The Setting
Let be a von Neumann algebra on a complex Hilbert space , in standard form with a cyclic and separating unit vector , vector state , modular operator and modular automorphism group , and let the centralizer be trivial. Take and a bounded operator on , and average and along the modular flow over finitely many times with weights summing to one. The question is whether one choice of times and weights does two things at once: it keeps two moments of the averages in bounded in terms of and , and it keeps the weighted average of at least as large as a given level below the mean of over . A second question is whether, for and in a narrow spectral band of around a point , the moment bound can be taken to depend on alone.
Standard Modular Data
A real spectral calculus on is a -homomorphism from the bounded continuous functions into that is nondegenerate: the vectors , with compactly supported and , are dense in .✦
For real , the spectral character is the function ✦, and the unitary group of is (1)✦ For , the half exponential cutoff is the function ✦, and the graph of is the set of pairs (2)✦
A vector has -spectral support in a set when for every bounded continuous function vanishing on ; this stands for .✦
Standard modular data on consist of a von Neumann algebra on , a cyclic and separating unit vector , a real spectral calculus , standing for the logarithm of the modular operator , and an antiunitary involution of , subject to the Tomita graph identity (3)✦ the closure taken in . This is the statement for the Tomita operator , with .
The modular flow of standard modular data and their vector state are, for bounded operators on , (4) with the inner product conjugate-linear in its first argument. Standard modular data have trivial centralizer, , when every with for all real is a scalar multiple of the identity.✦
Means and Finite Averages
For finitely many real weights and real times , the finite modular average of a bounded operator on is the operator .✦ For , the translates and the finite modular averages of and of are (5)
For an ultrafilter on and a function , the mean of along is the limit along of the symmetric averages of over , (6)✦ For a bounded operator on and , the orbit defect of under is the function on . A free ultrafilter is an ultrafilter on containing every cofinite set; the mean of the orbit defect of along a free ultrafilter is the quantity that enters the hypotheses of Theorem 1.
For real , the recovery moment constant is (7)✦
The Claim
Theorem 1. Let standard modular data on be given, with von Neumann algebra , unit vector , vector state , modular flow and unitary group .
(a) Suppose and . Then there are finitely many times and weights with such that, with , and as in (5), (8)✦
(b) Let be a free ultrafilter on , let and be bounded operators on , let be real times and real weights with , and let . Then there is a real number with (9)✦
(c) Suppose , let be a free ultrafilter on , a bounded operator on and , and let . Then there are finitely many times and weights with such that, with , and as in (5), (10)✦
(d) Let and be real numbers with , and let be such that and has -spectral support in . Then (11)✦
In the paper, the result reads: for fixed , the paper’s parameter, there are finitely many times and weights summing to with and , where , and , with the constant of (7).
The library reaches this bound in two steps, (10) in real parts of and then (11) under its band hypotheses.
What Is Checked
The claim is stated over standard modular data on : a von Neumann algebra with a cyclic and separating unit vector , a real spectral calculus with its unitary group (1) and the graph (2) of , and an antiunitary involution satisfying (3); and over the modular flow and vector state (4), trivial centralizer, the finite modular averages (5), the mean (6) of the orbit defect along a free ultrafilter, -spectral support and the constant (7).
About these constructions the library proves that the orbit defect is a bounded continuous function with for every . The cutoff and the character are bounded continuous functions as well, with and .