A single-qubit state is a point r = (x,y,z) in the Bloch ball, ρ = (I + r·σ)/2. Any physical process (gate + noise) acts as an affine map r → R·r + t. Standard quantum process tomography reconstructs R and t by sending four known probe states through the channel — the maximally mixed state (r=0) gives t directly, and the +x, +y, +z probes give the three columns of R once t is subtracted:
t = Λ(0)
R·x̂ = Λ(x̂) − t
R·ŷ = Λ(ŷ) − t
R·ẑ = Λ(ẑ) − t
The reconstructed map is exactly what the rotating point cloud shows: every point started on the unit Bloch sphere (a pure state) and is redrawn at R·r + t. Since |det R| is the volume-scale factor of a linear map, it directly reports how much of the state space the channel destroys — a value near 1 means the sphere survives intact; near 0 means almost every state collapses toward a single mixed point.
- Depolarizing p — Bloch vector shrinks isotropically: r → (1−p)r.
- Dephasing γ — kills coherence (x,y) while leaving the measured z-population alone: (x,y) → √(1−γ)(x,y), z unchanged. Models T2 decay.
- Amplitude damping γ — models energy relaxation toward |0⟩: (x,y) → √(1−γ)(x,y), z → (1−γ)z + γ. This is the only non-unital channel here, so it is the one that shifts t away from the origin.
Average gate fidelity is estimated by a deterministic Fibonacci-sphere sample of probe states r: each is sent through the ideal gate (r_ideal = R_gate·r, still on the unit sphere) and through the full reconstructed channel (r_out = R·r + t), then averaged using F(ψ) = (1 + r_ideal·r_out)/2 — the standard overlap between a pure state and a general mixed state on the Bloch ball.
Drag anywhere on the canvas to rotate the view — the projection is a genuine 3D→2D orthographic projection of the Bloch ball with independent azimuth and elevation, computed from scratch every frame, not a flattened picture.