Infinite-Dimensional Spaces

Infinite-dimensional linear algebra is the backbone of functional analysis: Banach and Hilbert spaces (spaces of functions, sequences, distributions) are vector spaces whose Hamel dimension is always uncountable but whose useful structure comes from a Schauder basis or an orthonormal basis, replacing finite Hamel sums with convergent infinite series. This is essential in Fourier analysis (the basis $\{e^{inx}\}$ for $L^2$), quantum mechanics (the state space of a quantum system is an infinite-dimensional Hilbert space, and observables are operators, not matrices), and numerical analysis (finite element methods approximate infinite-dimensional solution spaces by finite-dimensional ones). In probability theory, the existence of a Hamel basis for the space of all real-valued functions is used to construct pathological, non-measurable objects (Hamel-basis-derived functions solving Cauchy's functional equation without being linear, related to Vitali's non-measurable sets). In theoretical computer science and cryptography, infinite-dimensional vector spaces over finite fields (formal power series, infinite-dimensional codes) appear in coding theory and lattice-based cryptography, where the finite-versus-infinite distinction governs which algorithms terminate. In every one of these settings, the moral of this article recurs: the algebraic theorem (a basis exists, dimension is well-defined) is true and useful for abstract classification, but the practical tools (orthonormal expansions, generating functions, approximation schemes) are topological refinements built on top of it, not substitutes for it.

:::quiz[Why is the discontinuous linear functional constructed on $C[0,1]$ in Example 2 not something one can write down with an explicit formula?]

• Because $C[0,1]$ is finite-dimensional
• Because its construction depends on a Hamel basis of $C[0,1]$, whose existence uses the Axiom of Choice and has no explicit description ✓
• Because linear functionals on normed spaces are always continuous
• Because $C[0,1]$ has no linearly independent uncountable subsets

Ask the tutor: Why must a Hamel basis of an infinite-dimensional Banach space always be uncountable, and how does the Baire Category Theorem prove this?

In a finite-dimensional space, a basis is a finite list of vectors, and every element of the space is reached by a single, finite linear combination — this is baked into the definition above, and it does not change when the space becomes infinite. What changes is only the size of the list. The word "infinite" in "infinite-dimensional" refers purely to the cardinality of the basis, never to the length of the linear combinations used to reach a vector: a Hamel basis combination is always a finite sum, even in a space of dimension the continuum. This is the single most important, and most frequently misunderstood, fact in the subject. It is tempting to think that in the space of all sequences, an element like $(1, \tfrac12, \tfrac14, \tfrac18,\dots)$ is "the infinite linear combination" $\sum_{n\ge 0} 2^{-n} e_n$ of the standard unit vectors $e_n$. This is true in the sense of a convergent series — but convergence is a topological, analytic notion, requiring a norm or metric, that plain vector-space (Hamel) theory does not have access to. As a linear-algebraic object with no topology, this sequence is not a finite combination of the $e_n$, so the $e_n$ alone do not Hamel-span the sequence space; the Hamel basis must be enlarged by an enormous, non-constructive infinite (indeed uncountable) set of further vectors to reach it algebraically. The picture below contrasts the two notions on the sequence space $\mathbb{R}^\mathbb{N}$: the "easy" subspace of finitely-supported sequences (Hamel-spanned by the $e_n$) versus the much bigger full space, whose Hamel basis needs uncountably many extra vectors that have no simple description.

Example 1: The space of all real sequences has dimension $2^{\aleph_0}$ (the continuum), realized abstractly but not concretely. Let $V = \mathbb{R}^{\mathbb{N}}$, the vector space of all sequences $(x_0,x_1,x_2,\dots)$ of real numbers, with coordinatewise addition and scalar multiplication. We determine $\dim V$. Step 1: $\dim V \ge 2^{\aleph_0}$. Consider the family of sequences $\{r^{(t)}\}_{t \in \mathbb{R}}$ defined by $r^{(t)}_n = t^n$ for $n = 0,1,2,\dots$ (so $r^{(t)} = (1, t, t^2, t^3, \dots)$). We claim any finite sub-collection $r^{(t_1)},\dots,r^{(t_k)}$ with distinct $t_1,\dots,t_k$ is linearly independent. Suppose

Looking at coordinate $n$ gives $\sum_{i=1}^k c_i t_i^n = 0$ for every $n \ge 0$. Truncating to $n = 0,1,\dots,k-1$ gives a linear system with matrix $(t_i^{\,n})_{0\le n \le k-1,\ 1\le i\le k}$, whose transpose is the Vandermonde matrix on distinct nodes $t_1,\dots,t_k$; its determinant is $\prod_{i<j}(t_j - t_i) \ne 0$ since the $t_i$ are distinct. Hence the system forces $c_1=\cdots=c_k=0$. So the uncountable family $\{r^{(t)}: t\in\mathbb{R}\}$ is linearly independent, and since $|\mathbb{R}| = 2^{\aleph_0}$, we get $\dim V \ge 2^{\aleph_0}$. Step 2: $\dim V \le 2^{\aleph_0}$. As a set, $V = \mathbb{R}^{\mathbb{N}}$ has cardinality $|\mathbb{R}|^{\aleph_0} = (2^{\aleph_0})^{\aleph_0} = 2^{\aleph_0 \cdot \aleph_0} = 2^{\aleph_0}$. Any linearly independent subset of $V$, in particular any Hamel basis, has cardinality at most $|V| = 2^{\aleph_0}$. Combining both bounds, $\dim V = 2^{\aleph_0}$. But note: the proof of Step 1 shows a specific uncountable linearly independent set, not a basis — none of the $r^{(t)}$ span $V$ (for instance the sequence $(1,0,1,0,1,0,\dots)$ is not obviously a finite combination of finitely many $r^{(t)}$'s in any explicit way, and completing $\{r^{(t)}\}$ to an actual Hamel basis requires invoking Zorn's Lemma again, adding further vectors with no closed-form description). This illustrates concretely why $\dim(\mathbb{R}^\mathbb{N}) = 2^{\aleph_0}$ is a true, provable cardinal number, yet no one can write down an actual basis — existence is guaranteed by pure set theory, not by construction.

Example 2: A discontinuous linear functional on an infinite-dimensional normed space, built from a Hamel basis. Let $V = C[0,1]$, the vector space of continuous real functions on $[0,1]$, equipped with the sup-norm $\|f\|_\infty = \sup_{x\in[0,1]} |f(x)|$. We construct a linear functional $\varphi: V \to \mathbb{R}$ that is linear but not continuous — a phenomenon impossible for linear maps out of a finite-dimensional normed space. By the Existence Theorem, extend the linearly independent singleton $\{1\}$ (the constant function $1$) to a Hamel basis $B$ of $V$ with $1 \in B$. Since $C[0,1]$ is infinite-dimensional (it contains the linearly independent set $\{1,x,x^2,x^3,\dots\}$, independent by the same Vandermonde-type / degree argument as for polynomials), $B$ is infinite, so we may pick some other element $g \in B$ with $g \ne 1$. Define $\varphi: V \to \mathbb{R}$ on the basis $B$ by

where we additionally arrange (by rescaling within the construction, or just by picking, using the Hamel-basis extension freedom, a sequence $g_k \in B$ with $\|g_k\|_\infty \le 1$ for each $k$ — this is possible because we may replace any basis vector $b$ by $b/\|b\|_\infty$, which remains a valid basis) that $\|g_k\|_\infty \le 1$ for all $k$. Extend $\varphi$ to all of $V$ by linearity: every $f \in V$ has a unique finite basis expansion $f = \sum c_b b$, and we set $\varphi(f) = \sum c_b \varphi(b)$ (a finite sum, well-defined by uniqueness of Hamel coordinates). This $\varphi$ is a well-defined linear functional by the same coordinate argument as in the Direct Sum Corollary. Now $\varphi$ is unbounded: $\varphi(g_k) = k$ while $\|g_k\|_\infty \le 1$, so

A linear functional on a normed space is continuous if and only if it is bounded on the unit ball (a standard equivalence for linear maps between normed spaces). Since $\varphi$ is unbounded on the unit ball, $\varphi$ is discontinuous — indeed, discontinuous at every point, since linear maps are continuous everywhere or nowhere. Note this construction is entirely non-constructive: it relies on the existence of a Hamel basis of $C[0,1]$, which (as in Example 1) cannot be exhibited explicitly.