Glossary

A quantum computer large enough and reliable enough to break deployed public-key cryptography.

Published estimates put this in the range of millions of physical qubits with sustained error correction. Nothing within several orders of magnitude of that exists. The term is useful precisely because it names a threshold rather than a date.

The process by which a quantum system loses its quantum character to its environment. Stray fields, thermal noise, vibration, and the control wiring itself all carry information away, after which the superposition is gone and what remains is classical noise.

Coherence time is how long you have before this dominates. What matters is not the absolute figure but its ratio to gate time: how many operations fit inside the window.

Post-processing techniques that reduce the bias noise introduces into an expectation value, without correcting the underlying errors.

Mitigation is not error correction. It does not protect a computation, and its sampling cost generally grows exponentially with circuit size, which limits it to shallow circuits. It is a way to extract more from noisy hardware, not a path to fault tolerance.

The property of a computing scheme that lets it produce a correct result even though its components fail during the computation, by correcting errors faster than they accumulate.

Fault tolerance is not a feature added at the end. It shapes the architecture, determines the physical qubit budget, and dominates the cost of every serious resource estimate.

A measure of how closely an operation or a state matches what was intended, usually quoted as a percentage or as an error rate.

Two-qubit gate fidelity is the number that constrains circuit depth, and it should be read as measured in a full circuit with many gates running at once, not in isolation. A gate characterised alone will always look better than the same gate embedded in a busy device.

One unit of quantum information encoded across many physical qubits, protected by error correction that detects and fixes errors while the computation runs.

The term is used loosely. It can mean an encoded state that was prepared and measured, a memory that outperformed its physical parts, a memory that improves as the code grows, or a qubit that actually participates in logical operations. Only the last is computation. Ask which is meant.

A specially prepared resource state consumed to perform the non-Clifford operations that make a quantum computer universal.

Clifford operations alone are classically simulable, so magic states are not optional. They are produced by distillation, which consumes many noisy copies to yield one good one, and in most resource estimates distillation dominates the cost of the entire machine.

The current era of hardware: processors with tens to thousands of physical qubits, no working error correction, and circuit depths limited by accumulated noise.

The term was introduced to describe what these machines are, and specifically to set expectations about what they cannot do. It has since been used in marketing to suggest a stage on a roadmap. It is a description of a constraint.

A single two-level quantum system that a laboratory can control: a superconducting circuit, a trapped ion, a neutral atom, a photon, a spin in silicon. It is unstable by default and requires continuous effort to preserve.

Qubit count in an announcement almost always means physical qubits. On its own it says how much hardware exists, not what the hardware can do.

Cryptographic algorithms designed to resist attack by a large quantum computer, based on mathematical problems Shor’s algorithm does not solve.

Standardised replacements exist and are deployable. Migration timelines are driven by data lifetime and the duration of infrastructure change, not by predictions about quantum hardware.

The set of techniques for protecting quantum information without copying it and without reading it.

Information is spread across many physical qubits so it lives in their correlations. Parity checks are measured, chosen to reveal where an error occurred without revealing the stored state. A decoder infers the most likely error and corrects it. This repeats continuously for the whole computation.

A family of codes that encode more logical qubits per physical qubit than the surface code, by giving up the requirement that interacting qubits be neighbours.

The trade is overhead for connectivity. Whether hardware can supply the long-range connections these codes need, at the fidelity they need, is one of the open questions worth watching.

The means by which someone outside the manufacturer runs a circuit on real hardware, generally through a cloud service with a queue.

Access terms matter more than they appear to. What is exposed — calibration data, pulse-level control, error rates per qubit — determines whether outside work can verify anything or only observe outputs.

A demonstration that a quantum processor performed a task no known classical method can perform in comparable time.

The claim has two halves: a measured quantum experiment and an estimated classical baseline. Only the first is measured. There is no proof that the tasks used are classically hard, so an advantage claim is a statement about the state of classical algorithms on the day it was made.

A softer claim than advantage: that a quantum processor produced a useful result at a scale where classical verification is difficult, without asserting that classical methods cannot do it.

The distinction matters because utility claims have repeatedly been answered by classical methods that could in fact do it. Read the classical follow-up before accepting the framing.

A calculation of what a fault-tolerant machine would need to run a specific algorithm: physical qubits, code distance, runtime, distillation factories.

These are the numbers that actually track progress toward applications, and they move for algorithmic and architectural reasons rather than because a new chip was announced. When an estimate falls by an order of magnitude, that is news.

The statistical uncertainty that comes from estimating a quantum expectation value by repeated measurement. Halving the error bar requires roughly four times as many shots.

This is a hard constraint on variational algorithms, where the number of measurements needed for chemical accuracy can exceed what is practical regardless of how good the hardware becomes.

The most studied family of quantum error-correcting codes. Qubits sit on a two-dimensional lattice and only neighbouring qubits interact, which suits hardware that cannot connect distant qubits.

Its advantages are a relatively high threshold and a layout real chips can build. Its cost is overhead: roughly a thousand physical qubits per logical qubit for useful logical error rates, and more under pessimistic assumptions.

The physical error rate below which error correction starts to help. Below it, adding physical qubits to a logical qubit lowers the logical error rate. Above it, the same addition raises it.

The threshold is a sign change, not a gradient, which is why below threshold is the phrase to look for in any error-correction result. For the surface code it sits near one per cent per operation, depending on the noise model and the decoder.

A hybrid method in which a shallow quantum circuit prepares a state, a classical optimiser adjusts the circuit’s parameters, and the loop repeats.

These were the main hope for useful work on noisy hardware. In practice they are limited by measurement cost, by optimisation landscapes that flatten as systems grow, and by noise that biases the result. They remain worth watching and have not yet beaten good classical methods on a problem anyone has.