Noncommutative Integration

What happens to measure theory when 'functions' become matrices - and why does quantum mechanics need it?

• **Quantum mechanics:** observables are self-adjoint operators in B(H); probability and integration in QM are noncommutative measure theory
• **Quantum information:** relative entropy D(rho||sigma) is the noncommutative analog of KL; foundation of quantum state estimation
• **Neural networks:** Schatten matrix norms S_p regularize weights; the nuclear norm (S_1) induces low rank in Netflix recommender
• **Random matrices:** Voiculescu free probability is the noncommutative analog of classical probability for limiting spectral measures

Предварительные знания

• Operator algebras and C*-algebras
• Functional analysis and Hilbert spaces
• Classical L^p spaces and duality

• Previous measure theory lesson

Von Neumann Algebras

In 1936, John von Neumann sought a mathematical language for quantum mechanics. The classical answer is the algebras of operators closed in the weak operator topology. His bicommutant theorem (M')' = M turned a topological condition into an algebraic one. By 2020, these algebras had become the foundation of quantum information and tensor network methods modeling entanglement in Google's Sycamore.

The algebra B(H) for infinite-dimensional H is a type I_infty factor. Algebras of local observables in quantum field theory are type III_1 factors (Connes). The free group algebra F_n is a type II_1 factor; isomorphisms between them remain open.

Tracial States and Noncommutative L^p

The idea of von Neumann and Segal: replace the integral against a measure by a trace on an operator algebra. Cyclicity tau(ab) = tau(ba) is the noncommutative analog of integral symmetry. On matrices M_n the trace gives Schatten L^p norms used in neural network regularization: the nuclear norm (S_1) induces low rank, critical for matrix factorization in the Netflix recommender.

Nuclear norm ||W||_1 = sum sigma_i in Netflix Prize matrix factorization replaces rank (NP-hard) with a convex relaxation. A canonical case of a noncommutative L^1 norm in ML.

Modular Theory and Applications

In 1973 Minoru Tomita and Masamichi Takesaki built modular theory: every normal state on a von Neumann algebra generates a one-parameter group of automorphisms sigma_t. In quantum physics this is the time evolution of KMS states (thermal equilibrium). Today modular flows govern computation of quantum relative entropy in quantum chemistry at Google Quantum AI.