Skip to content
🔬 ScientistPure mathematics — no exam syllabus

Real Analysis — The Foundations of Calculus

Real analysis is calculus made rigorous. For centuries, mathematicians used derivatives and integrals without a precise understanding of what they were. Limits were "approaching" but never arriving. Infinitesimals were "infinitely small" but nobody could define them. In the 19th century, Cauchy, Weierstrass, Dedekind, and Riemann rebuilt calculus from the ground up — replacing vague intuition with epsilon-delta precision. The result is real analysis: the study of the real number system, sequences, continuity, differentiation, and integration, all built on the solid foundation of the completeness axiom.

"God created the integers, all else is the work of man."
— Leopold Kronecker

📏 The Real Numbers — Filling the Gaps

Why the Rationals Have Holes and How We Filled Them
The ancient Greeks discovered that √2 is irrational — it can't be written as a fraction. But geometrically, √2 is the length of the diagonal of a unit square. It must exist. This gap between the rational numbers (which have holes) and the geometric continuum was a scandal for 2000 years. In 1872, Dedekind solved it with his cut: he defined a real number as a partition of the rationals into two sets L and R where every element of L is less than every element of R, and L has no largest element. The cut for √2 puts all rationals with x² < 2 into L and all with x² > 2 into R. This construction fills every gap in the rationals, creating the continuum of real numbers — the completeness of ℝ.
Real numbers are not given by nature — they are constructed to make the number line continuous.

The Completeness Axiom

Every non-empty set of real numbers that is bounded above has a least upper bound (supremum). This is the completeness axiom, and it's the single property that distinguishes ℝ from ℚ. In ℚ, the set {x : x² < 2} has no supremum (√2 is missing). In ℝ, it does. From completeness, everything else follows: the intermediate value theorem, the extreme value theorem, the convergence of Cauchy sequences, the existence of limits.

Completeness Axiom
Every non-empty set bounded above has a supremum
This one property guarantees the continuum has no gaps.

Archimedean Property

For any real number x, there exists a natural number n such that n > x. This seemingly obvious statement — that there are no infinite real numbers — is a consequence of completeness. It implies the rationals are dense in ℝ: between any two real numbers, there is a rational.

🔢 Sequences — Approaching Infinity

What Does It Mean for a Sequence to Converge?
A sequence (aₙ) converges to L if for every ε > 0, there exists N such that for all n ≥ N, |aₙ − L| < ε. This epsilon-definition, perfected by Weierstrass, removes all vagueness from "approaches." The sequence 1, 1/2, 1/3, 1/4, ... converges to 0 because for any ε, choose N > 1/ε and all subsequent terms are within ε. A Cauchy sequence is one where terms get arbitrarily close to each other: for every ε > 0, there exists N such that for all m,n ≥ N, |aₘ − aₙ| < ε. In ℝ, every Cauchy sequence converges — this is equivalent to completeness and is the standard way to define completeness in metric spaces.
A sequence converges not when it "reaches" its limit, but when it can get arbitrarily close and stay there.

Key Theorems

Monotone Convergence
Bounded monotone sequences converge
Bolzano-Weierstrass
Every bounded sequence has a convergent subsequence
Squeeze Theorem
If aₙ ≤ bₙ ≤ cₙ and aₙ,cₙ → L then bₙ → L
Cauchy Criterion
A sequence converges iff it is Cauchy
✨
Proof That √2 Exists (Using Completeness)▶
Let S = {x ∈ ℝ : x² < 2}. S is non-empty (1 ∈ S) and bounded above (2 is an upper bound). By completeness, S has a supremum s. We show s² = 2.

If s² < 2, then for small h > 0, (s+h)² = s² + 2sh + h² < 2 for sufficiently small h, so s+h ∈ S — contradiction since s is an upper bound.

If s² > 2, then for small h > 0, (s−h)² > 2, so s−h is a smaller upper bound — contradiction.

Therefore s² = 2, and s = √2. This proof works only because of completeness — in ℚ, no such s exists.

Series — Adding Infinitely Many Numbers

A series Σ aₙ converges if the sequence of partial sums converges. The geometric series Σ rⁿ converges to 1/(1−r) when |r| < 1. The harmonic series Σ 1/n diverges (slowly — like ln n). The alternating harmonic series Σ (−1)ⁿ⁺¹/n converges to ln 2. A series converges absolutely if Σ |aₙ| converges; otherwise it converges conditionally. The Riemann rearrangement theorem says conditionally convergent series can be rearranged to sum to any value — absolute convergence is the only kind that's well-behaved.

🎯 Continuity — Functions Without Gaps

Cauchy and Weierstrass Tame the Infinite
A function f is continuous at a if for every ε > 0, there exists δ > 0 such that |x − a| < δ implies |f(x) − f(a)| < ε. This is the famous epsilon-delta definition. Intuitively: if you want the output to stay within a tiny window ε, you can find a tiny input window δ that guarantees it. Continuous functions on closed intervals have the Intermediate Value Property (they take every value between f(a) and f(b)) and the Extreme Value Property (they attain their maximum and minimum). Both rely on completeness — they fail on ℚ.
Continuity means small changes in input produce small changes in output — and the size of "small" can be chosen uniformly.

Uniform Continuity

A function is uniformly continuous on a set if δ can be chosen independently of the point — the same δ works everywhere. Continuous functions on closed bounded intervals are automatically uniformly continuous (Heine-Cantor theorem). On unbounded intervals, this fails: f(x) = x² is continuous on ℝ but not uniformly continuous — the δ required to keep changes small shrinks as x grows.

Differentiability — The Best Linear Approximation

A function is differentiable at a if the limit f'(a) = lim_{h→0} (f(a+h) − f(a))/h exists. Geometrically, f'(a) is the slope of the tangent line — the best linear approximation to f near a. Differentiability implies continuity, but not vice versa (f(x) = |x| is continuous but not differentiable at 0). The Mean Value Theorem says there's a point where the instantaneous slope equals the average slope: f'(c) = (f(b)−f(a))/(b−a).

Mean Value Theorem
f'(c) = (f(b) − f(a)) / (b − a) for some c ∈ (a,b)
If f(a) = f(b), then f'(c) = 0 (Rolle's theorem).

Taylor's Theorem — Power Series Approximations

If f is n+1 times differentiable, then f(x) can be approximated by its Taylor polynomial with an error term: f(x) = Σ_{k=0}^{n} f^{(k)}(a)(x−a)ᵏ/k! + Rₙ(x). The remainder Rₙ(x) can be expressed in various forms (Lagrange, Cauchy, integral). For analytic functions, the Taylor series converges to f within its radius of convergence.

📊 Integration — Measuring Area Under Curves

How to Make Area Rigorous
The Riemann integral partitions the domain into subintervals, forms rectangles whose heights are the function values at sample points, and takes the limit as the partition becomes finer. A function is Riemann integrable if this limit exists and is independent of the sample points chosen. Continuous functions are Riemann integrable, as are monotone functions and functions with countably many discontinuities. The Fundamental Theorem of Calculus connects integration and differentiation: the derivative of the integral is the original function, and the integral of the derivative is the change in the original function.
Integration is not just area — it's the inverse of differentiation, and its existence depends on continuity (which depends on completeness).
FTC Part 1
d/dx ∫ₐˣ f(t) dt = f(x)
FTC Part 2
∫ₐᵇ f'(x) dx = f(b) − f(a)
Integration by Parts
∫ u dv = uv − ∫ v du
Improper Integral
∫ₐ^∞ f(x) dx = lim_{t→∞} ∫ₐᵗ f(x) dx

Sequences of Functions — When Can We Swap Limits?

When does lim_{n→∞} lim_{x→a} fₙ(x) = lim_{x→a} lim_{n→∞} fₙ(x)? Not always! Pointwise convergence — each x individually — is too weak to allow swapping. Uniform convergence — where the rate of convergence is independent of x — preserves continuity, differentiability, and integrability, allowing limits to be swapped. The Weierstrass M-test gives a sufficient condition for uniform convergence of series.

Beyond Riemann — The Lebesgue Integral

The Riemann integral has limitations: it doesn't handle functions with dense discontinuities well, and it doesn't play nicely with limits. The Lebesgue integral partitions the range instead of the domain, leading to a more powerful theory that integrates more functions and allows better limit theorems (dominated convergence, monotone convergence). This is the foundation of measure theory and modern analysis, essential for probability theory and partial differential equations.

"Analysis is the art of taking limits, and real analysis is the science of when you're allowed to."
— Anonymous analyst

🎯 Practice — Check Your Analysis Intuition

These puzzles build real analysis intuition — think epsilon!

← Back to Math Map

Metric Spaces and Completeness

A metric space is a set X equipped with a distance function d : X × X → ℝ satisfying positivity, symmetry, and the triangle inequality. The real line ℝ with the usual distance d(x,y) = |x−y| is the prototype. A sequence (xₙ) in a metric space is Cauchy if for every ε > 0 there exists N such that for all m,n ≥ N, d(xₘ,xₙ) < ε. The space is called complete when every Cauchy sequence converges to a point of the space. ℝ is complete; ℚ is not because a Cauchy sequence of rationals can converge to an irrational limit. Completeness is the analytic counterpart of the least‑upper‑bound property and underpins the existence of limits for iterative processes such as Newton’s method or the contraction mapping principle. In a complete metric space, the Banach fixed‑point theorem guarantees a unique fixed point for any contraction mapping, a tool used repeatedly in differential equations and numerical analysis.

Important examples of complete metric spaces include ℝⁿ with the Euclidean metric, the space ℓ² of square‑summable sequences, and the space C([a,b]) of continuous functions on a closed interval with the supremum norm. Incomplete spaces arise naturally: the space of polynomials on [0,1] with the supremum norm is not complete; its completion is C([0,1]). The process of completing a metric space by adding limits of all Cauchy sequences yields a complete space unique up to isometry. This construction mirrors the passage from ℚ to ℝ and shows that completeness is not a peculiarity of the real numbers but a structural property that can be imposed on any metric space.

Compactness in ℝⁿ

Compactness captures the idea that a set is “small” in a topological sense while retaining the completeness of closed bounded intervals. In ℝⁿ the Heine‑Borel theorem states that a subset is compact iff it is closed and bounded. Equivalently, every open cover admits a finite subcover. Sequential compactness — every sequence has a convergent subsequence — coincides with the covering definition in metric spaces. The Bolzano‑Weierstrass theorem is the one‑dimensional incarnation: every bounded sequence in ℝ has a convergent subsequence. Compactness is preserved under continuous images: if f is continuous and K is compact, then f(K) is compact. Consequently a continuous real‑valued function on a compact set attains its maximum and minimum (extreme value theorem) and is uniformly continuous (Heine‑Cantor theorem). These facts are the backbone of many existence proofs in optimisation and the calculus of variations.

In infinite‑dimensional normed spaces closed and bounded no longer implies compact; the unit ball of ℓ² is closed and bounded but not compact. This failure motivates the study of weak topologies and the notion of relative compactness. The Arzelà‑Ascoli theorem characterises compact subsets of C([a,b]) as those that are uniformly bounded and equicontinuous. It provides a practical criterion for extracting convergent subsequences from families of functions, a technique essential in the proof of the Peano existence theorem for ordinary differential equations and in the theory of integral equations.

Worked Example: Uniform Convergence on a Compact Set
Let fₙ(x) = xⁿ on [0,1]. Show that (fₙ) converges pointwise but not uniformly. Then prove that on any compact subinterval [0,a] with 0 ≤ a < 1 the convergence is uniform.
Pointwise limit: for x ∈ [0,1) we have xⁿ → 0; for x = 1 we have 1ⁿ = 1. Hence the pointwise limit function f(x) is 0 on [0,1) and 1 at x = 1. This limit is not continuous, so the convergence cannot be uniform on [0,1] (uniform limit of continuous functions is continuous). On [0,a] with a < 1, supx∈[0,a] |xⁿ−0| = aⁿ → 0 as n → ∞. Given ε > 0 choose N such that aᴺ < ε; then for all n ≥ N and all x ∈ [0,a], |fₙ(x)−0| ≤ aⁿ < ε. Thus convergence is uniform on [0,a].
Exam‑Day Strategy
When a question asks about uniform convergence, first check continuity of the pointwise limit. If the limit is discontinuous on a compact domain, uniform convergence is impossible. For positive results, bound the supremum of the difference by a quantity independent of x; the Weierstrass M‑test is the standard tool for series of functions.

The Baire Category Theorem

The Baire category theorem is a cornerstone of modern analysis. In a complete metric space (or a locally compact Hausdorff space) the intersection of countably many dense open sets is dense. Equivalently, a non‑empty complete metric space cannot be expressed as a countable union of nowhere‑dense closed sets. A set is nowhere dense if its closure has empty interior. The theorem implies that “most” points of a complete space belong to a residual set (a countable intersection of dense open sets). This abstract statement has concrete consequences: the set of continuous nowhere‑differentiable functions is residual in C([0,1]); the set of points where a separately continuous function fails to be jointly continuous is of first category. The proof uses a nested‑ball construction: given dense open sets Uₖ, pick a closed ball B₁ ⊂ U₁, then B₂ ⊂ U₂ ∩ interior(B₁) with radius halved, and so on. Completeness guarantees the intersection of the balls contains a point belonging to all Uₖ.

Applications extend to functional analysis. The uniform boundedness principle (Banach‑Steinhaus) follows from Baire: if a family of continuous linear operators is pointwise bounded on a Banach space, it is uniformly bounded on some neighbourhood. The open mapping theorem and the closed graph theorem are also derived via Baire category arguments. In measure theory, the theorem shows that a non‑empty complete metric space without isolated points is uncountable, providing a topological proof that ℝ is uncountable without diagonalisation.

Differentiation of Integrals and the Lebesgue Differentiation Theorem

The Lebesgue differentiation theorem states that for any locally integrable function f on ℝⁿ, the average value of f over balls centred at x converges to f(x) for almost every x as the radius tends to zero. Formally, limr→0 (1/|B(x,r)|) ∫B(x,r) f(y) dy = f(x) a.e. This result generalises the fundamental theorem of calculus to the Lebesgue setting and shows that the indefinite integral of an L¹ function is differentiable almost everywhere with derivative equal to the original function. The proof relies on the Hardy‑Littlewood maximal function Mf(x) = supr>0 (1/|B(x,r)|) ∫B(x,r) |f(y)| dy and the weak‑type (1,1) inequality |{x : Mf(x) > λ}| ≤ Cₙ ‖f‖₁ / λ. The maximal inequality controls the set where the averages deviate significantly from f, allowing a covering argument (Vitali covering lemma) to establish almost‑everywhere convergence.

Consequences include the Lebesgue density theorem: for any measurable set E ⊂ ℝⁿ, almost every point of E has density 1, and almost every point of the complement has density 0. This underpins the structure theory of sets of finite perimeter and the theory of functions of bounded variation. In probability, the differentiation theorem justifies the almost‑sure convergence of conditional expectations with respect to a filtration generated by shrinking σ‑algebras. The theorem fails for the Riemann integral because Riemann integrable functions need not be defined on a set of full measure; the Lebesgue framework is essential.

Common Traps

  • Confusing completeness of a metric space with closedness of a subset; a closed subset of a complete space is complete, but a complete subset need not be closed in a larger incomplete space.
  • Assuming that closed and bounded implies compact in every metric space; this holds only in ℝⁿ (Heine‑Borel) and fails in infinite‑dimensional normed spaces.
  • Thinking pointwise convergence of continuous functions preserves continuity; uniform convergence is required, and the limit of a pointwise convergent sequence of continuous functions can be highly discontinuous.
  • Misapplying the Baire category theorem to non‑complete spaces; the conclusion can fail dramatically, e.g., ℚ is a countable union of singletons, each nowhere dense in ℚ.
  • Believing the Lebesgue differentiation theorem holds for every point; it holds only almost everywhere, and counterexamples exist on sets of measure zero.
  • Using the Riemann integral to differentiate under the integral sign without verifying uniform convergence of the difference quotients; the Lebesgue dominated convergence theorem provides the correct hypothesis.
  • Overlooking that the Hardy‑Littlewood maximal function is not integrable even when f ∈ L¹, which is why the weak‑type estimate is the sharp available bound.