Chapter 8 · Introduction

Chapter 8 — Functional Analysis

Linear algebra in infinite dimensions: spaces of sequences and functions with a norm, the bounded operators between them, the three great theorems that completeness buys, Hilbert spaces, and the spectral theorem for compact self-adjoint operators.

Around 1900 Ivar Fredholm and David Hilbert were solving integral equations such as $u(x)-\int_0^1K(x,t)\,u(t)\,dt=f(x)$, which arise from vibrating strings, heat flow and potential theory. Fredholm noticed that these equations behave like systems of linear equations with infinitely many unknowns; Hilbert made the idea precise by working in the space of square-summable sequences, where a function is replaced by its list of Fourier coefficients. Two decades later, in Lwów, Stefan Banach and his colleagues — famously working at the tables of the Scottish Café — turned this into a general theory: a function is a point of a vector space, its size is a norm, and an integral or differential operator is a linear map between such spaces.

What makes the subject worth a chapter is that infinite dimensions are genuinely different. A linear map can be discontinuous; a bounded sequence can have no convergent subsequence; a norm on a space of functions can fail to be complete. Yet when the space is complete — a Banach space — three remarkable theorems hold: pointwise bounded families of operators are uniformly bounded, bounded bijections have bounded inverses, and operators with closed graphs are bounded. Add an inner product and the geometry of the plane comes back: orthogonal projections, orthonormal bases, and the fact that every bounded linear functional is an inner product with a fixed vector. The chapter ends with the infinite-dimensional version of "a symmetric matrix can be diagonalised", the result that powers the theory of integral equations and Sturm–Liouville problems.

Norms & Banach spaces→Bounded operators→Dual spaces→Hahn–Banach→Baire→UBP · OMT · CGT→Hilbert spaces→Projection & Riesz→Spectral theorem
🎯 By the end of this chapter you can…
  • decide whether a given normed space ($\ell^p$, $c_0$, $c_{00}$, $C[a,b]$, $L^p$, a subspace, a new norm) is complete, and prove it;
  • compute norms of matrices, integral operators, multiplication operators and linear functionals, and decide whether the norm is attained;
  • identify the dual spaces of $\ell^p$, $c_0$ and $L^p$, and use the Hahn–Banach theorem in its extension and separation forms;
  • apply the uniform boundedness principle, the open mapping and bounded inverse theorems and the closed graph theorem — and explain, with counterexamples, why each needs completeness;
  • work in Hilbert spaces: orthogonal complements, projections, Bessel's inequality and Parseval's identity, orthonormal bases and the Riesz representation theorem;
  • recognise compact and self-adjoint operators and use the spectral theorem for compact self-adjoint operators to find eigenvalues and solve operator equations.
Teaser — The Unit Balls of $\ell^p$ and Their Duals

The solid curve is the unit sphere $|x|^p+|y|^p=1$ of the $p$-norm on $\mathbb{R}^2$; the dashed curve is the unit sphere of the conjugate norm, $\tfrac1p+\tfrac1q=1$, which measures the size of linear functionals (§8.1). Watch $p$ run from $1$ to $\infty$. Only at $p=2$ do the two curves coincide — and only then does the parallelogram law hold, which is exactly the condition for a norm to come from an inner product (§8.3).

p = 1.33
p (solid)—
q (dashed)—
‖x+y‖² + ‖x−y‖²—
2‖x‖² + 2‖y‖²—

How this is examined

GATE MA regularly asks about functional analysis, usually in a form that rewards knowing the standard examples cold. Typical MCQs ask which of several normed spaces is complete, or which of several maps is a bounded linear operator. NAT questions ask for a number: the norm of a functional such as $f\mapsto\int_0^1 g(t)f(t)\,dt$ on $C[0,1]$, the norm of a functional on $\ell^p$ (where the conjugate exponent does the work), the norm of a matrix or an integral operator, a distance to a subspace in a Hilbert space, or an eigenvalue of a compact integral operator. MSQs list four statements about the big theorems and their hypotheses — "if $X$ and $Y$ are Banach spaces and $T$ is a bounded bijection, then…", "every Cauchy sequence in $(C[0,1],\|\cdot\|_1)$…", "the Hahn–Banach extension is unique…" — and every false option is one of the counterexamples collected in the ⚠️ boxes of the four sections. The same material appears in CSIR NET Part C and in the functional-analysis part of university MSc examinations.

Sections in this chapter

✅ Before you start

You need linear algebra — vector spaces, linear maps, matrices, eigenvalues and inner products (§3.1, §3.3) — and the metric-space language of §4.1: open and closed sets, Cauchy sequences, completeness and compactness. Uniform convergence and the Weierstrass approximation theorem (§4.2) are used for $C[a,b]$, and the Lebesgue integral with the monotone and dominated convergence theorems (§4.3) is needed for $L^p$. For a slower first pass through norms, $C[a,b]$, equivalent norms and the Baire category theorem with full proofs, see the CSIR NET course: CSIR Unit 1 · §7.1 Metric Spaces and Completeness and CSIR Unit 1 · §7.4 Normed Spaces and $C[a,b]$.