning probability of \(G\) taken over all non-signaling strategies for Alice and Bob.

Relationships between different strategies and values

For an extended nonlocal game, \(G\), the values have the following relationship:

Note

\[0 \leq \omega(G) \leq \omega^*(G) \leq \omega_{ns}(G) \leq 1.\]

Example: The BB84 extended nonlocal game

The BB84 extended nonlocal game is defined as follows. Let \(\Sigma_A = \Sigma_B = \Gamma_A = \Gamma_B = \{0,1\}\), define

\[\begin{split}\begin{equation} \begin{aligned} V(0,0|0,0) = \begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix}, &\quad V(1,1|0,0) = \begin{pmatrix} 0 & 0 \\ 0 & 1 \end{pmatrix}, \\ V(0,0|1,1) = \frac{1}{2}\begin{pmatrix} 1 & 1 \\ 1 & 1 \end{pmatrix}, &\quad V(1,1|1,1) = \frac{1}{2}\begin{pmatrix} 1 & -1 \\ -1 & 1 \end{pmatrix}, \end{aligned} \end{equation}\end{split}\]

define

\[\begin{split}V(a,b|x,y) = \begin{pmatrix} 0 & 0 \\ 0 & 0 \end{pmatrix}\end{split}\]

for all \(a \not= b\) or \(x \not= y\), define \(\pi(0,0) = \pi(1,1) = 1/2\), and define \(\pi(x,y) = 0\) if \(x \not=y\).

We can encode the BB84 game, \(G_{BB84} = (\pi, V)\), in numpy arrays where prob_mat corresponds to the probability distribution \(\pi\) and where pred_mat corresponds to the operator \(V\).

# Define the BB84 extended nonlocal game.
import numpy as np
from toqito.states import basis

# The basis: {|0>, |1>}:
e_0, e_1 = basis(2, 0), basis(2, 1)

# The basis: {|+>, |->}:
e_p = (e_0 + e_1) / np.sqrt(2)
e_m = (e_0 - e_1) / np.sqrt(2)

# The dimension of referee's measurement operators:
dim = 2
# The number of outputs for Alice and Bob:
a_out, b_out = 2, 2
# The number of inputs for Alice and Bob:
a_in, b_in = 2, 2

# Define the predicate matrix V(a,b|x,y) \in Pos(R)
bb84_pred_mat = np.zeros([dim, dim, a_out, b_out, a_in, b_in])

# V(0,0|0,0) = |0><0|
bb84_pred_mat[:, :, 0, 0, 0, 0] = e_0 @ e_0.conj().T
# V(1,1|0,0) = |1><1|
bb84_pred_mat[:, :, 1, 1, 0, 0] = e_1 @ e_1.conj().T
# V(0,0|1,1) = |+><+|
bb84_pred_mat[:, :, 0, 0, 1, 1] = e_p @ e_p.conj().T
# V(1,1|1,1) = |-><-|
bb84_pred_mat[:, :, 1, 1, 1, 1] = e_m @ e_m.conj().T

# The probability matrix encode \pi(0,0) = \pi(1,1) = 1/2
bb84_prob_mat = 1 / 2 * np.identity(2)

The unentangled value of the BB84 extended nonlocal game

It was shown in [3] and [4] that

\[\omega(G_{BB84}) = \cos^2(\pi/8).\]

This can be verified in |toqito⟩ as follows.

# Calculate the unentangled value of the BB84 extended nonlocal game.
from toqito.nonlocal_games.extended_nonlocal_game import ExtendedNonlocalGame
import numpy as np

# Define an ExtendedNonlocalGame object based on the BB84 game.
bb84 = ExtendedNonlocalGame(bb84_prob_mat, bb84_pred_mat)

# The unentangled value is cos(pi/8)**2 \approx 0.85356
print("The unentangled value is ", np.around(bb84.unentangled_value(), decimals=2))

The unentangled value is  0.85

The BB84 game also exhibits strong parallel repetition. We can specify how many parallel repetitions for |toqito⟩ to run. The example below provides an example of two parallel repetitions for the BB84 game.

# The unentangled value of BB84 under parallel repetition.
from toqito.nonlocal_games.extended_nonlocal_game import ExtendedNonlocalGame
import numpy as np

# Define the bb84 game for two parallel repetitions.
bb84_2_reps = ExtendedNonlocalGame(bb84_prob_mat, bb84_pred_mat, 2)

# The unentangled value for two parallel repetitions is cos(pi/8)**4 \approx 0.72855
print("The unentangled value for two parallel repetitions is ", np.around(bb84_2_reps.unentangled_value(), decimals=2))

The unentangled value for two parallel repetitions is  0.73

It was shown in [4] that the BB84 game possesses the property of strong parallel repetition. That is,

\[\omega(G_{BB84}^r) = \omega(G_{BB84})^r\]

for any integer \(r\).

The standard quantum value of the BB84 extended nonlocal game

We can calculate lower bounds on the standard quantum value of the BB84 game using |toqito⟩ as well.

# Calculate lower bounds on the standard quantum value of the BB84 extended nonlocal game.
from toqito.nonlocal_games.extended_nonlocal_game import ExtendedNonlocalGame
import numpy as np

# Define an ExtendedNonlocalGame object based on the BB84 game.
bb84_lb = ExtendedNonlocalGame(bb84_prob_mat, bb84_pred_mat)

# The standard quantum value is cos(pi/8)**2 \approx 0.85356
print("The standard quantum value is ", np.around(bb84_lb.quantum_value_lower_bound(), decimals=2))

The standard quantum value is  0.85

From [4], it is known that \(\omega(G_{BB84}) = \omega^*(G_{BB84})\), however, if we did not know this beforehand, we could attempt to calculate upper bounds on the standard quantum value.

There are a few methods to do this, but one easy way is to simply calculate the non-signaling value of the game as this provides a natural upper bound on the standard quantum value. Typically, this bound is not tight and usually not all that useful in providing tight upper bounds on the standard quantum value, however, in this case, it will prove to be useful.

The non-signaling value of the BB84 extended nonlocal game

Using |toqito⟩, we can see that \(\omega_{ns}(G) = \cos^2(\pi/8)\).

# Calculate the non-signaling value of the BB84 extended nonlocal game.
from toqito.nonlocal_games.extended_nonlocal_game import ExtendedNonlocalGame
import numpy as np

# Define an ExtendedNonlocalGame object based on the BB84 game.
bb84 = ExtendedNonlocalGame(bb84_prob_mat, bb84_pred_mat)

# The non-signaling value is cos(pi/8)**2 \approx 0.85356
print("The non-signaling value is ", np.around(bb84.nonsignaling_value(), decimals=2))

The non-signaling value is  0.85

So we have the relationship that

\[\omega(G_{BB84}) = \omega^*(G_{BB84}) = \omega_{ns}(G_{BB84}) = \cos^2(\pi/8).\]

It turns out that strong parallel repetition does not hold in the non-signaling scenario for the BB84 game. This was shown in [1], and we can observe this by the following snippet.

# The non-signaling value of BB84 under parallel repetition.
from toqito.nonlocal_games.extended_nonlocal_game import ExtendedNonlocalGame
import numpy as np

# Define the bb84 game for two parallel repetitions.
bb84_2_reps = ExtendedNonlocalGame(bb84_prob_mat, bb84_pred_mat, 2)

# The non-signaling value for two parallel repetitions is cos(pi/8)**4 \approx 0.73825
print(
    "The non-signaling value for two parallel repetitions is ", np.around(bb84_2_reps.nonsignaling_value(), decimals=2)
)

The non-signaling value for two parallel repetitions is  0.74

Note that \(0.73825 \geq \cos(\pi/8)^4 \approx 0.72855\) and therefore we have that

\[\omega_{ns}(G^r_{BB84}) \not= \omega_{ns}(G_{BB84})^r\]

for \(r = 2\).

Example: The CHSH extended nonlocal game

Let us now define another extended nonlocal game, \(G_{CHSH}\).

Let \(\Sigma_A = \Sigma_B = \Gamma_A = \Gamma_B = \{0,1\}\), define a collection of measurements \(\{V(a,b|x,y) : a \in \Gamma_A, b \in \Gamma_B, x \in \Sigma_A, y \in \Sigma_B\} \subset \text{Pos}(\mathcal{R})\) such that

\[\begin{split}\begin{equation} \begin{aligned} V(0,0|0,0) = V(0,0|0,1) = V(0,0|1,0) = \begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix}, \\ V(1,1|0,0) = V(1,1|0,1) = V(1,1|1,0) = \begin{pmatrix} 0 & 0 \\ 0 & 1 \end{pmatrix}, \\ V(0,1|1,1) = \frac{1}{2}\begin{pmatrix} 1 & 1 \\ 1 & 1 \end{pmatrix}, \\ V(1,0|1,1) = \frac{1}{2} \begin{pmatrix} 1 & -1 \\ -1 & 1 \end{pmatrix}, \end{aligned} \end{equation}\end{split}\]

define

\[\begin{split}V(a,b|x,y) = \begin{pmatrix} 0 & 0 \\ 0 & 0 \end{pmatrix}\end{split}\]

for all \(a \oplus b \not= x \land y\), and define \(\pi(0,0) = \pi(0,1) = \pi(1,0) = \pi(1,1) = 1/4\).

In the event that \(a \oplus b \not= x \land y\), the referee’s measurement corresponds to the zero matrix. If instead it happens that \(a \oplus b = x \land y\), the referee then proceeds to measure with respect to one of the measurement operators. This winning condition is reminiscent of the standard CHSH nonlocal game.

We can encode \(G_{CHSH}\) in a similar way using numpy arrays as we did for \(G_{BB84}\).

# Define the CHSH extended nonlocal game.
import numpy as np

# The dimension of referee's measurement operators:
dim = 2
# The number of outputs for Alice and Bob:
a_out, b_out = 2, 2
# The number of inputs for Alice and Bob:
a_in, b_in = 2, 2

# Define the predicate matrix V(a,b|x,y) \in Pos(R)
chsh_pred_mat = np.zeros([dim, dim, a_out, b_out, a_in, b_in])

# V(0,0|0,0) = V(0,0|0,1) = V(0,0|1,0).
chsh_pred_mat[:, :, 0, 0, 0, 0] = np.array([[1, 0], [0, 0]])
chsh_pred_mat[:, :, 0, 0, 0, 1] = np.array([[1, 0], [0, 0]])
chsh_pred_mat[:, :, 0, 0, 1, 0] = np.array([[1, 0], [0, 0]])

# V(1,1|0,0) = V(1,1|0,1) = V(1,1|1,0).
chsh_pred_mat[:, :, 1, 1, 0, 0] = np.array([[0, 0], [0, 1]])
chsh_pred_mat[:, :, 1, 1, 0, 1] = np.array([[0, 0], [0, 1]])
chsh_pred_mat[:, :, 1, 1, 1, 0] = np.array([[0, 0], [0, 1]])

# V(0,1|1,1)
chsh_pred_mat[:, :, 0, 1, 1, 1] = 1 / 2 * np.array([[1, 1], [1, 1]])

# V(1,0|1,1)
chsh_pred_mat[:, :, 1, 0, 1, 1] = 1 / 2 * np.array([[1, -1], [-1, 1]])

# The probability matrix encode \pi(0,0) = \pi(0,1) = \pi(1,0) = \pi(1,1) = 1/4.
chsh_prob_mat = np.array([[1 / 4, 1 / 4], [1 / 4, 1 / 4]])

Example: The unentangled value of the CHSH extended nonlocal game

Similar to what we did for the BB84 extended nonlocal game, we can also compute the unentangled value of \(G_{CHSH}\).

# Calculate the unentangled value of the CHSH extended nonlocal game
from toqito.nonlocal_games.extended_nonlocal_game import ExtendedNonlocalGame
import numpy as np

# Define an ExtendedNonlocalGame object based on the CHSH game.
chsh = ExtendedNonlocalGame(chsh_prob_mat, chsh_pred_mat)

# The unentangled value is 3/4 = 0.75
print("The unentangled value is ", np.around(chsh.unentangled_value(), decimals=2))

The unentangled value is  0.75

We can also run multiple repetitions of \(G_{CHSH}\).

# The unentangled value of CHSH under parallel repetition.
from toqito.nonlocal_games.extended_nonlocal_game import ExtendedNonlocalGame
import numpy as np

# Define the CHSH game for two parallel repetitions.
chsh_2_reps = ExtendedNonlocalGame(chsh_prob_mat, chsh_pred_mat, 2)

# The unentangled value for two parallel repetitions is (3/4)**2 \approx 0.5625
print("The unentangled value for two parallel repetitions is ", np.around(chsh_2_reps.unentangled_value(), decimals=2))

The unentangled value for two parallel repetitions is  0.56

Note that strong parallel repetition holds as

\[\omega(G_{CHSH})^2 = \omega(G_{CHSH}^2) = \left(\frac{3}{4}\right)^2.\]

Example: The non-signaling value of the CHSH extended nonlocal game

To obtain an upper bound for \(G_{CHSH}\), we can calculate the non-signaling value.

# Calculate the non-signaling value of the CHSH extended nonlocal game.
from toqito.nonlocal_games.extended_nonlocal_game import ExtendedNonlocalGame
import numpy as np

# Define an ExtendedNonlocalGame object based on the CHSH game.
chsh = ExtendedNonlocalGame(chsh_prob_mat, chsh_pred_mat)

# The non-signaling value is 3/4 = 0.75
print("The non-signaling value is ", np.around(chsh.nonsignaling_value(), decimals=2))

The non-signaling value is  0.75

As we know that \(\omega(G_{CHSH}) = \omega_{ns}(G_{CHSH}) = 3/4\) and that

\[\omega(G) \leq \omega^*(G) \leq \omega_{ns}(G)\]

for any extended nonlocal game, \(G\), we may also conclude that \(\omega^*(G) = 3/4\).

Example: An extended nonlocal game with quantum advantage

So far, we have only seen examples of extended nonlocal games where the standard quantum and unentangled values are equal. Here we’ll see an example of an extended nonlocal game where the standard quantum value is strictly higher than the unentangled value.

Example: A monogamy-of-entanglement game with mutually unbiased bases

Let \(\zeta = \exp(\frac{2 \pi i}{3})\) and consider the following four mutually unbiased bases:

\[\begin{split}\begin{equation}\label{eq:MUB43} \begin{aligned} \mathcal{B}_0 &= \left\{ e_0,\: e_1,\: e_2 \right\}, \\ \mathcal{B}_1 &= \left\{ \frac{e_0 + e_1 + e_2}{\sqrt{3}},\: \frac{e_0 + \zeta^2 e_1 + \zeta e_2}{\sqrt{3}},\: \frac{e_0 + \zeta e_1 + \zeta^2 e_2}{\sqrt{3}} \right\}, \\ \mathcal{B}_2 &= \left\{ \frac{e_0 + e_1 + \zeta e_2}{\sqrt{3}},\: \frac{e_0 + \zeta^2 e_1 + \zeta^2 e_2}{\sqrt{3}},\: \frac{e_0 + \zeta e_1 + e_2}{\sqrt{3}} \right\}, \\ \mathcal{B}_3 &= \left\{ \frac{e_0 + e_1 + \zeta^2 e_2}{\sqrt{3}},\: \frac{e_0 + \zeta^2 e_1 + e_2}{\sqrt{3}},\: \frac{e_0 + \zeta e_1 + \zeta e_2}{\sqrt{3}} \right\}. \end{aligned} \end{equation}\end{split}\]

Define an extended nonlocal game \(G_{MUB} = (\pi,R)\) so that

\[\pi(0) = \pi(1) = \pi(2) = \pi(3) = \frac{1}{4}\]

and \(R\) is such that

\[{ R(0|x), R(1|x), R(2|x) }\]

represents a measurement with respect to the basis \(\mathcal{B}_x\), for each \(x \in \{0,1,2,3\}\).

Taking the description of \(G_{MUB}\), we can encode this as follows.

# Define the monogamy-of-entanglement game defined by MUBs.
prob_mat = 1 / 4 * np.identity(4)

dim = 3
e_0, e_1, e_2 = basis(dim, 0), basis(dim, 1), basis(dim, 2)

eta = np.exp((2 * np.pi * 1j) / dim)
mub_0 = [e_0, e_1, e_2]
mub_1 = [
    (e_0 + e_1 + e_2) / np.sqrt(3),
    (e_0 + eta**2 * e_1 + eta * e_2) / np.sqrt(3),
    (e_0 + eta * e_1 + eta**2 * e_2) / np.sqrt(3),
]
mub_2 = [
    (e_0 + e_1 + eta * e_2) / np.sqrt(3),
    (e_0 + eta**2 * e_1 + eta**2 * e_2) / np.sqrt(3),
    (e_0 + eta * e_1 + e_2) / np.sqrt(3),
]
mub_3 = [
    (e_0 + e_1 + eta**2 * e_2) / np.sqrt(3),
    (e_0 + eta**2 * e_1 + e_2) / np.sqrt(3),
    (e_0 + eta * e_1 + eta * e_2) / np.sqrt(3),
]

# List of measurements defined from mutually unbiased basis.
mubs = [mub_0, mub_1, mub_2, mub_3]

num_in = 4
num_out = 3
pred_mat = np.zeros([dim, dim, num_out, num_out, num_in, num_in], dtype=complex)

pred_mat[:, :, 0, 0, 0, 0] = mubs[0][0] @ mubs[0][0].conj().T
pred_mat[:, :, 1, 1, 0, 0] = mubs[0][1] @ mubs[0][1].conj().T
pred_mat[:, :, 2, 2, 0, 0] = mubs[0][2] @ mubs[0][2].conj().T

pred_mat[:, :, 0, 0, 1, 1] = mubs[1][0] @ mubs[1][0].conj().T
pred_mat[:, :, 1, 1, 1, 1] = mubs[1][1] @ mubs[1][1].conj().T
pred_mat[:, :, 2, 2, 1, 1] = mubs[1][2] @ mubs[1][2].conj().T

pred_mat[:, :, 0, 0, 2, 2] = mubs[2][0] @ mubs[2][0].conj().T
pred_mat[:, :, 1, 1, 2, 2] = mubs[2][1] @ mubs[2][1].conj().T
pred_mat[:, :, 2, 2, 2, 2] = mubs[2][2] @ mubs[2][2].conj().T

pred_mat[:, :, 0, 0, 3, 3] = mubs[3][0] @ mubs[3][0].conj().T
pred_mat[:, :, 1, 1, 3, 3] = mubs[3][1] @ mubs[3][1].conj().T
pred_mat[:, :, 2, 2, 3, 3] = mubs[3][2] @ mubs[3][2].conj().T

Now that we have encoded \(G_{MUB}\), we can calculate the unentangled value.

import numpy as np

g_mub = ExtendedNonlocalGame(prob_mat, pred_mat)
unent_val = g_mub.unentangled_value()
print("The unentangled value is ", np.around(unent_val, decimals=2))

The unentangled value is  0.65

That is, we have that

\[\omega(G_{MUB}) = \frac{3 + \sqrt{5}}{8} \approx 0.65409.\]

However, if we attempt to run a lower bound on the standard quantum value, we obtain.

import numpy as np

g_mub = ExtendedNonlocalGame(prob_mat, pred_mat)
q_val = g_mub.quantum_value_lower_bound()
print("The standard quantum value lower bound is ", np.around(q_val, decimals=2))

The standard quantum value lower bound is  0.65

Note that as we are calculating a lower bound, it is possible that a value this high will not be obtained, or in other words, the algorithm can get stuck in a local maximum that prevents it from finding the global maximum.

It is uncertain what the optimal standard quantum strategy is for this game, but the value of such a strategy is bounded as follows

\[2/3 \geq \omega^*(G) \geq 0.6609.\]

For further information on the \(G_{MUB}\) game, consult [1].

References

[1] (1,2,3,4)

Vincent Russo. Extended nonlocal games. 2017. arXiv:1704.07375.

[2]

D. Cavalcanti, P. Skrzypczyk, G. H. Aguilar, R. V. Nery, P.H. Souto Ribeiro, and S. P. Walborn. Detection of entanglement in asymmetric quantum networks and multipartite quantum steering. Nature Communications, August 2015. URL: http://dx.doi.org/10.1038/ncomms8941, doi:10.1038/ncomms8941.

[3] (1,2)

Marco Tomamichel, Serge Fehr, Jędrzej Kaniewski, and Stephanie Wehner. A monogamy-of-entanglement game with applications to device-independent quantum cryptography. New Journal of Physics, 15(10):103002, October 2013. URL: http://dx.doi.org/10.1088/1367-2630/15/10/103002, doi:10.1088/1367-2630/15/10/103002.

[4] (1,2,3,4)

Nathaniel Johnston, Rajat Mittal, Vincent Russo, and John Watrous. Extended non-local games and monogamy-of-entanglement games. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 472(2189):20160003, May 2016. URL: https://arxiv.org/abs/1510.02083.