Extensions: exact local expectations from a product-observable evaluator
OpenQAlgorithms.Papers.Independent.Intakethird100.P57.exact_network_extensionsFix . Suppose a normalized base vector has an exact evaluator for , uniformly for arbitrary complex local matrices , including non-Hermitian factors. For every observable supported on at most sites, one graph enhancement followed by local unitaries admits an exact scalar reduction with
when each base evaluation costs at most operations uniformly in the factors. Alternatively, for a supplied unitary Clifford circuit of length , followed by local unitaries, there is a separate reduction with
The programs precede all numerical data; their calls, factor formation, and scalar output evaluation have explicit semantics. Thus an efficient exact base evaluator yields efficient fixed-support expectations under either enhancement. Clifford conjugation may spread support. These reductions do not assert arbitrary alternation of the two enhancement layers.
import Definitions.Def_QAlgorithms_Papers_Independent_Intakethird100_P57
noncomputable section
open scoped ComplexConjugate Matrix
namespace QAlgorithms.Papers.Independent.Intakethird100.P57
attribute [local instance] Classical.propDecidable
/-- PDF p.4 Extensions: exact conditional reductions, separately for one graph layer or
one Clifford circuit. For support ≤s, at most 4^s arbitrary product evaluations are used;
all other scalar work is polynomial in N and circuit length. Each product call is charged
B in addition to explicit factor formation. Thus any efficient exact base evaluator gives
an efficient local evaluator. No approximate-evaluation contract or alternating enhancement
layers are asserted, and Clifford propagation need not preserve small support. -/
theorem exact_network_extensions :
∀ s : ℕ, ∃ C : ℕ, 0 < C ∧
∃ (graphPrograms : (N : ℕ) → (S : Finset (Fin N)) → EnhancementProgram N S)
(cliffordPrograms : (N : ℕ) → (S : Finset (Fin N)) →
List (QubitCliffordGate N) → EnhancementProgram N S),
∀ (N : ℕ) (S : Finset (Fin N)), 1 ≤ N → S.card ≤ s →
let gp := graphPrograms N S
(gp.scalar.calls ≤ 4 ^ s ∧
(∀ B : ℕ, gp.scalar.work B ≤ C * (N + 1) ^ 3 + 4 ^ s * B) ∧
(∀ (psi : EuclideanSpace ℂ (Qubits N)) (hpsi : IsState psi)
(base : (Fin N → Matrix Bool Bool ℂ) → ℂ), ExactProductEvaluator psi base →
∀ (G : PhaseGraph N) (V : Fin N → Matrix.unitaryGroup Bool ℂ)
(O : Matrix (S → Bool) (S → Bool) ℂ),
∃ mem : ℕ → ℂ,
gp.scalar.execute (enhancementData G V O) base = some mem ∧
gp.output < gp.scalar.code.length ∧
mem gp.output = inner ℂ (enhance G V psi)
(act (supportedOp S O) (enhance G V psi)))) ∧
(∀ circuit : List (QubitCliffordGate N),
let cp := cliffordPrograms N S circuit
cp.scalar.calls ≤ 4 ^ s ∧
(∀ B : ℕ, cp.scalar.work B ≤ C * (N + circuit.length + 1) ^ 3 + 4 ^ s * B) ∧
(∀ (psi : EuclideanSpace ℂ (Qubits N)) (hpsi : IsState psi)
(base : (Fin N → Matrix Bool Bool ℂ) → ℂ), ExactProductEvaluator psi base →
∀ (G : PhaseGraph N) (V : Fin N → Matrix.unitaryGroup Bool ℂ)
(O : Matrix (S → Bool) (S → Bool) ℂ),
let chi := act (productOp fun k => (V k : Matrix Bool Bool ℂ))
(act (cliffordMatrix circuit) psi)
∃ mem : ℕ → ℂ,
cp.scalar.execute (enhancementData G V O) base = some mem ∧
cp.output < cp.scalar.code.length ∧
mem cp.output = inner ℂ chi (act (supportedOp S O) chi))) := by sorry
end QAlgorithms.Papers.Independent.Intakethird100.P57
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
For every natural number , there exist a positive natural number and two families of enhancement programs, and , defined for every natural number , every finite set of sites , and, for the second family, every finite list of the Clifford gates described below, such that the following guarantees hold whenever and . A scalar program here is a finite instruction list operating on complex memory initially zero, writing successive instruction results at successive addresses starting at zero; its operations are complex constants, input reads, addition, multiplication, division, conjugation, , and calls to a supplied function on a family of complex matrices assembled from memory entries. Reading an unwritten address gives zero, and division by zero fails. Its call count is the number of these function calls, and its work at parameter charges for each call and for each other instruction. The program has at most calls and, for every , work at most . For every vector on the Boolean strings of length with , and every function satisfying for every family , where , the following holds for every real weight array satisfying , , and , every family of unitary matrices , and every complex matrix indexed by Boolean assignments on . Define the diagonal matrix by , define , and define when and agree outside , and otherwise. Give the program input values , , and at their respective phase, rotation, and observable labels. Execution with succeeds with some memory function , its designated output address is smaller than its instruction-list length, and the value there is . For every circuit , the program likewise has at most calls and, for every , work at most . Under the same quantification over normalized , exact , , , and , its execution on those same input labels succeeds with an in-range output address whose value is , where . The permitted circuit gates are a Hadamard on one site, with entries for bits , a one-site phase gate , and a controlled bit flip with distinct control and target sites; the latter sends a Boolean string to the string obtained by replacing its target bit by the exclusive-or of its target and control bits. One-site gates act as the identity on all other sites, and is the product of the gate matrices in reverse list order, so the first listed gate acts first and the empty circuit gives the identity. The second output formula contains no dependence on . Empty , including , is allowed, and no Hermitian assumption is made on ; the program families are defined also at and , but the stated guarantees apply only under the displayed bounds.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.