The Setting
Conjugation by a shallow circuit is measured here through transitions from a product projection’s zero-mismatch subspace to sectors with many mismatches. Lemma 6 concerns such transitions when the conjugated product projection has a restricted support.
For total qubits, a classical bit belongs to ✦, and a bit string of length assigns a classical bit to each wire.✦ A quantum operator on these qubits is a complex matrix indexed by pairs of bit strings✦; a single-qubit operator has its rows and columns indexed by classical bits.✦ For a family of single-qubit operators and bit strings of length , the tensor product has the entries in (1).✦ For quantum operators , conjugation is given by (1), where denotes the adjoint.✦ All norms are Euclidean operator norms. (1)
Product projection data on qubits consist of a support and local vectors , normalized on as in (2).✦ The outer product is the single-qubit operator with the matrix entries in (2); for , it is the rank-one projector onto . (2)
The mismatch pattern projection , for a set of wire indices, uses the complementary local projector on the supported wires in , as specified in (3).✦ The product projection selects the empty mismatch pattern.✦ (3)
For a real cutoff , the high spectral projection sums the mismatch patterns of cardinality at least ✦, and the low spectral projection is its complement, as in (4).✦ (4)
Circuit Layers and Localization
A local unitary layer assigns a one-qubit unitary to each wire.✦ A disjoint reflection layer is an ordered list of product projection data with pairwise disjoint supports.✦ Their layer matrices are those in (5); the rightmost matrix acts first. (5)
For a nonnegative integer , the enlarged reflection-circuit class consists of operators of depth at most , with identity layers allowed. Its recursive definition is (6), where ranges over nonnegative integers, over local unitary layers and over disjoint reflection layers. (6)✦
Uniform localization , for real exponents , is the assertion that (7) holds for all sufficiently large , simultaneously for all and all product projection data .✦ (7)
For an ordered list of product projection data, its reflection product expansion is a list of pairs with real coefficient and product projection data . The reflection product expansion of the empty gate list is the single-term list , where is product projection data with empty support. Its coefficient list is indexed by selections of gate positions as in (8); for disjoint gate supports, the term indexed by uses the selected gates’ local vectors on the union of their supports. (8)
The Claim
Theorem 1. Let be a nonnegative integer, and assume for every pair of real exponents . Let and . For all sufficiently large , every and all product projection data with satisfy (9).✦ (9)✦
Let be a unitary operator on qubits, a disjoint reflection layer, a local unitary layer, and product projection data. Let be real cutoffs and . Suppose that every product projection datum satisfies the three estimates in (10). (10) Then the finite layer estimate is (11). (11)✦
For every ordered list of product projection data, the absolute coefficient weight of its reflection product expansion is (12), with the sum taken over list entries.✦ (12)✦
What Is Checked
The statements are over finite quantum operators, normalized product projection data and layers with disjoint reflection supports, with preceding-depth uniform localization assumed in (9) and the three transition estimates (10) assumed in (11).