Skip to content

Solvers

Combinatorial optimization via SC-native Ising machines.

  • StochasticIsingGraph — Quantum-inspired Ising solver. Spins S_i in {-1, +1} (mapped to 0/1 for SC). Energy: E = -Sum(J_ij * S_i * S_j) - Sum(h_i * S_i). Finds minimum-energy configuration via simulated annealing with SC arithmetic.

Maps to SC hardware: spin products = AND gates, energy accumulation = popcount.

Python
import numpy as np
from sc_neurocore.solvers import StochasticIsingGraph

J = np.random.randn(10, 10)
J = (J + J.T) / 2
np.fill_diagonal(J, 0)
solver = StochasticIsingGraph(num_spins=10, J=J)
solution = solver.solve(n_steps=1000)

sc_neurocore.solvers.ising

StochasticIsingGraph dataclass

Quantum-Inspired Ising Machine Solver.

Spins S_i in {-1, 1} (mapped to 0, 1 for SC). Energy E = -Sum(J_ij * S_i * S_j) - Sum(h_i * S_i). Goal: Find configuration that minimizes E.

Source code in src/sc_neurocore/solvers/ising.py
Python
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
@dataclass
class StochasticIsingGraph:
    """
    Quantum-Inspired Ising Machine Solver.

    Spins S_i in {-1, 1} (mapped to 0, 1 for SC).
    Energy E = -Sum(J_ij * S_i * S_j) - Sum(h_i * S_i).
    Goal: Find configuration that minimizes E.
    """

    num_spins: int
    J: np.ndarray[Any, Any]  # Coupling matrix (symmetric, zero diagonal)
    h: np.ndarray[Any, Any]  # Bias vector
    temperature: float = 1.0
    anneal_rate: float = 0.99

    def __post_init__(self) -> None:
        # Initialize spins randomly
        self.spins = np.random.randint(0, 2, self.num_spins).astype(np.int8)
        # Convert 0/1 to -1/1 for physics calc
        self.bipolar_spins = 2 * self.spins - 1

    def step(self) -> None:
        """
        Perform one Metropolis-Hastings update step (parallel / cellular automaton style).
        """
        # Calculate local field H_i = Sum(J_ij * S_j) + h_i
        # Using matrix multiplication
        local_field = np.dot(self.J, self.bipolar_spins) + self.h

        # Calculate Energy Difference Delta_E if we flip S_i
        # Delta_E = 2 * S_i * H_i
        # (Physics convention)

        delta_E = 2 * self.bipolar_spins * local_field

        # Probability of flipping: P = min(1, exp(-Delta_E / T))
        # If Delta_E < 0 (flip reduces energy), P=1 (always flip, greedy)
        # If Delta_E > 0 (flip increases energy), P = exp(...)

        # Vectorized probability calculation
        flip_prob = np.exp(-delta_E / self.temperature)
        flip_prob = np.minimum(1.0, flip_prob)

        # Determine flips
        random_draws = np.random.random(self.num_spins)
        should_flip = random_draws < flip_prob

        # Apply flips
        # Flip -1 to 1 and 1 to -1: S_new = -S_old
        self.bipolar_spins[should_flip] *= -1

        # Update 0/1 representation
        self.spins = (self.bipolar_spins + 1) // 2

        # Anneal
        self.temperature *= self.anneal_rate

        return self.get_energy()  # type: ignore[return-value]

    def get_energy(self) -> float:
        """Calculate global energy."""
        # E = -0.5 * S^T * J * S - h^T * S
        # Factor 0.5 because J_ij is counted twice in full matrix sum
        interaction = -0.5 * np.dot(self.bipolar_spins, np.dot(self.J, self.bipolar_spins))
        bias = -np.dot(self.h, self.bipolar_spins)
        return interaction + bias

    def get_config(self) -> np.ndarray[Any, Any]:
        return self.spins

step()

Perform one Metropolis-Hastings update step (parallel / cellular automaton style).

Source code in src/sc_neurocore/solvers/ising.py
Python
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
def step(self) -> None:
    """
    Perform one Metropolis-Hastings update step (parallel / cellular automaton style).
    """
    # Calculate local field H_i = Sum(J_ij * S_j) + h_i
    # Using matrix multiplication
    local_field = np.dot(self.J, self.bipolar_spins) + self.h

    # Calculate Energy Difference Delta_E if we flip S_i
    # Delta_E = 2 * S_i * H_i
    # (Physics convention)

    delta_E = 2 * self.bipolar_spins * local_field

    # Probability of flipping: P = min(1, exp(-Delta_E / T))
    # If Delta_E < 0 (flip reduces energy), P=1 (always flip, greedy)
    # If Delta_E > 0 (flip increases energy), P = exp(...)

    # Vectorized probability calculation
    flip_prob = np.exp(-delta_E / self.temperature)
    flip_prob = np.minimum(1.0, flip_prob)

    # Determine flips
    random_draws = np.random.random(self.num_spins)
    should_flip = random_draws < flip_prob

    # Apply flips
    # Flip -1 to 1 and 1 to -1: S_new = -S_old
    self.bipolar_spins[should_flip] *= -1

    # Update 0/1 representation
    self.spins = (self.bipolar_spins + 1) // 2

    # Anneal
    self.temperature *= self.anneal_rate

    return self.get_energy()  # type: ignore[return-value]

get_energy()

Calculate global energy.

Source code in src/sc_neurocore/solvers/ising.py
Python
75
76
77
78
79
80
81
def get_energy(self) -> float:
    """Calculate global energy."""
    # E = -0.5 * S^T * J * S - h^T * S
    # Factor 0.5 because J_ij is counted twice in full matrix sum
    interaction = -0.5 * np.dot(self.bipolar_spins, np.dot(self.J, self.bipolar_spins))
    bias = -np.dot(self.h, self.bipolar_spins)
    return interaction + bias