Two-Mode Binomial Codes

• Two-mode binomial codes are bosonic quantum error-correcting encodings that distribute binomially structured Fock-space amplitudes across two oscillator modes.
• They are constructed via spin-coherent Schwinger-boson mappings or group-theoretic, beamsplitter-based rotational symmetries, each defining distinct logical qubit architectures.
• These codes enhance error resilience by leveraging fixed-total-excitation properties to mitigate amplitude-damping and dephasing, with practical implementations in high-fidelity entangling gate experiments.

Two-mode binomial codes are bosonic quantum error-correcting encodings in which binomially structured Fock-space amplitudes are distributed across two oscillator modes. In the literature, the term has acquired two non-equivalent meanings. In the original structural sense, it denotes a single logical qubit jointly encoded in two bosonic modes, typically inside a fixed-total-excitation manifold and naturally described by a Schwinger-boson spin representation (Albert et al., 2017). Earlier amplitude-damping constructions already exhibited this form, for example with logical states $(|0,4\rangle+|4,0\rangle)/\sqrt2$ and $|2,2\rangle$ (Michael et al., 2016). More recently, a distinct two-mode rotationally symmetric/binomial family has been constructed group-theoretically and implemented with passive linear optics (Ahmed et al., 28 Aug 2025). By contrast, a recent controlled-phase-gate experiment concerns a two-mode binomial-code processor in which two separate storage modes each host one lowest-order single-mode binomial logical qubit, so “two-mode” there refers to two encoded qubits rather than one qubit delocalized across two modes (Xu et al., 9 Nov 2025).

1. Terminology and conceptual scope

The ambiguity of the phrase “two-mode binomial codes” is now part of the subject itself. The most important distinction is between a joint two-mode encoding and a two-mode processor built from two single-mode encodings. These are related by hardware and control methodology, but they are not the same code construction.

Usage	Representative states	Distinguishing feature
Joint two-mode binomial code	$|\mu_{\mathrm{bin2}}\rangle$	One logical qubit spread across two modes
Early two-mode amplitude-damping code	$(|0,4\rangle+|4,0\rangle)/\sqrt2,\ |2,2\rangle$	Fixed-total-excitation entangled codewords
Two-mode rotationally symmetric/binomial family	$$\hat U_{BS}((|0\rangle+|2N\rangle)/\sqrt2\otimes|N\rangle)$$	Group-theoretic multimode construction
Two-mode binomial-code processor	Each mode: $$|0_L\rangle=(|0\rangle+|4\rangle)/\sqrt2,\ |1_L\rangle=|2\rangle$$	Two logical qubits, one per mode

This distinction matters because many structural statements about fixed-total-excitation two-mode codes do not automatically apply to processor architectures where each mode independently carries a single-mode binomial qubit. Conversely, recent two-logical-qubit gate experiments exploit properties of the lowest-order single-mode binomial code that are not statements about the older jointly encoded two-mode family (Albert et al., 2017, Michael et al., 2016, Ahmed et al., 28 Aug 2025, Xu et al., 9 Nov 2025).

2. Spin-coherent and Schwinger-boson construction

The canonical structural definition of a two-mode binomial code is the one introduced through the spin-coherent-state framework. In that formulation, a spin-$J$ irrep is realized in two bosonic modes $a_1,a_2$ by the Jordan–Schwinger mapping

$$J_{[x,y,z]}=\frac12\sum_{j,k=0}^{1} a_j^\dagger [X,Y,Z]_{jk} a_k, \qquad J_z=\frac12(a_1^\dagger a_1-a_2^\dagger a_2),$$

with the spin-$J$ sector identified with the fixed-total-excitation manifold

$|2,2\rangle$0

The two-mode binomial codewords are then

$|2,2\rangle$1

with

$|2,2\rangle$2

These states occupy basis vectors of the form

$|2,2\rangle$3

so each mode occupation is a multiple of $|2,2\rangle$4 and the total excitation is fixed across the entire codeword superposition (Albert et al., 2017).

This formulation makes explicit that single-mode binomial codes, two-mode binomial codes, and permutation-invariant qubit codes share the same spin-coherent coefficient pattern

$|2,2\rangle$5

but differ in how the spin basis is embedded into physical Hilbert space. In the two-mode case, the embedding is into a two-oscillator Fock basis rather than a sparse ladder of one-mode Fock states. The same spin picture also supplies a logical/check-operator structure: $|2,2\rangle$6 acts as a logical $|2,2\rangle$7-type operator on the antipodal spin-coherent codewords, while $|2,2\rangle$8 acts as a check operator (Albert et al., 2017).

The fixed-total-excitation property is the defining algebraic feature. It sharply distinguishes the joint two-mode construction from standard single-mode binomial codes, where the codewords occupy a sparse one-mode number lattice but do not have exactly fixed excitation number. It also explains why two-mode binomial codes are naturally compared to NOON-type amplitude-distribution codes and to other collective-spin encodings (Albert et al., 2017).

3. Error structure, backaction, and recovery logic

Two-mode binomial codes are closely tied to amplitude-damping protection because the fixed-total-excitation constraint trivializes the no-jump backaction under equal losses on the two modes. For two-mode pure loss with identical loss rates, the no-jump factor is

$|2,2\rangle$9

and any code confined to a fixed-$|\mu_{\mathrm{bin2}}\rangle$0 sector is an eigenstate of this operator. In other words, the no-loss backaction acts as a scalar on the code subspace rather than differentially distorting the logical superposition (Albert et al., 2017).

This feature is already visible in the explicit two-mode amplitude-damping analogue discussed in the original binomial-code paper: $|\mu_{\mathrm{bin2}}\rangle$1 For equal decay rates, the no-jump operator is

$|\mu_{\mathrm{bin2}}\rangle$2

and on this code space it acts as

$|\mu_{\mathrm{bin2}}\rangle$3

A single loss in mode $|\mu_{\mathrm{bin2}}\rangle$4 or $|\mu_{\mathrm{bin2}}\rangle$5 moves the codewords into distinct error spaces,

$|\mu_{\mathrm{bin2}}\rangle$6

$|\mu_{\mathrm{bin2}}\rangle$7

so parity measurement on each mode identifies which mode lost the photon, after which a mode-specific unitary correction can return the state to the code space (Michael et al., 2016).

The same paper makes an important comparative point: the single-mode binomial code was presented as a replacement for these earlier two-mode amplitude-damping constructions, with the claimed advantages that only one bosonic mode is required and that gain and dephasing errors can also be addressed (Michael et al., 2016). At the same time, the later structural treatment of two-mode binomial codes is more limited than its single-mode analysis. It defines the codewords, the Schwinger-boson representation, and the inherited $|\mu_{\mathrm{bin2}}\rangle$8-based logical/check operators, but it does not provide a dedicated QEC matrix, explicit recovery circuit, entanglement-fidelity benchmark, or hashing-bound analysis specialized to the two-mode family (Albert et al., 2017).

A recurrent misconception is therefore that fixed total excitation by itself amounts to a complete QEC characterization. The literature supports a narrower statement: fixed total excitation removes joint no-jump distortion under equal loss, but full performance analysis and recovery design are separate questions, and for the canonical $|\mu_{\mathrm{bin2}}\rangle$9 family they were not worked out to the same depth as for single-mode binomial codes (Albert et al., 2017).

4. Rotationally symmetric two-mode binomial families

A more recent line of work constructs two-mode binomial codes as instances of a multimode rotationally symmetric bosonic framework derived from group covariance. In the two-mode case, the general logical basis is

$(|0,4\rangle+|4,0\rangle)/\sqrt2,\ |2,2\rangle$0

where $(|0,4\rangle+|4,0\rangle)/\sqrt2,\ |2,2\rangle$1 is an even order of rotational symmetry and

$(|0,4\rangle+|4,0\rangle)/\sqrt2,\ |2,2\rangle$2

is a beamsplitter-mode-mixing transformation generated by

$(|0,4\rangle+|4,0\rangle)/\sqrt2,\ |2,2\rangle$3

The main explicit “two-mode binomial” instance is the $(|0,4\rangle+|4,0\rangle)/\sqrt2,\ |2,2\rangle$4 family

$(|0,4\rangle+|4,0\rangle)/\sqrt2,\ |2,2\rangle$5

with bare states $(|0,4\rangle+|4,0\rangle)/\sqrt2,\ |2,2\rangle$6 and $(|0,4\rangle+|4,0\rangle)/\sqrt2,\ |2,2\rangle$7 before beamsplitter encoding (Ahmed et al., 28 Aug 2025).

In this construction, logical Pauli operators are implemented by simple physical operations. The logical $(|0,4\rangle+|4,0\rangle)/\sqrt2,\ |2,2\rangle$8 is

$(|0,4\rangle+|4,0\rangle)/\sqrt2,\ |2,2\rangle$9

which acts as a SWAP on the logical basis, and the logical $$\hat U_{BS}((|0\rangle+|2N\rangle)/\sqrt2\otimes|N\rangle)$$0 is

$$\hat U_{BS}((|0\rangle+|2N\rangle)/\sqrt2\otimes|N\rangle)$$1

The same framework supports an $$\hat U_{BS}((|0\rangle+|2N\rangle)/\sqrt2\otimes|N\rangle)$$2 gate through self-Kerr and a $$\hat U_{BS}((|0\rangle+|2N\rangle)/\sqrt2\otimes|N\rangle)$$3 through cross-Kerr, while $$\hat U_{BS}((|0\rangle+|2N\rangle)/\sqrt2\otimes|N\rangle)$$4 and $$\hat U_{BS}((|0\rangle+|2N\rangle)/\sqrt2\otimes|N\rangle)$$5 are implemented by gate teleportation using encoded ancillas (Ahmed et al., 28 Aug 2025).

The error-analysis results are unusually concrete. For the explicit $$\hat U_{BS}((|0\rangle+|2N\rangle)/\sqrt2\otimes|N\rangle)$$6 and $$\hat U_{BS}((|0\rangle+|2N\rangle)/\sqrt2\otimes|N\rangle)$$7 codes, loss is exactly correctable to first order. For $$\hat U_{BS}((|0\rangle+|2N\rangle)/\sqrt2\otimes|N\rangle)$$8, dephasing is also exactly correctable to first order at

$$\hat U_{BS}((|0\rangle+|2N\rangle)/\sqrt2\otimes|N\rangle)$$9

More broadly, for the full two-mode family the paper constructs an exact recovery for correlated dephasing of the form

$$|0_L\rangle=(|0\rangle+|4\rangle)/\sqrt2,\ |1_L\rangle=|2\rangle$$0

Numerically, the optimal dephasing performance occurs at

$$|0_L\rangle=(|0\rangle+|4\rangle)/\sqrt2,\ |1_L\rangle=|2\rangle$$1

and the authors argue that, unlike single-mode rotationally symmetric bosonic codes, these two-mode binomial instances exhibit no trade-off between improving photon-loss protection and improving dephasing protection as $$|0_L\rangle=(|0\rangle+|4\rangle)/\sqrt2,\ |1_L\rangle=|2\rangle$$2 increases (Ahmed et al., 28 Aug 2025).

This modern family is conceptually distinct from the older fixed-$$|0_L\rangle=(|0\rangle+|4\rangle)/\sqrt2,\ |1_L\rangle=|2\rangle$$3 Schwinger-boson bin2 construction. The former is group-theoretic, beamsplitter-encoded, and optimized around linear-optics realizations of the Pauli group; the latter is the direct spin-coherent embedding into a fixed-total-excitation manifold. Both are two-mode binomial codes, but they emphasize different physical symmetries and different operational advantages.

5. Encoded entangling gates in two-mode binomial-code processors

A different usage of the topic appears in the recent experimental realization of a controlled-phase gate between two separate lowest-order binomial logical qubits, each stored in its own superconducting cavity. Here the logical basis in each cavity is

$$|0_L\rangle=(|0\rangle+|4\rangle)/\sqrt2,\ |1_L\rangle=|2\rangle$$4

so each logical qubit lives in the single-mode subspace $$|0_L\rangle=(|0\rangle+|4\rangle)/\sqrt2,\ |1_L\rangle=|2\rangle$$5. The two-mode system is therefore the tensor product of two independent single-mode binomial encodings, not a single jointly encoded two-mode qubit. The key structural simplification is that

$$|0_L\rangle=(|0\rangle+|4\rangle)/\sqrt2,\ |1_L\rangle=|2\rangle$$6

so a logical CZ,

$$|0_L\rangle=(|0\rangle+|4\rangle)/\sqrt2,\ |1_L\rangle=|2\rangle$$7

is physically equivalent to adding a $$|0_L\rangle=(|0\rangle+|4\rangle)/\sqrt2,\ |1_L\rangle=|2\rangle$$8 phase only to the joint Fock component $$|0_L\rangle=(|0\rangle+|4\rangle)/\sqrt2,\ |1_L\rangle=|2\rangle$$9. The interaction is mediated by a central coupler transmon with dispersive Hamiltonian

$J$0

A selective coupler drive resonant with the $J$1 photon-number manifold induces a conditional cyclic evolution and hence a geometric phase on the target branch. The main methodological contribution is “geometric phase engineering”: rather than using a long spectrally selective pulse, the protocol keeps the coupler-only geometric ansatz and numerically optimizes the waveform with GRAPE while constraining the $J$2 branch to accumulate a net $J$3 phase and all other relevant branches zero net phase. Because the gate is photon-number preserving, the optimization can be confined to

$J$4

an 18-dimensional effective Hilbert space. The reported gate duration is $J$5, while the encode and decode GRAPE pulses each take $J$6. Quantum process tomography gives a logical CZ process fidelity of $J$7 without post-selection and $J$8 with post-selection on the coupler ground state, with post-selection success probability $J$9. The experiment also prepares an encoded Bell state visualized by joint Wigner tomography. The gate preserves cavity photon-number distributions and logical code-space occupations, but the authors explicitly distinguish this from a fully fault-tolerant or error-transparent entangling gate; coupler leakage and cavity dephasing remain the dominant limitations (Xu et al., 9 Nov 2025).

This experiment materially changes the operational significance of “two-mode binomial codes.” It shows that two separate binomially encoded modes can support a high-fidelity entangling gate directly in encoded space, using a control primitive far simpler than generic multimode optimal control. At the same time, it should not be conflated with the older notion of a two-mode binomial code as a single logical qubit jointly encoded across two oscillators (Xu et al., 9 Nov 2025).

6. Relation to adjacent bosonic-code programs

Two-mode binomial codes occupy an intermediate position within bosonic coding theory. Structurally, the canonical $a_1,a_2$0 family is closely related to NOON-type constructions, and its spin-coherent coefficients are the same ones that also generate permutation-invariant qubit codes; what changes across these families is the physical embedding of the common spin data (Albert et al., 2017).

They should also be distinguished from other two-mode bosonic encodings. A prominent counterexample is the pair-cat code, whose codewords are infinite-support superpositions in fixed photon-number-difference sectors,

$a_1,a_2$1

stabilized by the dissipative jump

$a_1,a_2$2

Pair-cat codes are therefore two-mode bosonic codes, but not binomial codes in the standard finite-support sense; their organizing symmetry is photon-number difference rather than binomial weighting across a fixed-total-excitation manifold (Albert et al., 2018).

Recent advances in logical control and state synthesis are likewise adjacent rather than identical to two-mode binomial-code theory. “A Mathematical Structure for Amplitude-Mixing Error-Transparent Gates for Binomial Codes” develops parity-nested, error-transparent amplitude-mixing gates for single-mode binomial encodings (Wetherbee et al., 2024). “Noise-resilient nonadiabatic geometric quantum computation for bosonic binomial codes” studies single-mode geometric control of the lowest-order binomial code in a cavity–qutrit architecture (Li et al., 18 Mar 2026). “Arbitrary high-fidelity binomial codes from multiphoton spin-boson interactions” presents a single-mode MPJC-based synthesis method for arbitrary finite-support binomial codewords (Laha et al., 11 Jul 2025). A plausible implication is that future two-mode binomial-code implementations will borrow their control ideas—error-transparent sector engineering, geometric-path design, and finite-support state synthesis—while retaining genuinely two-mode syndrome structures and symmetry constraints.

Taken together, the literature presents two-mode binomial codes not as a single monolithic object but as a family of closely related ideas: fixed-total-excitation joint encodings, dual-rail-like rotationally symmetric multimode constructions, and two-mode processor architectures built from separate single-mode binomial qubits. Their common thread is the use of highly structured bosonic support to convert otherwise large oscillator Hilbert spaces into analytically tractable and experimentally addressable encoded manifolds.