⚛️ quantum physics

Estimating the performance boundary of Gottesman-Kitaev-Preskill codes and number-phase codes

This paper establishes a quantitative performance boundary between Gottesman-Kitaev-Preskill and number-phase bosonic codes under general photon loss and dephasing noise, revealing that the crossover in their relative advantages occurs when dephasing is approximately two orders of magnitude weaker than loss, thereby providing a practical methodology for selecting and optimizing these encodings in realistic experimental environments.

Original authors: Kai-Xuan Wen, Dong-Long Hu, Shengyong Li, Ze-Liang Xiang

Published 2026-03-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Kai-Xuan Wen, Dong-Long Hu, Shengyong Li, Ze-Liang Xiang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to send a delicate message across a stormy ocean. The message is your quantum information, and the boat carrying it is a harmonic oscillator (a fancy physics term for a vibrating system, like a spring or a light wave).

The problem? The ocean is full of two types of storms:

Photon Loss: Waves knocking things off the boat (energy disappearing).
Dephasing: The wind spinning the boat around randomly (losing the direction or "phase" of the message).

To survive, you need a special lifeboat design. In the world of quantum computing, scientists have invented two main types of lifeboats: GKP codes and NP codes.

This paper is essentially a massive survival guide that tells you: "If the storm looks like X, use Boat A. If it looks like Y, use Boat B."

Here is the breakdown of their findings using simple analogies:

1. The Two Lifeboat Designs

The GKP Code (The Grid Anchor)

How it works: Imagine drawing a perfect grid on the ocean surface. Your message is a tiny dot sitting exactly in the center of a grid square.
Strength: If a small wave pushes your dot slightly (a "displacement" error), the grid acts like a magnet. It pulls your dot back to the center of the square. It is incredibly good at handling Photon Loss (waves knocking things away).
Weakness: If the wind spins the whole ocean map (dephasing), the grid gets confused. The "center" moves, and the dot gets lost.

The NP Code (The Compass Rose)

How it works: Imagine your message isn't a dot on a grid, but a spinning compass needle. The code is designed so that the needle points in specific, safe directions (like North, East, South, West).
Strength: If the wind spins the compass slightly (a "rotation" error), the code snaps the needle back to the nearest safe direction. It is incredibly good at handling Dephasing (spinning).
Weakness: If a giant wave knocks the whole boat off course (photon loss), the compass might spin too far to recover.

2. The Big Discovery: The "Tipping Point"

For a long time, scientists knew both boats were good, but they didn't know exactly when to switch from one to the other. They were guessing.

This paper ran thousands of computer simulations (using a super-smart AI called an "Evolutionary Algorithm") to find the exact boundary between the two.

The Result:
They found a "Tipping Point" in the storm.

Scenario A: If the "spinning wind" (dephasing) is 100 times weaker than the "knocking waves" (photon loss), the GKP Grid is the winner. It handles the waves perfectly.
Scenario B: If the "spinning wind" gets stronger (even just a little bit relative to the waves), the NP Compass takes over. It handles the spinning much better.

The Analogy:
Think of it like choosing between a sailboat and a submarine.

If the problem is just rough waves (loss), a sailboat (GKP) is fine.
But if the water starts swirling violently (dephasing), you need a submarine (NP) that can stabilize itself.
The paper tells you the exact moment the waves get so turbulent that you must switch from the sailboat to the submarine. That moment happens when the turbulence is about 1% of the wave height.

3. How They Did It (The "Magic" Tool)

Usually, testing these codes is like trying to find the best route through a maze while blindfolded. It takes forever.

The authors built a super-fast GPS for this maze.

They used GPUs (the powerful chips in gaming computers) to calculate how well a code would survive a storm in just a few seconds.
They used an Evolutionary Algorithm (like natural selection). They created thousands of "virtual boats," let the worst ones crash, and kept the best ones to "breed" new, better designs.
This allowed them to map out the entire ocean and draw a clear line on the map showing exactly where one boat stops being the best and the other starts winning.

4. Why This Matters

Before this paper, engineers building quantum computers had to guess which code to use. They might have picked the wrong one, wasting money and time.

Now, they have a rulebook:

Measure your environment (How much is the system losing energy? How much is it spinning?).
Check the ratio.
If the spinning is tiny compared to the loss: Use the GKP code.
If the spinning is significant: Use the NP code.

Summary

This paper is the ultimate weather forecast for quantum computers. It tells us that while both GKP and NP codes are excellent, they are specialists. GKP is the master of the waves, and NP is the master of the wind. The paper draws the line in the sand, telling us exactly when the wind becomes strong enough that we need to switch from the wave-specialist to the wind-specialist to keep our quantum information safe.

1. Problem Statement

Bosonic quantum error-correcting (QEC) codes encode logical information in the infinite-dimensional Hilbert space of a harmonic oscillator. Two primary paradigms exist:

Gottesman-Kitaev-Preskill (GKP) codes: Encode information in a phase-space lattice protected by translational symmetry (position and momentum). They are robust against small displacement errors (dominant in photon loss).
Number-Phase (NP) codes: (Including cat and binomial codes) Encode information in a lattice within the number-phase domain, protected by discrete rotational symmetry. They are robust against phase-rotation errors (dominant in dephasing).

The Core Challenge: While both codes are well-studied individually, it remains unclear under what specific physical noise conditions (combinations of photon loss and dephasing) one code intrinsically outperforms the other. Previous studies often relied on fixed lattice geometries or specific noise assumptions, failing to map the global performance landscape or quantitatively define the "crossover" boundary where the optimal encoding strategy shifts.

2. Methodology

The authors developed a scalable, systematic framework to optimize code parameters and benchmark performance across a broad range of noise conditions.

A. Optimization Framework

Objective: Maximize the near-optimal channel fidelity ( $\tilde{F}^{opt}$ ) under a generalized noise channel combining photon loss ( $\gamma$ ) and pure dephasing ( $\kappa$ ).
Metric: Instead of computationally expensive semidefinite programming (SDP) required for exact optimal fidelity, they use the near-optimal fidelity derived from the QEC matrix. This metric satisfies a rigorous two-sided bound ( $1 - \tilde{F}^{opt}/2 \le 1 - F^{opt} \le 1 - \tilde{F}^{opt}$ ) and allows for direct evaluation via matrix operations.
Algorithm: They employ the Covariance Matrix Adaptation Evolution Strategy (CMA-ES), a derivative-free evolutionary algorithm suitable for high-dimensional, non-convex, and multimodal parameter spaces.
Acceleration: The QEC matrix calculations are implemented using PyTorch with GPU acceleration, reducing fidelity evaluation time from hours (convex optimization) to seconds, enabling large-scale parameter sweeps.

B. Code Parameterization

GKP Codes: Optimized over lattice shape parameters ( $\alpha, \beta$ ) and the energy envelope width ( $\Delta$ ). The lattice geometry is allowed to deviate from the ideal hexagonal structure to adapt to noise.
NP Codes: Optimized over a generalized parameter set $T_{NP} = \{f, s, r, n\}$ $T_{N P} = {f, s, r, n}$ , where:
- $s, f$ : Determine the lattice geometry (spacing and skewness).
- $r, n$ : Parameters of a Gaussian envelope (squeezing and mean photon number) used to approximate finite-energy states.
- This framework generalizes standard cat and binomial codes, allowing for broader optimization.

C. Noise Model

The system evolves under a Lindblad master equation with simultaneous photon loss and dephasing. The performance is evaluated using the dimensionless parameters $\gamma t$ (loss strength) and $\kappa t$ (dephasing strength).

3. Key Results

A. Optimal Parameter Trends

GKP Adaptation: Under combined noise, the optimal GKP lattice remains close to the hexagonal configuration but exhibits slight rotations and deviations in $\alpha$ and $\beta$ when both loss and dephasing are strong. The energy parameter $\Delta$ increases (reducing mean photon number) as dephasing strength increases, indicating a trade-off where high photon numbers penalize phase stability.
NP Adaptation: The optimal NP lattice geometry ( $f$ ) shifts away from the highly symmetric diamond lattice ( $f=0.5$ ) in high-loss/weak-dephasing regimes to increase the effective code distance against excitation loss. The squeezing parameter $r$ becomes negative in high-loss regimes to reduce phase uncertainty.

B. Performance Boundary

The authors mapped the near-optimal fidelity for both codes across $\gamma t \in [0.01, 0.15]$ and $\kappa t \in [0.0001, 0.0072]$ .

GKP Dominance: GKP codes significantly outperform NP codes in regimes dominated by photon loss.
NP Dominance: NP codes exhibit superior resilience in regimes dominated by dephasing.
The Crossover: A sharp performance boundary was identified. The crossover occurs when the dephasing strength ( $\kappa t$ ) is approximately two orders of magnitude smaller than the photon loss strength ( $\gamma t$ ).
- Below this boundary (higher dephasing relative to loss), NP codes are superior.
- Above this boundary (lower dephasing relative to loss), GKP codes are superior.
Sensitivity: GKP codes are found to be substantially more sensitive to dephasing than NP codes are to photon loss.

C. Fidelity Scaling

The study observed that fidelity degradation scales differently with noise types: it is concave with respect to photon loss but convex with respect to dephasing. This difference arises because dephasing scales with the square root of the noise strength relative to displacement errors.

4. Significance and Contributions

Quantitative Boundary Definition: The paper provides the first quantitative estimation of the performance boundary between GKP and NP codes, offering a concrete rule of thumb (the 100:1 loss-to-dephasing ratio) for experimentalists to select the appropriate encoding.
Methodological Advancement: The work establishes a practical, extensible methodology for benchmarking bosonic codes. By combining GPU-accelerated near-optimal fidelity evaluation with CMA-ES, it overcomes the computational bottlenecks of previous SDP-based approaches, allowing for the optimization of complex, high-dimensional code parameters.
Insight into Symmetry Transition: The results suggest that near the performance boundary, the optimal encoding may require a transition between pure translational symmetry (GKP) and pure rotational symmetry (NP). This hints at the potential existence of hybrid bosonic codes with mixed symmetry characteristics that could outperform both standard families in the crossover regime.
Experimental Guidance: The findings provide direct guidance for the deployment of bosonic QEC in realistic hardware (e.g., superconducting cavities), where noise is rarely purely one type. It enables the selection of code families and the tuning of parameters (like lattice geometry and squeezing) based on the specific noise profile of the device.

In summary, this work bridges the gap between theoretical code design and practical implementation by systematically mapping the performance landscape of bosonic codes, revealing a distinct operational boundary driven by the competition between displacement and rotation errors.