PaperlayerView exact revision on arXiv

Source-linked AI summary

Hybrid Oscillator-Qubit Quantum Processors: Instruction Set Architectures, Abstract Machine Models, and Applications

Yuan Liu, Shraddha Singh, Kevin C. Smith, Eleanor Crane, John M. Martyn, Alec Eickbusch, Alexander Schuckert, Richard D. Li, Jasmine Sinanan-Singh, Micheline B. Soley, Takahiro Tsunoda, Isaac L. Chuang, Nathan Wiebe, Steven M. Girvin

arXiv:2407.10381v3quant-ph

TL;DR

Systematic, rigorous compilation schemes for hybrid oscillator-qubit algorithms are not yet available, and tracking compilation rules becomes difficult and less efficient for complex algorithms. This paper develops frameworks for hybrid CV-DV processors, including abstract machine models, instruction sets, and compilation approaches; it concludes that hybrid digital architectures may offer substantial computational advantage for simulating matter with bosonic degrees of freedom.

Abstract

from arXiv · show

Quantum computing with discrete variable (DV, qubit) hardware is approaching the large scales necessary for computations beyond the reach of classical computers. However, important use cases such as quantum simulations of physical models containing bosonic modes, and quantum error correction are challenging for DV-only systems. Separately, hardware containing native continuous-variable (CV, oscillator) systems has received attention as an alternative approach, yet the universal control of such systems is non-trivial. In this work, we show that hybrid CV-DV hardware offers a great advantage in meeting these challenges, offering a powerful computational paradigm that inherits the strengths of both DV and CV processors. We provide a pedagogical introduction to CV-DV systems and the multiple abstraction layers needed to produce a full software stack connecting applications to hardware. We present a variety of new hybrid CV-DV compilation techniques, algorithms, and applications, including the extension of quantum signal processing concepts to CV-DV systems and strategies to simulate systems of interacting spins, fermions, and bosons. To facilitate the development of hybrid CV-DV processor systems, we introduce formal Abstract Machine Models and Instruction Set Architectures -- essential abstractions that enable developers to formulate applications, compile algorithms, and explore the potential of current and future hardware for realizing fault-tolerant circuits, modules, and processors. Hybrid CV-DV quantum computations are beginning to be performed in superconducting, trapped ion, and neutral atom platforms, and large-scale experiments are set to be demonstrated in the near future. We present a timely and comprehensive guide to this relatively unexplored yet promising approach to quantum computation and providing an architectural backbone to guide future development.

I. INTRODUCTION

This section introduces a systematic architectural framework for hybrid CV-DV processors, spanning hardware, programming, compilation, and applications. It develops abstract machine models and instruction sets that distinguish the roles of oscillators and qubits and situate these abstractions alongside resource analysis and optional quantum error correction.

I. INTRODUCTION

  • The tutorial addresses challenges in treating oscillators and qubits in a common computational framework and in programming hybrid processors, reasoning about their power, and estimating their resources.It identifies abstract machine models and universal instruction set architectures as key abstractions for resource estimation and co-design of algorithms, software, and hardware.Source: §I · p. 3

A. Outline of this Manuscript

  • The manuscript proceeds from physical mechanisms through models, architectures, programming and compilation, and algorithms and applications across seven abstraction layers.Its outline covers bosonic representations and gates, error correction, compilation and circuit synthesis, applications, and comparisons with qubit-only hardware.Source: §I · p. 3, A. Outline of this Manuscript · p. 4

• Programming Languages and Compilation:

  • The compilation and applications overview includes entangling-gate techniques, quantum signal processing for hybrid systems, bosonic error-correction compilation, CV-DV state transfer, tomography, and simulation of oscillator-qubit models.The listed applications include state transfer for the quantum Fourier transform and simulation of models involving oscillators and qubits.Source: • Programming Languages and Compilation · p. 3

B. Reader’s Guide

  • Highlighted contributions include hybrid Wigner process tomography error bounds, QSP with non-commuting oscillator position and momentum operators, exact entangling-gate synthesis, and oscillator-based simulation and compilation methods.The overview also describes scalable compilation for representing oscillators on qubit-only hardware and costing its gate count.Source: B. Reader’s Guide · p. 5

II. HYBRID CV-DV QUANTUM COMPUTING ARCHITECTURE

  • The architecture overview spans physical models, three abstract machine models, instruction set architectures, compilation, error correction, and potential applications.Its seven-layer stack connects user applications and programs to compilation, logical and physical gates, instructions, pulses, and hardware.Source: §II · p. 5, Figure 2 · p. 6

A. Physical Model of Hybrid CV-DV Quantum Computation

  • The tutorial focuses on coupled oscillators and qubits, assuming direct measurement of qubits and hardware-constrained interactions, with beam splitters as the main connectivity mechanism.The discussion often specializes to a near-neighbor layout pairing each qubit with an oscillator; ion traps and optical tweezers can dynamically reconfigure geometry.Source: A. Physical Model of Hybrid CV-DV… · p. 7

B. Abstract Machine Models

  • Abstract machine models bridge hardware architecture and instruction sets, enabling analysis of design tradeoffs and computational complexity under resource constraints; quantum models must also account for entanglement and coherence.The passage identifies communication, storage, and computation as quantum-specific areas where these resources matter.Source: B. Abstract Machine Models · p. 7
  • Hybrid AMMs combine quantum operations with classical measurement and feed-forward, distinguishing user-visible abstractions from transpiler-visible models and organizing the latter as QCMM, OCMM, and HMM.The user-visible layer can hide hardware details, while transpiler-visible models support mapping instructions to physical connectivity and error properties and resource estimation.Source: B. Abstract Machine Models · p. 9

1. Qubit-centric Abstract Machine Model (QCMM)

  • QCMM uses qubits for computation and bosonic subsystems for memory, where long resonator coherence and lower-overhead bosonic error correction can support memory without direct oscillator logic gates.Qubit states can be written to and read from bosonic memories, and different codes can provide memory tiers with different decoding and read-write characteristics.Source: §1 · pp. 9–10

2. Oscillator-centric Abstract Machine Model (OCMM)

  • OCMM uses bosonic resonators for computation and qubits sparsely for state preparation, measurement, and induced interactions, fitting computations such as boson sampling and Bose-Hubbard simulation.Coupled oscillator systems are of computational interest, although realistic optical boson-sampling imperfections can simplify classical simulation.Source: §2 · p. 11

3. Hybrid Abstract Machine Model (HMM)

  • HMM uses both qubit and bosonic gates for arbitrary transformations on both systems and is the paper’s main focus.The table characterizes HMM applications as underexplored while identifying potential simulations beyond present-day technology.Source: Table II · p. 10, §3 · p. 11

4. User-visible AMMs as an Alternative Taxonomy

  • An alternative user-visible taxonomy classifies AMMs by exposed quantum degrees of freedom rather than physical components, with QCMM exposing qubits, OCMM oscillator operations, and HMM both CV and DV degrees of freedom.These abstractions may hide auxiliary qubits, physical connectivity, or the implementation of logical qubits.Source: §4 · p. 11

C. Instruction Set Architectures

  • The ISA is the AMM-based hardware–software interface; the paper presents Fock-space and phase-space ISAs, conditioning primitive operations on oscillator excitations or position and momentum, respectively.The tutorial assumes tightly coupled qubits and oscillators, synchronized operations, parallel gates across distinct qumodes and qubits, and measurement-based conditional control.Source: C. Instruction Set Architectures · pp. 11–13

D. Quantum Error Correction

  • Quantum error correction is an optional compilation stage because fault tolerance can require prohibitive overhead; bosonic codes can encode logical qubits in oscillators, while omitting QEC preserves modes’ full CV characteristics for bosonic simulations.With QEC, logical layout, connectivity, and residual errors depend on the hardware and codes; without it, physical oscillator-qubit instructions serve directly as logical-level instructions.Source: D. Quantum Error Correction · pp. 13–14
  • Hybrid CV-DV quantum error correction can encode both oscillator and qubit components, while auxiliary transmon levels enable semi-fault-tolerant SNAP and controlled-displacement gates.In NISQ operation, the compiler instead outputs physical-gate sequences directly to the control pulse generator.Source: Figure 4 · p. 14, D. Quantum Error Correction · p. 14

E. Programming Languages and Compilation

  • Programming and compilation require oscillator, qubit, and hybrid-operation abstractions; a hardware-agnostic QIR can connect high-level programs to device-specific logical gates, with bosonic QIR representing oscillator controls as Hamiltonian polynomials.A conditional beam-splitter term coupling two oscillators through Pauli Z is one example, noted as useful for Z2 lattice gauge theory simulation.Source: E. Programming Languages and Compilation · pp. 14–15

F. Algorithms and Applications

  • Native bosonic modes have a relatively simple error model and enabled superconducting-cavity memory quantum error correction to exceed break-even using bosonic codes.Source: F. Algorithms and Applications · p. 15
  • Native bosonic modes offer hardware-efficient simulation of bosons, whose field operators are difficult to synthesize with qubits; experiments simulated molecular vibrational spectra and dissipative dynamics.With a boson-number cutoff, each mode can be represented using about log2[Nmax + 1] qubits, but accurately synthesizing the square-root factors in field-operator matrix elements requires many gates.Source: F. Algorithms and Applications · p. 16
  • CV state and process tomography using Wigner- or characteristic-function measurements can inform hardware error models, abstract machine models, and software tools such as Bosonic Qiskit.These models can describe errors associated with gates, measurements, and hardware time evolution within an instruction set architecture.Source: F. Algorithms and Applications · p. 16
  • Hybrid oscillator-qubit systems can use qubits to control complex non-Gaussian resonator states and deterministic bosonic gates; controllable beam splitters support rapid, parallel routing, while QND photon counting can flag loss as an erasure [233].Flagged erasures have lower entropy and are easier to correct than errors whose locations are unknown.Source: F. Algorithms and Applications · p. 16

III. STATES AND OPERATORS

  • The states-and-operators section introduces oscillator and multi-oscillator Hilbert spaces, Fock and phase-space representations, and truncation of the infinite-dimensional space as foundations for later hybrid instruction sets.The phase-space representation has no one-to-one mapping to a discrete-variable architecture.Source: §III · p. 17

A. Qubit States and Operations

B. The Hilbert Space of Oscillators

The section develops discrete Fock-state and continuous-variable descriptions of the infinite-dimensional oscillator Hilbert space, using phase-space and stellar representations to relate state structure and evolution to operations. It connects stellar evolution to root transformations, notes that the effects of arbitrary gates remain unresolved, and proposes root-manipulation protocols and a stellar instruction set as future compilation directions.

B. The Hilbert Space of Oscillators

  • Constant energy-level spacing makes the harmonic oscillator isochronous, enabling a rotating-frame interaction representation that removes the rapid free evolution.The work uses this representation almost exclusively to study slower evolution under additional control terms.Source: B. The Hilbert Space of Oscillators · p. 18

C. Representation and Bases

  • Oscillator information resides in the joint Hilbert space of qubits and harmonic oscillators, whose oscillator vector space is called a qumode.The qumode differs from a qubit’s two-level state space and supports multiple state and operator representations.Source: C. Representation and Bases · p. 18

1. Fock Basis

  • Fock states form the number-operator eigenbasis, while coherent states are a continuous, over-complete family of annihilation-operator eigenstates that reveal oscillators’ continuous-variable features.Fock states can be built from the vacuum with raising operators; coherent states are characterized as simple displacements of the Gaussian vacuum, while Fock states are highly non-Gaussian.Source: §1 · pp. 18–19

2. Position-Momentum Basis

  • Position and momentum provide continuous oscillator representations; the vacuum has Gaussian wavefunctions and electric-field fluctuations called vacuum noise.In Wigner units, coherent states have position and momentum variances of 1/4, and their complex amplitude shifts the Gaussian in position or momentum.Source: §2 · pp. 19–20

3. Introduction to Phase Space

  • Classical phase-space flow preserves area; oscillator Hamiltonians illustrate squeezing into an equal-area ellipse, large-|x| displacement, and four-fold rotational symmetry useful for bosonic 4-leg cat-code error correction.The examples show how classical flow-field intuition can guide quantum unitary design; oscillator nonlinearities can be implemented with non-Gaussian controls or hybrid oscillator-qubit gates.Source: §3 · pp. 21–22

4. Stellar-Holomorphic Representation

  • The stellar representation maps bosonic operators to a complex variable and derivative, visualizing state dynamics through analytic-function zeros and quantifying non-Gaussianity by stellar rank.Stellar rank also quantifies state-preparation difficulty; root movement, including shifting, merging, and splitting, represents qumode dynamics.Source: §4 · pp. 22–23
  • The special SNAP gate is free oscillator evolution that globally rotates stellar-function roots in phase space without changing their number.How arbitrary SNAP gates or other arbitrary unitaries move the roots remains an open question.Source: §4 · p. 23
  • Shifting, deleting, or creating individual stellar-function roots remains an open problem that could inform CV algorithm compilation and enable a stellar instruction set.The proposed stellar ISA would manipulate individual roots, drawing on the instruction-set architecture introduced in Sec. II.Source: §4 · p. 23
  • An oscillator entangled with the two qubit basis states requires two separate stellar functions, and the stellar representation has been generalized to multiple modes.The passage cites for the multi-mode generalization.Source: §4 · p. 23

D. Truncation of the Hilbert Space

The section explains practical truncation of oscillator Hilbert spaces and cutoff selection, then introduces oscillator-qubit instruction sets and Gaussian control operations. It describes how Gaussian transformations compose, including noncommutative displacements and their geometric phase, and how phase-space rotations can implement a quantum Fourier transform or be realized virtually through software control.

D. Truncation of the Hilbert Space

  • Because oscillator states cannot reach infinite energy and anharmonicity grows with boson number, practical and rigorous algorithm analysis requires a finite Hilbert-space truncation.Truncation can use phase-space Gaussian damping, a stellar-function root or polynomial-degree limit, or an energy cutoff in the Fock basis; smoother damping is also possible.Source: D. Truncation of the Hilbert Space · p. 24
  • For time-independent H = ωa†a + V with positive semi-definite V, conserved mean energy supports a simple, generally applicable but crude cutoff estimate for many quadratic field Hamiltonians.Tighter bounds are possible when more information is available.Source: D. Truncation of the Hilbert Space · pp. 24–25

IV. INSTRUCTION SETS FOR OSCILLATOR-QUBIT SYSTEMS

  • The instruction-set discussion classifies gates as qubit-only, oscillator-only, or hybrid entangling, and shows that hybrid gates can yield universal unitaries even when the oscillator component is at most linear.These gates can be viewed as qubit rotations depending on oscillator position and momentum, or as qubit-conditioned phase-space displacements, extending quantum signal processing possibilities.Source: §IV · pp. 25–26

A. Gaussian Oscillator Control

  • Gaussian unitaries, generated by Hamiltonians at most quadratic in position and momentum, induce linear symplectic transformations and preserve Gaussian states.The discussion develops this control framework and its composition for continuous-variable systems.Source: §1 · pp. 26–27, A. Gaussian Oscillator Control · p. 26

2. Composition of Two Gaussian Transformations

  • Successive Gaussian transformations act on oscillator operators in the same order as the corresponding active unitaries act on the state.The section cautions that reversing the passive transformation order in the Heisenberg picture gives an incorrect result.Source: §2 · p. 28

B. Gaussian Operations

  • Displacement, phase-space rotation, single-mode squeezing, and beam splitting or two-mode squeezing form a generating set for Gaussian gates, which can be compiled efficiently using Bloch-Messiah decomposition.The cited discussion notes that the Gaussian oscillator operations considered are natively available in superconducting platforms.Source: §1 · p. 26, B. Gaussian Operations · p. 28

1. Displacement Gate

  • The displacement gate translates oscillator states in phase space by Δx = Re α and Δp = Im α, and maps coherent states to coherent states.Its normal-ordered form is preferred for numerical calculations with a truncated oscillator Hilbert space.Source: §1 · p. 29
  • Displacements form a non-Abelian Weyl-Heisenberg group, and closed phase-space loops produce a geometric phase used in GKP error correction and compiling entangling gates.In the chosen Wigner units, displacements commute when their parallelogram area is an integer multiple of π.Source: §1 · p. 29

2. Phase-Space Rotation Gate

  • A π/2 phase-space rotation performs the position-to-momentum quantum Fourier transform; free oscillator evolution realizes rotations virtually by shifting the control time reference and tracking later drive phases in software.No perturbation or drive is needed to carry out the rotation gate.Source: §2 · p. 30

3. Single-Mode Squeezing Gate

This section describes Gaussian oscillator gates, including squeezing, beam-splitting, SUM, and translations, and relates them to qubit Clifford operations, classical simulation, and Pauli-like operations in a GKP code space. It also outlines Gaussian operations’ Clifford-like properties, their ability to entangle Gaussian states, and the limitation that Gaussian resources alone cannot provide continuous-variable error correction.

3. Single-Mode Squeezing Gate

  • Single-mode squeezing reduces fluctuations in a chosen quadrature while increasing the conjugate quadrature, preserving the commutation relation and saturating the Heisenberg inequality.Quadratic terms generate the transformation; three- or four-wave mixing and dissipation engineering can generate them. Superconducting circuits have reached squeezing up to e^2r ∼ 10 dB, with maximum squeezing limited in part by the self-Kerr effect.Source: §3 · p. 30

4. Two-Mode Beam-Splitter Gates

  • The beam-splitter gate preserves total excitation number and can swap cavity states independently of their encoded bosonic state, enabling routing across hardware.A 50:50 beam splitter has demonstrated the Hong-Ou-Mandel effect; with information encoded in the n = 0, 1 Fock states, that effect can populate n = 2 and leave the code space, so the gate is not iSWAP.Source: §4 · pp. 31–32

5. Two-Mode Squeezing Gate

  • Two-mode squeezing transforms pairs of modes and produces an entangled state with equal photon numbers, while its gate can be decomposed into beam splitters and single-mode squeezers.For φ = 0, the two-mode squeezed vacuum is described as a CV analog of a DV Bell state; a pumped Josephson ring modulator achieved g_TMS/2π = 12.6 MHz.Source: §5 · p. 32

6. Two-Mode Sum Gate

  • The SUM gate displaces one oscillator according to the other oscillator’s position and can be generated from beam-splitter and two-mode-squeezing interactions with identical rates.Its Bloch-Messiah decomposition replaces two-mode squeezing with single-mode squeezers.Source: §6 · p. 33

C. Gaussian Operations are Analogous to Clifford Gates

  • Gaussian operations are the CV analog of Clifford operations: they map phase-space displacements to displacements and permit efficient classical tracking of Gaussian-state evolution.As in qubit Clifford circuits, this classical tractability is limited to the Gaussian operations; non-Gaussian states and Wigner negativity are associated with the breakdown of that simplicity.Source: C. Gaussian Operations are Analogous… · pp. 33, 35, 36
  • The oscillator phase-space rotation R(θ) can be implemented virtually by resetting the phase of subsequent oscillator drives in software.The gate summary is not comprehensive, and experimental realization of many oscillator-only gates requires an auxiliary nonlinear system such as a qubit.Source: C. Gaussian Operations are Analogous… · p. 36, Table IV · p. 36
  • Gaussian operations can entangle Gaussian states, but cannot produce Wigner negativity from them.Two-mode squeezing and the SUM gate are examples of Gaussian operations that produce entanglement.Source: C. Gaussian Operations are Analogous… · p. 36
  • Because Gaussian states are not displacement eigenstates, one proposed stabilizer framework restricts to GKP code states, where phase-space translations obey the Pauli algebra.The two GKP code words are stabilizer states, and the code compactifies phase space to a two-dimensional Hilbert space.Source: C. Gaussian Operations are Analogous… · p. 37
  • The CNOT operation on GKP oscillator qubits is implemented by the two-mode Gaussian SUM gate.Mode b is displaced by an odd number of half lattice constants exactly when mode a is in state |1L⟩.Source: C. Gaussian Operations are Analogous… · p. 37
  • Randomized benchmarking techniques for Clifford gates cannot be directly applied to Gaussian operations because they do not form a unitary t-design.Source: C. Gaussian Operations are Analogous… · p. 38
  • Gaussian resources, comprising Gaussian operations and homodyne measurement, are insufficient for continuous-variable quantum error correction.This is a no-go result for the continuous-variable analogue of carrying out error correction with Clifford operations and measurements.Source: C. Gaussian Operations are Analogous… · p. 38

D. Hybrid Non-Gaussian Operations

This section covers hybrid non-Gaussian gate mechanisms and instruction sets for oscillator-qubit control, including how they can be matched to hardware and applications. It also discusses classical control and tailored tomography for processor operation and characterization, alongside the challenges of benchmarking oscillator systems.

D. Hybrid Non-Gaussian Operations

  • Combining qubit-assisted universal oscillator control with qubit readout or reset enables universal oscillator-channel control using a single auxiliary qubit.Experiments demonstrated cooling to squeezed vacuum and dissipation to superpositions of squeezed states; channel methods can remove excess entropy or excitations from bosonic codes.Source: D. Hybrid Non-Gaussian Operations · p. 38

1. Conditional Displacement

  • Conditional displacements shift oscillator phase space according to the qubit state, while conditional rotations and parity gates enable qubit-mediated control and photon-number measurements.Parity-gate phase kickback supports QND photon-number parity measurement and, with phase estimation or related methods, full photon-number distribution measurement.Source: §1 · p. 39, §2 · pp. 39–40

3. Photon Number Selective Qubit Rotation (SQR) Gate

  • Photon-number-selective qubit rotations combined with displacements give universal hybrid control, while SNAP phases combined with displacements give universal oscillator control.SNAP applies user-chosen phases to individual Fock states using an auxiliary qubit as a catalyst.Source: §3 · p. 40, §4 · pp. 40–41

5. Exponential-SWAP Gate

  • The eSWAP gate entangles two modes and remains code agnostic because both identity and SWAP act independently of the encoded cavity states.At θ = π/2, it yields an iSWAP gate in a two-mode encoding, including when the logical states are binomial code states.Source: §5 · p. 41

1. Oscillator-only ISAs

  • Gaussian oscillator operations generate only linear and quadratic controls and cannot create Wigner negativity; adding a nonlinear resource such as a cubic Hamiltonian or non-Gaussian resource states enables universality.A Gaussian-only computation can instead consume fixed non-Gaussian states supplied by a specialized state factory.Source: §1 · p. 42, §5 · p. 44

2. Phase-Space Instruction Set

  • The phase-space instruction set achieves universal hybrid control using qubit-controlled displacements, qubit rotations, and beam-splitters, even with only linear oscillator-qubit coupling.Commutator constructions generate higher-degree polynomial controls, extending to multiple oscillators and qubits.Source: §2 · pp. 42–43

3. Fock-Space Instruction Set

  • The Fock-space instruction set combines photon-number-conditioned qubit rotations with unconditional displacements and beam-splitters, making it dual to the phase-space set in which oscillator operations are qubit-conditioned.SQR and SNAP exploit native dispersive coupling in superconducting systems; optimized instruction sequences have been demonstrated theoretically for expressive, efficient state preparation.Source: §3 · p. 43

4. Sideband Instruction Set

  • The Sideband instruction set uses qubit rotations, Jaynes–Cummings sideband transitions, and beam-splitters; trapped-ion implementations achieved universal control under a bounded energy budget that prevents leakage beyond a chosen Fock level.Sideband operations have also been demonstrated in neutral atoms for cooling and manipulating the lowest two Fock levels.Source: §4 · pp. 43–44

6. Choice of Instruction Set

  • Instruction sets should match hardware parameters and applications: phase-space gates suit concise x,p and GKP-state descriptions, while Fock-space gates suit photon-number tasks and SNAP gates conserve photon number.Source: §6 · p. 44

1. Control Flow and Feedback

  • Measurement-derived classical outcomes can adapt later measurements or gates, support variational algorithms and error correction, and enable runtime feedforward to help overcome hardware connectivity limits.Source: §1 · pp. 44–45

2. Benchmarking and Certification

  • Oscillator benchmarking is challenging because the Hilbert space is infinite-dimensional, and randomized benchmarking remains open: Gaussian operations do not form a unitary t-design, though regularized-rigged designs offer approximate progress.Source: §2 · pp. 45–46
  • O(σ_i^2 log(1/δ)/ϵ^2) samples suffice to learn one conditional-Wigner phase-space point under the stated exponential-tail assumptions; full reconstruction is comparable to tomography when variance and second derivatives are modest.Source: §2 · pp. 46–47
  • For hybrid process characterization, finite-squeezing Choi-state constructions replace unphysical infinite squeezing with a finite-region approximation, while Wigner tomography can support gate-set tomography up to unobservable gauges.Source: §2 · pp. 48–49

V. QUANTUM ERROR CORRECTION WITH OSCILLATORS

This section surveys oscillator-based quantum error correction, including bosonic encodings, noise models, and logical-state preparation and readout. It also covers compiling logical operations with hybrid phase-space and Fock-space instruction sets, with circuit duration as an efficiency metric.

A. Dissipation and Decoherence

  • At 20 mK, typical 5-GHz microwave resonators have excitations frozen out, unlike qubits whose first-excited-state probability can reach ~1–10%.Packing many qubits into one oscillator makes mean boson number and excitation loss rate scale exponentially with qubit count, though oscillator encoding can benefit modest counts. Oscillator noise includes damping and heating; cavity dephasing is often caused by dispersive coupling to control transmons.Source: A. Dissipation and Decoherence · pp. 49–50

B. Overview of Bosonic QEC

  • Bosonic logical memories have achieved break-even, where the corrected logical qubit’s coherence exceeds that of the best available uncorrected qubit.Fault-tolerant computation still requires noise-resilient logical operations and manageable overhead; progress includes fault-tolerant error detection and error-transparent and path-independent gates.Source: B. Overview of Bosonic QEC · p. 51

1. Single-Mode Encodings

  • Under the same photon-loss model, GKP codes significantly outperform cat and binomial codes for most loss rates and approach the Gaussian thermal-loss channel’s quantum capacity within a constant factor.Bosonic codes generally gain protection as average photon number increases, despite photon-loss rates increasing linearly with photon number.Source: §1 · p. 51

2. Error Correction Techniques and Break-Even Performance:

  • Autonomous GKP stabilization in trapped ions produced an encoded-state lifetime about 3× longer, while a semi-autonomous protocol achieved error-correction gain G ~2.3.Auxiliary-qubit decay limits GKP performance; biased-noise auxiliary qubits can help, and the gain was preserved in a first error-corrected d = 3, 4-level GKP qudit encoding.Source: §2 · p. 52

3. Multi-mode Encodings

  • Dual-rail qubits use two oscillator modes and can convert photon-loss leakage into heralded erasures through joint photon-number parity measurement.Arbitrary single-qubit rotations use beam-splitter couplings; erasures are more readily corrected than errors of unknown location, but practical scaling of multimode codes remains unresolved under circuit-level noise.Source: §3 · pp. 52–53

4. Oscillator-to-Oscillator Encoding

  • Oscillator-to-oscillator codes aim to protect an entire bosonic Hilbert space, but Gaussian errors cannot be corrected using only Gaussian resources and proposed code families lack a threshold for classical displacement noise.Proposals use GKP states as non-Gaussian resources or auxiliary qubits, including a GKP-Two-Mode-Squeezing code.Source: §4 · p. 53

5. Code Concatenation

  • CV-DV concatenation can exploit analog GKP syndrome information: a demonstrated three-qubit hybrid code corrected double-bit-flip errors, beyond the single-bit-flip correction of usual DV codes.A comparison of concatenation strategies found CV-DV encoding performed best for correcting both logical X and Z errors; practical multimode alternatives remain to be compared.Source: §5 · pp. 53–54

C. Logical ISA for Bosonic QEC

  • The section compiles logical bosonic-QEC instructions with qubits as ancillas and oscillators as data, matching translation-symmetric GKP codes to phase-space ISA and rotationally symmetric cat codes to Fock-space ISA.The examples cover preparation, readout, stabilization, single-qubit rotations, and entangling operations, drawing mainly on non-abelian QSP methods and also Fock-space schemes.Source: C. Logical ISA for Bosonic QEC · p. 54

1. Preliminary Information

  • Because conditional-displacement gate errors grow with gate duration, circuit efficiency should account for total circuit time rather than gate count alone.Typical qubit rotations take 24 ns, whereas a conditional displacement CD(α) lasts approximately |α| µs.Source: Figure of merit · p. 54

2. Logical State Preparation

  • For Δ=0.302 and N=3 peaks, the phase-space GKP preparation method yields infidelity O(10^-2) with circuit duration 5.84 µs.It is described as the only fully deterministic GKP preparation scheme known that requires no numerical optimization; ancilla post-selection is available every 1.95 µs.Source: GKP Code · p. 55
  • For a four-legged cat state with α = 5, single-variable QSP using BB190 disentangles the ancilla with fidelity 0.99 at circuit depth N = 6.The passage notes that a non-Abelian QSP correction can be found when the two cat sizes have a ratio that is an even integer.Source: Four-Legged Cat Code · p. 56
  • Binomial codes can be prepared with the sideband-ISA Law-Eberly protocol or numerically optimized phase-space gates, while dual-rail codes can use Law-Eberly or phase-space non-Abelian QSP.The simplest binomial code corrects at most one photon loss; the cited preparation routes include,, and [456].Source: Four-Legged Cat Code · p. 56

3. Logical Readout

  • For square GKP codes with Δ=0.302, logical readout has failure probability 10^-4 and oscillator back action O(ϵ^2); GKP codes support all three logical Pauli bases with equal ease.The circuit entangles the logical oscillator state with an auxiliary qubit, whose measurement implements the logical measurement.Source: §3 · pp. 56–57

4. Stabilization and Error Correction

This section covers hybrid instruction sets for stabilizing and correcting bosonic codes, including ancilla-based error control and logical-qubit operations. It also describes compilation methods for hybrid CV-DV processors, complementary representations of hybrid unitaries, and architecture-specific compilation challenges and strategies.

4. Stabilization and Error Correction

  • Cat-state amplification reverses energy-relaxation shrinkage by entangling the ancilla with cat-state blobs, displacing them outward, then disentangling at the enlarged size.The protocol uses the adjoint of the non-abelian QSP cat-preparation sequence; for amplification to 2α, BB1 outperforms non-abelian QSP above approximately α+δ ≈ 7.5.Source: Cat State Amplification · p. 57, Figure 17 · p. 58
  • The sBs dissipative GKP channel autonomously stabilizes the code space, and has been used in beyond-break-even QEC memory experiments for GKP qubits and qudits.In trapped ions, the channel was demonstrated using the sideband ISA; ancilla resets move the oscillator closer to the code space with high probability, removing up to one bit of entropy.Source: Dissipative GKP Stabilization · pp. 58–59

5. Logical Single-Qubit Rotations

  • GKP phase transfer teleports arbitrary logical rotations that classical phase-space translations cannot provide, and can correct errors simultaneously for GKP qubits and, with adaptation, qudits.A piece-wise construction can reduce ancilla dephasing; the protocol is described as yielding high-fidelity gates with biased-noise ancillas.Source: §5 · p. 60
  • For binomial and four-legged cat codes, hybrid ancilla circuits implement logical rotations and can detect or tolerate ancilla errors, while efficient analytic QSP sequences for universal binomial control remain open.The binomial-code circuit can herald ancilla or photon-loss errors; SNAP gates and qutrit ancillas offer fault-tolerant alternatives.Source: §5 · p. 60

6. Entangling Bosonic Logical Qubits

  • Bosonic logical entangling gates can use ancilla CPHASE gates or beam-splitters, enabling controlled logical operations and, with the former approach, deterministic ancilla states for error detection.The CPHASE approach uses two ancillas per operation; the beam-splitter approach can use one ancilla and supports gates such as CZL, CXL, ZZ(θ), and XX(θ).Source: §6 · p. 61
  • Bosonic logical-qubit entanglers can use shared instruction sets across codes, with excitation-preserving gates and controlled-SWAP constructions described for multiple code families.The passages describe Fock-space-ISA implementations for four-legged cat and binomial codes, and controlled-SWAP resources for dual-rail, four-legged cat, and binomial codes.Source: Binomial Code · p. 62, §6 · p. 62, Dual-Rail Code · p. 62

VI. COMPILATION METHODS

  • Hybrid CV-DV compilation combines analytical frameworks, superconducting-specific entangling-gate synthesis, numerical methods, and an extension of quantum signal processing.The section also treats product formulas and linear combinations of unitaries as pedagogical background for bosonic compilation.Source: §VI · p. 62

A. Introduction and Overview of Compilation Schemes

  • Hybrid compilation decomposes multi-qubit–oscillator unitaries into native primitives or single-qubit–oscillator operations combined with entangling gates, while accounting for platform-specific layouts and interactions.For superconducting cavity-mediated interactions, fast, high-fidelity SWAPs enable logical all-to-all connectivity despite near-neighbor physical constraints.Source: §VI · p. 62, A. Introduction and Overview of… · p. 63

B. Representations of Hybrid CV-DV Algorithms

  • Hybrid unitaries can be understood through either their Hermitian generators or the unitaries accessible from the Lie-bracket algebra of the control set.These are presented as complementary pictures for describing generic multi-qubit–oscillator operations.Source: B. Representations of Hybrid CV-DV… · p. 63

1. Hermitian Generator Picture

  • The Hermitian-generator picture represents multi-qubit, multi-oscillator generators using qubit Pauli operators and oscillator polynomials, which are truncated to finite degree in practice.The representation also connects oscillator transformations to phase-space flows and the stellar representation, including characterization of non-Gaussianity through polynomial zeros.Source: §1 · pp. 63–64

2. Unitary Picture

This section presents a unitary representation of hybrid qubit–oscillator dynamics and exact oscillator-mediated compilation techniques for physical-qubit and oscillator entangling gates. It covers long-range gates, flexible connectivity, parallel-mode optimizations, and ancilla-mediated synthesis of nonlinear oscillator interactions.

2. Unitary Picture

  • The unitary picture generalizes single-qubit quantum signal processing to coupled multi-qubit–oscillator dynamics, with finite polynomial and displacement-scale truncations enabling coarse-grained compilation.Its unitary is expressed using oscillator polynomials and momentum-boost and position-displacement operators; for a single qubit–oscillator, it provides a Fourier-series view of phase space.Source: §2 · p. 64

C. Exact Analytical Compilation of Entangling Gates

  • Exact analytical entangling-gate compilation is developed for superconducting hybrid hardware, with techniques inspired by trapped-ion methods and potentially extendable to other CV-DV platforms.Source: C. Exact Analytical Compilation of… · p. 64

1. Useful Primitives

  • Beam-splitter SWAP networks route oscillator states to synthesize nonlocal hybrid and oscillator–oscillator gates, while remaining agnostic to encoded bosonic information.Multiple SWAP strings can be parallelized with scheduling to avoid collisions; the text also notes improved speed and fidelity of bosonic SWAP operations in recent experiments.Source: Beam-Splitter SWAP Networks · pp. 64–65
  • Conditional displacements and beam-splitter networks can compile oscillator operations conditioned on high-weight Pauli operators, enabling many-qubit entangling gates such as Toffoli.Source: Synthesizing Hybrid Oscillator-Qubit… · pp. 65–66
  • Composed conditional displacements split an initial coherent state into 2^n phase-space blobs that may overlap and interfere, generally entangling the qubit and oscillator.The four-displacement sequence has qualitatively distinct phase-space behavior at α = 2, 1, and 0.3; at small α it approximates conditional squeezing.Source: Composing Conditional Displacements · pp. 66–67, Figure 22 · p. 68

2. Compilation of Entangling Gates between Physical Qubits

  • A phase-space compilation strategy targets entangling gates between physical qubits coupled to independent oscillators, using controlled displacements and beam-splitters; a closed loop also realizes an oscillator-state-independent Rz(θ).The proposed strategy uses the phase-space instruction set and can use remote beam-splitters synthesized by SWAP networks.Source: §2 · p. 67, Two-Qubit Entangling Gates · p. 69, Single-Qubit Gate · p. 69
  • Arbitrary-weight Pauli-conditioned displacements extend the two-qubit compilation strategy to complex n-qubit entangling gates, with Toffoli as an illustrative example.A more efficient Toffoli construction uses controlled-parity operations and displacements conditioned on both qubits being |1⟩.Source: Many-Qubit Gates · pp. 69–70
  • Distributing a geometric phase across n modes reduces each displacement’s size and duration by 1/√n, cutting multi-qubit gate time by √n and lowering photon number per cavity.The total photon number across modes remains unchanged, so the approach does not reduce photon-loss rate.Source: Strategies for Optimization · p. 70
  • Rectangular phase-space loops can shorten gate runtime by alternating short conditional with long unconditional displacements, trading that speedup for increased photon number and photon loss.The best balance depends on the experimental platform and compiled gate.Source: Strategies for Optimization · p. 71
  • Unlike MS and RIP gates’ fixed bus-mode couplings, the proposed scheme selectively couples chosen qubits to chosen oscillators, enabling flexible multi-qubit gates and parallel operations while avoiding direct-coupling cross-talk.RIP gates are proposed as superconducting analogs to MS gates; the proposed scheme uses oscillator dispersive coupling and beam-splitter routing instead of direct qubit coupling.Source: Advantages over MS and RIP Gates · pp. 71–72

3. Compilation of Entangling Gates between Oscillators

  • For oscillator entangling gates, an ancillary qubit mediates interactions bit by bit, enabling operations on one mode conditioned on Fock-subspace information from another.The method uses two SQR gates and two applications of a mode operation; the latter can be a conditional displacement, controlled phase-space rotation, or SNAP gate.Source: §3 · p. 72
  • Iterating these conditioned operations synthesizes nonlinear two-mode gates, with depth O(Nmax) generally, O(log(Nmax)) for certain functions, and O(Nmax^2) for a general two-mode function; the scheme extends to multiple oscillators.To leave the qubit and oscillators disentangled, the ancillary qubit must begin in a known Zanc eigenstate.Source: §3 · pp. 72–73

D. Systematic Compilation of Single-Qubit-Oscillator Unitaries

This section develops systematic compilation methods for single-qubit-oscillator unitaries, extending product-formula, LCU, and QSP techniques to hybrid CV-DV systems, with exactness reached only in the limit of infinitely many infinitesimal steps. It also covers hybrid QSP for oscillator quadratures, including polynomial-function unitaries and non-Abelian phase-space control that can improve cat-state preparation, while noting that its general approximation power lacks a complete theorem.

1. Trotter-Suzuki and Product Formulas

  • Trotter-Suzuki and BCH product formulas synthesize hybrid unitaries from Hamiltonian terms and commutators, while higher-order variants support arbitrary polynomial interactions using one auxiliary qubit.The methods are useful for conditional squeezing, Jaynes-Cummings interactions, and conditional cross-Kerr interactions; the latter is synthesized with error scaling as η^3.Source: §1 · p. 73, Example: Conditional Squeezing Operation · p. 74

2. LCU

  • Hybrid LCU uses either qubits or oscillators as auxiliary registers for linear-combination coefficients, with oscillator auxiliaries enabling continuous integrals of unitaries.The oscillator-controlled rotation scheme requires measurement post-selection; in the cat-state example, qubit measurement in the Z basis yields even or odd cats, while X-basis measurement yields coherent states.Source: §2 · p. 75

3. Quantum Signal Processing for Hybrid Systems

  • Hybrid QSP promotes classical rotation angles to commuting oscillator operators such as position or momentum, allowing qubit-based QSP techniques to apply without modification.The section also describes controlled displacement as a block-encoding of an oscillator operator, enabling QSP sequences to implement polynomial transformations.Source: §3 · pp. 75, 76, 78
  • Hybrid QSP deterministically prepares oscillator cat states, and the BB190,y protocol corrects spin-rotation errors caused by oscillator-position fluctuations.The comparison also includes a non-abelian QSP correction; the BB190,y protocol is described as efficient at correcting these errors.Source: §3 · pp. 76–77, Figure 29 · p. 78
  • Hybrid QSP implements oscillator unitaries that are exponentials of imaginary polynomial functions of position, momentum, or a fixed linear combination of both.The position-kick construction can be adapted to momentum or combined quadratures; hybrid QSP has also enabled a displacement-sensing interferometer with Heisenberg-like scaling for single-shot binary decisions.Source: §3 · p. 78
  • Non-Abelian QSP [456] processes non-commuting position and momentum through bivariable transformations, but its universality is supported by numerical evidence rather than a complete theorem.The momentum and position kicks set the phase-space approximation resolution, and proving efficient universality is challenging because the quadratures do not commute.Source: Non-Abelian QSP · pp. 78–79
  • For cat-state preparation, non-Abelian QSP uses momentum-dependent spin rotations to cancel position-fluctuation rotation errors to first order [456].Compared with the BB190,y scheme, it requires fewer operations and avoids large momentum boosts, improving robustness to ancilla and cavity decay and residual cavity anharmonicity; a similar circuit is used experimentally for GKP states and was independently found by numerical optimization.Source: Non-Abelian QSP · pp. 79–80
  • The abelian and non-Abelian QSP cat-state circuits are deterministic but approximate, whereas the LCU circuits are exact but non-deterministic.Fig. 29 compares the fidelities of the two QSP protocols, but the supplied passage gives no numerical values or winner.Source: Non-Abelian QSP · p. 80

E. Approximate Numerical Compilation

This section presents numerical optimization for approximate compilation of bosonic operations and CV-DV algorithm applications focused on qubit–oscillator state transfer. It covers single-variable and non-Abelian QSP approaches, including analysis of transfer complexity and fidelity and a reinterpretation of the DV-to-CV protocol as non-Abelian QSP operations.

E. Approximate Numerical Compilation

  • Numerical compilation optimizes layered universal-gate circuits by back-propagation until a batch circuit reaches target fidelity; the N = 2 protocols are reported as the best known numerically optimized protocols.The discussion also identifies barren plateaus—exponentially small gradient variance—as a failure mode; cited analyses find variance decays as E^-νM, with ν = 1 for shallow and ν = 2 for deep circuits.Source: E. Approximate Numerical Compilation · pp. 81–82

VII. ALGORITHMS AND APPLICATIONS

  • The section uses CV-DV instruction sets to motivate applications including arbitrary CV-to-DV and DV-to-CV state transfer, a state-transfer-based quantum Fourier transform, bosonic Hamiltonian simulation, and quantum random walks.It positions these applications as addressing challenges in developing and evaluating algorithms for CV systems.Source: §VII · p. 82

A. Arbitrary Qubit-Oscillator State Transfer

  • Two DV-to-CV protocols transfer qubit basis states to oscillator states localized at positions corresponding to their integer representations, using single-variable or non-Abelian QSP.The protocols leave the qubits and oscillator disentangled and can serve as subroutines for qubit-centric, oscillator-centric, and hybrid CV-DV abstract machine models.Source: A. Arbitrary Qubit-Oscillator State… · pp. 82–83

1. State Transfer Method 1: Single-variable QSP

  • Single-variable QSP first applies controlled displacements, then uses QSP sequences approximating square-wave filters to disentangle qubits from the oscillator.The displacement stage has time complexity O(∆2^n); the QSP stage scales as O(2^n log(1/ϵ)). Finite-width Gaussian basis states and imperfect polynomial approximations contribute transfer error, while greater spacing ∆ or stronger squeezing can improve fidelity.Source: §1 · pp. 83–84

2. State Transfer Method 2: Non-Abelian QSP

  • The non-Abelian QSP subsection reviews the protocol of Ref., bounds its performance, and presents it as a special case of non-Abelian QSP.The protocol transfers oscillator states to n qubits through repeated controlled-displacement operations; reversing it gives a DV-to-CV transfer protocol.Source: §2 · p. 85
  • The exact DV-to-CV output state is unphysical because it has infinite energy, but it can be well approximated by a squeezed vacuum.The output is encoded in displaced sinc states.Source: §2 · p. 86
  • With fixed coupling, the protocol uses O(n) gates but has O(λ^2n) total displacement and time complexity; sufficiently tunable, strong coupling can reduce this overhead.The total displacement is the source of the exponential time overhead under fixed coupling.Source: §2 · p. 86
  • Infidelity transferring |3⟩ to n qubits is ≈0.2 at n = 4 and ≈9 · 10−4 at n = 10; λ must be tuned for each n to maximize fidelity.Reference [514] derives fidelity bounds for both transfer directions, whereas gives no explicit tuning bounds.Source: §2 · p. 86
  • The [514] analysis attributes CV-to-DV infidelity to wavefunction support beyond |q| ≤ λ(2n−1) and variation, while DV-to-CV infidelity is primarily limited by out-of-range support.The derivative-dependent contribution is absent in the DV-to-CV direction because its associated approximation is naturally satisfied there [514].Source: §2 · pp. 86–87
  • The DV-to-CV protocol can be expressed as a product of n degree-1 non-Abelian QSP sequences, one acting between the oscillator and each qubit, and may generalize to higher-degree sequences.The re-expression identifies each qubit operation with a degree-1 sequence; using higher-degree sequences is proposed as a possible generalization.Source: §2 · p. 87

B. Quantum Fourier Transform

This section describes how oscillator Fourier evolution and qubit–oscillator state transfers implement the QFT, and surveys hybrid approaches to simulating bosonic, spin, and fermionic systems. It covers dynamics, state preparation, programmable operations and measurement, including bosonic-interaction and light–matter models, large-spin encoding, and photon-loss detection for that encoding.

B. Quantum Fourier Transform

  • The QFT implementation transfers a qubit state to an oscillator, applies free evolution as a continuous Fourier transform, and transfers the state back to qubits.It uses DV-to-CV state transfer and its inverse alongside bosonic operations, making it applicable to qubit-centric, oscillator-centric, and hybrid CV-DV abstract machine models.Source: B. Quantum Fourier Transform · p. 87

1. Continuous-Discrete Fourier Transform Correspondence

  • Periodic encoding of a discrete signal in a localized basis function makes its continuous Fourier transform a sum over the signal’s discrete Fourier transform, weighted by the basis function’s Fourier transform.Applying this correspondence to qubit amplitudes lets oscillator Fourier evolution produce coefficients from which the QFT can be recovered, provided the basis function has a suitable Fourier transform.Source: §1 · pp. 87–88

2. QFT Method 1: Single-variable QSP

  • O(2^n + a log(1/ϵ)) gates and O(2^n + a(∆ + log(1/ϵ))) time characterize the single-variable QSP QFT protocol.Its fidelity is maximized with small QSP error, large relative oscillator-basis spacing, and many ancilla qubits.Source: §2 · p. 88

3. QFT Method 2: Non-Abelian QSP

  • O(n + a) gates and fidelity F = 1 − O(1/2^a) characterize the non-Abelian QSP QFT protocol.With fixed qubit–oscillator coupling, its total displacement O(λ2^(n+a)) gives time complexity O(λ2^(n+a)); sufficiently tunable and strong coupling can reduce this time.Source: §3 · p. 88

C. Quantum Simulation

  • Hybrid oscillator–qubit hardware natively simulates bosonic degrees of freedom, while its universal instruction sets implement arbitrary Hamiltonian terms and enable local and nonlocal observable measurements.All-qubit implementations of bosonic operations can be costly; for example, one displacement on a qubit-encoded oscillator containing 10 bosons requires on the order of 100 two-qubit gates with Fock-binary encoding.Source: C. Quantum Simulation · p. 90

1. Dynamics

  • Digital quantum simulation synthesizes time evolution by decomposing it into gates, and hybrid oscillator–qubit hardware can facilitate Trotter-Suzuki formulas, linear combinations of unitaries, quantum walks, qubitization, and randomized approaches.For a Hamiltonian sum, Trotter-Suzuki reduces evolution to implementing each term’s unitary; higher-order formulas can improve error scaling at the cost of more gates.Source: §1 · p. 90

2. State Preparation

  • Ground-state preparation remains an open efficiency challenge; near-term methods include VQE and QAOA, while fault-tolerant approaches include QPE and QSP-based energy filtering.Adiabatic preparation can yield arbitrarily accurate ground-state approximations for gapped systems, but shrinking excitation gaps can make it inefficient; QPE can prepare excited states when the trial state has non-negligible overlap.Source: §2 · pp. 90–91

3. Observables

  • Hybrid hardware can measure a wide range of observables; parity-based binary search measures boson occupation with logarithmic rather than linear scaling in nmax.Conditional momentum boosts can estimate position quadrature, while beam-splitter conjugation maps hopping terms to photon-number differences.Source: §3 · p. 92, Number Operator · p. 92, Field Operators · p. 92
  • Conditional cross-Kerr gates can probe bosonic interactions through an ancilla without disentangling the modes, whereas terminal number measurements correlate n_i and n_j but collapse them to Fock states.Ancilla-based phase estimation could search for n_i n_j, but requires repeated fast, accurate synthesis of cross-Kerr unitaries at varied rotation angles.Source: Bosonic Interaction Terms · pp. 92–93

4. One Qubit, One Mode: The Rabi Model

  • Hybrid oscillator-qubit instructions simulate programmable Rabi dynamics using oscillator and qubit rotations plus conditional displacement; when g ≪ {ωm, ωq}, the rotating-wave approximation yields Jaynes-Cummings excitation exchange.At stronger coupling, the text identifies ultra-strong coupling at g ≳ 0.1{ωm, ωq} and deep strong coupling at g ≳ {ωm, ωq}, regimes predicted to display ground-state virtual excitations.Source: §4 · p. 93

5. Many Modes: The Bose-Hubbard Model and the Fractional Quantum Hall Effect

  • Digital oscillator-qubit simulation supports programmable bosonic Hamiltonians, including magnetic-field hopping via beam-splitter phases, and broader observable measurement than analog approaches readily provide.For Bose-Hubbard evolution, hopping maps to BS(−2J∆t, 0), the chemical-potential term to R(−µ∆t), and onsite interactions to a SNAP gate.Source: §5 · pp. 93–94

6. Large Spin Models: The Transverse-field Ising Model

  • Restricting two oscillators to N = 2S photons encodes spin S; beam splitters supply SU(2) rotations, while a SNAP nonlinearity extends control to SU(d).Spin measurements and tomography use photon-number measurements and beam-splitter rotations, independent of S.Source: §6 · p. 95
  • The transverse-field Ising example uses beam splitters for field terms and auxiliary-qubit conditional cross-Kerr gates for spin interactions; fixed photon-number encoding enables parity-based loss detection and post-selection.The spin-spin interaction terms are products of number operators of distinct oscillators and are implemented using conditional cross-Kerr gates approximately synthesized with BCH techniques.Source: §6 · pp. 95–96

7. Many Qubits, Many Modes: Z2 Lattice Gauge Theory

This section examines hybrid oscillator-qubit simulations, including lattice gauge theories, and approaches to representing and compiling bosonic dynamics on qubit hardware. It covers binary and unary encodings, scalable compilation of displacement operations, and encoding-related noise, error-detection, and gate-complexity considerations.

7. Many Qubits, Many Modes: Z2 Lattice Gauge Theory

  • At g/J = 0.1, a central boson spreads in a light cone, whereas at g/J = 5 it remains mostly trapped in the five-site Trotterized simulation.The initial state has one boson at the central site, other sites vacant, and Gauss-law value Gj = +1.Source: Figure 39 · p. 96, §7 · p. 97
  • The model’s hopping and field terms produce deconfined dynamics for |J/g| ≫ 1 and confined dynamics for |J/g| ≪ 1; its ground state remains confining for g/J > 0.Hopping flips an intermediary link’s field orientation, while the field energy favors aligned qubits; conserved Gauss-law generators can support error mitigation by post-selection.Source: §7 · pp. 96–97

8. Outlook: Fermion-Boson Problems

  • Hybrid CV-DV systems support fermion-boson simulation by encoding fermions in qubits or high-Q oscillators [621, 622] and compiling fSWAP gates [625, 626].The fSWAP is a SWAP with an additional minus sign on |11⟩ to account for fermionic exchange statistics.Source: §8 · p. 97

D. Quantum Random Walks

  • Repeated conditional displacements and qubit rotations implement oscillator-walker quantum random walks; experiments used phase-space operations to explore a topological effect, and oscillator measurements can instead drive qubit walks.The latter construction can produce multi-qubit dynamics through measurement back-action in an oscillator-parameterized configuration space.Source: D. Quantum Random Walks · pp. 98, 100
  • Interference from a local phase of π between coherent states |±ϵ⟩ yields an effective displacement δx−∼O(1), much larger than ϵ, but this is extremely rare; changing the qubit measurement axis can increase its probability.Source: D. Quantum Random Walks · p. 99

A. Mapping between Oscillators and Qubits

  • Because an oscillator has an infinite-dimensional Hilbert space, mapping it to qubits requires truncation; the section reviews alternative encodings for different operations.The discussed options include mappings such as the field-basis representation and Gray code.Source: A. Mapping between Oscillators and… · p. 100

1. Hardware-Efficient Mapping (Fock Binary Encoding)

  • Fock binary encoding maps L qubits to the lowest 2^L oscillator states, with Nmax = 2^L − 1; conversely, one resonator can replace L qubits for that truncated space.The Fock-basis matrices of a and a† carry over directly, but implementing their square-root matrix elements on qubit hardware is difficult.Source: §1 · pp. 100–101

VIII. ADVANTAGES OVER QUBIT-ONLY ARCHITECTURES

  • The hardware comparison centers on bosonic gate complexity: qubit-only systems struggle to implement bosonic operations despite logarithmic encoding cost, while hybrid systems natively provide oscillator raising and lowering operators.The section frames gate-count comparison as an advantage of hybrid CV-DV hardware over qubit-only hardware.Source: §VIII · p. 100, §1 · p. 101

1. Hardware-Efficient Mapping (Fock Binary Encoding)

  • With γ>κ, the binary qubit encoding’s mean boson-number loss follows oscillator damping with κ replaced by γ, even when its state-leaving rate is bounded by γL.A 103-photon cat state in the cited experiment had photon-loss coherence time ∼35 µs, versus ∼11 µs for its approximately 10-qubit representation.Source: §1 · pp. 101–102
  • A rigorous noisy-oscillator versus many-qubit comparison must account for error times relative to gate durations; an amplitude-damping analysis remains future work.Ref. found regimes where noisy qudits outperform logarithmically many qubits under dephasing, while the authors expect an oscillator advantage but leave amplitude damping for future analysis.Source: §1 · p. 102

2. Mapping with Redundancy (Unary Encoding)

  • A unary encoding uses Nmax+1 qubits with one excited qubit marking photon number, making some bit flips detectable while leaving flips to the all-zero state undetectable.The redundant state space enables simple repetition encoding for bit-flip correction.Source: §2 · p. 102

B. Compilation of Bosonic Operators into and from Standard Qubit Gate Sets

  • A Newton-iteration compilation technique computes square-root matrix elements for arbitrary bosonic operations and estimates displacement CNOT cost against brute-force compilation.For a cutoff K = 2^n − 1, the displacement strategy uses n qubits plus a helper qubit to encode adjacent-level coefficients for controlled rotations.Source: B. Compilation of Bosonic Operators… · pp. 102–103
  • Bosonic operators can become high-weight qubit operations, motivating native oscillator hardware for bosonic simulations; Gray-code encoding does not remove this complexity.In the example, the destruction operator includes a weight-8 expression and 15 similar terms.Source: B. Compilation of Bosonic Operators… · p. 103

2 Provide a Trotter decomposition of the form

The section examines hybrid CV-DV architectures and their applications, including the costs of simulating bosonic displacement operations with qubit-only hardware versus native bosonic operations. It also outlines compilation methods and open questions across theory, software, hardware architecture, and fault tolerance.

2 Provide a Trotter decomposition of the form

  • The binary-Fock method maps neighboring Fock-state amplitudes onto a main qubit using an LSB-controlled CNOT and controlled incrementer, applies a rotation, then transfers amplitudes back.A binary register represents the truncated bosonic state; an incrementer controlled on the main qubit rearranges each even-odd pair into a two-dimensional subspace for the operation.Source: Mapping the Bosonic State to… · pp. 104–105
  • The method does not account for the generally nontrivial preparation of the encoded qubit state; its resource estimate also includes ancillary registers for incrementers and controlled swaps.The stated register total is 3n qubits, including the system register and main qubit.Source: Mapping the Bosonic State to… · pp. 104–105
  • Reversible Newton iterations compute the required square roots, with the controlled-rotation cost negligible in most settings compared with the O(n^2) Toffoli cost of square-root computation.Newton iterations converge quadratically, so the iteration count scales doubly logarithmically with the desired error tolerance.Source: Computation of the Square-Roots · p. 105, Controlled Rotations (CR) · p. 106
  • A single Trotter step uses (270m + 126)n^2 + (144 + 228m)n − 12m − 22 two-qubit operations, where n is register size and m is the Newton-iteration count.The estimate combines amplitude-transfer, square-root, and controlled-rotation costs.Source: Cost of simulating a single exponential · p. 106
  • More than a million CNOT gates are estimated even for small cutoffs, increasing approximately linearly with cutoff; Fig. 41 contrasts the estimate with brute-force upper bounds.The authors caution that their Trotter-step estimate is likely grossly overestimated and that the gap does not establish a fundamental advantage; alternative encodings can trade gate count for qubit count.Source: Cost of Simulating the Displacement… · p. 107, Figure 41 · p. 107, C. Efficiency of Bosonic Instruction… · p. 108

C. Efficiency of Bosonic Instruction Set Relative to Qubit-only Simulations

  • 36,788 CNOT gates are estimated for one displacement at Fock cutoff 63 with six qubits and m=2, versus one native bosonic instruction-set operation.The two Newton-Raphson iterations correspond to errors on the order of 10^-4; qubit damping errors are an additional complication.Source: C. Efficiency of Bosonic Instruction… · p. 108

IX. CONCLUSIONS AND OUTLOOK

  • The work presents hybrid processor architectures, compilation techniques, error-correction strategies, and applications including state transfer, quantum Fourier transforms, bosonic error correction, and matter simulation.For matter simulation, hybrid digital architectures may avoid costly arithmetic by using equivalent native bosonic operations.Source: §IX · p. 108

A. Challenges and Opportunities for AMMs and ISAs

  • The paper frames hybrid AMMs and ISAs around five concerns: bosonic-state representation, universal bosonic gates, hardware and architectures, compilation, and processor use cases.Bosonic information may be represented in position or momentum variables or in the Fock basis, complicating mapping between oscillators and qubits.Source: A. Challenges and Opportunities for… · p. 108
  • Hybrid hardware supports Fock-space, phase-space, and sideband ISAs, whose instructions can compose universal computation and enable continuous-variable gate-set tomography.The proposed physical interactions are discussed for superconducting, trapped-ion, and neutral-atom platforms; bosonic error-correcting codes may address dissipation and noise.Source: A. Challenges and Opportunities for… · p. 109
  • The paper defines qubit-centric, oscillator-centric, and hybrid oscillator-qubit universal AMMs for user-facing abstractions or transpiler-visible hardware, supporting resource estimation.The models specify either the abstract Hilbert space or available physical degrees of freedom, connectivity, and gates.Source: A. Challenges and Opportunities for… · p. 109
  • The authors report evidence of near-term hybrid advantage for qubit-based simulation methods and extend quantum signal processing to non-Abelian operations and Fourier-transform applications.In the non-Abelian extension, qubit rotation angles depend on non-commuting bosonic position and momentum variables.Source: A. Challenges and Opportunities for… · p. 109

B. Open Questions and Directions for Future Work

  • Future work should identify applications where hybrid hardware outperforms qubit-only or classical approaches, develop variational CV-DV algorithms, and complete non-Abelian QSP theory.Potential applications include coupled nuclei-electron processes and light-matter interactions; classical signal-processing methods may guide quantum signal processing.Source: B. Open Questions and Directions for… · p. 110
  • CV gate synthesis still lacks polylogarithmic error scaling, optimal sequences for arbitrary non-commuting-variable QSP, and a complete complexity theory for hybrid circuit synthesis.The Trotter-Suzuki approaches considered here and in offer sub-polynomial scaling with error tolerance, whereas Solovay-Kitaev [647, 648] achieves polylogarithmic scaling over a discrete gate set.Source: B. Open Questions and Directions for… · pp. 110, 112
  • Hybrid architectures need memory hierarchies and more capable quantum ALUs, while compilers should map idle operations to memories and exploit native bosonic gates.The paper also calls for application-tailored resource-estimation software and more advanced primitives such as modular exponentiation and matrix arithmetic.Source: B. Open Questions and Directions for… · pp. 110–112
  • Reliable fault tolerance remains challenging: CV channel tomography lacks a general high-dimensional method, auxiliary qubits complicate bosonic error correction, and oscillator protection against photon loss remains open.The sole cited oscillator-code attempt corrects classical displacement errors, and code families based on that noise model lack a threshold.Source: B. Open Questions and Directions for… · p. 111
  • Oscillator-based codes can quadratically reduce magic-state-distillation overhead, while oscillator-encoded qudits can enable several-orders-of-magnitude improvements in distillation resource overhead.The paper also notes that high-fidelity qudits via oscillator error correction could further improve distillation.Source: B. Open Questions and Directions for… · p. 111

Appendix A: Physical Implementations

This appendix describes physical control mechanisms for hybrid oscillator–qubit systems in superconducting circuits and trapped ions, linking hardware interactions to conditional gates and instruction sets. It also explains how a spare qubit prepared as an ancilla in |+⟩ can implement an unconditional displacement through a conditional displacement.

b. Conditional Cavity Gates

  • An echoed double-drive sequence cancels unwanted Stark-shift terms to implement conditional cavity displacement, while amplified two-mode squeezing enables fast conditional entangling gates in the weak-dispersive regime.The echoed-conditional-displacement construction uses opposite large displacements with an ancilla flip; the two-mode squeezing gate’s speed scales with χt sinh 2α.Source: b. Conditional Cavity Gates · pp. 115–116

c. Cavity-Controlled Qubit Gates

  • Driving the qubit at a photon-number-shifted transition frequency enables photon-number-selective qubit rotations, which can be combined into SNAP gates.The selection assumes χ is sufficiently large relative to the drive amplitude; slow pulses use total time Tgate > 2π/χ, and the lab-frame gate includes a conditional cavity rotation that can be canceled or tracked in software.Source: SQR Gates · pp. 116–117

2. Trapped Ion Quantum Processors

  • In trapped-ion processors, mechanical modes provide bosonic modes, typically at MHz frequencies, and laser or microwave interactions support phase-space instructions.Multi-chromatic laser drives can simultaneously couple many qubits to many mechanical modes.Source: §2 · p. 117

a. Qubit Rotations

  • Trapped-ion qubit rotations can use microwave-driven hyperfine transitions, with near-field gradients enabling individual addressing, or stimulated Raman transitions driven by detuned laser beams.Stimulated Raman transitions generally require relatively strong laser intensities because they are second-order processes.Source: a. Qubit Rotations · p. 117

b. Sideband Gates

  • A red sideband gate combined with an arbitrary XY-plane single-qubit rotation is universal, while red and blue sideband gates together can produce conditional displacement.Trapped-ion SNAP gates are not known to have been realized because dispersive qubit–mode coupling is absent, though sideband transitions can synthesize such coupling.Source: b. Sideband Gates · pp. 117–118

c. Conditional Displacements

  • Simultaneous equal-strength red and blue sideband drives yield a displacement conditioned on the qubit’s Pauli-X eigenstate, and a Hadamard conjugation converts it to the controlled-displacement gate.The combined interaction is proportional to σx x.Source: c. Conditional Displacements · p. 118
  • Preparing an unused qubit in |+⟩ as an ancilla and applying a conditional displacement realizes an unconditional displacement.The ancilla qubit is not otherwise involved in the circuit.Source: c. Conditional Displacements · p. 118

3. Neutral Atom Quantum Processors

This section covers neutral-atom oscillator hardware and gate implementations, including oscillator decompositions, noise models, ISA cross-compilation and universality, and alternative architectures with their operational constraints. It also develops QSP-based SNAP and superparity compilation and compiles phase-space ISA gates from Fock-space ISA primitives.

3. Neutral Atom Quantum Processors

  • Neutral-atom tweezers provide up to three motional modes per atom, avoiding mode crowding at the cost of reduced mode connectivity and lower, tens-of-kHz frequencies.The tweezer potential supports Fock states up to 100–1000 and has a tunable Kerr nonlinearity of 0.1%–1% of the mode frequency; combined qubit-boson operations and motional coherence exceeding 100 ms have been demonstrated.Source: §3 · p. 118

b. Multi Qubit Gates

  • Neutral-atom qubits use Raman or microwave-plus-AC-Stark-shift single-qubit gates, Rydberg-blockade CZ and multi-controlled Z gates, and tweezer-potential modulation for universal single-mode control.During blockade entangling gates, atoms do not simultaneously occupy Rydberg states, preventing those gates from affecting motional states; free evolution under the tweezer potential implements phase-space rotations.Source: a. Single Qubit Gates · pp. 118–119, b. Multi Qubit Gates · p. 119, c. Single Mode Gates · p. 119

d. Multi Mode Gates

  • Inter-mode beam splitters and two-mode squeezing can be synthesized by concatenating qubit-controlled operations with the BCH formula, by Rydberg-state dipolar forces, or through tunneling between tweezers.Different phases and Trotterization can isolate beam-splitter and two-mode-squeezing terms in the BCH construction.Source: d. Multi Mode Gates · p. 119

e. Sideband Gates

  • Sideband gates couple qubits to motion in the Lamb-Dicke regime, with sideband resolution requiring Ωη < ωr; optical lattices and bosonic atoms offer alternative architectures.Lattice shaking can excite oscillators and superlattices can engineer beam-splitter tunneling; bosonic-atom approaches naturally suit number-conserving problems, while displacements may be engineered through coupling to a Bose-Einstein condensate.Source: e. Sideband Gates · p. 119, f. Alternative Architectures · p. 119

Appendix B: Bloch-Messiah Decomposition

  • The Bloch-Messiah decomposition expresses symplectic transformations as beam splitters surrounding single-mode squeezers, enabling two-mode squeezing synthesis and nonlocal beam splitters via SWAP networks.The SUM gate’s symplectic representation preserves mode-operator commutation relations, while the two-mode-squeezing transformation correlates output position quadratures and anticorrelates momentum quadratures for large r and φ = 0.Source: Appendix B · p. 120, §1 · pp. 120–121, §2 · p. 121

Appendix C: Oscillator Noise Models

  • Bosonic channels are CPTP maps with formally infinite Kraus dimension; amplitude damping includes photon-loss and no-jump backaction, and truncation can be made trace-preserving by modifying the no-jump operator.Bosonic dephasing randomizes phase, preserving photon-number diagonal elements while reducing off-diagonal magnitudes; tweezer amplitude or phase fluctuations can also cause mode dephasing.Source: §1 · p. 121, §2 · p. 122, §3 · p. 122

4. Heating

  • Ion-trap heating is driven by electric-field noise with debated microscopic origins, whereas neutral-atom-array heating mainly arises from atom movement and can be mitigated by slower transport or fixed architectures with SWAP gates.Typical ion heating rates range from ∼10^1 phonons/sec at 4 K to 10^3 phonons/sec at room temperature for a 1 MHz trap and 50 µm distance; other than center-of-mass modes, heating is generally lower.Source: §4 · pp. 122–123

Appendix D: Universality of ISAs and Cross Compilation

  • Cross-compilation among phase-space, Fock-space, and sideband ISAs establishes single qubit-oscillator universality and, with compiled multiqubit entangling gates, recursive universality for multiqubit-oscillator systems.The phase-space ISA can simulate a dispersive Hamiltonian with an enhanced effective coupling to compile SNAP gates, while the reverse direction compiles controlled displacements using the Fock-space ISA.Source: Appendix D · p. 123, §1 · p. 123, a. Fock-Space ISA from Phase-Space… · p. 123

b. SNAP Gate Compilation by Quantum Hamiltonian Simulation

  • Rational choices of θ and δ enable QSP superparity operations that phase-mark Fock states by photon number modulo ℓ, with worst-case phase error scaling as 1/N.These operations nearly disentangle the ancilla, but phases on states meant to receive zero can be erroneous; the stated error excludes errors in the QSP realization of dispersive evolution. A SNAP gate can be built from two selective qubit rotations, with the second imparting the desired Fock-state phase.Source: b. SNAP Gate Compilation by Quantum… · p. 124, Table D.1 · p. 124

c. Phase-Space ISA from Fock-Space ISA

  • The phase-space ISA can be compiled from the Fock-space ISA by reducing selective rotations to qubit rotations and synthesizing conditional displacements using controlled parity.An SQR gate with θ_n = nπ, conjugated by Ry(π/2) rotations, realizes controlled parity; hybrid oscillator-qubit synthesis then realizes conditional displacement.Source: c. Phase-Space ISA from Fock-Space ISA · pp. 124–125

2. Compilation between Sideband ISA and Phase-Space ISA

This section develops compilation across sideband, phase-space, and Fock-space instruction sets and presents universality constructions for multi-qubit–oscillator systems and oscillator-only control. It also describes cavity-conditioned qubit-flip measurements for selecting photon-number states and composing photon-number tests, including thresholds, into efficient binary search.

2. Compilation between Sideband ISA and Phase-Space ISA

a. Sideband Gates from Fock-Space ISA

  • For finite photon cutoff n_max, sideband gates can be compiled directly from SQR gates and unconditional displacements, although finding an efficient shallow circuit remains open.The construction is formally correct and convergent; higher-order product formulas could improve its accuracy.Source: a. Sideband Gates from Fock-Space ISA · p. 126

b. Fock-Space ISA from Sideband Gates

  • SQR gates and unconditional displacements can be synthesized from sideband gates by composing the subspace rotations constructed in Ref..The cited constructions provide arbitrary clean SU(2) rotations within qubit–Fock-level subspaces and between oscillator Fock levels.Source: b. Fock-Space ISA from Sideband Gates · p. 126

4. Universality of Multi-Qubit-Oscillator Systems

5. Oscillator-only Universality: Controllability and the Cubic Instruction Set

  • Gaussian operations augmented by a single cubic term such as x^3 generate arbitrary-order oscillator polynomials and enable universal oscillator control.The cubic commutator raises polynomial degree, unlike quadratic terms; multimode control additionally requires beam-splitter gates between physically adjacent oscillators.Source: §5 · pp. 127–129
  • For infinite-dimensional oscillators, an infinite generated algebra alone does not ensure controllability if symmetries such as photon-number parity commute with every generator.The section defines controllability in terms of generating arbitrary bounded-degree polynomial Hamiltonians and notes that this subtlety can be checked and avoided.Source: §5 · p. 128

3. Cavity Photon Number Measurement

  • Measuring the ancilla after a cavity-conditioned flip collapses the state according to the measurement outcome, with qubit readout inferred from a dispersive shift in cavity frequency.The readout cavity may be separate from the cavity being measured.Source: §3 · p. 130
  • Composing cavity-conditioned bit flips implements a projector for any binary-valued selection of photon numbers.Two simultaneous drive tones can flip the qubit for either of two photon numbers; the construction generalizes to a product of gates.Source: §3 · p. 130
  • Photon-number conditions can drive threshold tests, conditional cavity displacements, and an efficient binary search for the photon number.For example, the ancilla can be flipped when the photon number is less than or greater than a selected value, then used to condition a cavity displacement.Source: §3 · p. 130

a. Cavity Photon Number Parity Measurement

The section reviews methods for oscillator-state characterization and preparation, from ancilla-based cavity measurements and photon-number extraction to compilation approaches and practical limits on tomography. It also discusses qubit QSP/QSVT and composite-pulse error robustness.

a. Cavity Photon Number Parity Measurement

  • A controlled parity rotation entangles an ancilla with cavity photon-number parity for QND readout; parity also supports cavity tomography and bosonic-code error correction.The practical parity pulse sequence [237] uses dispersive conditional cavity rotation.Source: a. Cavity Photon Number Parity… · p. 130, Figure 43 · p. 131

4. Cavity Photon Number ‘Super-Parity’ Measurement

  • Commuting super-parity measurements recover all photon-number bits in O(log Nmax) steps with classical feed-forward, exponentially more efficiently than sequentially testing each number.An alternative binary search starts from the most-significant bit and does not require active feed-forward.Source: §4 · p. 131, Figure 44 · p. 132, Figure 45 · p. 132

5. Homodyne Detection

  • Balanced homodyne detection measures a phase-selected quadrature; measurements at different phases reconstruct the Wigner function by Radon transform, while stationary microwave modes can instead map onto an auxiliary qubit.At microwave frequencies, amplifier noise complicates state reconstruction.Source: §5 · p. 133, Figure 46 · p. 133

6. Wigner and Characteristic Function Tomography

  • A Wigner-function value at β is obtained by displacing that phase-space point to the origin and measuring photon-number parity expectation.Parity alone suffices for Wigner tomography, although the full photon-number distribution can improve robustness.Source: §6 · pp. 133–134
  • Full Wigner-function reconstruction requires estimating a grid of phase-space values, and its measurement-setting count scales exponentially with oscillator number unless derivative norms shrink.The passage describes interpolation from grid points and notes that prior knowledge about the Wigner function may enable more efficient reconstruction.Source: §6 · p. 134

7. The Husimi Q-Function

  • The Husimi Q-function is always positive, making it convenient in some cases but less suited to studying quantum-state nonclassicality.The P-function is described as unnormalizable and difficult to use.Source: §7 · p. 135

8. Calibrating Controlled Displacements.

  • A controlled-displacement calibration produces Berry-fringe qubit oscillations with rotation angle φ = −4|α|^2, allowing the displacement α to be inferred from qubit measurements.An alternative uses random displacements followed by inversion and measures vacuum probability or photon-number distribution to assess displacement noise and errors.Source: §8 · p. 135, Figure 47 · p. 135

Appendix F: Comparison between Analytical and Numerical Compilation

  • Appendix F compares numerically optimized circuits with analytical approximate compilations using circuit depth or number of layers as a figure of merit.The supplied passage introduces the comparison but does not provide its results.Source: Appendix F · p. 135
  • Numerical compilation achieves lower circuit depth than Trotterization for Fock-state preparation, but requires longer circuit duration at equal infidelity.The numerical circuits were optimized for depth rather than duration.Source: Trotterization · p. 136, Figure 49 · p. 136
  • QSP-based squeezing [456] matches numerical methods in duration: at comparable Tcircuit ≈7.6, it achieves 8.5 dB squeezing, −8.40 dB anti-squeezing, and 0.9978 fidelity, versus ’s −9.9 dB and 0.99 fidelity.The cited numerical result also reports 8.5 dB squeezing; its optimization targeted circuit depth, unlike the duration-optimized QSP scheme.Source: Trotterization · pp. 136–137, Figure 49 · p. 136

Appendix G: Brief Review of QSP, QSVT Formalism for Qubits

  • QSP transforms a signal encoded in a 2 × 2 unitary, while QSVT [488] extends polynomial transformations to singular values of larger, potentially rectangular block-encoded matrices without explicit SVD.The passage notes that performing an SVD of an n-qubit operator has O(23n) cost.Source: Appendix G · p. 137

1. Intuition from Composite Pulses

  • For small rotation-angle errors, the BB1 composite pulse gives ⟨σz⟩≈1 − 3.7ϵ^6, indicating an error of O(ϵ^3).Its additional rotations cancel small over- or under-rotation errors when the error varies slowly during the pulse sequence.Source: §1 · pp. 137–138

2. Qubit QSP

Qubit QSP constructs constrained polynomial transformations from rotation sequences, and this work extends QSP to bosonic systems for hybrid oscillator-qubit compilation. The appendix details two QFT constructions that transfer states between qubits and an oscillator, apply oscillator Fourier evolution, and transfer the result back.

2. Qubit QSP

  • QSP extends from single-qubit transformations to bosonic systems, enabling its use for compilation on hybrid oscillator-qubit hardware.The attainable polynomial transformations obey parity constraints, and efficient classical algorithms can find the required QSP angles.Source: §2 · p. 138

1. Details on QFT Method 1: Single-Variable QSP

  • The single-variable QFT method transfers a periodic qubit encoding to oscillator states, symmetrizes it, applies oscillator Fourier evolution, and transfers it back.The construction appends ancillas to make coefficients periodic and improve fidelity with the exact QFT.Source: §1 · p. 139
  • The Dirichlet-kernel limit yields an oscillator state concentrated at q_l with infidelity O(1/2a), carrying the phase that reproduces QFT coefficients.The final qubit output is nearly the QFT, with an additional infidelity O(σ/∆) in the state-transfer step.Source: §1 · p. 140

2. Details on QFT Method 2: Non-Abelian QSP

  • The non-Abelian method changes to the |ϕ_s⟩ basis using local operations, transfers the state to an oscillator, applies Fourier evolution, and reverses the transfer.Ancillas make the coefficients periodic, and the final qubit state is nearly the QFT up to a phase factor e^(-iπx/2).Source: §2 · pp. 141–142
  • In the non-Abelian method, the finite-a Dirichlet-kernel approximation has infidelity O(1/2a) before the oscillator and qubits decouple.The resulting qubit state is nearly the QFT in the |ϕ_s=2^al⟩ basis.Source: §2 · p. 141

Related papers

Loading 2407.10381v3…