Analysis I — Chapter-by-Chapter Outline

Author: Terence Tao First published: 2006 (Hindustan Book Agency) Edition covered: Fourth Edition (2022), Hindustan Book Agency / Springer (ISBN 978-981-19-7261-4). The fourth edition incorporates corrections reported since the third edition (2016) plus additional exercises. The chapter and section structure is identical to the third edition. Two appendices — on mathematical logic and the decimal system — appear at the end of the volume and are not numbered as main chapters.

---

Central thesis

Analysis I argues that calculus as typically taught — a body of powerful rules applied without proof — rests on foundations that are shakier than they appear, and that those foundations can and must be laid from scratch. Tao's organizing claim is that every object manipulated in first-year calculus (natural numbers, integers, rationals, reals, sequences, series, continuous functions, derivatives, integrals) can be constructed from more primitive material, and that understanding those constructions is what transforms mechanical facility with formulas into genuine mathematical competence.

The book begins with nothing beyond the bare assertion that 0 is a natural number and that every natural number has a successor, and builds upward through integers, rationals, reals, and the whole apparatus of single-variable analysis. This bottom-up architecture serves a precise pedagogical purpose: each layer's construction shows exactly which properties come from where, so students learn not just that a theorem holds but why it must hold given what has been established so far. The payoff is twofold — students gain the ability to write complete, gap-free proofs, and they gain the conceptual security to know when standard rules apply and when they break down.

Why do familiar-looking manipulations of limits, sums, and integrals sometimes yield nonsensical results — and what would it take to be truly certain they never do?

---

Chapter 1 — Introduction

Central question

What is mathematical analysis, and why should a student who already knows calculus bother studying it rigorously?

Main argument

What analysis is

Section 1.1 answers the question directly: analysis is the branch of mathematics that rigorously justifies the operations of calculus. It is not a collection of new computational techniques but a re-examination of old ones — a demand to prove, rather than merely accept, that limits, derivatives, and integrals behave as advertised.

Why analysis matters — the paradox strategy

Section 1.2 motivates the entire book through a string of instructive failures. Tao presents informal paradoxes — calculations that look valid step by step but produce absurd results like 1 = 0. A typical example: if one naively swaps the order of summation in a doubly infinite series without verifying absolute convergence, one can derive contradictory values for the same sum. Another involves applying L'Hôpital's rule to a limit that does not satisfy the rule's hypotheses, obtaining a wrong answer. These are not mere curiosities; they are exactly the errors that arise when the machinery of calculus is used without understanding the conditions under which it is valid.

The chapter makes an explicit case for going all the way back to the natural numbers rather than starting with, say, the axioms of an ordered field. The constructions in early chapters may feel unnecessary — surely we all know what 0, 1, 2, … are — but Tao argues that the logical gaps in our naive understanding compound as the subject develops, and patching them later is harder than building them right from the start.

The scope of the two-volume series

The introduction sketches the program: Analysis I covers the number systems through the Riemann integral; Analysis II continues with metric spaces, sequences of functions, power series, several-variable calculus, Fourier analysis, and the Lebesgue integral. Together they constitute a rigorous undergraduate course in real analysis.

Key ideas

• Calculus rules are theorems, not definitions — they require proof, and each proof rests on hypotheses that can fail.
• Swapping limits, sums, and integrals is the most common source of undetected error in informal analysis.
• Foundational rigor is not pedantry; it is the only reliable defense against category errors and hidden assumptions.
• The book's strategy is genetic: construct each number system from the one below it, so every property can be traced to its origin.

Key takeaway

The chapter establishes that "doing analysis" means accepting nothing without proof and building every concept from explicitly stated axioms — a discipline that pays off whenever standard calculus intuitions fail.

---

Chapter 2 — Starting at the beginning: the natural numbers

Central question

What is a natural number, and how can addition and multiplication be rigorously defined on a system whose only primitive is the successor operation?

Main argument

The Peano axioms

Section 2.1 introduces the five Peano axioms: (i) 0 is a natural number; (ii) the successor n++ of every natural number n is a natural number; (iii) 0 is not the successor of any natural number; (iv) the successor function is injective (n++ = m++ implies n = m); (v) the principle of mathematical induction — if a property P holds of 0 and, whenever it holds of n it also holds of n++, then it holds of every natural number. These five axioms completely characterize the natural numbers up to isomorphism.

Tao is careful to note what the axioms exclude: without axiom (iii) one could have a "wrap-around" system where the successors cycle back to 0; without axiom (iv) one could have multiple natural numbers sharing a successor; without the induction axiom, an "illegal" element outside the intended sequence could satisfy all other axioms.

Defining addition recursively

Section 2.2 defines addition by recursion on the second argument: 0 + m := m, and (n++) + m := (n + m)++. Every subsequent property of addition — commutativity, associativity, the cancellation law — is then proved by induction from this definition alone. There is no appeal to geometric or empirical intuitions. The proofs model exactly the discipline students are expected to internalize: state the claim, induct, use only previously established lemmas.

Defining multiplication and ordering

Section 2.3 parallels section 2.2: n × 0 := 0, and n × (m++) := n × m + n. Distributivity, commutativity, and the lack of zero divisors all follow by induction. The ordering n ≤ m is defined as: there exists a natural number k such that m = n + k. The chapter proves that ≤ is a linear order compatible with addition and multiplication.

Key ideas

• The Peano axioms are parsimonious: five statements suffice to pin down the entire structure of the natural numbers.
• Mathematical induction is not a technique bolted on from outside — it is one of the defining axioms.
• Addition and multiplication are not assumed; they are defined by recursion and proved to satisfy familiar laws.
• Every proof by induction in later chapters is a direct application of Axiom 2.5 in its general form.
• The order relation on ℕ is definable purely in terms of addition, requiring no new primitive.

Key takeaway

The natural numbers emerge entirely from five axioms and recursive definitions — no intuitive or geometric input is needed — establishing the template of axiomatic construction that the entire book follows.

---

Chapter 3 — Set theory

Central question

What is a set, and which axioms about sets are consistent, sufficient for mathematics, and free from paradox?

Main argument

The naive approach and its failure

Section 3.1 introduces sets informally: a set is a collection of objects, an element either belongs to it or does not, and two sets are equal if they have the same elements. The section introduces the key set-operations — union, intersection, set-difference, and the power set — and the subset relation, all governed by explicit axioms patterned on Zermelo-Fraenkel set theory.

Russell's paradox

Section 3.2 (marked optional) demonstrates that naive set theory — the theory obtained by allowing any predicate φ(x) to define a set {x : φ(x)} — is inconsistent. Russell's construction is the set R = {x : x ∉ x}. If R ∈ R then by definition R ∉ R; if R ∉ R then by definition R ∈ R. The resolution adopted by modern mathematics is to replace unrestricted comprehension with restricted comprehension: given an already existing set A and a predicate φ, one may form {x ∈ A : φ(x)}, but one may not form sets from scratch.

Functions

Section 3.3 defines a function f : X → Y as an assignment of exactly one element f(x) ∈ Y to each x ∈ X. Crucially, functions are defined by their input-output behavior, not by a formula; two functions are equal if and only if they agree on every input. The section distinguishes injections (one-to-one), surjections (onto), and bijections, and introduces function composition with its associativity and the invertibility criterion.

Images and inverse images

Section 3.4 defines the direct image f(S) = {f(x) : x ∈ S} and the inverse image f⁻¹(U) = {x ∈ X : f(x) ∈ U} for subsets S ⊆ X and U ⊆ Y. These operations behave differently: f⁻¹ preserves all Boolean operations (unions, intersections, complements), whereas f preserves unions and maps intersections only to subsets of the intersection of images.

Cartesian products and cardinality

Section 3.5 constructs the Cartesian product X × Y as the set of ordered pairs, and section 3.6 introduces finite cardinality: a set has cardinality n if it admits a bijection with {0, 1, …, n−1}. The section proves that cardinality is well-defined (the cardinality of a finite set is unique) and establishes basic cardinality arithmetic.

Key ideas

• Unrestricted comprehension leads to contradiction; all set-formation must be grounded in previously existing sets.
• Functions in rigorous mathematics are defined extensionally — by their graph — not by a formula.
• Inverse images are better-behaved than direct images because they commute with all set operations.
• Cardinality of finite sets is a theorem, not a definition, once bijections are the standard.

Key takeaway

A consistent foundation for mathematics requires that sets be built from other sets by controlled operations rather than summoned from arbitrary predicates — the resolution of Russell's paradox that makes the rest of the book's constructions safe.

---

Chapter 4 — Integers and rationals

Central question

How can the integers and the rational numbers be constructed rigorously from the natural numbers, and what new algebraic structure does each extension introduce?

Main argument

The integers as formal differences

Section 4.1 defines the integers as equivalence classes of pairs of natural numbers, where (a, b) represents the "formal difference" a − b. Two pairs (a, b) and (c, d) are equivalent when a + d = b + c. Addition, multiplication, and negation on integers are defined at the level of representatives and proved to be well-defined (independent of the choice of representative). The integers ℤ form a commutative ring with no zero divisors.

The rationals as formal quotients

Section 4.2 builds the rationals from the integers by the same strategy: a rational number is an equivalence class of pairs (a, b) with b ≠ 0, where (a, b) ~ (c, d) iff ad = bc. The resulting structure ℚ is a field — every nonzero element has a multiplicative inverse. The natural numbers embed into the integers, and the integers embed into the rationals, in a way that preserves all arithmetic operations.

Absolute value and exponentiation

Section 4.3 defines |x| for rational x as max(x, −x) and establishes the triangle inequality |x + y| ≤ |x| + |y|. Integer exponentiation xⁿ for rational x and integer n ≥ 0 is defined recursively; the familiar exponent laws follow by induction.

Gaps in the rational numbers

Section 4.4 proves that the rational numbers are incomplete: there is no rational number whose square is 2. The standard proof by contradiction (if p/q in lowest terms satisfies (p/q)² = 2, then p and q are both even, contradicting lowest terms) is the canonical demonstration that ℚ fails to capture all lengths. This section motivates the entire project of constructing ℝ in Chapter 5.

Key ideas

• Both ℤ and ℚ are built by the formal-object method: introduce equivalence classes of pairs and prove operations are well-defined on equivalence classes.
• Each extension preserves the arithmetic of the previous system and adds a new inverse operation: subtraction for ℤ, division for ℚ.
• The triangle inequality is not just a computational tool; it is the geometric soul of absolute value and the prototype for all metric-space inequalities in later chapters.
• The irrationality of √2 is the first concrete evidence that ℚ is insufficient and the construction in Chapter 5 is necessary.

Key takeaway

The integers and rationals are not given objects but constructed ones — the formal-difference and formal-quotient constructions make rigorous what it means to subtract and divide, while also revealing the fundamental gap that the rationals leave open.

---

Chapter 5 — The real numbers

Central question

How can the gap exposed in Chapter 4 — the irrationality of √2 and the absence of limits for many Cauchy sequences — be filled, and what is the minimal construction that yields a complete ordered field?

Main argument

Cauchy sequences of rationals

Section 5.1 defines a sequence (aₙ) of rationals to be a Cauchy sequence if for every rational ε > 0 there exists N such that |aₙ − aₘ| ≤ ε for all n, m ≥ N. The definition captures sequences "trying to converge" even when no rational limit exists. The section establishes that Cauchy sequences are bounded.

Equivalence of Cauchy sequences

Section 5.2 defines two Cauchy sequences (aₙ) and (bₙ) to be equivalent if (aₙ − bₙ) → 0, and proves this is an equivalence relation. The idea is that two sequences represent the same "potential limit" if they are eventually indistinguishable at any prescribed tolerance.

Constructing ℝ

Section 5.3 defines a real number as an equivalence class of Cauchy sequences of rationals. Addition and multiplication of real numbers are defined by operating termwise on representatives, and the section proves independence of representative. The real numbers form a field extending ℚ.

Ordering the reals

Section 5.4 defines x > 0 for a real number x by: x has a representative Cauchy sequence that is eventually bounded away from 0 in the positive direction. The ordering x < y iff y − x > 0 is shown to be a linear order compatible with field operations — making ℝ an ordered field.

The least upper bound property

Section 5.5 proves the central theorem: every nonempty set of real numbers that has an upper bound has a least upper bound (supremum). This is the completeness property that distinguishes ℝ from ℚ. The proof uses the Cauchy-sequence construction: one exhibits the supremum as the limit of an explicit sequence of upper bounds. Completeness is what makes it possible to define limits and integrals unambiguously.

Real exponentiation, part I

Section 5.6 defines xⁿ for real x and natural n, then extends to rational exponents by taking nth roots whose existence is guaranteed carefully via the properties of the ordering just established.

Key ideas

• The Cauchy-sequence construction is a quotient construction: ℝ = (Cauchy sequences of rationals) / (equivalence of sequences with difference → 0).
• Completeness — the least upper bound property — is the property that ℚ lacks and ℝ possesses by construction.
• Every Cauchy sequence of reals converges in ℝ; this is equivalent to completeness and is the version most used in later chapters.
• The ordered field axioms plus completeness uniquely characterize ℝ up to isomorphism — any complete ordered field is isomorphic to the one constructed here.

Key takeaway

The real numbers are not just "the rationals plus √2 and π": they are the unique complete ordered field, constructed as equivalence classes of rational Cauchy sequences, with completeness as the defining structural property.

---

Chapter 6 — Limits of sequences

Central question

Once the real numbers exist as a complete ordered field, how does the formal ε-N definition of convergence lead to a usable calculus of limits?

Main argument

Convergence and limit laws

Section 6.1 defines a sequence (aₙ) of real numbers to converge to a limit L ∈ ℝ if for every ε > 0 there exists N such that |aₙ − L| < ε for all n ≥ N. The section immediately establishes uniqueness of limits (a sequence has at most one limit) and the limit laws: if aₙ → x and bₙ → y then aₙ + bₙ → x + y, aₙbₙ → xy, and (if y ≠ 0 and all bₙ ≠ 0) aₙ/bₙ → x/y. Each law requires a careful ε/2 or ε/3 splitting argument. The squeeze theorem follows: if aₙ ≤ bₙ ≤ cₙ for all n and aₙ, cₙ → L, then bₙ → L.

The extended real number system

Section 6.2 adjoins +∞ and −∞ to ℝ to form the extended reals ℝ* = ℝ ∪ {+∞, −∞}, with arithmetic conventions (e.g., x + ∞ = ∞ for all real x, 0 × ∞ is left undefined). This system simplifies statements of theorems about suprema and infima of unbounded sets.

Suprema and infima of sequences

Section 6.3 defines the supremum and infimum of a sequence (aₙ) as the supremum and infimum of its range as a set.

Limsup, liminf, and limit points

Section 6.4 defines the limit superior lim sup aₙ = inf{N} sup{n≥N} aₙ and the limit inferior lim inf aₙ analogously. A real number L is a limit point of (aₙ) if it is a subsequential limit. The central result is: (aₙ) converges iff lim inf aₙ = lim sup aₙ, and their common value is the limit.

Standard limits

Section 6.5 derives the standard limits of analysis: xⁿ → 0 for |x| < 1, x^(1/n) → 1 for x > 0, n^(1/n) → 1. Each is computed from the definitions rather than assumed.

Subsequences and the Bolzano–Weierstrass theorem

Section 6.6 proves the Bolzano–Weierstrass theorem: every bounded sequence of real numbers has a convergent subsequence. The proof constructs the convergent subsequence by iteratively bisecting intervals. This theorem is the key compactness result used later to prove that continuous functions on closed bounded intervals attain their maxima.

Real exponentiation, part II

Section 6.7 uses limits to complete the definition of real exponentiation x^α for irrational α: define x^α = lim_{q→α, q rational} x^q, and prove convergence and the expected exponent laws.

Key ideas

• The ε-N definition is the formal substitute for the informal "aₙ gets arbitrarily close to L."
• Limit laws reduce all convergence proofs for algebraic combinations to finitely many ε/k arguments.
• Limsup and liminf give a convergence theory that does not require a candidate limit to be known in advance.
• Bolzano–Weierstrass says that boundedness alone forces partial convergence — a first instance of the compactness principle.
• Every Cauchy sequence of reals is convergent (by completeness), giving a criterion for convergence without knowing the limit.

Key takeaway

The ε-N definition of limits, together with the completeness of ℝ, yields a complete toolkit for deciding convergence, computing limits, and guaranteeing that bounded sequences have cluster points.

---

Chapter 7 — Series

Central question

When can an infinite sum Σaₙ be assigned a finite value, and what tests allow this question to be answered without computing the sum explicitly?

Main argument

Finite series

Section 7.1 lays the algebraic groundwork for finite sums, establishing summation notation, linearity, telescoping sums, and rearrangement identities. These results for finite sums are the prototypes for the more delicate infinite-sum results that follow.

Infinite series and partial sums

Section 7.2 defines the infinite series Σₙ₌₁^∞ aₙ as the limit (if it exists) of the partial sums Sₙ = a₁ + a₂ + … + aₙ. A series converges if the partial sum sequence converges; otherwise it diverges. The zero-limit test (if Σaₙ converges then aₙ → 0, though the converse fails — the harmonic series 1 + 1/2 + 1/3 + … diverges despite terms going to 0) is established here. Absolute convergence (Σ|aₙ| < ∞) is defined and shown to imply convergence; the converse fails (the alternating harmonic series converges conditionally).

Non-negative series and comparison

Section 7.3 develops tools for series with aₙ ≥ 0, exploiting monotonicity of partial sums. The comparison test (if 0 ≤ aₙ ≤ bₙ and Σbₙ converges, then Σaₙ converges) follows immediately. The p-series Σ1/nᵖ converges iff p > 1. The geometric series Σxⁿ = 1/(1−x) for |x| < 1 is proved rigorously.

Rearrangement

Section 7.4 addresses a subtle issue: if a series converges only conditionally (not absolutely), its sum can change under rearrangement of terms (Riemann's rearrangement theorem). Absolutely convergent series are stable under rearrangement — any reordering gives the same sum. Tao also proves that absolutely convergent double series can have their order of summation interchanged, directly resolving the limit-swap paradoxes of Chapter 1.

Root and ratio tests

Section 7.5 introduces the root test: if lim sup |aₙ|^(1/n) = L < 1 then Σaₙ converges absolutely; if L > 1 it diverges; if L = 1 the test is inconclusive. The ratio test: if lim |aₙ₊₁/aₙ| = L with L < 1 then Σaₙ converges absolutely. Both tests follow from comparison with a geometric series. The root test is strictly more powerful than the ratio test.

Key ideas

• An infinite sum is a limit of finite sums, not a separate primitive operation.
• Absolute convergence and conditional convergence are genuinely distinct: conditionally convergent series lack stability.
• The harmonic series is the canonical example that the vanishing of terms is necessary but not sufficient for convergence.
• Rearrangement of absolutely convergent series is safe; rearrangement of conditionally convergent series is dangerous.
• Root and ratio tests reduce convergence to comparison with a geometric series.

Key takeaway

An infinite series converges if and only if its partial sums converge, and the theory distinguishes the stable case (absolute convergence) from the fragile case (conditional convergence), with the rearrangement theorem marking the sharp boundary between them.

---

Chapter 8 — Infinite sets

Central question

Are all infinite sets the same "size," and if not, how is cardinality of infinite sets defined and compared?

Main argument

Countability

Section 8.1 defines a set X to be countably infinite if there is a bijection between X and ℕ. A set is at most countable if it is finite or countably infinite. The section proves: ℤ is countably infinite; ℚ is countably infinite (via the Cantor dovetail enumeration of pairs); finite unions and Cartesian products of countable sets are countable.

Summation on infinite sets

Section 8.2 extends the notion of series to sums over countably infinite index sets. For f : X → ℝ with f ≥ 0, the sum Σ_{x ∈ X} f(x) is defined as the supremum of all finite subsums. The key theorem is that for non-negative functions, the sum over a countably infinite set equals the sum over any enumeration — order is irrelevant. For absolutely summable families the same holds without the non-negativity assumption.

Uncountable sets

Section 8.3 introduces uncountability. Cantor's diagonal argument proves that the set ℝ (equivalently, the power set 2^ℕ) is uncountable: no enumeration of all reals in [0,1] can be complete, because one can always construct a real differing from the nth listed real in its nth decimal digit. A corollary: for any set X, the power set 2^X has strictly greater cardinality than X itself.

The axiom of choice

Section 8.4 states the axiom of choice (AC): for any collection of nonempty sets, there exists a function that selects one element from each. Tao explains that AC is independent of the other axioms of ZF and that it is used in Analysis I in specific places, which the text marks explicitly.

Ordered sets

Section 8.5 defines partial and linear orders, chains, minimal elements, and the well-ordering principle (every nonempty subset of ℕ has a least element). The section introduces well-ordered sets and their role in transfinite arguments.

Key ideas

• Two sets have the same cardinality iff they admit a bijection — this definition works for infinite sets without invoking a "number of elements."
• Countable sets are the smallest infinite sets; uncountable sets (like ℝ) are strictly larger.
• Cantor's diagonal argument is the canonical technique for proving that no list can exhaust all elements of a set.
• The axiom of choice, while intuitively obvious, has non-trivial implications and is logically independent of ZF.
• Summation over infinite index sets requires additional care to be order-independent.

Key takeaway

Infinity is not monolithic: countable and uncountable infinities are genuinely different sizes, with Cantor's diagonal argument proving that the real numbers are uncountably infinite and thus "more numerous" than the natural numbers.

---

Chapter 9 — Continuous functions on ℝ

Central question

What does it mean for a real-valued function to be continuous, and which properties of continuous functions follow purely from the topology of ℝ?

Main argument

Subsets of the real line

Section 9.1 reviews open intervals, closed intervals, open sets (arbitrary unions of open intervals), closed sets (complements of open sets), and bounded sets. The open/closed distinction governs which theorems hold.

Algebra of functions

Section 9.2 establishes that sums, products, and quotients (where the denominator is nonzero) of functions preserve structural properties, setting up the algebraic toolkit for the subsequent theorems.

Limits of functions

Section 9.3 defines the functional limit lim{x→x₀} f(x) = L via ε-δ: for every ε > 0 there exists δ > 0 such that |x − x₀| < δ and x ≠ x₀ implies |f(x) − L| < ε. A key sequential characterization: lim{x→x₀} f(x) = L iff for every sequence (aₙ) with aₙ → x₀ and aₙ ≠ x₀, f(aₙ) → L. This bridges Chapter 6 and Chapter 9.

Continuous functions

Section 9.4 defines f to be continuous at x₀ if lim_{x→x₀} f(x) = f(x₀). Continuity is preserved under composition. Polynomials and rational functions are continuous on their domains.

Left and right limits

Section 9.5 defines one-sided limits, allowing the study of continuity at endpoints and the classification of discontinuities.

The maximum principle

Section 9.6 proves the extreme value theorem: a continuous function on a closed bounded interval [a, b] attains its maximum and minimum. The proof uses Bolzano–Weierstrass: take a sequence at which the function approaches its supremum; extract a convergent subsequence; use continuity to show the limit achieves the supremum.

The intermediate value theorem

Section 9.7 proves the intermediate value theorem (IVT): if f is continuous on [a, b] and f(a) < c < f(b), then there exists x₀ ∈ (a, b) with f(x₀) = c. The proof uses the least-upper-bound property via bisection. The IVT is the rigorous foundation for existence arguments in calculus.

Monotonic functions

Section 9.8 studies increasing and decreasing functions. Monotonic functions have one-sided limits everywhere, and their discontinuities (if any) are jump discontinuities. A monotone function on an interval can have at most countably many discontinuities.

Uniform continuity

Section 9.9 defines f to be uniformly continuous on E if for every ε > 0 there exists δ > 0 (depending only on ε, not on any specific x) such that |f(x) − f(y)| < ε whenever |x − y| < δ. The Heine–Cantor theorem is proved: any continuous function on a closed bounded interval is uniformly continuous. Uniform continuity is the precise condition needed for Riemann integrability in Chapter 11.

Limits at infinity

Section 9.10 extends limit notation to lim{x→+∞} f(x) and lim{x→−∞} f(x).

Key ideas

• The ε-δ definition of continuity is the function-level analog of the ε-N definition for sequences.
• The extreme value theorem and IVT both require the closed, bounded interval — neither holds on open intervals.
• The sequential characterization of limits is the main bridge between sequence theory (Chapter 6) and function theory (Chapter 9).
• Uniform continuity is stronger than pointwise continuity and is required to ensure Riemann integrability.
• Monotone functions are discontinuous on at most a countable set — a non-trivial cardinality result.

Key takeaway

Continuity is defined precisely in terms of limits, and the completeness of ℝ — specifically Bolzano–Weierstrass and the least upper bound property — is what gives continuous functions their strong geometric properties on closed bounded intervals.

---

Chapter 10 — Differentiation of functions

Central question

What does it mean for a function to be differentiable at a point, and which classical theorems about derivatives follow rigorously from the definition?

Main argument

Basic definitions

Section 10.1 defines the derivative of f at x₀ as f'(x₀) = lim_{x→x₀} [f(x) − f(x₀)]/(x − x₀) when this limit exists. The section establishes: differentiability implies continuity (not the converse); the derivative is linear (sum rule) and satisfies the product rule; the chain rule (f ∘ g)'(x₀) = f'(g(x₀)) · g'(x₀) is proved using the Newton approximation reformulation f(x) = f(x₀) + f'(x₀)(x − x₀) + o(x − x₀).

Local extrema and Fermat's theorem

Section 10.2 proves that if f has a local maximum or minimum at an interior point x₀ and f is differentiable there, then f'(x₀) = 0 (Fermat's theorem). This connects the algebraic notion of a vanishing derivative to the geometric notion of a turning point and is the key lemma for the mean value theorem.

Monotone functions and the mean value theorem

Section 10.3 proves Rolle's theorem (if f is continuous on [a, b], differentiable on (a, b), and f(a) = f(b), then f'(c) = 0 for some c ∈ (a, b)), and then the mean value theorem (MVT): if f is continuous on [a, b] and differentiable on (a, b), then f'(c) = [f(b) − f(a)]/(b − a) for some c ∈ (a, b). The MVT implies that a function with zero derivative is constant, that a positive derivative implies strict monotonicity, and provides linear approximation error bounds.

Inverse functions and derivatives

Section 10.4 proves the one-dimensional inverse function theorem: if f is continuous and strictly monotone on an interval, its inverse f⁻¹ is differentiable wherever f' ≠ 0, and (f⁻¹)'(y) = 1/f'(f⁻¹(y)). This is used to differentiate root functions and logarithms.

L'Hôpital's rule

Section 10.5 proves L'Hôpital's rule: if f(x₀) = g(x₀) = 0, both are differentiable near x₀, and g'(x₀) ≠ 0, then lim_{x→x₀} f(x)/g(x) = f'(x₀)/g'(x₀). More general versions (0/0 and ∞/∞ forms, limits at infinity) are also treated. The chapter explicitly identifies the conditions under which the rule applies, addressing the misapplication flagged in Chapter 1.

Key ideas

• Differentiability is strictly stronger than continuity: |x| is continuous but not differentiable at 0.
• The chain rule in its Newton-approximation form generalizes naturally to higher dimensions and is the conceptual basis for the multivariable chain rule in Analysis II.
• The mean value theorem is the primary tool for translating information about derivatives into information about functions (monotonicity, constancy, bounds).
• L'Hôpital's rule has specific prerequisites that must be verified; applying it to non-qualifying limits is one of the errors from Chapter 1.
• The inverse function theorem in dimension 1 requires only strict monotonicity and differentiability.

Key takeaway

The derivative is defined as a limit, and all classical differentiation theorems — chain rule, mean value theorem, L'Hôpital's rule — follow from that definition with no additional assumptions beyond differentiability, making precise exactly which conditions each result requires.

---

Chapter 11 — The Riemann integral

Central question

What is the right definition of the integral of a function on an interval, and which functions are integrable by this definition?

Main argument

Partitions

Section 11.1 introduces a partition of [a, b]: a finite set of points a = x₀ < x₁ < … < xₙ = b. Refinements (adding partition points) yield tighter approximations.

Piecewise constant functions and their integrals

Section 11.2 defines a piecewise constant function as one constant on each sub-interval of some partition. The integral of such a function is the finite sum of (constant value) × (sub-interval length). The section proves this is independent of which compatible partition is used, making the definition well-posed.

Upper and lower Riemann integrals

Section 11.3 defines the upper Riemann integral U(f) = inf{I(g) : g piecewise constant, g ≥ f} and the lower Riemann integral L(f) = sup{I(g) : g piecewise constant, g ≤ f}. A function is Riemann integrable on [a, b] if U(f) = L(f), and the common value is ∫ₐᵇ f.

Basic properties

Section 11.4 proves linearity: ∫(f + g) = ∫f + ∫g, ∫(cf) = c∫f; monotonicity: f ≤ g implies ∫f ≤ ∫g; and additivity over sub-intervals: ∫ₐᵇ = ∫ₐᶜ + ∫ᶜᵇ for any c ∈ [a, b].

Riemann integrability of continuous functions

Section 11.5 proves that every continuous function on [a, b] is Riemann integrable. The key ingredient is uniform continuity: given ε, choose δ from uniform continuity; then any partition with mesh < δ yields upper and lower sums within ε(b − a) of each other, so U(f) − L(f) < ε.

Riemann integrability of monotone functions

Section 11.6 proves that every monotone function on [a, b] is Riemann integrable, even if it has a countably infinite set of jump discontinuities.

A non-Riemann integrable function

Section 11.7 constructs the Dirichlet function — the indicator 1_ℚ of the rationals in [0,1], equal to 1 on rationals and 0 on irrationals. For any partition, every upper approximation sums to 1 and every lower approximation sums to 0, so U(f) = 1 ≠ 0 = L(f). This example shows the Riemann integral cannot handle all bounded functions and motivates the Lebesgue integral of Analysis II.

The Riemann–Stieltjes integral

Section 11.8 generalizes the Riemann integral by replacing length dx with dα for a monotone function α. When α is a step function the Stieltjes integral reduces to a discrete sum; when α(x) = x it reduces to the ordinary Riemann integral. This unification foreshadows measure-theoretic integration.

The two fundamental theorems of calculus

Section 11.9 proves both fundamental theorems. FTC1: if f is Riemann integrable on [a, b] and F(x) = ∫ₐˣ f, then F is continuous; if f is also continuous at x₀, then F'(x₀) = f(x₀). FTC2: if f is differentiable on [a, b] and f' is Riemann integrable, then ∫ₐᵇ f'(x) dx = f(b) − f(a). Together these theorems establish that differentiation and integration are mutually inverse operations under appropriate hypotheses.

Consequences of the fundamental theorems

Section 11.10 derives integration by parts, the change of variables formula (substitution rule), and integral estimates. These are the computational rules of calculus, now resting on a rigorous foundation.

Key ideas

• The Riemann integral is defined via upper and lower approximations by piecewise constant functions, making the definition completely explicit.
• Integrability requires upper and lower integrals to agree — this is precisely what fails for the Dirichlet function.
• Uniform continuity (from Chapter 9) is the key ingredient that makes continuous functions integrable.
• The FTC closes the logical circle: differentiation (Chapter 10) and integration are proved inverse rather than assumed so.
• The Riemann–Stieltjes integral and the Dirichlet function both point toward the need for a more general theory.

Key takeaway

The Riemann integral is constructed from scratch via upper and lower sums, the fundamental theorem of calculus is proved rather than assumed, and a concrete non-integrable function shows exactly where the Riemann theory ends and a more powerful theory begins.

---

The book's overall argument

1. Chapter 1 (Introduction) — establishes the need for rigorous foundations by demonstrating that informal applications of limit operations can lead to contradictions; sets the agenda for the entire book.
2. Chapter 2 (Starting at the beginning: the natural numbers) — constructs ℕ from five Peano axioms, defines addition and multiplication by recursion, and introduces mathematical induction as the primary proof technique.
3. Chapter 3 (Set theory) — provides the logical language (sets, functions, equivalence classes) needed to make the subsequent number-system constructions precise, and motivates restricted comprehension via Russell's paradox.
4. Chapter 4 (Integers and rationals) — extends ℕ to ℤ via formal differences and to ℚ via formal quotients, establishing a field; proves the irrationality of √2 to show ℚ is insufficient.
5. Chapter 5 (The real numbers) — constructs ℝ as equivalence classes of rational Cauchy sequences, proves the least upper bound property (completeness), and establishes ℝ as the unique complete ordered field.
6. Chapter 6 (Limits of sequences) — develops the ε-N theory of convergence, limsup and liminf, and the Bolzano–Weierstrass theorem on top of the completed ℝ.
7. Chapter 7 (Series) — translates convergence theory into a theory of infinite sums, distinguishes absolute from conditional convergence, and proves the root and ratio tests.
8. Chapter 8 (Infinite sets) — extends cardinality from finite to infinite, proves ℝ is uncountable via Cantor's diagonal argument, and introduces the axiom of choice.
9. Chapter 9 (Continuous functions on ℝ) — gives the ε-δ definition of continuity and proves the extreme value theorem and intermediate value theorem, both requiring completeness.
10. Chapter 10 (Differentiation of functions) — defines the derivative as a limit and proves the chain rule, mean value theorem, and L'Hôpital's rule from first principles.
11. Chapter 11 (The Riemann integral) — constructs the integral from piecewise constant approximations, proves both fundamental theorems of calculus, and exhibits a non-integrable function to delimit the theory's scope.

---

Common misunderstandings

Misunderstanding: The book re-derives things "everyone already knows," so it is just slower calculus.

The book's purpose is not to rederive computational formulas but to establish why those formulas hold and under which conditions they hold. The construction of ℝ, for example, is not an alternative algorithm for arithmetic — it is a proof that a complete ordered field exists and is unique. Without this, statements like "every Cauchy sequence of reals converges" are not theorems but hopes.

Misunderstanding: Proofs by induction are only needed for the chapters on natural numbers.

Mathematical induction appears throughout the book — in the construction of ℤ and ℚ (where well-ordering of ℕ is used), in the proof of the binomial theorem, in the study of finite series, and in the construction of convergent subsequences. The induction principle introduced in Chapter 2 is the backbone of the entire text.

Misunderstanding: The Cauchy sequence construction of ℝ is one of several equally good approaches.

There are two standard constructions (Cauchy sequences and Dedekind cuts), but both produce the same object: a complete ordered field. Once you have proved the construction satisfies the complete ordered field axioms, you may use those axioms freely without ever referring back to the construction. The choice of construction is a matter of exposition, not of mathematical content.

Misunderstanding: Absolute convergence and convergence are the same thing for "nice" series.

They are not. The alternating harmonic series Σ(−1)ⁿ/n converges conditionally but not absolutely, because Σ1/n diverges. This distinction matters: conditionally convergent series can be rearranged to converge to any real number (Riemann's rearrangement theorem), a phenomenon with no analog for absolutely convergent series.

Misunderstanding: L'Hôpital's rule always applies to 0/0 indeterminate forms.

L'Hôpital's rule requires both f and g to be differentiable near the point, g' ≠ 0 near (but perhaps not at) the point, and in the most general form the limit of f'/g' must exist. The rule is not applicable when f'/g' itself is undefined or oscillates without a limit. Chapter 1 flags this as one of the standard errors; Chapter 10 provides the precise conditions.

Misunderstanding: The Riemann integral handles all bounded functions.

It does not. The indicator of ℚ ∩ [0,1] is bounded but not Riemann integrable. The correct statement is that continuous (or monotone) functions on closed bounded intervals are Riemann integrable. Handling a larger class of functions requires the Lebesgue integral, introduced in Analysis II.

---

Central paradox / key insight

The central paradox of Analysis I is that the real numbers — the medium in which all of calculus takes place — are not given but constructed, and the construction requires an infinite process (equivalence classes of infinite sequences) to produce a single number. Every real number is, at bottom, an equivalence class of infinite approximations.

This might appear circular: to define √2 as the limit of a Cauchy sequence, one needs the notion of a limit, but limits are defined using real numbers. Tao resolves the circularity by building the limit concept twice: first as a notion for rational Cauchy sequences (Chapter 5, to construct ℝ), and then as a full ε-N limit for real sequences (Chapter 6, using the already-constructed ℝ). The two uses of "limit" are logically distinct.

The deeper insight the book delivers is a completeness–compactness duality: ℝ is complete (Cauchy sequences converge), which is equivalent to the least upper bound property, which powers Bolzano–Weierstrass (bounded sequences have convergent subsequences), which powers the extreme value theorem (continuous functions on closed bounded intervals achieve their extrema). Every theorem in Chapters 9–11 that says "something nice happens" ultimately traces back to the completeness built into ℝ in Chapter 5.

A complete ordered field is not a description of something we observe; it is a construction we make — and every theorem in analysis is a consequence of that construction's exactness.

---

Important concepts

Peano axioms

The five axioms that characterize the natural numbers: 0 exists; every natural number has a successor; 0 is not a successor; the successor function is injective; the induction principle. Together they pin down ℕ up to isomorphism.

Mathematical induction

The proof technique flowing from Peano Axiom 5: to prove P(n) for all n ∈ ℕ, prove P(0) and prove P(n) ⟹ P(n+1). Strong induction (P(0), …, P(n) all imply P(n+1)) is equivalent and often more convenient.

Formal difference / formal quotient

The equivalence-class construction used to extend number systems: integers are equivalence classes of pairs (a, b) of naturals under (a, b) ~ (c, d) iff a + d = b + c; rationals are equivalence classes of pairs (p, q) of integers with q ≠ 0 under (p, q) ~ (r, s) iff ps = qr.

Cauchy sequence

A sequence (aₙ) in ℚ (or ℝ) such that for every ε > 0 there exists N with |aₙ − aₘ| < ε for all n, m ≥ N. In ℝ, Cauchy sequences are exactly the convergent sequences (by completeness).

Completeness (least upper bound property)

Every nonempty subset of ℝ that is bounded above has a supremum in ℝ. Equivalent to: every Cauchy sequence of reals converges in ℝ. This is the property that distinguishes ℝ from ℚ.

Supremum / infimum

The supremum (least upper bound) of a set S ⊆ ℝ is the smallest number ≥ every element of S. The infimum (greatest lower bound) is the largest number ≤ every element. Denoted sup S and inf S; they exist in ℝ (but not necessarily in ℚ) by completeness.

ε-N definition of convergence

A sequence (aₙ) converges to L if for every ε > 0 there exists N ∈ ℕ such that for all n ≥ N, |aₙ − L| < ε. The quantifier order (∀ε ∃N ∀n ≥ N) is crucial: ε comes first, N depends on ε.

Limsup and liminf

lim sup aₙ = lim{N→∞} sup{n≥N} aₙ (the eventual supremum), lim inf aₙ = lim{N→∞} inf{n≥N} aₙ (the eventual infimum). A sequence converges iff lim inf = lim sup.

Absolute convergence

A series Σaₙ is absolutely convergent if Σ|aₙ| < ∞. Absolute convergence implies convergence, and absolutely convergent series are stable under arbitrary rearrangement.

Countable / uncountable

A set is countably infinite if it admits a bijection with ℕ; it is uncountable if it is infinite but not countably infinite. ℕ, ℤ, ℚ are countably infinite; ℝ is uncountable (Cantor's diagonal argument).

ε-δ definition of continuity

A function f : ℝ → ℝ is continuous at x₀ if for every ε > 0 there exists δ > 0 such that |x − x₀| < δ implies |f(x) − f(x₀)| < ε.

Uniform continuity

f is uniformly continuous on E if for every ε > 0 there exists δ > 0 such that for all x, y ∈ E with |x − y| < δ, |f(x) − f(y)| < ε. The Heine–Cantor theorem: every continuous function on a closed bounded interval is uniformly continuous.

Riemann integrable

A bounded function f on [a, b] is Riemann integrable if its upper and lower Riemann integrals agree: U(f) = L(f). The common value is ∫ₐᵇ f.

Fundamental theorem of calculus

Two related theorems: (FTC1) the antiderivative of a continuous Riemann integrable function is its integral as a function of the upper limit; (FTC2) if f is differentiable and f' is integrable, then ∫ₐᵇ f'(x) dx = f(b) − f(a).

Axiom of choice

For any family of nonempty sets, there exists a function selecting one element from each. Independent of ZF; used in Analysis I in specific places marked by the text.

---

References and Web Links

Primary book and edition information

• Tao, Terence. Analysis I, Fourth Edition. Texts and Readings in Mathematics 37. Hindustan Book Agency / Springer, 2022.

  • Fourth Edition on Springer
  • Fourth Edition at AMS Bookstore
  • Amazon listing (Hindustan Book Agency, 4th ed.)

• Tao, Terence. Analysis I, Third Edition. Texts and Readings in Mathematics 37. Hindustan Book Agency / Springer, 2016.

  • Third Edition on Springer
  • Third Edition on Internet Archive (open access)
  • Amazon listing (3rd ed.)

Author's official page and errata

• Tao's official book page for Analysis I — errata for all editions, sample chapters, and links to the companion Analysis II.
• Terence Tao's homepage at UCLA

Background and overview

• Terence Tao — Wikipedia
• Peano axioms — Wikipedia
• Russell's paradox — Wikipedia

Lean formalization of Analysis I

• A Lean companion to Analysis I (Tao's blog post, 2025)
• GitHub: teorth/analysis — Lean companion repository

Additional chapter summaries and study resources

These are secondary summaries and should be used alongside, rather than instead of, the original book.

• Tao Analysis Solutions blog — worked solutions to exercises from all chapters.
• Analysis (Tao, 3rd ed.) — section-by-section outline — used to cross-check this document's structure.
• solverer.com — solution manual index for Analysis I