How the simulator computes what it shows.

Circuits evolve a full density matrix ρ, not a state vector, so mixed states from decoherence are represented exactly rather than sampled. Unitary gates act as ρ → UρU†. Noise enters as Kraus channels, which is what a Lindblad master equation reduces to over a fixed gate duration.

• Amplitude damping from T₁, applied per gate against its real duration, with γ = 1 − e^−t/T₁.
• Pure dephasing from T₂, with the rate following 1/T_φ = 1/T₂ − 1/2T₁.
• Depolarizing, separate rates for one- and two-qubit gates.
• Readout error as a symmetric bit flip over the outcome distribution.

Reported quantities: state fidelity F = ⟨ψ|ρ|ψ⟩ against the noiseless reference, purity Tr(ρ²), and per-qubit von Neumann entropy against the rest. The density matrix is 4ⁿ, so above six qubits the engine falls back to the pure-state path rather than becoming unusable.

Connectivity and routing

Real processors are not all-to-all. Each QPU carries a coupling map — linear, ring, grid, heavy-hex or all-to-all. A two-qubit gate between non-adjacent qubits is routed with SWAPs found by breadth-first search over the coupling graph, and those SWAPs are inserted into the circuit and genuinely simulated. Each SWAP is three CNOTs, so its cost lands in the depth and the fidelity rather than being counted and ignored.

Qubits are typed data, ancilla or communication. Communication qubits are the ones coupled to the optical interface; using one for local work raises a warning, and in distributed mode a QPU with no comm qubit pays an interface penalty on every teleported gate.

The optical layer

A link is sized as an optical engineer would size it. The loss budget is itemised: fibre attenuation at the chosen grade and wavelength, connectors at 0.3 dB a pair, a fusion splice every 5 km at 0.05 dB, and 1 dB coupling at each end. Transmittance is 10^−dB/10.

• Fibre grades: ultra-low-loss 0.16, standard SMF-28 0.20, legacy field 0.25, hollow-core 0.50 dB/km at 1550 nm.
• Detectors: SNSPD, warm SNSPD, Si APD, InGaAs APD, each with efficiency, dark count rate, timing jitter and dead time. Dead time caps the achievable rate; jitter and dispersion set the coincidence window, which decides how many dark counts get in.
• Sources: SPDC, quantum dot, atomic ensemble, with heralding efficiency and a g² multi-photon quality that feeds the fidelity.

This reproduces, from physics alone, why quantum networks use superconducting detectors: at 50 km detector choice barely matters, but at 500 km an InGaAs APD falls below the entanglement threshold while an SNSPD holds, because signal-to-noise collapses as dark counts overtake a weak signal.

Networks and distributed computing

Entanglement distribution over a repeater chain multiplies Werner parameters through each swap and applies the coincidence penalty for waiting on multiple links. QBER follows from the end-to-end fidelity, and the secret key rate collapses to zero past the 11 % QBER limit of BB84. Routing between QPUs uses Dijkstra over fibre length, so a path through repeaters is found the way a control plane would find it. End-to-end fidelity of a distributed circuit is the product of every local circuit fidelity and every entangled pair consumed.

What it does not model

The models are standard and the physics is real, but they are idealised. There is no detector afterpulsing, no wavelength multiplexing beyond a simple mode count, no quantum error correction, and a uniform depolarising noise model rather than a device-specific one. Use it to build intuition and size a problem. For a procurement decision, we model your actual topology with the noise characteristics of the devices you would deploy.

Verification

The engine ships with 100 regression tests, each asserting against a known analytic answer: trace preservation, hermiticity, channel limits, noise monotonicity, fibre attenuation, the 11 % QBER threshold, that SWAP routing permutes qubits without changing the noiseless distribution, and invariants that no input produces a NaN or a fidelity outside the physical range. They run on every commit.