Binomial and Numerical Bosonic Codes
Binomial codes use weighted photon-number superpositions to catch photon loss exactly, and numerical optimization extends the idea to tailored noise.
Photon loss as the enemy
The leading error in a bosonic mode is loss of a photon, the annihilation operator acting on the state. A code word made of a superposition of Fock (photon-number) states can be designed so that losing one, or a fixed number, of photons maps every code word onto a distinct, orthogonal error space that a measurement can identify and undo.
The binomial construction
Binomial codes use binomial coefficients as the weights of the Fock-state superposition. By spacing the occupied photon numbers correctly, the code guarantees that up to L photon losses, plus dephasing up to some order, keep the logical information recoverable. The name comes directly from those binomial weights, which balance the state so error words stay orthogonal.
- Code words are finite superpositions of evenly spaced Fock states.
- Loss of a photon is detected as a change in the photon-number parity structure.
- Parameters can be tuned to protect against a chosen number of losses and dephasing events.
- Unlike GKP, states are finite-energy and normalizable exactly.
Numerical, or optimized, bosonic codes take this further. Rather than fixing an analytic form, one specifies the actual noise channel of a device and numerically searches for the code words that best resist it. This can outperform any closed-form family for a specific, well-characterized oscillator.
The trade is generality versus tailoring. Binomial codes are clean, analytic, and easy to reason about, while numerically optimized codes squeeze out extra performance at the cost of being device-specific and harder to interpret. Both share the practical requirement of precise oscillator state preparation and parity measurement, typically via a coupled ancilla qubit.