Hilbert space

A Hilbert space is a complete inner product space — a vector space equipped with an inner product (a generalization of the dot product) that induces a norm, and which is complete with respect to that norm. Every Hilbert space is a Banach space, but the additional structure of the inner product gives Hilbert spaces a geometry that Banach spaces lack: orthogonality, orthogonal projection, and self-duality via the Riesz representation theorem. This geometry makes Hilbert spaces the natural habitat of quantum mechanics, Fourier analysis, and stochastic processes.

Definition and Basic Properties

Formally, a Hilbert space H is a vector space over the real or complex numbers equipped with an inner product ⟨·, ·⟩ that satisfies:

• Conjugate symmetry: ⟨x, y⟩ = ⟨y, x⟩*
• Linearity in the first argument: ⟨ax + by, z⟩ = a⟨x, z⟩ + b⟨y, z⟩
• Positive-definiteness: ⟨x, x⟩ ≥ 0 with equality iff x = 0

The inner product induces a norm ||x|| = √⟨x, x⟩, and completeness requires that every Cauchy sequence converges in this norm. The simplest infinite-dimensional Hilbert space is l², the space of square-summable sequences. The prototypical example in analysis is L²(ℝ), the space of square-integrable functions, which underlies quantum mechanics and signal processing.

Orthogonality and Projection

The inner product structure allows a notion of orthogonality: two vectors x and y are orthogonal if ⟨x, y⟩ = 0. This leads to the orthogonal decomposition: for any closed subspace M of a Hilbert space, every vector can be written uniquely as the sum of a vector in M and a vector in its orthogonal complement M⊥. The map that sends a vector to its component in M is the orthogonal projection, and it is the unique closest-point projection in the norm.

This property is not shared by general Banach spaces. In a Banach space, a closed subspace may not admit any complement at all, let alone an orthogonal one. The existence of orthogonal projection is what makes Hilbert spaces computationally tractable: optimization problems, least-squares approximation, and the Gram-Schmidt process all rely on it.

The Riesz Representation Theorem

A fundamental result is that every continuous linear functional on a Hilbert space can be represented as an inner product with a unique vector. That is, for every bounded linear functional φ, there exists a unique y ∈ H such that φ(x) = ⟨x, y⟩ for all x. This establishes a canonical isomorphism between H and its dual space H*, making Hilbert spaces self-dual.

This self-duality is not merely a convenience. It is the mathematical reason why quantum mechanical states and observables can be represented in the same space, why the bra-ket notation of Dirac is coherent, and why the spectral theorem takes its particularly elegant form in Hilbert spaces.

Examples and Applications

• l² — the space of sequences (xₙ) with Σ|xₙ|² < ∞
• L²(Ω) — the space of square-integrable functions on a measure space Ω
• Sobolev space|Hᵏ — the space of functions whose derivatives up to order k are square-integrable
• Reproducing kernel Hilbert space|RKHS — spaces where point evaluation is a continuous functional

These spaces appear throughout physics and engineering: quantum state spaces, signal processing, control theory, and machine learning. The Karhunen-Loève theorem provides the optimal basis for stochastic processes in Hilbert spaces, while the Sobolev embedding theorem governs how functions in these spaces behave in classical settings.

The obsession with Hilbert spaces in physics education often obscures a deeper truth: the inner product is not a gift from nature but a modeling choice. We use Hilbert spaces because we can compute in them, not because the universe insists on orthogonality. There are physical systems — notably certain dissipative and non-equilibrium systems — where the natural state space is Banach but not Hilbert, and forcing an inner product onto such systems produces artifacts rather than insights. Hilbert spaces are the crown jewels of functional analysis, but they are not the whole kingdom.