Real Analysis

The Baire category theorem states that in a complete metric space, the intersection of any countable collection of dense open sets is still dense; equivalently, a complete metric space cannot be written as a countable union of nowhere-dense sets. This gives a rigorous, topological notion of 'most' points or 'most' functions having a given property, without appealing to measure — a set is called meager (or of first category) if it is a countable union of nowhere-dense sets, and Baire's theorem says a complete metric space is never itself meager. The theorem is the engine behind several surprising existence proofs, most famously that the set of continuous nowhere-differentiable functions is 'topologically generic' (comeager) in C[0,1] — such pathological functions are not rare exceptions but overwhelmingly the norm from a category standpoint.

The Bolzano–Weierstrass theorem states that every bounded sequence of real numbers has a convergent subsequence. The sequence itself need not converge — it may oscillate forever, like aₙ = (-1)ⁿ — but boundedness alone guarantees that some subsequence settles down. The standard proof uses repeated bisection: halve an interval containing infinitely many terms, keep the half that still contains infinitely many terms, and iterate, producing nested intervals whose lengths shrink to 0 and which (by completeness of ℝ) pin down a single point that a subsequence converges to. The theorem is foundational for compactness arguments in ℝⁿ and underlies proofs of the Extreme Value Theorem and existence results throughout analysis.

A sequence (aₙ) of real numbers is Cauchy if its terms eventually become arbitrarily close to each other, not merely close to some fixed limit. Formally, for every ε > 0 there is an index N beyond which any two terms aₘ, aₙ (both indexed at least N) satisfy |aₘ − aₙ| < ε. The remarkable fact, due essentially to Cauchy and made rigorous by Weierstrass and Cantor, is that in the real numbers a sequence converges if and only if it is Cauchy — so one can verify convergence without ever knowing the limit in advance. This equivalence is exactly the completeness property of ℝ, and it fails over the rationals, where a Cauchy sequence of rationals (e.g. successive decimal truncations of √2) can fail to converge to a rational limit.

Completeness is the property distinguishing the real numbers ℝ from the rationals ℚ: informally, ℝ has 'no gaps.' It can be stated in several logically equivalent ways — the least-upper-bound (supremum) property, the Cauchy completeness property, the Monotone Convergence property, and the Nested Interval property — all of which hold in ℝ but fail in ℚ. The most common formulation, due to Dedekind, is the least-upper-bound property: every nonempty subset of ℝ that is bounded above has a least upper bound (supremum) in ℝ. Completeness is the single axiom (beyond the ordered field axioms shared with ℚ) that makes calculus work: it underlies the Intermediate Value Theorem, the Extreme Value Theorem, and the existence of limits of Cauchy sequences.

The Monotone Convergence Theorem (for sequences) states that every bounded monotone sequence of real numbers converges. If (aₙ) is increasing and bounded above, it converges to sup{aₙ}; if decreasing and bounded below, it converges to inf{aₙ}. This is one of the most direct consequences of the completeness of ℝ — the least-upper-bound property guarantees the supremum exists, and monotonicity forces the sequence to actually approach it. The theorem is a workhorse for proving convergence when the limit is not known in advance, such as showing (1+1/n)ⁿ converges (defining e) or establishing convergence of recursively defined sequences.

The Mean Value Theorem (MVT) states that if f is continuous on [a,b] and differentiable on (a,b), then there exists at least one point c ∈ (a,b) where the instantaneous rate of change f'(c) equals the average rate of change over [a,b]. Rigorously, this depends critically on both hypotheses — continuity on the closed interval and differentiability on the open interval — and is proved via Rolle's theorem, itself a consequence of the Extreme Value Theorem applied to a continuous function on a compact set. The MVT is the analytic engine behind many foundational results: it proves that a function with zero derivative everywhere on an interval is constant, that increasing/decreasing behavior is controlled by the sign of f', and it underlies the proof of Taylor's theorem with remainder.

Taylor's theorem quantifies the error made when a function f is approximated by its n-th degree Taylor polynomial Pₙ at a point a: the difference f(x) − Pₙ(x), called the remainder Rₙ(x), can be written explicitly in several equivalent forms. The Lagrange form expresses Rₙ(x) using the (n+1)-th derivative evaluated at some unknown point between a and x, directly generalizing the Mean Value Theorem (which is exactly the n=0 case). This remainder formula converts an approximation into a rigorous inequality: bounding the (n+1)-th derivative on the relevant interval bounds the error, which is what justifies using truncated Taylor series for numerical computation with guaranteed precision.

A metric space (X, d) is complete if every Cauchy sequence in X converges to a limit that lies in X itself. This generalizes the completeness of ℝ (with the usual absolute-value metric) to arbitrary spaces equipped with a notion of distance: ℝⁿ with the Euclidean metric is complete, the space C[a,b] of continuous functions with the sup-norm metric is complete, but ℚ (with the usual metric) is not, since Cauchy sequences of rationals can converge to irrational 'holes' outside ℚ. Completeness is not a purely topological property (two metrics can induce the same topology while only one is complete), but it is preserved by uniform equivalence of metrics, and every metric space embeds isometrically into a complete one, its completion — exactly how ℝ itself is constructed as the completion of ℚ.

Pointwise convergence of a sequence of functions (fₙ) to f only requires that, at each individual x, the numerical sequence fₙ(x) converges to f(x) — different x's are allowed to converge at wildly different rates. Uniform convergence demands a single rate that works across the whole domain at once. Every uniformly convergent sequence converges pointwise, but the converse is false, and the gap between the two notions matters enormously: pointwise limits can destroy continuity, integrability, and differentiability in ways uniform limits cannot.

A sequence of functions (fₙ) converges uniformly to f on a set D if the maximum gap between fₙ and f over the entire set D shrinks to zero — a single N controls the approximation everywhere in D simultaneously, unlike pointwise convergence where N may depend on the point. Uniform convergence is the stronger, more useful notion: it is exactly the condition needed to guarantee that limits of continuous functions are continuous, that limits and integrals can be interchanged, and (under mild extra hypotheses) that limits and derivatives can be interchanged. The standard tool for proving uniform convergence of series of functions is the Weierstrass M-test.

A function f is continuous at each point of its domain if, near every point a, a suitable δ can be found for each ε (δ may depend on a). Uniform continuity strengthens this: a single δ must work simultaneously for every point in the domain, for a given ε. Every uniformly continuous function is continuous, but the converse fails — f(x) = 1/x on (0,1) is continuous at every point yet not uniformly continuous, because the required δ shrinks without bound as x approaches 0. A key theorem (Heine–Cantor) guarantees that continuity and uniform continuity coincide on compact sets, such as closed bounded intervals [a,b].

The epsilon-delta (ε-δ) definition is the rigorous formalization of what it means for a function to approach a limit. Instead of relying on the intuitive but vague idea of a variable 'getting arbitrarily close' to a value, it converts the statement into a precise back-and-forth challenge: for every tolerance ε (epsilon) someone names around the limit L, you must be able to produce a corresponding neighborhood δ (delta) around the input a such that every x within δ of a (excluding a itself) makes f(x) land within ε of L. This definition, due to Bolzano, Cauchy, and finally Weierstrass, replaced two centuries of informal 'infinitesimal' reasoning and put the whole of calculus and real analysis on solid logical footing.

A sequence is a function from the natural numbers to the real numbers, written a₁, a₂, a₃, ... or (aₙ). The limit of a sequence is the value its terms 'tend to' as n grows without bound — a sequence with a finite limit is called convergent; otherwise it diverges. The rigorous ε-N definition of sequential convergence is one of the two central limiting notions of analysis (alongside the ε-δ limit of a function), and the two are intimately connected: sequential limits give the cleanest route to defining continuity, series, and completeness, and are often called the fundamental notion on which the whole of mathematical analysis ultimately rests.

Not every sequence of real numbers converges, but every bounded sequence has two canonical 'boundary values' that always exist: the limit superior (limsup) and limit inferior (liminf). The limsup is the largest value that infinitely many terms of the sequence get arbitrarily close to (the largest subsequential limit); the liminf is the smallest such value. When these two values coincide, the sequence converges to their common value; when they differ, the sequence oscillates between (at least) two distinct accumulation points and diverges. Limsup and liminf extend to unbounded sequences by allowing the values +∞ or −∞, so they are always defined for any real sequence — unlike the ordinary limit.

One of the first structural theorems relating the two central notions of single-variable calculus is that differentiability is a strictly stronger property than continuity: if a function f is differentiable at a point a, then f must also be continuous at a. The proof is a short, purely algebraic manipulation of the difference quotient, but the result has real teeth — its converse is dramatically false. Continuous functions need not be differentiable at a point (the corner of |x| at 0 is the elementary example), and in one of analysis's most startling constructions, Weierstrass exhibited a function that is continuous everywhere on ℝ yet differentiable nowhere, showing just how much weaker continuity is than differentiability.

A Fourier series represents a periodic function as an infinite sum of sines and cosines. Where a Taylor series approximates a function using powers of x near a single point, a Fourier series approximates using waves of different frequencies across the whole period — making it the natural tool whenever a signal repeats.

An infinite series Σaₙ is said to converge if its sequence of partial sums Sₙ = a₁ + a₂ + ... + aₙ converges to a finite limit as n → ∞; otherwise the series diverges. Because a series is defined entirely in terms of a sequence (its partial sums), the rigorous ε-N machinery of sequential convergence transfers directly. The central question of this topic is not computing sums in closed form (only a few special series, like geometric series, allow that) but determining WHETHER a given series converges at all — a question answered by a toolkit of convergence tests (comparison, ratio, root, integral, alternating series) developed throughout the 18th and 19th centuries.

A series Σaₙ is absolutely convergent if the series of absolute values Σ|aₙ| converges; it is conditionally convergent if Σaₙ itself converges but Σ|aₙ| diverges. Absolute convergence is the stronger notion — it implies ordinary convergence, and moreover guarantees the sum is completely robust: absolutely convergent series can be rearranged, regrouped, and manipulated in ways that are otherwise unsafe. Conditional convergence, by contrast, arises purely from delicate cancellation between positive and negative terms; the Riemann rearrangement theorem shows that a conditionally convergent series of reals can be reordered to converge to ANY target value, or to diverge, making the distinction one of the most consequential subtleties in the theory of infinite series.

The Riemann integral is the rigorous definition, due to Bernhard Riemann, of the integral of a function on an interval — the definition that makes precise the intuitive picture used throughout introductory calculus of 'area under a curve' as a limit of sums of rectangle areas. It defines the integral by approximating the region under the graph of f with finite sums of areas of thin vertical rectangles (Riemann sums), then asking whether these sums converge to a single value as the rectangles are made arbitrarily thin (the partition is refined). For every continuous function on a closed, bounded interval, this limiting value exists and defines the integral; Riemann sums that are close to the limit also serve as numerical approximations to it.

The Riemann–Stieltjes integral generalizes the Riemann integral by replacing the increment (xᵢ₊₁ - xᵢ) of the integration variable with the increment (α(xᵢ₊₁) - α(xᵢ)) of a second function, the integrator α. Written ∫ₐᵇ f dα, it reduces to the ordinary Riemann integral when α(x)=x, but allows integrating against functions with jumps, kinks, or other non-smooth behavior — in particular, when α is a step function, the Riemann–Stieltjes integral collapses to a weighted sum evaluated exactly at the jump points, unifying discrete sums and continuous integrals into a single framework. Developed by Thomas Joannes Stieltjes in 1894 while studying continued fractions, it became foundational for probability theory (integrating against a cumulative distribution function handles discrete, continuous, and mixed random variables uniformly) and for the later, more general Lebesgue–Stieltjes integral.

A function f on [a,b] has bounded variation if the total up-and-down 'wiggle' of its graph — measured by summing |f(xᵢ₊₁)-f(xᵢ)| over a partition and taking the supremum over all partitions — is finite. Introduced by Camille Jordan in 1881 while studying the convergence of Fourier series, the class of functions of bounded variation (BV) sits strictly between continuous functions and arbitrary bounded functions: every monotone function and every function with a continuous derivative on a compact interval is BV, but BV functions can still have jump discontinuities (just only countably many, and each of finite size). The single most important structural fact is the Jordan decomposition: every BV function can be written as the difference of two monotone increasing functions — reducing many questions about general BV functions to the much simpler monotone case.

A Banach space is a normed vector space that is complete with respect to the metric induced by its norm. Completeness means every Cauchy sequence of vectors converges to a limit within the space. Banach spaces are the natural setting for most of functional analysis: they include all finite-dimensional normed spaces, the sequence spaces \ell^p (1 \leq p \leq \infty), the function spaces L^p(\mu) for a measure \mu, and the space C([a,b]) of continuous functions with the sup-norm. The structure of Banach spaces underlies the three pillars of functional analysis: the Hahn-Banach theorem, the Open Mapping theorem, and the Uniform Boundedness Principle.

A Hilbert space is a complete inner product space — an infinite-dimensional generalisation of Euclidean space. The inner product provides notions of angle, orthogonality, and projection that go beyond a bare norm. The most important examples are L^2(\mu) (square-integrable functions) and \ell^2 (square-summable sequences). Hilbert spaces are the natural setting for quantum mechanics, Fourier analysis, and the theory of PDEs.

The Arzela-Ascoli theorem characterises the compact subsets of C(K) — the space of continuous functions on a compact metric space K with the sup-norm. A set F of functions is relatively compact (has compact closure) if and only if it is uniformly bounded and equicontinuous. Equicontinuity is the condition that all functions in F are uniformly continuous with a single modulus of continuity. The theorem is the function-space analogue of the Bolzano-Weierstrass theorem (bounded sequences in R^n have convergent subsequences) and is a key tool in the existence theory for ODEs, PDEs, and integral equations.

The Stone-Weierstrass theorem is a vast generalisation of Weierstrass's classical result that every continuous function on [a,b] can be uniformly approximated by polynomials. Marshall Stone's theorem (1948) identifies exactly which subalgebras of C(K) are dense: a subalgebra A of C(K) (K compact Hausdorff) is dense in the sup-norm if and only if it separates points and vanishes nowhere (or, for the real version, if A is a real subalgebra that separates points and contains constants). This single result unifies the density of polynomials, trigonometric polynomials, and many other approximating families.

A compact operator is a bounded linear operator between Banach spaces that maps bounded sets to precompact sets (sets whose closure is compact). Compact operators are in many respects the closest infinite-dimensional analogue of finite-rank (matrix) operators: their spectrum behaves like that of a finite matrix (at most countably many nonzero eigenvalues accumulating only at 0), and the Fredholm alternative holds—either \(T - \lambda I\) is invertible or \(\lambda\) is an eigenvalue with finite-dimensional eigenspace. Compact operators arise naturally as integral operators, as solution operators for elliptic PDEs via Sobolev embeddings, and in the spectral theory of differential operators.

The theory of distributions (generalized functions), developed by Laurent Schwartz in the 1940s, extends the notion of a function to objects that may not be defined pointwise but act on smooth test functions through integration. Every locally integrable function defines a distribution, but distributions also include objects like the Dirac delta \(\delta_0\)—which assigns to each test function its value at the origin—and derivatives of any order of any distribution. This framework resolves the problem that many natural 'functions' arising in physics and PDE theory (impulse forces, point charges, fundamental solutions) are not classical functions. Every distribution is infinitely differentiable in the distributional sense.

Functional analysis is the branch of mathematics concerned with the study of vector spaces equipped with a topology—typically induced by a norm or inner product—and the linear operators acting between them. It unifies and vastly generalises the classical theories of ordinary and partial differential equations, Fourier analysis, and linear algebra. The central objects are infinite-dimensional normed spaces (Banach and Hilbert spaces), bounded linear operators between them, and the dual spaces of continuous linear functionals. Three cornerstone theorems—the Hahn–Banach theorem, the open mapping theorem, and the uniform boundedness principle—give the field its distinctive flavour and power.

The Hahn–Banach theorem is one of the three pillars of functional analysis. In its most basic form it asserts that a bounded linear functional defined on a subspace of a normed space can be extended to the entire space without increasing its norm. The theorem has two flavours: the analytic extension form and the geometric separation form. The latter—that two disjoint convex sets can be separated by a hyperplane—is the cornerstone of convex analysis, optimization, and mathematical economics. The theorem requires no completeness (it holds for all normed spaces) and is proved via Zorn's lemma.

The open mapping theorem (Banach–Schauder theorem) and the closed graph theorem are two of the three fundamental principles of functional analysis, alongside the Hahn–Banach theorem. The open mapping theorem states that a surjective bounded linear operator between Banach spaces is an open map—it sends open sets to open sets. The closed graph theorem provides a complementary criterion: a linear operator between Banach spaces is bounded if and only if its graph is closed. Both theorems rest on the Baire category theorem applied to the Banach space structure and have profound consequences for operator theory and PDE analysis.

Sobolev spaces \(W^{k,p}(\Omega)\) are function spaces that combine \(L^p\) integrability with control on weak derivatives up to order \(k\). They are the natural setting for the variational (weak) formulation of partial differential equations: classical solutions require pointwise smoothness, but Sobolev spaces allow 'solutions' that satisfy an integrated version of the PDE, dramatically expanding existence theory. The Sobolev embedding theorems control how much regularity is gained or preserved as one varies \(k\) and \(p\), and trace theorems allow boundary conditions to be imposed rigorously on domains with boundary.

Spectral theory extends the notion of eigenvalues and eigenvectors from finite-dimensional linear algebra to operators on infinite-dimensional Banach and Hilbert spaces. The spectrum of an operator \(T\) replaces the finite set of eigenvalues and decomposes into the point spectrum (eigenvalues), continuous spectrum, and residual spectrum. For bounded self-adjoint operators on Hilbert spaces, the spectral theorem provides a complete decomposition analogous to diagonalisation: every such operator can be realised as a multiplication operator on an \(L^2\) space, encoded by a projection-valued measure.

The weak topology on a Banach space \(X\) is the coarsest topology that makes every bounded linear functional \(f \in X^*\) continuous. It is strictly coarser than the norm topology in infinite dimensions, producing a richer landscape of convergent sequences and compact sets. Weak convergence is fundamental in the calculus of variations, PDE theory, and optimization: minimizing sequences often converge weakly but not in norm, and weak compactness of bounded sets (in reflexive spaces) is the key tool for extracting convergent subsequences.