Weak Topologies and Compactness

Machine learning in infinite-dimensional spaces (RKHS, neural networks as functionals) requires guarantees that a minimum exists. The Banach-Alaoglu theorem and reflexivity are the mathematical foundation of those guarantees.

• SVM in RKHS: weak convergence and kernel compactness
• Tikhonov regularization: existence of minimizer via Banach-Alaoglu
• Optimal control: admissible controls in L²
• PDE equations: weak solutions in Sobolev spaces

Weak Topology: Convergence Through Functionals

The **weak topology** on a Banach space X is the coarsest topology under which all continuous linear functionals f ∈ X* remain continuous. Weak convergence x_n ⇀ x means: f(x_n) → f(x) for all f ∈ X*.

**Weak* topology:** On the dual space X* there is another topology-weak*, where f_n →_* f means f_n(x) → f(x) for all x ∈ X. It is coarser than the weak topology on X*. This is key in the Banach-Alaoglu theorem.

The Banach-Alaoglu Theorem: Compactness of the Unit Ball

**Banach-Alaoglu theorem:** The unit ball in the dual space X* is compact in the weak* topology. This is a fundamental result-in infinite-dimensional spaces the ball is not compact in norm, but it is compact in a weaker topology.

**Sequential compactness:** In metric spaces compactness = sequential compactness. But the weak* topology on X* is metrizable only when X is separable (has a countable dense subset). In general, the proof requires ultrafilters (Tychonoff's theorem).

Reflexive Spaces and Duality

A **reflexive Banach space** X is one that is isomorphic to its second dual X**. This means the canonical embedding J: X → X** (J(x)(f) = f(x)) is a surjective isomorphism.

**James's theorem (1964):** A Banach space X is reflexive if and only if every f ∈ X* attains its norm on the unit ball. This is a deep characterization: reflexivity ↔ norm-attaining functionals.

Key Takeaways

• Weak convergence: x_n ⇀ x ⟺ f(x_n) → f(x) for all f ∈ X*
• In infinite dimensions: weak ≠ strong (example: e_n ⇀ 0 in l²)
• Banach-Alaoglu: B_{X*} is compact in the weak* topology
• Reflexivity: X ≅ X**-B_X is compact in the weak topology
• L^p is reflexive for 1 < p < ∞; L¹ and L∞ are not

Related Topics

Weak topologies connect to fixed-point theorems and variational methods.

• Applications — Variational methods use weak convergence
• Banach Spaces — Duality in Banach spaces

Вопросы для размышления

• Why does weak convergence always equal strong convergence in finite-dimensional spaces?
• How is Banach-Alaoglu used in the proof of existence of a functional minimizer?
• What is the difference between weak* (weak*) convergence of measures and ordinary weak convergence?

Связанные уроки

• la-01-vectors-intro