"""Two-player nonlocal game."""
from collections import defaultdict
import cvxpy
import numpy as np
from toqito.helper import update_odometer
from toqito.matrix_ops import tensor
from toqito.matrix_ops.tensor_unravel import tensor_unravel
from toqito.nonlocal_games.binary_constraint_system_game import check_perfect_commuting_strategy
from toqito.rand import random_povm
from toqito.state_opt.npa_hierarchy import npa_constraints
[docs]
class NonlocalGame:
r"""Create two-player nonlocal game object.
*Nonlocal games* are a mathematical framework that abstractly models a
physical system. This game is played between two players, Alice and Bob, who
are not allowed to communicate with each other once the game has started and
who play cooperative against an adversary referred to as the referee.
The nonlocal game framework was originally introduced in [@Cleve_2010_Consequences].
A tutorial is available in the documentation. For more info, see
[Nonlocal Games](../../../generated/gallery/nonlocal_games/index.md).
"""
def __init__(self, prob_mat: np.ndarray, pred_mat: np.ndarray, reps: int = 1) -> None:
"""Construct nonlocal game object.
Args:
prob_mat: A matrix whose (x, y)-entry gives the probability
that the referee will give Alice the value `x` and Bob
the value `y`.
pred_mat: A four-dimensional matrix whose (a,b,x,y)-entry gives
the outcome for answers "a" and "b" given questions
"x" and "y".
reps: Number of parallel repetitions to perform. Default is 1.
"""
if reps == 1:
self.prob_mat = prob_mat
self.pred_mat = pred_mat
self.reps = reps
else:
num_alice_out, num_bob_out, num_alice_in, num_bob_in = pred_mat.shape
self.prob_mat = tensor(prob_mat, reps)
pred_mat2 = np.zeros(
(
num_alice_out**reps,
num_bob_out**reps,
num_alice_in**reps,
num_bob_in**reps,
)
)
i_ind = np.zeros(reps, dtype=int)
j_ind = np.zeros(reps, dtype=int)
for i in range(num_alice_in**reps):
for j in range(num_bob_in**reps):
to_tensor = np.empty([reps, num_alice_out, num_bob_out])
for k in range(reps - 1, -1, -1):
to_tensor[k] = pred_mat[:, :, i_ind[k], j_ind[k]]
pred_mat2[:, :, i, j] = tensor(to_tensor)
j_ind = update_odometer(j_ind, num_bob_in * np.ones(reps))
i_ind = update_odometer(i_ind, num_alice_in * np.ones(reps))
self.pred_mat = pred_mat2
self.reps = reps
# _raw_constraints will store the original 1D BCS constraints (if provided) for later analysis.
self._raw_constraints = None
[docs]
@classmethod
def from_bcs_game(cls, constraints: list[np.ndarray], reps: int = 1) -> "NonlocalGame":
r"""Convert constraints that specify a binary constraint system game to a nonlocal game.
Binary constraint system games (BCS) games were originally defined in [@Cleve_2014_Characterization].
Args:
constraints: List of binary constraints that define the game.
reps: Number of parallel repetitions to perform. Default is 1.
Returns:
A NonlocalGame object arising from the variables and constraints that define the game.
"""
if (num_constraints := len(constraints)) == 0:
raise ValueError("At least 1 constraint is required")
num_variables = constraints[0].ndim
# Retrieve dependent variables for each constraint.
dependent_variables = np.zeros((num_constraints, num_variables))
for j in range(num_constraints):
for i in range(num_variables):
# Identifying independent variables based on equality check.
dependent_variables[j, i] = np.diff(constraints[j], axis=i).any()
# Compute the probability matrix.
prob_mat = np.zeros((num_constraints, num_variables))
for j in range(num_constraints):
p_x = 1.0 / num_constraints
num_dependent_vars = dependent_variables[j].sum()
if num_dependent_vars == 0:
raise ValueError(f"Constraint {j} is degenerate (has no dependent variables).")
else:
p_y = dependent_variables[j] / num_dependent_vars
prob_mat[j] = p_x * p_y
# Compute the prediction matrix.
pred_mat = np.zeros((2**num_variables, 2, num_constraints, num_variables))
for x_ques in range(num_constraints):
for a_ans in range(pred_mat.shape[0]):
# Convert Alice's truth assignment to binary.
bin_a = np.array(list(map(int, np.binary_repr(a_ans, num_variables))))
# Convert truth assignment to a tuple for easy indexing.
truth_assignment = tuple(bin_a)
for y_ques in range(num_variables):
# Bob’s assignment is Alice’s truth assignment for the current variable.
b_ans = truth_assignment[y_ques]
# Check if this satisfies the constraint.
if constraints[x_ques][truth_assignment] == 1:
pred_mat[a_ans, b_ans, x_ques, y_ques] = 1
game = cls(prob_mat, pred_mat, reps)
game._raw_constraints = constraints
return game
[docs]
def is_bcs_perfect_commuting_strategy(self) -> bool:
r"""Determine if the BCS game admits a perfect commuting-operator strategy.
This method checks whether the binary constraint system game, from which the current
nonlocal game was constructed, has a perfect quantum strategy in the commuting-operator model.
It converts the raw BCS tensor constraints (if needed) into matrix form and evaluates
their satisfiability using a helper function.
Raises:
If no constraints are stored (i.e., if the game was not created from a BCS game).
Returns:
True if a perfect commuting-operator strategy exists; False otherwise.
"""
# If the stored constraints are tensor-form (i.e. not 1D), convert them to raw (1D) form.
if self._raw_constraints[0].ndim != 1:
converted = []
for tensor_constraint in self._raw_constraints:
converted.append(tensor_unravel(tensor_constraint))
raw_constraints = converted
else:
raw_constraints = self._raw_constraints
# Now, for each raw constraint (which should be a 1D array of length n+1),
# extract M (all entries except the last) and b (derived from the last entry).
M_list = [c[:-1] for c in raw_constraints]
b_list = [0 if c[-1] == 1 else 1 for c in raw_constraints]
M_array = np.array(M_list, dtype=int)
b_array = np.array(b_list, dtype=int)
return check_perfect_commuting_strategy(M_array, b_array)
[docs]
def classical_value(self) -> float:
r"""Compute the classical value of the nonlocal game using Numba acceleration."""
A_out, B_out, A_in, B_in = self.pred_mat.shape
pm = np.copy(self.pred_mat)
pm *= self.prob_mat[np.newaxis, np.newaxis, :A_in, :B_in]
# Align dimensions for Bob's strategies
if A_out**A_in < B_out**B_in:
pm = pm.transpose((1, 0, 3, 2))
A_out, B_out, A_in, B_in = pm.shape
# Reorder axes to (A_out, A_in, B_out, B_in)
pm = pm.transpose((0, 2, 1, 3))
# Build power array for decoding Bob's outputs
pow_arr = np.array([B_out ** (B_in - 1 - y) for y in range(B_in)], dtype=np.int64)
# === Begin fast classical value logic ===
p_win = 0.0
total = B_out**B_in
for i in range(total):
best_sum = 0.0
for x in range(A_in):
best_for_x = 0.0
for a in range(A_out):
acc = 0.0
for y in range(B_in):
b_q = (i // pow_arr[y]) % B_out
acc += pm[a, x, b_q, y]
best_for_x = max(best_for_x, acc)
best_sum += best_for_x
p_win = max(p_win, best_sum)
return p_win
[docs]
def quantum_value_lower_bound(
self,
dim: int = 2,
iters: int = 5,
tol: float = 10e-6,
):
r"""Compute a lower bound on the quantum value of a nonlocal game [@Liang_2007_Bounds].
Calculates a lower bound on the maximum value that the specified
nonlocal game can take on in quantum mechanical settings where Alice and
Bob each have access to `dim`-dimensional quantum system.
This function works by starting with a randomly-generated POVM for Bob,
and then optimizing Alice's POVM and the shared entangled state. Then
Alice's POVM and the entangled state are fixed, and Bob's POVM is
optimized. And so on, back and forth between Alice and Bob until
convergence is reached.
Note that the algorithm is not guaranteed to obtain the optimal local
bound and can get stuck in local minimum values. The alleviate this, the
`iter` parameter allows one to run the algorithm some pre-specified
number of times and keep the highest value obtained.
The algorithm is based on the alternating projections algorithm as it
can be applied to Bell inequalities as shown in [@Liang_2007_Bounds].
The alternating projection algorithm has also been referred to as the
"see-saw" algorithm as it goes back and forth between the following two
semidefinite programs:
\[
\begin{equation}
\begin{aligned}
\textbf{SDP-1:} \quad & \\
\text{maximize:} \quad & \sum_{(x,y \in \Sigma)} \pi(x,y)
\sum_{(a,b) \in \Gamma}
V(a,b|x,y)
\langle B_b^y, A_a^x \rangle \\
\text{subject to:} \quad & \sum_{a \in \Gamma_{\mathsf{A}}}=
\tau, \qquad \qquad
\forall x \in \Sigma_{\mathsf{A}}, \\
\quad & A_a^x \in \text{Pos}(\mathcal{A}),
\qquad
\forall x \in \Sigma_{\mathsf{A}}, \
\forall a \in \Gamma_{\mathsf{A}}, \\
& \tau \in \text{D}(\mathcal{A}).
\end{aligned}
\end{equation}
\]
\[
\begin{equation}
\begin{aligned}
\textbf{SDP-2:} \quad & \\
\text{maximize:} \quad & \sum_{(x,y \in \Sigma)} \pi(x,y)
\sum_{(a,b) \in \Gamma} V(a,b|x,y)
\langle B_b^y, A_a^x \rangle \\
\text{subject to:} \quad & \sum_{b \in \Gamma_{\mathsf{B}}}=
\mathbb{I}, \qquad \qquad
\forall y \in \Sigma_{\mathsf{B}}, \\
\quad & B_b^y \in \text{Pos}(\mathcal{B}),
\qquad \forall y \in \Sigma_{\mathsf{B}}, \
\forall b \in \Gamma_{\mathsf{B}}.
\end{aligned}
\end{equation}
\]
Examples:
The CHSH game
The CHSH game is a two-player nonlocal game with the following
probability distribution and question and answer sets.
\[
\begin{equation}
\begin{aligned}
\pi(x,y) = \frac{1}{4}, \qquad (x,y) \in \Sigma_A \times \Sigma_B,
\qquad \text{and} \qquad (a, b) \in \Gamma_A \times \Gamma_B,
\end{aligned}
\end{equation}
\]
where
\[
\begin{equation}
\Sigma_A = \{0, 1\}, \quad \Sigma_B = \{0, 1\}, \quad \Gamma_A =
\{0,1\}, \quad \text{and} \quad \Gamma_B = \{0, 1\}.
\end{equation}
\]
Alice and Bob win the CHSH game if and only if the following equation is
satisfied.
\[
\begin{equation}
a \oplus b = x \land y.
\end{equation}
\]
Recall that \(\oplus\) refers to the XOR operation.
The optimal quantum value of CHSH is
\(\cos(\pi/8)^2 \approx 0.8536\) where the optimal classical value
is \(3/4\).
```python exec="1" source="above"
import numpy as np
from toqito.nonlocal_games.nonlocal_game import NonlocalGame
dim = 2
num_alice_inputs, num_alice_outputs = 2, 2
num_bob_inputs, num_bob_outputs = 2, 2
prob_mat = np.array([[1 / 4, 1 / 4], [1 / 4, 1 / 4]])
pred_mat = np.zeros((num_alice_outputs, num_bob_outputs, num_alice_inputs, num_bob_inputs))
for a_alice in range(num_alice_outputs):
for b_bob in range(num_bob_outputs):
for x_alice in range(num_alice_inputs):
for y_bob in range(num_bob_inputs):
if np.mod(a_alice + b_bob + x_alice * y_bob, dim) == 0:
pred_mat[a_alice, b_bob, x_alice, y_bob] = 1
chsh = NonlocalGame(prob_mat, pred_mat)
print(chsh.quantum_value_lower_bound())
```
Args:
dim: The dimension of the quantum system that Alice and Bob have
access to (default = 2).
iters: The number of times to run the alternating projection
algorithm.
tol: The tolerance before quitting out of the alternating
projection semidefinite program.
Returns:
The lower bound on the quantum value of a nonlocal game.
"""
# Get number of inputs and outputs.
_, num_outputs_bob, _, num_inputs_bob = self.pred_mat.shape
best_lower_bound = float("-inf")
for _ in range(iters):
# Generate a set of random POVMs for Bob. These measurements serve
# as a rough starting point for the alternating projection
# algorithm.
bob_tmp = random_povm(dim, num_inputs_bob, num_outputs_bob)
bob_povms = defaultdict(int)
for y_ques in range(num_inputs_bob):
for b_ans in range(num_outputs_bob):
bob_povms[y_ques, b_ans] = bob_tmp[:, :, y_ques, b_ans]
# Run the alternating projection algorithm between the two SDPs.
it_diff = 1
prev_win = -1
best = float("-inf")
while it_diff > tol:
# Optimize over Alice's measurement operators while fixing
# Bob's. If this is the first iteration, then the previously
# randomly generated operators in the outer loop are Bob's.
# Otherwise, Bob's operators come from running the next SDP.
alice_povms, lower_bound = self.__optimize_alice(dim, bob_povms)
bob_povms, lower_bound = self.__optimize_bob(dim, alice_povms)
it_diff = lower_bound - prev_win
prev_win = lower_bound
# As the SDPs keep alternating, check if the winning probability
# becomes any higher. If so, replace with new best.
best = max(best, lower_bound)
best_lower_bound = max(best, best_lower_bound)
return best_lower_bound
def __optimize_alice(self, dim, bob_povms) -> tuple[dict, float]:
"""Fix Bob's measurements and optimize over Alice's measurements."""
# Get number of inputs and outputs.
(
num_outputs_alice,
num_outputs_bob,
num_inputs_alice,
num_inputs_bob,
) = self.pred_mat.shape
# The cvxpy package does not support optimizing over 4-dimensional
# objects. To overcome this, we use a dictionary to index between the
# questions and answers, while the cvxpy variables held at this
# positions are `dim`-by-`dim` cvxpy variables.
alice_povms = defaultdict(cvxpy.Variable)
for x_ques in range(num_inputs_alice):
for a_ans in range(num_outputs_alice):
alice_povms[x_ques, a_ans] = cvxpy.Variable((dim, dim), hermitian=True)
tau = cvxpy.Variable((dim, dim), hermitian=True)
# .. math::
# \sum_{(x,y) \in \Sigma} \pi(x, y) V(a,b|x,y) \ip{B_b^y}{A_a^x}
win = 0
for x_ques in range(num_inputs_alice):
for y_ques in range(num_inputs_bob):
for a_ans in range(num_outputs_alice):
for b_ans in range(num_outputs_bob):
if isinstance(bob_povms[y_ques, b_ans], np.ndarray):
win += (
self.prob_mat[x_ques, y_ques]
* self.pred_mat[a_ans, b_ans, x_ques, y_ques]
* cvxpy.trace(bob_povms[y_ques, b_ans].conj().T @ alice_povms[x_ques, a_ans])
)
if isinstance(
bob_povms[y_ques, b_ans],
cvxpy.expressions.variable.Variable,
):
win += (
self.prob_mat[x_ques, y_ques]
* self.pred_mat[a_ans, b_ans, x_ques, y_ques]
* cvxpy.trace(bob_povms[y_ques, b_ans].value.conj().T @ alice_povms[x_ques, a_ans])
)
objective = cvxpy.Maximize(cvxpy.real(win))
constraints = []
# Sum over "a" for all "x" for Alice's measurements.
for x_ques in range(num_inputs_alice):
alice_sum_a = 0
for a_ans in range(num_outputs_alice):
alice_sum_a += alice_povms[x_ques, a_ans]
constraints.append(alice_povms[x_ques, a_ans] >> 0)
constraints.append(alice_sum_a == tau)
constraints.append(cvxpy.trace(tau) == 1)
constraints.append(tau >> 0)
problem = cvxpy.Problem(objective, constraints)
lower_bound = problem.solve()
return alice_povms, lower_bound
def __optimize_bob(self, dim, alice_povms) -> tuple[dict, float]:
"""Fix Alice's measurements and optimize over Bob's measurements."""
# Get number of inputs and outputs.
(
num_outputs_alice,
num_outputs_bob,
num_inputs_alice,
num_inputs_bob,
) = self.pred_mat.shape
# Now, optimize over Bob's measurement operators and fix Alice's
# operators as those are coming from the previous SDP.
bob_povms = defaultdict(cvxpy.Variable)
for y_ques in range(num_inputs_bob):
for b_ans in range(num_outputs_bob):
bob_povms[y_ques, b_ans] = cvxpy.Variable((dim, dim), hermitian=True)
win = 0
for x_ques in range(num_inputs_alice):
for y_ques in range(num_inputs_bob):
for a_ans in range(num_outputs_alice):
for b_ans in range(num_outputs_bob):
win += (
self.prob_mat[x_ques, y_ques]
* self.pred_mat[a_ans, b_ans, x_ques, y_ques]
* cvxpy.trace(bob_povms[y_ques, b_ans].H @ alice_povms[x_ques, a_ans].value)
)
objective = cvxpy.Maximize(cvxpy.real(win))
constraints = []
# Sum over "b" for all "y" for Bob's measurements.
for y_ques in range(num_inputs_bob):
bob_sum_b = 0
for b_ans in range(num_outputs_bob):
bob_sum_b += bob_povms[y_ques, b_ans]
constraints.append(bob_povms[y_ques, b_ans] >> 0)
constraints.append(bob_sum_b == np.identity(dim))
problem = cvxpy.Problem(objective, constraints)
lower_bound = problem.solve()
return bob_povms, lower_bound
[docs]
def nonsignaling_value(self) -> float:
"""Compute the non-signaling value of the nonlocal game.
Returns:
A value between [0, 1] representing the non-signaling value.
"""
alice_out, bob_out, alice_in, bob_in = self.pred_mat.shape
dim_x, dim_y = 2, 2
constraints = []
# Define K(a,b|x,y) variable.
k_var = defaultdict(cvxpy.Variable)
for a_out in range(alice_out):
for b_out in range(bob_out):
for x_in in range(alice_in):
for y_in in range(bob_in):
k_var[a_out, b_out, x_in, y_in] = cvxpy.Variable((dim_x, dim_y), hermitian=True)
constraints.append(k_var[a_out, b_out, x_in, y_in] >> 0)
# Define \sigma_a^x variable.
sigma = defaultdict(cvxpy.Variable)
for a_out in range(alice_out):
for x_in in range(alice_in):
sigma[a_out, x_in] = cvxpy.Variable((dim_x, dim_y), hermitian=True)
# Define \rho_b^y variable.
rho = defaultdict(cvxpy.Variable)
for b_out in range(bob_out):
for y_in in range(bob_in):
rho[b_out, y_in] = cvxpy.Variable((dim_x, dim_y), hermitian=True)
# Define \tau density operator variable.
tau = cvxpy.Variable((dim_x, dim_y), hermitian=True)
p_win = cvxpy.Constant(0)
for a_out in range(alice_out):
for b_out in range(bob_out):
for x_in in range(alice_in):
for y_in in range(bob_in):
p_win += self.prob_mat[x_in, y_in] * cvxpy.trace(
self.pred_mat[a_out, b_out, x_in, y_in].conj().T * k_var[a_out, b_out, x_in, y_in]
)
objective = cvxpy.Maximize(cvxpy.real(p_win))
# The following constraints enforce the so-called non-signaling
# constraints.
# Enforce that:
# \sum_{b \in \Gamma_B} K(a,b|x,y) = \sigma_a^x
for x_in in range(alice_in):
for y_in in range(bob_in):
for a_out in range(alice_out):
b_sum = 0
for b_out in range(bob_out):
b_sum += k_var[a_out, b_out, x_in, y_in]
constraints.append(b_sum == sigma[a_out, x_in])
# Enforce non-signaling constraints on Alice marginal:
# \sum_{a \in \Gamma_A} K(a,b|x,y) = \rho_b^y
for x_in in range(alice_in):
for y_in in range(bob_in):
for b_out in range(bob_out):
a_sum = 0
for a_out in range(alice_out):
a_sum += k_var[a_out, b_out, x_in, y_in]
constraints.append(a_sum == rho[b_out, y_in])
# Enforce non-signaling constraints on Bob marginal:
# \sum_{a \in \Gamma_A} \sigma_a^x = \tau
for x_in in range(alice_in):
sig_a_sum = 0
for a_out in range(alice_out):
sig_a_sum += sigma[a_out, x_in]
constraints.append(sig_a_sum == tau)
# Enforce that:
# \sum_{b \in \Gamma_B} \rho_b^y = \tau
for y_in in range(bob_in):
rho_b_sum = 0
for b_out in range(bob_out):
rho_b_sum += rho[b_out, y_in]
constraints.append(rho_b_sum == tau)
# Enforce that tau is a density operator.
constraints.append(cvxpy.trace(tau) == 1)
constraints.append(tau >> 0)
problem = cvxpy.Problem(objective, constraints)
ns_val = problem.solve()
return ns_val
[docs]
def commuting_measurement_value_upper_bound(self, k: int | str = 1) -> float:
r"""Compute an upper bound on the commuting measurement value of the nonlocal game.
This function calculates an upper bound on the commuting measurement value by
using k-levels of the NPA hierarchy [@Navascues_2008_AConvergent]. The NPA hierarchy is a uniform
family of semidefinite programs that converges to the commuting measurement value of
any nonlocal game.
You can determine the level of the hierarchy by a positive integer or a string
of a form like '1+ab+aab', which indicates that an intermediate level of the hierarchy
should be used, where this example uses all products of one measurement, all products of
one Alice and one Bob measurement, and all products of two Alice and one Bob measurements.
Args:
k: The level of the NPA hierarchy to use (default=1).
Returns:
The upper bound on the commuting strategy value of a nonlocal game.
"""
alice_out, bob_out, alice_in, bob_in = self.pred_mat.shape
mat = defaultdict(cvxpy.Variable)
for x_in in range(alice_in):
for y_in in range(bob_in):
mat[x_in, y_in] = cvxpy.Variable((alice_out, bob_out), name=f"M(a, b | {x_in}, {y_in})")
p_win = cvxpy.Constant(0)
for a_out in range(alice_out):
for b_out in range(bob_out):
for x_in in range(alice_in):
for y_in in range(bob_in):
p_win += (
self.prob_mat[x_in, y_in]
* self.pred_mat[a_out, b_out, x_in, y_in]
* mat[x_in, y_in][a_out, b_out]
)
npa = npa_constraints(mat, k)
objective = cvxpy.Maximize(p_win)
problem = cvxpy.Problem(objective, npa)
cs_val = problem.solve()
return cs_val