Remark 13: Optimal Cheeger Constant #
Statement #
Picking a graph with Cheeger constant h(G) = 1 is optimal for distance preservation:
- h(G) ≥ 1 is sufficient: By Lemma 3, d* ≥ d.
- h(G) > 1 doesn't help further: Any nontrivial logical of the original code, supported only on vertex qubits, commutes with all deformed code checks since it acts trivially on edge qubits. Hence it is also a nontrivial logical of the deformed code, so d* ≤ d. Combined: d* = d when h(G) ≥ 1.
- h(G) < 1 causes distance loss: d* ≥ h(G) · d < d.
Main Results #
liftToExtended: lifts a Pauli operator on V to V ⊕ E(G) with trivial edge actionliftToExtended_weight: the lifted operator has the same weightliftToExtended_isLogical: lifting a pure-X original logical gives a deformed code logicaldeformed_distance_le_original: d* ≤ d (under existence of suitable pure-X logical)deformed_distance_eq: d* = d when h(G) ≥ 1cheeger_one_is_optimal: h(G) = 1 is the optimal thresholddistance_loss_when_cheeger_lt_one: h(G) < 1 may cause distance loss
Lifting original operators to the extended system #
Lift a Pauli operator on V to V ⊕ E(G) with trivial (identity) action on edge qubits.
Equations
Instances For
The lifted operator has the same weight as the original.
The restriction of a lifted operator back to V recovers the original.
The lift preserves multiplication.
The lift sends identity to identity.
The lift is injective: it preserves non-identity.
Commutation of lifted operators with deformed code checks #
The lift commutes with flux checks: flux is pure Z on edges, lift has no support on edges.
The lift commutes with deformed original checks when the original operator commutes with the original check.
⟨lift(P), A_v⟩ = P.zVec(v): the symplectic inner product with a gaussLaw check.
The lift commutes with A_v iff P has no Z-support at v.
A pure-X operator (zVec = 0) lifted to V ⊕ E commutes with all gaussLaw checks.
Pure-X logicals lift to deformed code centralizer #
A pure-X operator in the centralizer of the original code lifts to the centralizer of the deformed code.
Point 1: h(G) ≥ 1 is sufficient for d* ≥ d #
Point 1: h(G) ≥ 1 is sufficient for d ≥ d.* Direct application of Lemma 3.
Point 2: d* ≤ d via lifting #
The deformed distance is at most the weight of any deformed code logical.
A pure-X logical P of the original code, when lifted, is a logical of the deformed code (given that the lift is not in the deformed stabilizer group).
Point 2: d ≤ d.* If there exists a pure-X logical of minimum weight whose lift is not a deformed stabilizer, then d* ≤ d.
Point 2 combined with Point 1: d* = d when h(G) ≥ 1 #
d = d when h(G) ≥ 1* (combining Points 1 and 2).
Point 3: h(G) < 1 causes distance loss #
Point 3: h(G) < 1 causes distance loss. The Lemma 3 bound gives d* ≥ h(G) · d, which is strictly less than d.
When h(G) < 1, the lower bound h(G) · d is strictly less than d (if d > 0 and h(G) > 0).
Optimality of h(G) = 1 #
Increasing h(G) beyond 1 doesn't improve the Lemma 3 bound: min(h(G), 1) = 1.
h(G) ≥ 1 implies min(h(G), 1) = 1.
h(G) > 1 doesn't help further: the lower bound from Lemma 3 is still d.
Summary #
The distance bound trichotomy:
- h(G) ≥ 1 → the bound gives d (saturates)
- h(G) < 1 → the bound gives h(G)·d < d (distance loss)