Analysis

The Core Insight

Every concept in analysis — continuity, the derivative, the integral, the sum of a series — is secretly a statement about limits. To make any of them rigorous, you must first pin down what "$x_n\to L$" means with no appeal to motion or infinitesimals: for every tolerance $\epsilon >0$, eventually all terms lie within $\epsilon$ of $L$. The completeness axiom then guarantees that limits which "ought" to exist actually do. Master the quantifier dance $\forall \epsilon \exists N$ and you hold the master key to the entire subject.

Formal Statement

Limit of a Sequence (ε–N)

A sequence $(x_n{)}_{n\in \mathbb{N}}$ of real numbers converges to $L\in \mathbb{R}$ if:

Here $\epsilon$ is the error tolerance (any positive real), $N$ is the threshold index (may depend on $\epsilon$), $|x_n-L|$ is the distance from the $n$-th term to the candidate limit $L$. We write ${\lim }_{n\to \infty }{x}_n=L$ or $x_n\to L$.

Limit of a Function (ε–δ)

Let $f$ be defined on a punctured neighbourhood of $a$. Then ${\lim }_{x\to a}f(x)=L$ means:

Symbols: $\delta$ is the input radius (depends on $\epsilon$ and $a$), $a$ the limit point, $L$ the target value. Continuity at $a$ is the special case $L=f(a)$, which additionally requires $f$ to be defined at $a$.

Completeness — Least Upper Bound Property

Every nonempty subset $S\subseteq \mathbb{R}$ that is bounded above has a least upper bound (supremum) $\sup S\in \mathbb{R}$:

This axiom is what distinguishes $\mathbb{R}$ from $\mathbb{Q}$: the set $\{q\in \mathbb{Q}:q^2<2\}$ has no rational supremum, but has supremum $\sqrt{2}$ in $\mathbb{R}$.

The Derivative as a Limit

A function $f$ is differentiable at $a$ with derivative $f^{\prime }(a)$ if the limit

exists. Differentiability at $a$ implies continuity at $a$ (but not conversely).

Riemann Integral as a Limit of Sums

For a bounded $f$ on $[a,b]$ with partition $P=\{a=x_0<\cdots <x_n=b\}$ and tags ${\xi }_i\in [x_{i-1},x_i]$, $f$ is Riemann integrable with integral $I$ if

where $\Vert P\Vert ={\max }_i(x_i-x_{i-1})$ is the mesh. Equivalently, the upper and lower Darboux sums share a common limit.

Why It Holds / Derivation

Bounded Monotone Sequences Converge (from Completeness)

1. The set $S=\{x_n:n\in \mathbb{N}\}$ is nonempty and bounded above, so by Equation [3] it has a supremum $L=\sup S$.
2. Fix $\epsilon >0$. By the second clause of [3], there exists an index $N$ with $x_N>L-\epsilon$.
3. Since the sequence is increasing, for all $n\ge N$ we have $x_n\ge x_N>L-\epsilon$.
4. Since $L$ is an upper bound, $x_n\le L<L+\epsilon$ for all $n$.
5. Combining steps 3–4: for $n\ge N$, $L-\epsilon <x_n\le L$, hence $|x_n-L|<\epsilon$.
6. This is exactly Equation [1], so $x_n\to L$. $■$

This single proof shows completeness is the engine: without it, increasing bounded sequences (like rational approximations to $\sqrt{2}$) could fail to converge inside the space.

Continuity ⟹ Boundedness on a Closed Interval (sketch)

1. Suppose not; then for each $n$ there is $x_n\in [a,b]$ with $|f(x_n)|>n$.
2. $[a,b]$ is bounded, so by Bolzano–Weierstrass $(x_n)$ has a convergent subsequence $x_{n_k}\to c$.
3. Since $[a,b]$ is closed, $c\in [a,b]$, where $f$ is defined and continuous.
4. Continuity gives $f(x_{n_k})\to f(c)$, so $(f(x_{n_k}))$ is bounded.
5. But $|f(x_{n_k})|>n_k\to \infty$ — contradiction.
6. Hence $f$ is bounded; a second supremum argument yields it attains its max and min. $■$

Karl Weierstrass | 1815–1897

The "father of modern analysis," Weierstrass spent years as a provincial schoolteacher before his rigorous lectures at Berlin transformed the field. He banished vague talk of "infinitely small quantities" by founding everything on the ε–δ formalism (refining ideas of Bolzano and Cauchy). His 1872 construction of a continuous, nowhere-differentiable function $\sum a^n\cos (b^n\pi x)$ stunned mathematicians and proved that geometric intuition is an unreliable guide. He also formalized uniform convergence, the Bolzano–Weierstrass theorem, and the Weierstrass Approximation Theorem.

Building Intuition

Think of $\epsilon$ as a target dartboard of shrinking radius around $L$. Convergence means: no matter how tiny I make the board, you can promise me that all darts from some point on land inside it. The cleverness is that $N$ is allowed to grow as $\epsilon$ shrinks — you get more "warm-up throws" for harder targets. Continuity is the same game played on inputs and outputs simultaneously: squeeze the output window to width $\epsilon$, and continuity guarantees a matching input window of width $\delta$ exists. Completeness is the silent referee ensuring the limit point is actually on the field — that the real line has no missing points where a sequence "wants to land" but cannot.

The ε–δ definition of a function limit: outputs trapped in the ε-band force inputs into the δ-band

Worked Examples

Proving a Limit by ε–N

1. Compute the distance: $\left|\frac{3n+1}{n+2}-3\right|=\left|\frac{3n+1-3n-6}{n+2}\right|=\frac{5}{n+2}$.
2. We need $\frac{5}{n+2}<\epsilon$, i.e. $n+2>\frac{5}{\epsilon }$, i.e. $n>\frac{5}{\epsilon }-2$.
3. Choose $N=\lceil \frac{5}{\epsilon }\rceil$ (any integer exceeding $\frac{5}{\epsilon }-2$ works).
4. Then $n\ge N⟹\frac{5}{n+2}<\epsilon$, verifying Equation [1].

For a numerical check, take $\epsilon =0.01$: then $N=\lceil 500\rceil =500$. At $n=500$, the term is $\frac{1501}{502}\approx 2.9900$, indeed within $0.01$ of $3$.

Derivative from First Principles

1. $f(2+h)=(2+h{)}^2+3(2+h)=4+4h+h^2+6+3h=10+7h+h^2$.
2. $f(2)=4+6=10$.
3. Difference quotient: $\frac{f(2+h)-f(2)}{h}=\frac{7h+h^2}{h}=7+h$ (for $h\ne 0$).
4. Take the limit: $\underset{h\to 0}{\lim }(7+h)=7$.

Sanity check against the rule $f^{\prime }(x)=2x+3$ at $x=2$: $2(2)+3=7$. ✓

Common Misconceptions

Continuous but Nowhere Differentiable

The Weierstrass function

Pointwise Limit Kills the Integral

Let $f_n(x)=n$ on $(0,1/n)$ and $0$ elsewhere on $[0,1]$. Then $f_n(x)\to 0$ pointwise for every fixed $x$ (eventually $x>1/n$), so ${\int }_0^1\lim f_n=0$. But ${\int }_0^1{f}_{n}dx=n\cdot \frac{1}{n}=1$ for all $n$, so $\lim \int f_n=1\ne 0$. The "spike escaping to infinity" carries area that vanishes pointwise but not in the integral.

Connections

Analysis is the trunk from which many branches grow. The ε–δ framework generalizes to abstract distances in Metric Spaces and to open-set structures in Topology, where continuity becomes "preimages of open sets are open." The Riemann integral (Equation [5]) is superseded by the more powerful Lebesgue Integration, which handles wild functions and limit-swapping gracefully. Differentiation extends to several variables in Multivariable Calculus and to the complex plane in Complex Analysis, where differentiability is astonishingly rigid. The study of infinite sums links to Power Series and Fourier Analysis, and the completeness of $\mathbb{R}$ is itself constructed in Construction of the Real Numbers via Dedekind cuts or Cauchy sequences.

Real-World Applications

1. Signal processing & audio compression — Fourier analysis (built on convergence of series) underlies MP3, JPEG, and digital filtering; convergence theorems guarantee reconstruction fidelity.
2. Numerical methods — Newton's method, quadrature, and ODE solvers (Runge–Kutta) all rest on convergence rates and error bounds proved in analysis.
3. Quantum mechanics — Hilbert spaces, $L^2$ functions, and self-adjoint operators are functional-analytic objects; the Schrödinger equation lives in this setting.
4. Probability & statistics — The Law of Large Numbers and Central Limit Theorem are convergence statements; measure theory (Lebesgue analysis) is the rigorous foundation of modern probability.
5. Financial mathematics — The Black–Scholes model uses stochastic calculus, an analysis built on Brownian motion and Itô integration.
6. Machine learning — Gradient descent convergence, neural network approximation (Universal Approximation Theorem descends from Weierstrass), and regularization all use analytic tools.
7. Engineering control systems — Stability analysis via Lyapunov functions and frequency-domain (Laplace transform) methods require limits, continuity, and analyticity.

Historical Development

• c. 1665–1684 — Newton and Leibniz independently develop calculus, using intuitive but logically shaky "infinitesimals" and "fluxions."
• 1734 — Bishop Berkeley's The Analyst attacks infinitesimals as "ghosts of departed quantities," exposing the foundational crisis.
• 1817 — Bolzano gives an early rigorous proof of the Intermediate Value Theorem and an ε-style definition of continuity, largely unnoticed.
• 1821 — Cauchy's Cours d'analyse systematizes limits, continuity, and convergence, bringing rigor toward the mainstream.
• 1854 — Riemann defines his integral precisely (Equation [5]) and studies trigonometric series, prompting questions about integrability.
• 1872 — Weierstrass publishes his continuous nowhere-differentiable function; Dedekind and Cantor independently construct $\mathbb{R}$ from $\mathbb{Q}$, securing completeness.
• 1902 — Lebesgue's thesis introduces measure and the Lebesgue integral, resolving the limit-swap defects of Riemann's theory.
• 1906–1932 — Fréchet, Hausdorff, and Banach abstract analysis into metric, topological, and normed spaces — the birth of functional analysis.

Subtleties & Limits

While the foundations of real analysis are settled, deep questions persist at its frontiers. The Riemann Hypothesis — concerning the zeros of the analytic continuation of $\zeta (s)=\sum n^{-s}$ — remains open and is the most famous problem in complex analysis. In measure theory, the existence of non-measurable sets depends on the Axiom of Choice, and the continuum hypothesis (is there a cardinality strictly between $|\mathbb{N}|$ and $|\mathbb{R}|$?) is independent of standard set theory, touching the very fabric of the real line. Nonstandard analysis (Robinson, 1960s) rehabilitated infinitesimals rigorously via hyperreal numbers, showing the "ghosts" can be exorcised in more than one way — a reminder that the ε–δ framework, while dominant, is not the only logically valid route.

See Also