## Does the Industry's Definition of "Logical Qubit" Need Raising?

A paper posted to arXiv by [Microsoft Quantum](https://quantumintel.tech/companies/microsoft) researchers Matthias Troyer and Chetan Nayak, together with QOLAB's John Martinis, argues that it does — and it proposes a four-part test to enforce a stricter standard. The framework introduces the term **"scalable logical qubit"** for an error-corrected qubit that can sustain repeated quantum error correction during computation, execute a universal set of [fault-tolerant quantum computing](https://quantumintel.tech/glossary/fault-tolerant-quantum-computing) operations, demonstrate a predictable path to lower logical error rates as physical qubits are added, and show a credible route to replication at the scale of hundreds or thousands of logical qubits.

The paper's practical stakes are explicit: the authors estimate that even the low end of useful quantum computation could demand more than 100 high-quality [logical qubits](https://quantumintel.tech/glossary/logical-qubit) with error rates around one failure in 10 billion operations. More demanding applications — large-scale chemistry and cryptanalysis — could require more than 1,000 logical qubits at error rates of one in 1 quadrillion operations or better. No current public demonstration is close to either threshold, which is precisely the authors' point.

---

## Why the Existing Terminology Is Causing Problems

The term "logical qubit" has done a lot of work across a lot of press releases over the past few years. Companies and academic groups have legitimately shown that encoding information across multiple physical qubits can improve quantum memory lifetime, suppress specific error types, or beat the performance of a single physical qubit under carefully defined conditions. Each of those results is real. None of them necessarily demonstrates a system capable of running a long and commercially useful quantum program.

The distinction matters to buyers and investors because it affects how to interpret competitive claims. A demonstration of an improved error rate in a small, isolated system may have limited bearing on whether that platform can operate hundreds of logical qubits simultaneously, execute circuits with mid-circuit measurement and classical feed-forward, or decode error syndromes fast enough to keep pace with gate operations.

The Troyer–Nayak–Martinis paper is framing this as a definitional problem the industry needs to solve collectively, rather than a critique of any single company. The choice of authors is notable: Nayak leads Microsoft's topological qubit program, Troyer is a leading figure in quantum algorithm resource estimation, and Martinis — formerly the architect of Google Quantum AI's superconducting roadmap — is now working through QOLAB. The combination spans hardware modalities and gives the proposal unusual cross-institutional credibility.

---

## The Four Criteria, Unpacked

The proposed framework requires that a scalable logical qubit satisfy all four of the following conditions simultaneously, not just the easiest one:

**1. Sustained through repeated QEC**
The logical qubit must be maintained during computation through continuous, active error correction cycles — not one-time error detection, not post-selection on error-free runs, and not classical post-processing of noisy data after the fact. Post-selection, the paper notes, becomes exponentially inefficient as circuit depth increases, because the probability of an entirely error-free run falls rapidly as computation grows longer. Error mitigation techniques face a similar scaling problem: they add sampling overhead that compounds quickly. The authors do not dismiss these tools, but argue they cannot substitute for genuine repeated QEC in any circuit that requires very low failure rates across billions of operations.

**2. Universal fault-tolerant operations**
The system must support a complete, universal set of fault-tolerant gates — including operations that respond to mid-circuit measurement outcomes. This is a higher bar than demonstrating that a logical qubit can sit quietly in memory with low error rates. Universal fault tolerance requires non-Clifford gates, typically delivered through magic state distillation or injection, which are resource-intensive and architecturally demanding.

**3. Predictable error suppression with scale**
The platform must belong to a code family and hardware architecture where adding more physical qubits reliably reduces the logical error rate. This requires demonstrating [below threshold](https://quantumintel.tech/glossary/below-threshold) operation — that the physical error rates in the system are low enough that a larger code distance actually helps rather than hurts. A system that shows a good logical error rate at small code distance but cannot improve it by scaling up has not demonstrated a credible path to fault tolerance.

**4. Credible replication to application-relevant scale**
The platform must have a realistic, engineering-grounded path to operating the number of logical qubits that applications actually require. The paper's own estimates place this at hundreds of logical qubits at the low end and more than 1,000 for the most demanding cases. This criterion forces an evaluation of control electronics, calibration infrastructure, interconnects, classical decoding latency, and physical hardware overhead — not just the performance of a single logical qubit in isolation.

---

## What This Means for the Broader QEC Field

The paper arrives as QEC demonstrations are proliferating across hardware modalities. Superconducting platforms, trapped-ion systems, neutral atom arrays, and photonic architectures have all published results showing encoded logical qubits with improved error rates. The Troyer–Nayak–Martinis framework is, in effect, a call to accompany those individual metrics with a fuller accounting.

For enterprise buyers evaluating quantum roadmaps, the four-criteria test offers a practical due-diligence checklist. A vendor that can show strong performance on criterion one (repeated QEC) but cannot articulate a hardware plan for criteria three and four (error suppression scaling and replication path) is further from commercial utility than headline qubit counts suggest.

For investors, the framework sharpens what to look for in technical milestones. A Series B or C raise tied to a "logical qubit demonstration" means something materially different depending on whether that demonstration satisfies all four criteria or only one.

The paper does not name specific companies or rank existing demonstrations against its criteria — a deliberate choice that keeps the framing collaborative rather than competitive. But the implicit scorecard is legible to anyone following the field closely.

---

## Skeptical Read

There is a reasonable objection to standards-setting exercises like this one: they can be written, intentionally or not, in ways that favor the approach of the institution proposing them. Microsoft's topological qubit program has been premised on the argument that achieving the physical error rates required for efficient QEC demands a fundamentally different qubit architecture. A framework that emphasizes reliable scaling of physical-qubit overhead and low-latency feedback could be read as favorable to that long-term bet.

That said, the specific numerical estimates in the paper — one failure in 10 billion operations for lower-end applications, one in 1 quadrillion for cryptanalysis — are not Microsoft-specific requirements. They follow from established resource estimation methodology that applies across hardware platforms. And Martinis's presence as a co-author, given his history building superconducting systems, makes a purely self-serving reading less persuasive.

The more substantive question is whether the field will actually adopt this vocabulary. The quantum computing industry has not historically converged quickly on shared benchmarking standards — [IBM Quantum](https://quantumintel.tech/companies/ibm) developed Quantum Volume and CLOPS partly because no universal metric existed. A framework paper on arXiv is a starting point, not a standard.

---

## Key Takeaways

- Microsoft Quantum's Matthias Troyer and Chetan Nayak, with QOLAB's John Martinis, have proposed a four-part definition of a "scalable logical qubit" via arXiv.
- The four criteria are: sustained repeated QEC during computation; universal fault-tolerant operations; predictable error suppression as physical qubits scale; and a credible replication path to application-relevant qubit counts.
- The paper estimates the low end of useful quantum computation requires more than 100 high-quality logical qubits at approximately one failure per 10 billion operations.
- Demanding applications such as large-scale chemistry and cryptanalysis could require more than 1,000 logical qubits at error rates of one in 1 quadrillion operations or better.
- Post-selection and error mitigation are explicitly identified as insufficient substitutes for repeated QEC at scale.
- The framework is modality-agnostic but the criteria are demanding enough to disqualify most current public demonstrations from the "scalable" label.

---

## Frequently Asked Questions

**What is a "scalable logical qubit" according to the Microsoft and QOLAB paper?**
The paper defines a scalable logical qubit as one that is maintained through repeated quantum error correction during computation, supports a universal set of fault-tolerant operations, belongs to an architecture where adding physical qubits reliably lowers the logical error rate, and has a credible path to replication at the hundreds-to-thousands scale needed for practical applications.

**Why isn't a protected quantum memory enough to count as a scalable logical qubit?**
A protected memory demonstrates that a logical qubit can survive with low error rates under passive or limited conditions. It does not demonstrate the ability to run deep circuits, perform mid-circuit measurements with classical feed-forward, or maintain performance when hundreds of such qubits operate simultaneously — all of which are required for useful quantum computation.

**What error rates does the paper say real applications will need?**
According to the paper, even lower-end useful quantum computations could require error rates around one failure in 10 billion operations. More demanding tasks like large-scale chemistry simulations and cryptanalysis could require rates of one failure in 1 quadrillion operations or better.

**How does post-selection differ from real quantum error correction?**
Post-selection discards experimental runs where errors are detected, keeping only the clean results. This can improve apparent performance in small demonstrations but becomes exponentially inefficient as circuits grow longer, because the probability of a fully error-free run drops rapidly with circuit depth. Genuine repeated QEC corrects errors in real time without discarding runs.

**Who are the authors and why does the institutional combination matter?**
The paper is authored by Matthias Troyer and Chetan Nayak of Microsoft Quantum and John Martinis of QOLAB. Nayak leads Microsoft's topological qubit effort, Troyer is a leading figure in quantum algorithm resource estimation, and Martinis was previously the principal architect of Google Quantum AI's superconducting program. The cross-institutional authorship spanning hardware and theory, and multiple modality backgrounds, gives the proposal broader credibility than a single-company white paper would carry.