Error correction for qudits

Short answer: Qudits are quantum systems with d levels instead of two, and they offer a larger local state space for gates and algorithms. Error correction for them has been mostly theoretical. A 2025 Nature paper reports the first experimental error-corrected logical qutrit and ququart using a bosonic code, with gains of 1.82 and 1.87 beyond break-even. Papers from 2024 and 2026 construct new qudit codes, including a four-qudit code and qudit LDPC codes. These are early-stage results.

Why qudits

A 2025 Nature paper explains that Hilbert space dimension is a key resource for quantum information processing. A large overall space is essential for error correction, and a large local space can make gates and algorithms more efficient. That has driven experimental work on platforms using qudits, systems of dimension d greater than 2. The authors note that, as with qubits, error correction will eventually be needed, but error correction of logical qudits had not been demonstrated experimentally. A 2026 paper in Quantum lists advantages of qudits: more efficient gate compilation, reduced resource requirements, improved error-correction primitives and better capabilities for communication and cryptography.

The 2025 experiment

The Nature paper reports an experimental error-corrected logical qutrit with d equal to 3 and ququart with d equal to 4, using the Gottesman-Kitaev-Preskill bosonic code. With a reinforcement learning agent the authors optimised the GKP qutrit and ququart as ternary and quaternary quantum memories, reaching error correction beyond break-even with gains of 1.82 plus or minus 0.03 for the qutrit and 1.87 plus or minus 0.03 for the ququart. They describe this as a new way of using the large Hilbert space of a harmonic oscillator for hardware-efficient error correction.

A noise-adapted code

A 2024 arXiv paper looks at the fact that most work on quantum codes has focused on qubits. The authors identify a four-qudit code, with parameters written [4,1], tailored to amplitude-damping noise. It satisfies the error correction conditions for all single-qudit and a few two-qudit damping errors to leading order in the damping parameter. They devise a syndrome extraction protocol that identifies those errors unambiguously and gives a noise-adapted recovery with a fidelity loss of order gamma squared. For d equal to 2 the scheme matches the known four-qubit code. They also assess the code with the Petz recovery map, noting differences from the qubit case, and generalise to a family of [2M+2, M] qudit codes.

Qudit LDPC codes

The 2026 Quantum paper observes that quantum low-density parity-check codes, one of the most promising code families, had been mostly restricted to qubits. The authors generalise recent LDPC advances to qudits with a general framework, applying it to several code types: bivariate bicycle codes, hypergraph product codes, subsystem hypergraph product codes, high-dimensional expander codes and fiber bundle codes. They numerically search for and decode novel qudit codes compatible with near-term hardware, and highlight qudit LDPC codes as a hardware-compatible route toward scalable error correction.

What this does not tell you

The experiment is a single logical qutrit or ququart memory in one device, and the other two papers are code constructions and numerical results. They do not show a fault-tolerant qudit computer.

FAQ

What is a qudit?

A quantum system with d levels, where d is greater than 2.

Has qudit error correction been demonstrated?

A 2025 Nature paper reports it for a qutrit and a ququart using a GKP code.

What does beyond break-even mean?

The encoded memory outlasts the best unencoded one, with gains of 1.82 and 1.87 reported.

Are there new qudit codes?

Yes. A 2024 paper gives a four-qudit amplitude-damping code and a 2026 paper generalises LDPC codes to qudits.

Is it a working computer?

No. These are memory experiments and code designs.