Let EE be a over K\mathbb K, with E′E'. The weak topology σ(E,E′)\sigma(E,E') is the weakest topology on EE making every φ∈E′\varphi\in E' continuous. It is the

pφ(x)=∣φ(x)∣,φ∈E′.p_\varphi(x)=|\varphi(x)|,\qquad \varphi\in E'.

Consequently, a net xix_i converges weakly to xx exactly when φ(xi)→φ(x)\varphi(x_i)\to\varphi(x) for every φ∈E′\varphi\in E'. The weak topology is no finer than the original topology because the functionals in E′E' were originally continuous.

Neighborhoods and separation

A basic weak neighborhood of xx restricts only finitely many functionals:

{y∈E:∣φj(y−x)∣<εj, 1≤j≤m}.\{y\in E:|\varphi_j(y-x)|<\varepsilon_j,\ 1\leq j\leq m\}.

The weak topology is Hausdorff exactly when E′E' of EE. This separation holds for Hausdorff , but can fail for general topological vector spaces whose continuous dual is too small.

Weak versus original convergence

Original-topology convergence implies weak convergence. The converse is usually false in infinite dimensions: weak convergence records every scalar observation φ(xi)\varphi(x_i), but not uniform control over all observations. For example, the standard in ℓ2\ell^2 converges weakly to 00 while every term has norm 11.

Weak versus weak-star

On a Banach dual space E′E', the weak topology is σ(E′,E′′)\sigma(E',E''), whereas the is σ(E′,E)\sigma(E',E), using only evaluations by elements of the specified predual. They coincide when the canonical image of EE exhausts E′′E'', as for , but not in general. Keeping the two entries in σ( ⋅ , ⋅ )\sigma(\,\cdot\,,\,\cdot\,) visible prevents this common ambiguity.

References
  1. Helmut H. Schaefer and Manfred P. Wolff, Topological Vector Spaces, 2nd ed., Springer, 1999. Publisher record. Relevant: Chapter IV on dual pairs and weak topologies.
  2. John B. Conway, A Course in Functional Analysis, 2nd ed., Springer, 1990. Publisher record. Relevant: Chapter V on weak convergence and duality.
Time-dependent functions

A applies this topology to continuity in time.