English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

A Physical Exam for Logical Qubits: Surface Codes II

Forum topic · 小凯 · 2026-11-07

Summary

This post, part of a quantum computing series, presents a 'health check report' for logical qubits built on surface codes, covering three key metrics. First, the logical error rate: surface-code error correction compounds physical error rates of ~10⁻³ down to 10⁻⁶ for chemistry simulations and 10⁻⁹–10⁻¹² for breaking RSA-2048. Second, the threshold: surface codes work when physical error rates fall below roughly 1%; today's ~10⁻³ error rates leave a tenfold margin, and each tenfold improvement shortens the required code distance. Third, overhead: a distance d≈25–30 patch needs about 2d² physical qubits—roughly 2,000 per logical qubit—so factoring RSA-2048, requiring thousands of logical qubits, totals around one million physical qubits. Beyond qubit counts, two 'time taxes' apply: magic state distillation (hundreds to thousands of impure states burned per usable T-gate state) and lattice surgery for logical CNOT gates, which doubles circuit depth. The author's takeaway: any 'quantum advantage' claim should be examined pre-tax or post-tax, since practical algorithms carry the full price of millions of physical qubits plus distillation factories.

> Prerequisite navigation: You need post 055 (the chessboard, chains, and channels; patrol-repair cycles), 054 (code distance formula), 051 (T1/T2 and thousand-gate lifetimes), and 038 (Shor's bill). > Next episode preview: 057, "The 2026 Logical Qubit Race: IBM, Google, and Neutral Atoms (November 2026 view)."

A Physical Exam for Logical Qubits: Surface Codes II

Post 051 gave physical qubits a health check: two metrics, T1 and T2, lifetimes on the order of a thousand gates, four to five orders of magnitude short of Shor's requirements—concluding "walk on two legs: engineering to extend lifetimes, error correction for compounding."

The chessboard (post 055) is that "compounding leg." This post issues a health check for the logical qubit—three metrics: logical error rate, threshold, and overhead. This report is the most "valuable" ledger in the whole column: whether quantum computing can deliver on its green-zone promises (breaking cryptography, simulating chemistry) is written in every line of it.

Metric 1: Logical Error Rate—Body Temperature After Compounding

The logical error rate ε_L: the probability that a logical qubit errs in each patrol round. It is "how much error remains after compounding."

How low must it go? Depends on the use case: for an hour-long chemistry simulation, ε_L around 10⁻⁶ suffices; for breaking RSA-2048 (post 038: tens of billions of gates, n³ scaling), the logical error rate must be pressed to 10⁻⁹–10⁻¹²—tens of billions of operations, with total error probability still below one. From 10⁻³ physical to 10⁻⁹ logical: compounding must span six orders of magnitude—that is error correction's reason to exist, and the whole source of its difficulty.

Metric 2: Threshold—The Passing Grade for Engineering

The threshold p_th: the dividing line of physical error rate below which the chessboard works. When the single-round physical error rate is below it (for the surface code, on the order of 1%), bigger codes get more stable (exactly the turning point experimentally verified in 2024, per post 055); above it, bigger boards make more errors—compounding becomes a compounding disaster.

Read the threshold with a plain-language gloss: it is a constant of the hardware's constitution—not a target, but a passing grade. Today's physical error rate of 10⁻³ (one in a thousand) sits a tenfold margin below the 10⁻² passing line—in bounds, but the margin isn't luxurious: every further tenfold reduction in physical error rate makes the later bills noticeably more comfortable (the value of post 051's "engineering extends lifetime" leg converts here into the chessboard's side length).

Metric 3: Overhead—The Chessboard's Size Bill

Finally, money. The ledger has three layers:

Layer one: what one chessboard costs. A d×d patch of data qubits plus sentinels ≈ 2d² physical qubits (post 055).

Layer two: how big d must be. Determined jointly by "how far the physical error rate sits below threshold" and "the target ε_L"—a rough plain-language formula: every tenfold improvement of physical error rate versus threshold saves several lattice units of side length. Plugging in today's numbers (physical error rate 10⁻³ against threshold 10⁻², target 10⁻⁹–10⁻¹²): d must be 25–30—one chessboard is about two thousand physical qubits supporting a single logical qubit. The "one logical qubit per thousand physical qubits" rumor in post 052 even looks half-optimistic against this arithmetic.

Layer three: how many chessboards. Factoring RSA-2048 requires thousands of logical qubits (post 038)—thousands × two thousand = a million physical qubits. Post 039's "three orders of magnitude short" now aligns at three places; the ledger closes.

The Time Tax: Distillery and Chessboard Surgery

Beyond the qubit bill there is a time bill, with two line items:

Tax one: magic state distillation (post 023's "yeast," now redeemed). Logical-level T gates aren't executed directly—large batches of "impure yeast states" are first sent into a distillery, and only high-purity magic states refined there can serve logical T gates. Distillation's consumption ratio is staggering: producing one合格 qualified yeast burns hundreds to thousands of defective ones—post 023's "yeast is most expensive" prophecy becomes the heaviest line in this ledger. Shor's tens of billions of yeast is, in essence, an order placed with the distillery.

Tax two: chessboard surgery. Logical CNOT (a gate between two logical qubits) doesn't rely on "making two chessboards fly gates at each other"; it relies on geometric splicing—sewing two chessboards edge-to-edge, sharing one round of sentinel patrol, then splitting apart (plain language: two machines must rehearse a full round to shake hands). Every logical two-qubit gate pays this time tax—circuit depth doubles, runtime doubles.

Qubit tax plus time tax together make the logical qubit's full body price. The green zone's big-ticket items (Shor, precise chemistry) all carry this price tag—after post 056, whenever readers see any "quantum advantage" news, they can instinctively ask: pre-tax or post-tax?

Taking the Analogy Apart

Where the health-check analogy borrows and where it doesn't:

  • Disanalogy one: a person's checkup metrics correlate but don't "compound" (blood pressure doesn't improve because checkups are frequent); the logical error rate is the residue after each round's clearing—the more frequent the patrols and the bigger the board, the exponentially lower it goes: a pure compounding structure.
  • Disanalogy two: a checkup reports "current health"; the threshold describes "constitutional fitness"—it's a device-level constant (set jointly by materials, control, and readout), unchanged by switching report sheets; change it only by changing hardware (post 051's "engineering leg").

Luggage to Take Home

1. Three health metrics: logical error rate (post-compounding body temperature; Shor needs 10⁻⁹–10⁻¹²), threshold (a surface-code passing line near 1%; today's margin is tenfold), overhead (today d≈25–30 → about two thousand physical qubits per logical qubit; thousands of logical qubits → a million physical qubits—the ledger aligns and closes at three points). 2. Two time taxes: magic state distillation (one pure yeast costs hundreds to thousands of defectives—post 023's "yeast is most expensive" redeemed) and chessboard surgery (logical CNOT via spliced handshakes, time tax charged per round). 3. Post-tax awareness when reading news: for any "quantum advantage," first ask pre-tax or post-tax—green-zone big items all carry the full body price of "a million qubits plus a distillery."

Next stop's hook: the ledger is on the table—now watch how the players climb. The 2026 logical qubit race among superconducting (IBM, Google), neutral atoms, and trapped ions: who's at which milestone, where the roadmaps diverge, and who plants the "million-qubit" signpost first? (November 2026 view, with an on-site link to the older IBM 70-logical-qubit article.) Post 057: "The 2026 Logical Qubit Race: IBM, Google, and Neutral Atoms (November 2026 view)."

Tags

#quantum-computing#surface-codes#error-correction#logical-qubits#threshold-theorem#magic-state-distillation#shors-algorithm

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178636890