Hahn-Banach Theorem: Statement, Meaning and Why It Matters

The Hahn-Banach theorem says a bounded linear functional defined on a subspace extends to the whole space without increasing its norm. If Y is a subspace of a normed space X and g is a bounded linear functional on Y, there is a bounded linear functional F on X agreeing with g on Y with ||F|| = ||g||. A functional on part of the space completes to the whole space at no extra cost.

It matters because it guarantees the dual is rich: enough functionals exist to separate points, to realize any vector's norm as f(x) = ||x||, and to extend measurements from a subspace. The geometric form separates disjoint convex sets by hyperplanes, founding duality in optimization and economics.

Key Concepts

• Analytic form (extension): a functional dominated by a sublinear function on a subspace extends with the same domination; the norm-preserving case is most used.
• Geometric form (separation): a nonempty open convex set and a disjoint convex set are separated by a closed hyperplane.
• Norming functional for every nonzero x some f with ||f|| = 1 attains f(x) = ||x||.
• Sublinear functional p(x + y) <= p(x) + p(y) and p(tx) = t p(x) for t >= 0; the norm is the model example.
• Zorn's lemma: the proof extends one dimension at a time; a maximal extension must cover all of X.
• Consequences the dual separates points of X, and ||x|| = sup_{||f|| <= 1} |f(x)|.

How It Works

The proof extends the functional one dimension at a time. Given g on Y and x_0 outside Y, each extension to span{Y, x_0} is fixed by the single value c = F(x_0), and domination constrains c to a nonempty interval; any choice there works. Zorn's lemma assembles these one-step extensions into a maximal one defined on all of X. For the norm-preserving version, dominate g by p(x) = ||g|| ||x||. The geometric form follows via the Minkowski functional of an open convex set. Since Zorn's lemma gives no formula, extensions are generally non-unique and non-constructive; the power lies in the consequences.

Worked Example

Extend a functional from a subspace of R^2. Let Y = {(a, 0) : a in R} be the x-axis, with g(a, 0) = a. On Y with the Euclidean norm, |g(a, 0)| = |a| = ||(a, 0)||, so ||g|| = 1. Define F(x, y) = x on all of R^2. Then F extends g since F(a, 0) = a, and |F(x, y)| = |x| <= sqrt(x^2 + y^2) = ||(x, y)||, so ||F|| <= 1; with F(1, 0) = 1 = ||(1,0)|| we get ||F|| = 1 = ||g||. The extension preserves the norm exactly, as Hahn-Banach promises. (Other extensions like F(x, y) = x + y fail: at (1, -1) the value 0 is fine, but ||F|| would exceed 1, e.g. F(1, 1) = 2 > sqrt(2).)

Common Misconceptions

• "The extension is unique." Extensions are typically highly non-unique; Hahn-Banach only guarantees that at least one norm-preserving extension exists.
• "Hahn-Banach is constructive." The proof uses Zorn's lemma, so in general there is no formula for the extension.
• "It works for nonlinear maps." Linearity is essential; there is no general extension theorem for arbitrary continuous functions preserving a Lipschitz constant without extra hypotheses.

Real-World Applications

• Optimization: separating hyperplane theorems prove strong duality: the primal and dual optimal values coincide under convexity.
• Economics: the second welfare theorem separates an efficient allocation from preferred allocations by a price hyperplane.
• Partial differential equations: existence of weak solutions often starts by defining a functional on a subspace and extending it.
• Approximation theory: norming functionals characterize best approximations and error formulas for quadrature rules.

Hahn-Banach says functionals extend without growing: whatever linear measurement you can make on a subspace, you can make on the whole space at the same cost. The analytic form gives the extension, the geometric form gives separating hyperplanes, and the corollaries, norming functionals and point separation, are what you will actually use.