A quantum collapse-inspired classical optimizer. Single file, NumPy only.
from TESA import TESA
best_spins, best_energy = TESA.optimize(energy_fn, delta_fn, n_spins=50)Quantum annealing uses an exponentially decaying tunneling strength to transition from exploration to exploitation. TESA translates this into a classical multi-restart strategy:
| Quantum | TESA |
|---|---|
| Γ(t) = G₀·(Gf/G₀)^t | K restarts, T₀ₖ = G₀·(Gf/G₀)^{k/K} |
| High Γ → global exploration | High T₀ → accept many uphill moves |
| Low Γ → local exploitation | Low T₀ → mostly downhill moves |
Each restart anneals from its own initial temperature. The first restart explores broadly; the last exploits deeply. This is the only element taken from quantum mechanics — no quantum hardware, no parallelism, just the exponential schedule.
for restart k in 0..K-1:
T0 = G0 * (Gf/G0)^(k/K) # decreasing initial temperatures
s = random spins
for step t in 0..steps_per_restart:
temp = T0 * (1 - t/steps)^2 # intra-round cooling
i = random spin
dE = energy_change(flip i)
if dE < 0 or random() < exp(-dE/temp):
flip i
local_search(s) # refine to local optimum
keep_best(s)Per-step cost is identical to standard simulated annealing: one spin flip evaluated. Same budget, better schedule.
from TESA import TESA
# Any Ising spin problem: define energy and flip delta
def energy(spins): # spins: numpy array of +/-1
...
def delta(spins, i): # energy change from flipping spin i
...
best_spins, best_E = TESA.optimize(
energy, delta,
n_spins=100, # number of binary variables
K=10, # restarts (default: 10)
G0=5.0, # first restart temp (default: 5.0)
Gf=0.005, # last restart temp (default: 0.005)
)
# Built-in: MAX-CUT
from TESA import TESA, random_graph
adj = random_graph(n=100, density=0.3)
cut, group_A, group_B = TESA.solve_maxcut(adj)| Parameter | Default | Role |
|---|---|---|
K |
10 | Number of restarts |
G0 |
5.0 | Initial temperature (first restart) |
Gf |
0.005 | Initial temperature (last restart) |
T |
n × 50 | Total spin flips |
MIT