Cat codes protect quantum information inside one oscillator by using superpositions of coherent states whose photon-number parity and residue structure respond predictably to loss. Instead of spreading one logical qubit across many two-level systems, the redundancy lives inside the large Hilbert space of a single bosonic mode.
The dominant error of many high-quality cavities is photon loss. A loss jump applies the annihilation operator a. Cat-code design turns that jump into a detectable change in photon-number parity while preserving enough information about the logical amplitudes to continue error correction.
There is not one unique “cat code”. Two-component cats, four-component cats, dissipatively stabilised cat qubits and squeezed-cat variants use related coherent-state geometry but protect against different error mechanisms and implement logic differently. This guide develops the common mathematics and then uses a four-component residue-class picture for explicit single-photon-loss correction.
Coherent-state superposition → photon-number parity structure → loss jump flips parity → syndrome monitor records the jump → recovery/frame update preserves logical amplitudes.
1. Coherent states are displaced vacuum
A coherent state satisfies
a|α⟩=α|α⟩
and has Fock expansion
|α⟩=e^{-|α|²/2}Σ_{n=0}∞ α^n|n⟩/√(n!).
The mean photon number is |α|². The overlap between opposite coherent states is
⟨α|−α⟩=e^{-2|α|²}.
For large |α|, the states |α⟩ and |−α⟩ become nearly orthogonal, allowing coherent-state components to act like well-separated phase-space lobes.
2. Even and odd cat states
Define normalised two-component cats
|C_α^+⟩=N_+(|α⟩+|−α⟩)
and
|C_α^-⟩=N_-(|α⟩−|−α⟩).
The plus state contains only even Fock numbers because odd powers of α cancel. The minus state contains only odd Fock numbers because even powers cancel.
The normalisation constants are
N_±=[2(1±e^{-2|α|²})]^{-1/2}.
3. Photon-number parity
Define the parity operator
Π=e^{iπN}=(-1)^N.
It acts on Fock states as
Π|n⟩=(-1)^n|n⟩.
Therefore
Π|C_α^+⟩=+|C_α^+⟩;Π|C_α^-⟩=−|C_α^-⟩.
Parity is a syndrome observable because a single photon loss changes photon number by one and therefore flips parity.
4. One photon loss maps even cat to odd cat
Use a|±α⟩=±α|±α⟩. Then
a(|α⟩+|−α⟩)=α(|α⟩−|−α⟩).
Thus, up to normalisation,
a|C_α^+⟩ ∝ |C_α^-⟩.
Similarly,
a|C_α^-⟩ ∝ |C_α^+⟩.
A single loss therefore toggles parity. Repeated nondemolition parity measurements can reveal the occurrence of photon jumps without measuring the logical amplitudes directly.
5. Why a two-component parity qubit is not automatically a full one-loss code
If logical information were encoded directly as arbitrary amplitudes of the even and odd cats, then a photon-loss jump would exchange those two basis states. The syndrome and logical transformation would be entangled with one another.
A genuine one-loss-correcting cat code therefore uses additional structure so the logical codewords initially occupy the same detectable parity sector and a loss moves them into corresponding orthogonal error sectors while preserving their relative amplitudes.
One convenient description uses photon-number residue classes modulo four.
6. Residue classes modulo four
Split the Fock basis into four sets:
- n≡0 mod 4;
- n≡1 mod 4;
- n≡2 mod 4;
- n≡3 mod 4.
A four-component cat code can choose logical codewords supported predominantly on the two even residue classes:
|0_L⟩ ∝ Σ_{k≥0} α^{4k}|4k⟩/√((4k)!)
and
|1_L⟩ ∝ Σ_{k≥0} α^{4k+2}|4k+2⟩/√((4k+2)!).
Both logical codewords have even parity, so a parity measurement does not distinguish their superposition.
7. A loss maps logical states to distinct odd error spaces
The annihilation operator lowers photon number by one.
Therefore
- the n≡0 mod 4 codeword maps to n≡3 mod 4;
- the n≡2 mod 4 codeword maps to n≡1 mod 4.
Both error states have odd parity. The parity flip reveals that a loss occurred, while the two logical amplitudes remain associated with distinct odd residue classes.
This is the code’s core error-syndrome geometry.
8. Worked residue-class example
Consider the truncated teaching state
|ψ_L⟩=c_0|4⟩+c_1|2⟩
as a simplified proxy for logical residue classes.
After one photon loss,
a|ψ_L⟩=2c_0|3⟩+√2 c_1|1⟩.
Parity has changed from even to odd. The two components remain distinguishable by their modulo-four residue but acquire unequal √n jump amplitudes.
A real code chooses amplitudes and recovery so that the logical information is preserved sufficiently well despite these n-dependent factors. This is why exact quantum error-correction conditions, not parity alone, determine performance.
9. Knill–Laflamme conditions for a bosonic code
For correctable error operators Ea, a code projector P should satisfy
P E_a†E_b P = c_ab P.
This means the environment cannot learn which logical codeword was present from the correctable error process.
For photon loss, relevant operators include I and a to first order. Exact finite-energy coherent-state codes may satisfy these conditions only approximately, so residual logical dephasing can remain even when the jump count is known.
10. Photon loss as an amplitude-damping process
For cavity energy-decay rate κ, the Lindblad master equation contains jump operator
L=√κ a.
Over a short interval dt, the probability of one photon loss is approximately
κ⟨N⟩dt
when this quantity is much less than one.
Higher-energy cats therefore lose photons more frequently. Increasing |α| separates coherent components better but raises mean photon number and hence loss-jump rate. Cat size creates a design trade-off.
11. Worked jump-rate estimate
Suppose a cavity has lifetime T1,cav=1/κ=1 ms and the encoded state has mean photon number n̄=4.
The mean jump rate is roughly
κn̄=4000 s^-1.
During a 10 μs parity-monitoring interval, the one-jump probability scale is approximately
4000×10^-5=0.04.
Four percent is already large enough that two-jump events and detector latency cannot always be ignored. Syndrome cadence must be matched to the physical jump rate.
12. No-jump evolution also matters
Conditioned on no detected photon jump, the state still evolves under a non-Hermitian effective operator containing
exp(-κN t/2).
Higher-number components shrink faster. A cat state’s coherent amplitude therefore contracts approximately as
α(t)=α(0)e^{-κt/2}
under pure loss.
Error correction must track this deterministic amplitude shrinkage as well as stochastic parity flips.
13. Parity measurement should be quantum nondemolition
A useful syndrome measurement distinguishes even from odd photon number without revealing which logical superposition occupies the code space.
In circuit QED, a dispersively coupled ancilla qubit can accumulate a phase dependent on cavity photon number. Proper pulse sequences map cavity parity onto the ancilla, which is then measured.
Repeated parity checks were central to the Ofek and colleagues experiment that demonstrated an encoded cat qubit whose corrected lifetime exceeded the lifetime of its constituent hardware elements. [2]
14. Syndrome measurement can itself introduce errors
The ancilla used to measure parity is noisy. A relaxation event during the parity-mapping sequence can propagate an uncontrolled phase into the cavity.
Thus measuring parity more frequently is not always better. Shorter intervals catch jumps sooner but expose the code to more ancilla operations and readout errors.
The optimal correction cadence balances uncorrected multiple-loss probability against measurement-induced backaction.
15. Error transparency and tracking jump times
Between jumps, different Fock components can accumulate different dynamical phases. If a jump changes which residue class is occupied, the time of the jump can affect the subsequent logical phase.
Repeated syndrome monitoring supplies approximate jump-time information, allowing software to update a logical phase frame rather than applying every correction physically.
As in qubit Pauli-frame tracking, part of bosonic recovery can be classical bookkeeping.
16. Cat-qubit noise bias
In dissipatively stabilised two-component cat qubits with large coherent separation, one logical error type can become exponentially suppressed in |α|² while another error type remains set by photon-loss processes.
This produces a strongly biased effective noise channel.
Bias-preserving gates and outer codes tailored to asymmetric X/Z error rates can exploit this structure. The architectural goal is not to make every physical error equally small but to create a simple, highly biased error model that higher-level error correction can handle efficiently.
17. Dissipative stabilisation
A two-photon driven-dissipative process can stabilise the manifold near coherent states |±α⟩. Schematically, an engineered jump operator may have the form
L_2∝a²−α².
Both |α⟩ and |−α⟩ satisfy (a²−α²)|±α⟩=0, so the dissipator attracts the oscillator toward their span while allowing logical superpositions inside that manifold.
This is an example of autonomous error suppression: engineered dissipation is used as a stabilising resource rather than treated only as noise.
18. Cat codes versus GKP codes
GKP codes arrange sharp lattice peaks across phase space and are naturally matched to small displacement errors. Cat codes arrange separated coherent-state components and photon-number parity sectors and are naturally matched to loss and biased-noise settings.
Both exploit one oscillator’s large Hilbert space, but their syndrome geometry is different:
- GKP: modular quadrature displacement syndrome;
- cat: photon-number parity/residue syndrome.
Hardware noise should influence code choice.
19. Cat codes versus binomial codes
Binomial bosonic codes use finite superpositions of Fock states chosen so error moments satisfy quantum error-correction conditions exactly to a target order.
Cat states use infinite coherent-state/Fock superpositions and can offer simpler phase-space preparation or autonomous stabilisation.
Neither family dominates universally. Average photon number, gate set, measurement access, dominant noise and available nonlinear control determine the practical trade-off.
20. Break-even is an architectural metric
A bosonic code demonstrates useful error correction when the protected logical memory outlives an appropriate uncorrected benchmark after including the overhead and new faults introduced by the correction machinery.
Comparing only the corrected logical lifetime with the worst component can make performance look stronger than it is. Strong experiments specify which bare encoding and hardware element provide the reference.
Ofek and colleagues demonstrated a full real-time cat-code correction protocol exceeding the best constituent lifetime in their system, an important break-even milestone. [2]
21. Common misconception: measuring parity measures the logical bit
In a properly designed one-loss cat code, both logical codewords begin in the same parity sector. Parity reveals the error syndrome rather than which logical superposition was stored.
22. Common misconception: a detected photon loss is automatically corrected
Detection supplies syndrome information. Logical recovery may still require a unitary, frame update, amplitude restoration and compensation for no-jump evolution and ancilla-induced phases.
23. Common misconception: larger cats are always better
Larger |α| makes coherent components more distinguishable and can strengthen some noise biases, but it increases mean photon number and therefore photon-loss rate. Optimal cat size is hardware and protocol dependent.
24. Worked synthesis problem
A bosonic logical state occupies only n≡0 and n≡2 modulo-four sectors and has even parity. One photon is lost.
Step 1: Parity. Every occupied Fock number decreases by one, so the state moves to odd parity.
Step 2: Residues. n≡0 maps to 3 mod 4; n≡2 maps to 1 mod 4.
Step 3: Syndrome. The even→odd parity flip signals that an odd number of losses has occurred since the previous reliable parity reference.
Step 4: Logical preservation. The amplitudes remain distributed across corresponding error residue classes, but √n jump weights and no-jump evolution mean an engineered recovery is still required.
Step 5: Repetition. If two losses occur between parity checks, parity returns to even and the simple parity record can miss the event. Syndrome cadence must keep multi-jump probability acceptably small.
25. Practice set
- Write the coherent-state eigenvalue equation.
- What is ⟨α|−α⟩?
- Why does the even cat contain only even photon numbers?
- Define photon-number parity.
- What does one photon loss do to parity?
- Why is a two-component even/odd basis not automatically a one-loss-correcting logical encoding?
- Which modulo-four residue classes can form the two even logical sectors?
- Where do they move after one loss?
- State the Knill–Laflamme condition.
- How does short-time loss probability scale with mean photon number?
- Why can frequent syndrome measurement become harmful?
- What trade-off appears when |α| is increased?
Answers
a|α⟩=α|α⟩.e^{-2|α|²}.- Odd Fock amplitudes from |α⟩ and |−α⟩ cancel.
Π=(-1)^N.- It flips even↔odd.
- A loss exchanges the parity sectors themselves, so syndrome and logical basis are not separated without extra structure.
- 0 mod 4 and 2 mod 4.
- 3 mod 4 and 1 mod 4, respectively.
PE_a†E_bP=c_abP.- Approximately
κ⟨N⟩dtfor short dt. - Ancilla and measurement operations can introduce additional faults and backaction.
- Component separation improves, but photon number and loss-jump rate increase.
Sources and further study
[1] Mazyar Mirrahimi and colleagues, Dynamically protected cat-qubits: a new paradigm for universal quantum computation. A foundational proposal for protected cat manifolds and bias-preserving bosonic computation.
[2] Nissim Ofek and colleagues, Demonstrating Quantum Error Correction that Extends the Lifetime of Quantum Information. A real-time cat-code experiment achieving a logical-memory break-even milestone in circuit QED.
[3] Marcel Bergmann and Peter van Loock, Quantum error correction against photon loss using multi-component cat states. A systematic treatment of multi-component cat codes and higher-order photon-loss correction.
[4] Jacob Hastrup and Ulrik Lund Andersen, All-optical cat-code quantum error correction. An optical approach to parity-sensitive cat-code recovery.
Continue through Quantum Mathematics
Guide 41: GKP Oscillator Codes, Lattice States, Modular Quadratures and Displacement Errors protects against small phase-space shifts. Guide 43: Continuous-Variable Cluster States, Graph Nullifiers and Measurement-Based Quantum Computation turns bosonic modes into a measurement-driven computing resource. Guide 44: Boson Sampling, Matrix Permanents, Linear Optics and Photonic Quantum Advantage studies passive linear-optical sampling complexity.