MerLean-example

38 Cor 1: Worst-Case Qubit Overhead

The worst-case qubit overhead for the fault-tolerant gauging measurement of a Pauli logical operator \(L\) of weight \(W\) in an \([\! [n,k,d]\! ]\) qLDPC stabilizer code is \(O(W \log ^2 W)\), where the implicit constant depends on the code parameters (the check weight bound \(w\) and the qubit participation bound \(c\)) but not on \(W\) or \(n\).

More precisely: there exists a choice of graph \(G\) such that the total number of additional qubits (edge qubits) introduced by the fault-tolerant gauging measurement procedure is \(O(W \log ^2 W)\), and the resulting deformed code satisfies all desiderata (bounded-weight checks, preserved distance \(d^* \geq d\), and LDPC property).

Edge Overhead from Sparsified Graph Construction

Theorem 1016 Edge Overhead from Degree

Let \(N, \Delta \) be natural numbers such that \(2N \leq \Delta \cdot N\). Then \(N \leq \Delta \cdot N / 2\).

Proof

This follows directly by integer arithmetic from the hypothesis \(2N \leq \Delta \cdot N\).

Theorem 1017 Edge Overhead is \(O(W \log ^2 W)\)

Let \(W, \Delta , K, R\) be natural numbers with \(R \leq K \cdot (\log _2 W)^2\) and \(\Delta {\gt} 0\). Then

\[ 2 \cdot (R+1) \cdot W \leq 2 \cdot (K \cdot (\log _2 W)^2 + 1) \cdot W. \]
Proof

We apply monotonicity of multiplication on the left by \(2\) and on the right by \(W\). It suffices to show \(R + 1 \leq K \cdot (\log _2 W)^2 + 1\), which follows by integer arithmetic from the hypothesis \(R \leq K \cdot (\log _2 W)^2\).

Theorem 1018 Per-Vertex Overhead

Let \(W, K, R\) be natural numbers with \(R \leq K \cdot (\log _2 W)^2\) and \(W {\gt} 0\). Then

\[ \frac{(R+1) \cdot W}{W} \leq K \cdot (\log _2 W)^2 + 1. \]
Proof

Rewriting \((R+1) \cdot W / W = R + 1\) using the cancellation lemma (since \(W {\gt} 0\)), the result follows by integer arithmetic from \(R \leq K \cdot (\log _2 W)^2\).

Desiderata Satisfied with \(O(W \log ^2 W)\) Overhead

Theorem 1019 All Desiderata with Overhead Bound

Let \(G\) be a constant-degree \(\Delta \) connected graph on \(V\) with \(|V| \geq 2\), sufficient expansion \(h(G) \geq 1\), \(Z\)-support matching edges, and a low-weight cycle basis with cycles of weight \(\leq 4\). Given the cycle rank property \(|C| + |V| = |E| + 1\), a layer cycle degree bound with \(M\) layers satisfying \(M \leq K \cdot (\log _2 |V|)^2\), and a per-layer parameter \(c {\gt} 0\), the following all hold:

  1. All desiderata are satisfied: \(\operatorname {AllDesiderataSatisfied}(G, \text{cycles}, \text{checks}, 1, 4)\).

  2. There exist \(R\) layers and a coloring \(f\) with \(\operatorname {LayerCycleDegreeBound}(\text{cycles}, f, c)\) and \(R \leq K \cdot (\log _2 |V|)^2 / c\).

  3. The edge overhead is linear: \(2|E| \leq \Delta \cdot |V|\).

  4. The expansion gives a boundary bound: for all nonempty \(S \subseteq V\) with \(2|S| \leq |V|\), we have \(|S| \leq |\partial S|\).

Proof

The four conclusions are established as a conjunction:

  1. All desiderata follow directly from construction_satisfies_all_desiderata applied to \(G\), the cycles, checks, \(\Delta \), the \(Z\)-support adjacency hypothesis, the sufficient expansion, the low-weight cycle basis, and the constant degree bound.

  2. The \(O(\log ^2 W)\) layer count follows from decongestion_log_squared applied to \(G\), \(\Delta \), the constant degree, connectivity, vertex count bound, the positivity of the expansion constant, the cycles, \(c\), the cycle rank property, and the König-Freedman-Hastings bound.

  3. The edge overhead bound \(2|E| \leq \Delta \cdot |V|\) follows from constant_degree_bounds_edges.

  4. The expansion-based boundary bound follows from expansion_gives_boundary_bound applied to each nonempty subset \(S\) with \(2|S| \leq |V|\).

Theorem 1020 Check Weights Bounded

Let \(G\) be a constant-degree \(\Delta \) graph with a low-weight cycle basis of weight \(\leq 4\). Then:

  1. For every vertex \(v \in V\), the Gauss law check \(A_v\) has weight at most \(1 + \Delta \).

  2. For every cycle \(p \in C\), the flux check \(B_p\) has weight at most \(4\).

Proof

The bound on Gauss law check weights follows from gaussLaw_checks_weight_bounded applied to \(G\), \(\Delta \), and the constant degree hypothesis. The bound on flux check weights follows from flux_checks_weight_bounded applied to \(G\), the cycles, \(4\), and the low-weight cycle basis hypothesis.

Distance Preservation

Theorem 1021 Distance Preserved with Overhead

Let \(G\) be a connected constant-degree \(\Delta \) graph on \(V\) with \(|V| \geq 2\) and sufficient expansion, equipped with a cycle basis satisfying the exactness conditions. Let \(\mathcal{C}\) be the original stabilizer code with logical operator \(L\) and let the deformed code have at least one logical operator. Given \(R \leq K \cdot (\log _2 |V|)^2\) layers, we have:

  1. The distance is preserved: \(d \leq d^*\), where \(d\) is the distance of the original code and \(d^*\) is the distance of the deformed code.

  2. The edge overhead is bounded: \(2|E| \leq \Delta \cdot |V|\).

Proof

The distance bound \(d \leq d^*\) follows directly from deformed_distance_ge_d_sufficient_expansion applied to \(G\), the cycles, checks, the original code, the identification of check indices, the deformed code data, the cycle parity condition, the commutativity of checks, connectivity, the vertex count bound, the exactness conditions, the sufficient expansion, the logical operator hypothesis, and the existence of deformed logicals. The edge overhead bound follows from constant_degree_bounds_edges.

Main Corollary

Theorem 1022 Corollary 1: Worst-Case Overhead

For a qLDPC stabilizer code with a Pauli logical operator \(L\) of weight \(W\), let \(G\) be a constant-degree \(\Delta \) connected graph on \(V\) with \(|V| = W\) (support of \(L\)), sufficient expansion \(h(G) \geq 1\), \(Z\)-support matching edges, a low-weight cycle basis with cycles of weight \(\leq 4\), and the cycle rank property \(|C| + |V| = |E| + 1\). Given a König coloring with \(M\) layers and the Freedman–Hastings bound \(M \leq K \cdot (\log _2 W)^2\), and a per-layer parameter \(c {\gt} 0\), the following all hold:

  1. All desiderata are satisfied: \(\operatorname {AllDesiderataSatisfied}(G, \text{cycles}, \text{checks}, 1, 4)\).

  2. There exist \(R\) layers and a coloring \(f\) with layer cycle degree bound \(c\) and \(R \leq K \cdot (\log _2 W)^2 / c\).

  3. The edge overhead is \(O(W)\): \(2|E| \leq \Delta \cdot W\).

  4. The distance is preserved: \(d \leq d^*\).

  5. Gauss law check weights are bounded: \(\operatorname {wt}(A_v) \leq 1 + \Delta \) for all \(v\).

  6. Flux check weights are bounded: \(\operatorname {wt}(B_p) \leq 4\) for all \(p\).

  7. The vertex overhead is bounded: for all \(R \leq K \cdot (\log _2 W)^2/c\), we have \((R+1) \cdot W \leq (K \cdot (\log _2 W)^2/c + 1) \cdot W\).

Proof

We prove each of the seven conclusions:

  1. All desiderata follow from construction_satisfies_all_desiderata applied to \(G\), the cycles, checks, \(\Delta \), the \(Z\)-support adjacency, the sufficient expansion, the low-weight cycle basis, and the constant degree.

  2. The layer count bound follows from decongestion_log_squared applied to \(G\), \(\Delta \), the constant degree, connectivity, vertex count bound, positivity of the expansion constant, the cycles, \(c\), the cycle rank property, and the König-Freedman-Hastings bound inputs.

  3. The edge overhead bound \(2|E| \leq \Delta \cdot |V|\) follows from constant_degree_bounds_edges.

  4. The distance preservation \(d \leq d^*\) follows from deformed_distance_ge_d_sufficient_expansion applied with the original code, the identification of check indices, the deformed code data, the cycle parity condition, the commutativity of checks, connectivity, the vertex count bound, the exactness conditions, the sufficient expansion, the logical operator hypothesis, and the existence of deformed logicals.

  5. The Gauss law check weight bound follows from gaussLaw_checks_weight_bounded for each vertex \(v\).

  6. The flux check weight bound follows from flux_checks_weight_bounded for each cycle \(p\).

  7. For the vertex overhead, let \(R \leq K \cdot (\log _2 W)^2/c\). We apply monotonicity of multiplication on the right by \(W\), and the bound \(R + 1 \leq K \cdot (\log _2 W)^2/c + 1\) follows by integer arithmetic.

Theorem 1023 Qubit Overhead is \(O(W \log ^2 W)\)

Let \(W, \Delta , K, c {\gt} 0\) be natural numbers, and let \(R \leq K \cdot (\log _2 W)^2 / c\). If \(2|E| \leq \Delta \cdot (R+1) \cdot W\), then

\[ |E| \leq \frac{\Delta \cdot (K \cdot (\log _2 W)^2/c + 1) \cdot W}{2}. \]
Proof

We first establish that \(R + 1 \leq K \cdot (\log _2 W)^2/c + 1\) by integer arithmetic from \(R \leq K \cdot (\log _2 W)^2/c\). Then \(\Delta \cdot (R+1) \cdot W \leq \Delta \cdot (K \cdot (\log _2 W)^2/c + 1) \cdot W\) by monotonicity of multiplication on the left by \(\Delta \) and on the right by \(W\). The result follows by integer arithmetic combining with the hypothesis \(2|E| \leq \Delta \cdot (R+1) \cdot W\).

Theorem 1024 Codespace Dimension with Overhead

Let \(G\) be a constant-degree \(\Delta \) graph with cycle rank property, and let the original code have parameters \([\! [n,k,d]\! ]\) with \(k \geq 1\). Given \(R \leq K \cdot (\log _2 n)^2\) layers, the deformed code satisfies:

  1. The number of encoded qubits decreases by one: \(n' - m' = k - 1\), where \(n'\) is the number of qubits and \(m'\) is the number of checks of the deformed code.

  2. The edge overhead is bounded: \(2|E| \leq \Delta \cdot |V|\).

Proof

The codespace dimension change follows directly from codespace_dimension_change_after_gauging applied to \(G\), the cycles, checks, the deformed code data, the cycle parity condition, the commutativity of checks, \(n\), \(k\), the cardinality identification, the dimension formula, \(k \geq 1\), and the cycle rank property. The edge overhead bound follows from constant_degree_bounds_edges.

Concrete Overhead Characterization

Theorem 1025 Overhead Ratio is \(O(\log ^2 W)\)

Let \(W, K, c {\gt} 0\) be natural numbers and \(R \leq K \cdot (\log _2 W)^2/c\). Then the overhead ratio satisfies \(R + 1 \leq K \cdot (\log _2 W)^2/c + 1\).

Proof

This follows by integer arithmetic from \(R \leq K \cdot (\log _2 W)^2/c\).

Theorem 1026 Total Qubit Count

Let \(G\) be a constant-degree \(\Delta \) graph on \(V\). Then the total qubit count satisfies

\[ 2 \cdot (|V| + |E|) \leq (2 + \Delta ) \cdot |V|. \]
Proof

From constant_degree_bounds_edges, we have \(2|E| \leq \Delta \cdot |V|\). Then \(2(|V| + |E|) = 2|V| + 2|E| \leq 2|V| + \Delta \cdot |V| = (2 + \Delta ) \cdot |V|\), which follows by nonlinear arithmetic using the distributivity \((2 + \Delta ) \cdot |V| = 2 \cdot |V| + \Delta \cdot |V|\).

Theorem 1027 Overhead Independent of \(n\)

Let \(W, \Delta , K, c {\gt} 0\) be natural numbers and \(R \leq K \cdot (\log _2 W)^2/c\). Then the overhead bound

\[ (R+1) \cdot W \leq (K \cdot (\log _2 W)^2/c + 1) \cdot W \]

depends on \(W\), \(\Delta \), \(K\), \(c\) but not on the code size \(n\).

Proof

We apply monotonicity of multiplication on the right by \(W\). The bound \(R + 1 \leq K \cdot (\log _2 W)^2/c + 1\) follows by integer arithmetic from the hypothesis \(R \leq K \cdot (\log _2 W)^2/c\).

Theorem 1028 Constant Depends on Code Parameters

Let \(W, \Delta , K, c {\gt} 0\) be natural numbers and \(R \leq K \cdot (\log _2 W)^2/c\). Then the multiplier satisfies

\[ \Delta \cdot (R+1) \cdot W \leq \Delta \cdot (K \cdot (\log _2 W)^2/c + 1) \cdot W. \]

The constant \(\Delta \cdot (K/c + 1)\) depends on the code parameters \(\Delta \) (maximum degree of \(G\), a function of check weight \(w\) and qubit degree), \(K\) (the Freedman–Hastings constant), and \(c\) (per-layer cycle-degree bound), but is independent of \(W\) and \(n\).

Proof

We apply monotonicity of multiplication on the left by \(\Delta \) and then on the right by \(W\). The inner bound \(R + 1 \leq K \cdot (\log _2 W)^2/c + 1\) follows by integer arithmetic from the hypothesis.

Theorem 1029 Deformed Code Total Qubits

Let \(G\) be a constant-degree \(\Delta \) graph on \(V\) with cycles and checks, and let the deformed stabilizer code be constructed from this data. Then the number of qubits of the deformed code satisfies

\[ n_{\text{deformed}} \leq |V| + \frac{\Delta \cdot |V|}{2}. \]
Proof

Rewriting the number of qubits of the deformed code using deformedStabilizerCode_numQubits, which gives \(n_{\text{deformed}} = |V| + |E|\), and applying the edge overhead bound \(2|E| \leq \Delta \cdot |V|\) from constant_degree_bounds_edges, we obtain \(|E| \leq \Delta \cdot |V|/2\) and thus \(n_{\text{deformed}} \leq |V| + \Delta \cdot |V|/2\) by integer arithmetic.