☾ LUNIBYTE

EXPERIMENT 001 / LOCAL SIMULATOR

LuniQ-Resilience AI

Two backup links. Six decision qubits. One transparent comparison.

Research prototype. “AI” is the product name; no AI model runs here. Explanations are fixed, and no network protection or quantum advantage is claimed.

01 / TOY NETWORK

Plan for a link outage

━ Existing chain ┄ Candidate backup ━ QAOA selection

Choose exactly 2 backup links. Maximize recovered pair-connectivity units across five separate, equally weighted single-link outages. Each candidate costs one budget unit.

See the exact problem

Base links: AB, BC, CD, DE, EF. Candidates: AC, BD, CE, DF, AD, CF. Outage weights: 5, 8, 9, 8, 5, equal to the disconnected node pairs when that chain edge fails. A bypass restores those pairs. Overlap counts once; maximum score 35. This sums separate scenarios, not simultaneous outages.

Minimize E = −coverage + 36(Σx − 2)². For feasible selections, coverage is the weighted union of bypassed edges. Six decision qubits encode candidate links, not network nodes.

02 / CIRCUIT SETTINGS

Make a measured run

No uploads, API calls or saved inputs. Local Web Worker, six qubits, 64 amplitudes. Device speed affects runtime.

Ready. Results appear only after a real run.

Actual QAOA

H initialization → diagonal Ising phase gates → RX mixer → classical angle search → seeded Born-rule measurements.

Honest constraints

The penalty encourages two links but does not force every shot to be feasible. No baseline solution is substituted into quantum samples.

Read the method

Farhi, Goldstone & Gutmann ↗. Ideal noiseless simulation; no hardware noise, latency model, traffic capacities or production security.