Reference

Glossary

Every technical word in the course and what it means in plain language, with flashcards for practice.

Every technical term in the course, with a plain-language meaning. On every page these words are marked with a dotted underline: hover or tap one to see its meaning. The “Used in” links point to the sections where each term appears.

Flashcards — term ⇄ meaning
Ch 1 · Sets, Real Numbers and Countability
TermMeaningUsed in
algebraic numberA number that is a root of a nonzero polynomial with integer coefficients, such as $\sqrt2$ or $\sqrt[3]{5}$. The algebraic numbers form a countable set; numbers that are not algebraic are transcendental.1.1 4.4
Archimedean propertyFor every $x\gt0$ and $y$ there is $n\in\mathbb{N}$ with $nx\gt y$; equivalently $\mathbb{N}$ is unbounded and $1/n\to0$. It holds in $\mathbb{R}$ and $\mathbb{Q}$ but fails in $\mathbb{R}(t)$.1.2 2.1
bijectionA function that is both one-one (injective) and onto (surjective); it pairs the elements of two sets perfectly and has an inverse function.1.1 2.3 3.1 3.3
cardinalityThe size of a set, compared through bijections: $|A|=|B|$ iff there is a bijection $A\to B$. $|\mathbb{N}|=\aleph_0$ and $|\mathbb{R}|=\mathfrak{c}$.1.1 4.1 4.4 7.3
completeness axiomEvery nonempty set of real numbers that is bounded above has a supremum in $\mathbb{R}$. This is what distinguishes $\mathbb{R}$ from $\mathbb{Q}$.1.2 2.1 2.2 7.3
countableA set is countable if it is finite or can be listed as $a_1,a_2,a_3,\dots$ (a bijection with $\mathbb{N}$). Examples: $\mathbb{Z}$, $\mathbb{Q}$, the algebraic numbers.1.1 1.2 3.1 3.2
denseA set $D\subseteq\mathbb{R}$ is dense if every open interval contains a point of $D$ (in a metric space: every open ball meets $D$). $\mathbb{Q}$ and the irrationals are dense in $\mathbb{R}$.1.1 1.2 3.1 3.2
equivalence relationA relation that is reflexive, symmetric and transitive. Its equivalence classes split the set into disjoint pieces (a partition).1.1 4.4 7.3 7.4
infimumThe greatest lower bound $\inf S$; equal to $-\sup(-S)$. For every $\varepsilon\gt0$ some element of $S$ is below $\inf S+\varepsilon$.1.2 2.2 3.2 4.1
nested interval propertyIf $[a_1,b_1]\supseteq[a_2,b_2]\supseteq\cdots$ are closed bounded intervals, their intersection is nonempty; it is a single point when the lengths shrink to $0$.1.2 2.2
ordered fieldA field with a set of positive elements closed under $+$ and $\times$ such that each element is positive, zero or negative. $\mathbb{Q}$ and $\mathbb{R}$ are ordered fields; $\mathbb{C}$ and $\mathbb{Z}_p$ cannot be ordered.1.2 2.1
power setThe set $\mathcal{P}(A)$ of all subsets of $A$. By Cantor's theorem it is always strictly larger than $A$.1.1
preimageFor $f:X\to Y$ and $B\subseteq Y$, the set $f^{-1}(B)=\{x: f(x)\in B\}$. Defined for every function, invertible or not; it respects unions, intersections and complements.1.1 3.1 4.4 7.1
supremumThe least upper bound $\sup S$: an upper bound such that, for every $\varepsilon\gt0$, some element of $S$ exceeds $\sup S-\varepsilon$. It need not belong to $S$.1.2 2.1 2.2 3.1
uncountableNot countable: no list $a_1,a_2,\dots$ contains all its elements. Examples: $\mathbb{R}$, every interval, $\{0,1\}^{\mathbb{N}}$, $\mathcal{P}(\mathbb{N})$.1.1 1.2 3.2 4.1
upper boundA number $u$ with $s\le u$ for every element $s$ of the set. A set may have many upper bounds or none.1.1 1.2 2.1 3.1
Ch 2 · Sequences and Series of Real Numbers
TermMeaningUsed in
absolutely convergentA series $\sum a_n$ is absolutely convergent if $\sum|a_n|$ converges. Absolute convergence implies convergence, and the sum is then unchanged by any rearrangement (similarly for improper integrals with $|f|$).2.3 4.2 4.5 5.1
Bolzano–Weierstrass theoremEvery bounded sequence of real numbers (or of points of $\mathbb{R}^n$) has a convergent subsequence. It depends on the completeness of $\mathbb{R}$.2.1 2.2 3.1 3.2
Cauchy condensation testFor decreasing $a_n\ge0$: $\sum a_n$ converges iff $\sum 2^ka_{2^k}$ converges. It settles the $p$-series and $\sum\frac1{n(\log n)^p}$.2.3
Cauchy productThe series $\sum c_n$ with $c_n=\sum_{k=0}^na_kb_{n-k}$. By Mertens' theorem it converges to $AB$ if one factor converges absolutely; for two conditionally convergent factors it may diverge.2.3 5.2
Cauchy sequenceA sequence with $|a_m-a_n|\lt\varepsilon$ for all $m,n\ge N(\varepsilon)$. In $\mathbb{R}$ a sequence converges iff it is Cauchy; in $\mathbb{Q}$ or $(0,1)$ a Cauchy sequence may have no limit in the set.1.2 2.1 2.2 2.3
Cesàro meanThe average $\sigma_n=(x_1+\cdots+x_n)/n$. If $x_n\to L$ then $\sigma_n\to L$, but not conversely: $(-1)^n$ has $\sigma_n\to0$.2.2 5.1
conditionally convergentConvergent but not absolutely convergent, like $1-\frac12+\frac13-\cdots$. Its positive and negative parts both diverge, and rearrangements can change its sum.2.3 4.2 4.5
contractionA map $f$ with $|f(x)-f(y)|\le c|x-y|$ for all $x,y$, where $0\le c\lt1$. On a closed interval mapped into itself it has a unique fixed point, the limit of every iteration $x_{n+1}=f(x_n)$.2.1 3.2 6.2 7.1
contractive sequenceA sequence with $|a_{n+2}-a_{n+1}|\le c\,|a_{n+1}-a_n|$ for all $n$, for one fixed constant $0\lt c\lt1$. Contractive sequences are Cauchy, so they converge in $\mathbb{R}$.2.1
convergent sequenceA sequence $(a_n)$ with a limit $L\in\mathbb{R}$: for every $\varepsilon\gt0$ there is $N$ such that $|a_n-L|\lt\varepsilon$ for all $n\ge N$. A convergent sequence is bounded and has only one limit.2.1 2.2 2.3 5.1
Dirichlet's testIf the partial sums of $\sum a_n$ are bounded and $b_n$ is monotone with $b_n\to0$, then $\sum a_nb_n$ converges; e.g. $\sum\frac{\sin nx}n$.2.3 4.2 5.1
harmonic seriesThe series $\sum\frac1n=1+\frac12+\frac13+\cdots$. Its terms tend to $0$ but it diverges; its partial sums $H_n$ grow like $\log n+\gamma$.2.3 4.2
Leibniz testIf $a_n$ decreases to $0$, then $\sum(-1)^{n+1}a_n$ converges, and the error after $n$ terms is at most $a_{n+1}$. Monotonicity cannot be dropped.2.3 4.2 5.1
limit inferior$\liminf x_n=\sup_n\inf_{k\ge n}x_k=-\limsup(-x_n)$, the smallest subsequential limit. A sequence converges iff $\liminf=\limsup$ (finite).2.2 2.3 4.2 4.4
limit superior$\limsup x_n=\inf_n\sup_{k\ge n}x_k$, the largest subsequential limit. For every $\varepsilon\gt0$ the terms are eventually below $\limsup x_n+\varepsilon$ and infinitely often above $\limsup x_n-\varepsilon$.2.2 2.3 4.1 4.4
monotone convergence theoremTwo results share this name. For sequences (§2.1): every bounded monotone sequence of real numbers converges — an increasing one to its supremum, a decreasing one to its infimum. For integrals (§4.5): if $0\le f_1\le f_2\le\cdots$ are measurable and $f_n\to f$ pointwise, then $\int f_n\to\int f$.2.1 2.2 2.3 4.5
monotone sequenceA sequence that is increasing ($a_n\le a_{n+1}$ for all $n$) or decreasing ($a_n\ge a_{n+1}$ for all $n$). A monotone sequence converges iff it is bounded.1.2 2.1 2.2 2.3
properly divergentA sequence that diverges to $+\infty$ or to $-\infty$: for every $M$ its terms are eventually $\gt M$ (or eventually $\lt M$). $(-1)^n$ and $n(-1)^n$ diverge but are not properly divergent.2.1 2.2
ratio testIf $\limsup|a_{n+1}/a_n|\lt1$ the series converges absolutely; if $|a_{n+1}/a_n|\ge1$ eventually it diverges; a limit equal to $1$ gives no information.2.2 2.3 5.2
Riemann rearrangement theoremA conditionally convergent series can be rearranged to converge to any prescribed real number, or to diverge to $+\infty$ or $-\infty$.2.3
root testWith $\rho=\limsup|a_n|^{1/n}$: $\rho\lt1$ gives absolute convergence, $\rho\gt1$ divergence, $\rho=1$ no information. It decides whenever the ratio test does, and sometimes when it does not.2.2 2.3 5.2
squeeze theoremIf $x_n\le y_n\le z_n$ eventually and $x_n\to L$, $z_n\to L$, then $y_n\to L$.2.1 3.1
Stolz–Cesàro theoremIf $b_n$ increases strictly to $+\infty$ and $(a_{n+1}-a_n)/(b_{n+1}-b_n)\to L$, then $a_n/b_n\to L$ — a discrete L'Hôpital rule.2.2
subsequenceA sequence $(x_{n_k})$ obtained from $(x_n)$ by choosing indices $n_1\lt n_2\lt\cdots$; the order is kept and $n_k\ge k$. Every subsequence of a convergent sequence converges to the same limit.2.1 2.2 2.3 3.1
subsequential limitA limit (in $\mathbb{R}$ or $\pm\infty$) of some subsequence of $(x_n)$; for real $L$: every neighbourhood of $L$ contains $x_n$ for infinitely many $n$. Also called a cluster value.2.2
Ch 3 · Continuity and Differentiability
TermMeaningUsed in
compact setA set for which every open cover has a finite subcover. In $\mathbb{R}$ (and $\mathbb{R}^n$) this is the same as closed and bounded, and the same as every sequence having a subsequence converging in the set.2.1 2.2 3.1 3.2
continuously differentiable$f$ is continuously differentiable ($C^1$) if $f'$ exists and is continuous. $x^2\sin\frac1x$ is differentiable but not $C^1$.3.3 4.3 5.2 6.1
convex function$f$ is convex on an interval if $f(\lambda x+(1-\lambda)y)\le\lambda f(x)+(1-\lambda)f(y)$: chords lie above the graph. For differentiable $f$ this means $f'$ is increasing; for twice differentiable $f$, $f''\ge0$.3.3
Darboux's theoremA derivative has the intermediate value property on every interval, even when it is discontinuous; so a derivative can have no jump or removable discontinuity.3.3
Dirichlet functionThe indicator function of $\mathbb{Q}$: $1$ at rational points and $0$ at irrational points. It has no limit at any point, so it is continuous nowhere.3.1 3.2 4.1 4.3
discontinuity of the second kindA discontinuity at which at least one one-sided limit does not exist as a real number, e.g. $\sin(1/x)$ or $e^{1/x}$ at $0$.3.2 3.3 4.3
extreme value theoremA continuous function on a nonempty compact set (such as $[a,b]$) is bounded and attains its maximum and minimum.2.2 3.1 3.3 4.1
Froda's theoremFor any function $f:\mathbb{R}\to\mathbb{R}$, the set of discontinuities of the first kind (removable or jump) is countable.3.2
Heine–Borel theoremA subset of $\mathbb{R}$ is compact if and only if it is closed and bounded. In particular every $[a,b]$ is compact.2.2 3.1 3.2 4.1
Heine–Cantor theoremA continuous function on a compact set is uniformly continuous; in particular every continuous function on $[a,b]$ is.3.2 7.2 7.4
Hölder continuous$|f(x)-f(y)|\le K|x-y|^\alpha$ for some $0\lt\alpha\le1$. $\sqrt x$ is Hölder of order $\tfrac12$ but not Lipschitz.3.2 7.1
intermediate value theoremIf $f$ is continuous on $[a,b]$ and $y$ lies between $f(a)$ and $f(b)$, then $f(c)=y$ for some $c\in[a,b]$. Continuous images of intervals are intervals.1.1 1.2 3.1 3.2
jump discontinuityA point where both one-sided limits exist but are different; their difference $f(c+)-f(c-)$ is the jump. Monotone functions have only jump discontinuities.3.2 3.3
L'Hôpital's ruleIf $f,g\to0$ (or $|g|\to\infty$), $g'\ne0$ near the point and $f'/g'\to L$, then $f/g\to L$. It is a sufficient condition only.2.2 3.3 4.1
Lagrange remainderIn Taylor's theorem, $f(x)-T_n(x)=\frac{f^{(n+1)}(\xi)}{(n+1)!}(x-c)^{n+1}$ for some $\xi$ between $c$ and $x$.3.3 5.1 6.1
limit pointA point $c$ such that every interval $(c-\delta,c+\delta)$ contains a point of the set other than $c$; equivalently, a limit of a sequence of points of the set different from $c$. It need not belong to the set.1.1 2.2 3.1 3.2
Lipschitz functionA function with $|f(x)-f(y)|\le K|x-y|$ for some constant $K$. A bounded derivative on an interval makes a function Lipschitz; Lipschitz functions are uniformly continuous.3.2 3.3 4.1 4.3
mean value theoremLagrange: if $f$ is continuous on $[a,b]$ and differentiable on $(a,b)$, then $f(b)-f(a)=f'(c)(b-a)$ for some $c\in(a,b)$. Cauchy's version: $(f(b)-f(a))g'(c)=(g(b)-g(a))f'(c)$.2.1 3.2 3.3 4.1
nowhere differentiable functionA continuous function with no derivative at any point, such as Weierstrass's $\sum a^n\cos(b^n\pi x)$ or Takagi's $\sum2^{-n}\operatorname{dist}(2^nx,\mathbb{Z})$.3.3 4.3 5.1 7.1
open coverA family of open sets whose union contains the given set. A subcover is a subfamily that still covers it.3.1 3.2 5.1 7.2
removable discontinuityA point where both one-sided limits exist and are equal but differ from the value of the function (or the function is undefined). Redefining one value removes it.3.2 3.3
Rolle's theoremIf $f$ is continuous on $[a,b]$, differentiable on $(a,b)$ and $f(a)=f(b)$, then $f'(c)=0$ for some $c\in(a,b)$.3.3
Taylor polynomial$T_n(x)=\sum_{k=0}^n\frac{f^{(k)}(c)}{k!}(x-c)^k$, the polynomial of degree at most $n$ matching $f$ and its first $n$ derivatives at $c$.3.3 5.1 6.1 7.4
Thomae's function$T(p/q)=1/q$ for $p/q$ in lowest terms and $T(x)=0$ for irrational $x$. Its limit is $0$ everywhere, so it is continuous exactly at the irrational points.3.1 3.2 4.1 4.3
uniformly continuous$f$ is uniformly continuous on $A$ if for every $\varepsilon\gt0$ one $\delta\gt0$ works for all points: $|x-y|\lt\delta\Rightarrow|f(x)-f(y)|\lt\varepsilon$. Example: $\sqrt x$ on $[0,\infty)$; non-example: $x^2$ on $\mathbb{R}$.2.2 3.1 3.2 3.3
Ch 4 · Integration: Riemann to Lebesgue
TermMeaningUsed in
almost everywhereExcept on a set of measure zero. For example $\mathbf 1_{\mathbb Q}=0$ almost everywhere (a.e.).4.3 4.4
Beta function$B(p,q)=\int_0^1x^{p-1}(1-x)^{q-1}dx$ for $p,q\gt0$; $B(p,q)=\Gamma(p)\Gamma(q)/\Gamma(p+q)$.4.2
Borel setA member of the smallest σ-algebra containing the open sets; e.g. open, closed, $G_\delta$ and $F_\sigma$ sets. Every Borel set is Lebesgue measurable, but not conversely.4.4
bounded convergence theoremOn a set of finite measure, if $f_n\to f$ a.e. and $|f_n|\le M$, then $\int f_n\to\int f$ — a special case of dominated convergence.4.5 5.2
bounded variation$f$ is of bounded variation on $[a,b]$, $f\in\mathrm{BV}[a,b]$, if $V_a^b(f)\lt\infty$. Monotone and Lipschitz functions are BV; $x\sin\frac1x$ is continuous but not BV.4.3 4.4 4.5
Cantor setThe set $C$ left after repeatedly removing open middle thirds from $[0,1]$; equivalently, the numbers with a ternary expansion using only the digits $0$ and $2$. Compact, perfect, nowhere dense, uncountable and of measure $0$.1.1 3.1 3.2 4.1
Cauchy principal valueThe symmetric limit $\lim_{R\to\infty}\int_{-R}^Rf$ (or $\lim_{\varepsilon\to0}$ of the integral with $(c-\varepsilon,c+\varepsilon)$ removed); it may exist when the improper integral diverges, as for $\int_{-1}^1\frac{dx}x$.4.2
Darboux sumThe upper sum $U(f,P)=\sum M_k\Delta x_k$ and lower sum $L(f,P)=\sum m_k\Delta x_k$ built from the supremum and infimum of $f$ on each subinterval.4.1 4.4 4.5 5.2
Dirichlet integral$\int_0^\infty\frac{\sin x}{x}\,dx=\frac\pi2$: convergent by Dirichlet's test, but not absolutely convergent.4.2
dominated convergence theoremIf $f_n\to f$ a.e. and $|f_n|\le g$ with $g$ integrable, then $\int|f_n-f|\to0$ and $\int f_n\to\int f$ (Lebesgue).4.5 5.2
Egorov's theoremOn a set of finite measure, a.e. convergence of measurable functions is uniform outside a set of arbitrarily small measure.4.4
fat Cantor setA Cantor-type set obtained by removing middle intervals whose total length is less than $1$; it is closed and nowhere dense but has positive measure, so its indicator is not Riemann integrable.4.1 4.4 4.5
Fatou's lemmaFor measurable $f_n\ge0$: $\int\liminf f_n\le\liminf\int f_n$. The inequality may be strict, and it needs $f_n\ge0$.4.5
Fubini's theoremIf $\iint|f|\lt\infty$, the double integral equals both iterated integrals. Without integrability the iterated integrals can differ.4.5
fundamental theorem of calculusPart I: $F(x)=\int_a^xf$ is Lipschitz and $F'(c)=f(c)$ wherever $f$ is continuous. Part II: if $F'$ exists on $[a,b]$ and is integrable, $\int_a^bF'=F(b)-F(a)$.4.1 4.3 4.5 5.2
Gamma function$\Gamma(s)=\int_0^\infty x^{s-1}e^{-x}dx$ for $s\gt0$; $\Gamma(s+1)=s\Gamma(s)$, $\Gamma(n+1)=n!$, $\Gamma(\tfrac12)=\sqrt\pi$.4.2
improper integralAn integral over an unbounded interval, or of a function unbounded near some point, defined as a limit of proper Riemann integrals, e.g. $\int_a^\infty f=\lim_{b\to\infty}\int_a^bf$. It converges when the limit exists and is finite.2.3 4.1 4.2 4.3
integratorThe function $\alpha$ in a Riemann–Stieltjes integral $\int f\,d\alpha$; its increments replace the lengths $\Delta x_k$, and its jumps act as point masses.4.3
Jordan decompositionJordan's theorem: $f$ is of bounded variation iff $f=g-h$ with $g,h$ increasing; the most economical choice is $g=f(a)+v^+$, $h=v^-$.4.3
Lebesgue integrableA measurable $f$ with $\int|f|\lt\infty$. Integrability is absolute: $f$ is integrable iff $|f|$ is, so $\frac{\sin x}{x}$ is not Lebesgue integrable on $(0,\infty)$.4.3 4.5
Lebesgue integralFor measurable $f\ge0$, $\int f=\sup\int\varphi$ over simple $0\le\varphi\le f$, where $\int\sum a_i\mathbf 1_{E_i}=\sum a_im(E_i)$; for integrable $f$, $\int f=\int f^+-\int f^-$.4.4 4.5 5.2 7.1
Lebesgue's criterion$f$ is Riemann integrable on $[a,b]$ iff it is bounded and its set of discontinuities has measure zero.4.1 4.3 4.4 4.5
locally integrableRiemann integrable on every closed bounded subinterval of the given interval; the standing hypothesis for improper Riemann integrals.4.2
Markov's inequalityFor measurable $f\ge0$ and $t\gt0$: $m(\{f\ge t\})\le\frac1t\int f$.4.5
measurable functionA function $f$ on a measurable set with $\{f\gt a\}$ measurable for every real $a$. Closed under sums, products, sup, inf, lim sup, lim inf and pointwise limits.4.4 4.5
measurable setA set $E$ that splits every set additively: $m^*(A)=m^*(A\cap E)+m^*(A\cap E^c)$ for all $A$ (Carathéodory). Measurable sets form a σ-algebra containing all Borel sets and all null sets.4.4 4.5
measure zeroA set coverable by countably many open intervals of total length less than any given $\varepsilon\gt0$. Countable sets and the Cantor set have measure zero.1.1 4.1 4.3 4.4
outer measureFor any $A\subseteq\mathbb{R}$, $m^*(A)=\inf\sum_k\ell(I_k)$ over all countable covers of $A$ by open intervals $I_k$. Defined for every set, countably subadditive, and equal to the length on intervals.4.4
positive and negative variation$v^\pm=\frac12\big(v\pm(f-f(a))\big)$: the total climb and the total descent of $f$ over $[a,x]$; both are increasing and $f=f(a)+v^+-v^-$.4.3
rectifiable curveA curve $\gamma:[a,b]\to\mathbb{R}^n$ whose inscribed polygons have bounded lengths; equivalently, every coordinate of $\gamma$ is of bounded variation. For $C^1$ curves $L=\int_a^b|\gamma'|$.4.3
Riemann integrableA bounded $f$ on $[a,b]$ whose upper and lower integrals agree; equivalently, $U(f,P)-L(f,P)$ can be made smaller than any $\varepsilon\gt0$ (Riemann's criterion).2.3 3.1 3.2 3.3
Riemann sumA sum $\sum f(\xi_k)\Delta x_k$ with one tag $\xi_k$ chosen in each subinterval; for an integrable $f$ it tends to $\int_a^bf$ as the mesh tends to $0$, whatever the tags.4.1 4.3
Riemann–Stieltjes integral$\int_a^bf\,d\alpha$, the limit of sums $\sum f(\xi_k)\big(\alpha(x_k)-\alpha(x_{k-1})\big)$; it exists for continuous $f$ and $\alpha$ of bounded variation, and equals $\int f\alpha'$ for $C^1$ integrators.4.3
simple functionA measurable function with finitely many values, $\varphi=\sum_{i}a_i\mathbf 1_{E_i}$ with disjoint measurable $E_i$. Every non-negative measurable function is an increasing limit of simple functions.4.4 4.5
tagged partitionA partition $a=x_0\lt x_1\lt\cdots\lt x_n=b$ of $[a,b]$ together with a chosen point (tag) in each subinterval; its mesh is the length of the longest subinterval.4.1 4.3
Tonelli's theoremFor measurable $f\ge0$ on $\mathbb R^2$, the double integral equals both iterated integrals (all possibly infinite); no integrability is needed.4.2 4.5
total variation$V_a^b(f)=\sup_P\sum_k|f(x_k)-f(x_{k-1})|$ over all partitions of $[a,b]$: the total distance travelled up and down by $f(x)$ as $x$ runs from $a$ to $b$.4.3
variation function$v(x)=V_a^x(f)$, the variation of $f$ over $[a,x]$; it is increasing and jumps exactly where $f$ does, by the same one-sided amounts.4.3
σ-algebraA family of sets containing $\varnothing$ and closed under complements and countable unions (hence also countable intersections).4.4
Ch 5 · Sequences and Series of Functions
TermMeaningUsed in
Abel's theoremIf $\sum a_n$ converges, then $\sum a_nx^n$ converges uniformly on $[0,1]$ and $\lim_{x\to1^-}\sum a_nx^n=\sum a_n$.2.3 5.1 5.2
Bernstein polynomial$B_nf(x)=\sum_{k=0}^nf(\tfrac kn)\binom nkx^k(1-x)^{n-k}$; for continuous $f$ on $[0,1]$, $B_nf\to f$ uniformly.4.1 5.2 7.4
Cauchy–Hadamard formulaThe formula $\frac1R=\limsup_{n\to\infty}|a_n|^{1/n}$ for the radius of convergence of $\sum a_n(x-c)^n$.2.2 5.2
Dini's theoremOn a compact set, a monotone sequence of continuous functions converging pointwise to a continuous function converges uniformly.5.1 7.2
pointwise convergence$f_n\to f$ pointwise on $D$ if $f_n(x)\to f(x)$ for each $x\in D$ separately; the index $N(\varepsilon,x)$ may depend on the point.4.4 4.5 5.1 5.2
power seriesA series of functions $\sum_{n\ge0}a_n(x-c)^n$; it converges on an interval centred at $c$ and can be differentiated and integrated term by term inside it.2.3 5.1 5.2
radius of convergenceThe number $R=1/\limsup|a_n|^{1/n}$: $\sum a_n(x-c)^n$ converges absolutely for $|x-c|\lt R$ and diverges for $|x-c|\gt R$; the end points must be tested separately.2.2 5.2
sup norm$\|g\|_D=\sup_{x\in D}|g(x)|$; uniform convergence is convergence in this norm.5.1 5.2 7.4 10.1
uniform convergence$f_n\to f$ uniformly on $D$ if one $N(\varepsilon)$ works for all $x\in D$; equivalently $\sup_{x\in D}|f_n(x)-f(x)|\to0$.3.3 4.3 4.4 4.5
uniform convergence on compact subsetsUniform convergence on every compact subset of an open set; it is what power series enjoy inside their interval of convergence, and it preserves continuity.5.2
uniformly boundedA family of functions is uniformly bounded on $D$ if one constant $K$ satisfies $|f_n(x)|\le K$ for all $n$ and all $x\in D$.5.1 7.4
uniformly Cauchy$(f_n)$ is uniformly Cauchy on $D$ if $|f_n(x)-f_m(x)|\lt\varepsilon$ for all $m,n\ge N(\varepsilon)$ and all $x\in D$; this is equivalent to uniform convergence.4.3 5.1 5.2 7.1
Weierstrass approximation theoremEvery continuous function on a closed bounded interval is a uniform limit of polynomials.5.1 5.2 7.1 7.4
Weierstrass M-testIf $|u_n(x)|\le M_n$ on $D$ and $\sum M_n\lt\infty$, then $\sum u_n$ converges absolutely and uniformly on $D$.3.3 4.3 5.1 5.2
Ch 6 · Functions of Several Variables
TermMeaningUsed in
critical pointA point where $\nabla f=0$. Every interior local extremum of a differentiable function is a critical point; the converse fails (saddle points).3.3 6.2 10.3
diffeomorphismA bijection between open sets which is $C^1$ and has a $C^1$ inverse. A local diffeomorphism is one near each point; $e^z$, written as a map of $\mathbb{R}^2$, is a local but not a global diffeomorphism.6.2
directional derivative$D_uf(a)=\lim_{t\to0}\frac{f(a+tu)-f(a)}{t}$ for a unit vector $u$. If $f$ is differentiable at $a$, then $D_uf(a)=\nabla f(a)\cdot u$.6.1
gradientThe vector $\nabla f(a)=(D_1f(a),\dots,D_nf(a))$ of partial derivatives. For differentiable $f$ it points in the direction of fastest increase, with rate $\|\nabla f(a)\|$, and is perpendicular to the level set through $a$.3.3 4.3 6.1 6.2
Hessian matrixThe symmetric matrix $H_f(a)=\left[\partial^2f/\partial x_i\partial x_j(a)\right]$ of second partial derivatives of a $C^2$ function; it gives the quadratic term $\frac12h^{\mathsf T}H_f(a)h$ of Taylor's formula.6.1 6.2 10.3
homogeneous function$f$ is homogeneous of degree $k$ if $f(tx)=t^kf(x)$ for $t\gt0$. Euler's theorem: for differentiable $f$ this holds iff $x\cdot\nabla f(x)=kf(x)$.6.1
implicit function theoremIf $F$ is $C^1$, $F(a,b)=0$ and $D_yF(a,b)$ is invertible, then near $(a,b)$ the equation $F(x,y)=0$ is equivalent to $y=g(x)$ for a unique $C^1$ function $g$, with $Dg=-(D_yF)^{-1}D_xF$.6.2
inverse function theoremIf $f$ is $C^1$ near $a$ and $\det J_f(a)\ne0$, then $f$ maps a neighbourhood of $a$ bijectively onto a neighbourhood of $f(a)$, with $C^1$ inverse $g$ and $Dg(f(x))=Df(x)^{-1}$. The result is local only.6.2
iterated limitA limit taken one variable at a time, such as $\lim_{x\to a}\lim_{y\to b}f(x,y)$. Equal iterated limits do not imply that the double limit exists: $\frac{xy}{x^2+y^2}$ at the origin.5.1 6.1
Jacobian matrixThe $m\times n$ matrix $\left[\partial f_i/\partial x_j(a)\right]$ of the total derivative of $f:\mathbb{R}^n\to\mathbb{R}^m$; for $m=n$ its determinant is the Jacobian determinant. The chain rule multiplies Jacobian matrices.6.1 6.2 8.2
Lagrange multiplierAt a local extremum of $f$ on $\{g_1=\dots=g_k=0\}$ with independent $\nabla g_i$, $\nabla f=\lambda_1\nabla g_1+\cdots+\lambda_k\nabla g_k$; the numbers $\lambda_i$ are the Lagrange multipliers.6.2
partial derivative$\frac{\partial f}{\partial x_j}(a)=\lim_{t\to0}\frac{f(a+te_j)-f(a)}{t}$: the derivative in one coordinate direction with the other variables frozen. Partial derivatives can exist at a point where $f$ is discontinuous.6.1 6.2
saddle pointA critical point that is neither a local maximum nor a local minimum, e.g. the origin for $x^2-y^2$. An indefinite Hessian gives a saddle point.6.2
Schwarz's theoremIf $f_x,f_y,f_{xy}$ exist near a point and $f_{xy}$ is continuous there, then $f_{yx}$ exists and equals $f_{xy}$; in particular mixed partials of a $C^2$ function are equal.6.1
second derivative testAt a critical point of a $C^2$ function: positive definite Hessian gives a strict local minimum, negative definite a strict local maximum, indefinite a saddle; a singular semidefinite Hessian decides nothing. In two variables use $D=f_{xx}f_{yy}-f_{xy}^2$.6.2
total derivativeThe linear map $Df(a)$ with $\|f(a+h)-f(a)-Df(a)h\|/\|h\|\to0$ as $h\to0$; $f$ is differentiable at $a$ when it exists.6.1 6.2
Ch 7 · Metric and Normed Spaces
TermMeaningUsed in
Arzelà–Ascoli theoremOn a compact metric space $K$, a family in $C(K)$ has every sequence possessing a uniformly convergent subsequence iff it is pointwise bounded and equicontinuous.7.2 7.4
Baire category theoremIn a nonempty complete metric space a countable intersection of dense open sets is dense; equivalently the space is not a countable union of closed sets with empty interior.1.1 3.1 3.2 7.1
Banach contraction principleA contraction $d(Tx,Ty)\le k\,d(x,y)$, $k\lt1$, of a nonempty complete metric space has exactly one fixed point, and the iterates $x_{n+1}=Tx_n$ converge to it with error at most $\tfrac{k^n}{1-k}d(x_0,x_1)$.2.1 6.2 7.1
Banach spaceA normed space that is complete in its norm metric, e.g. $\mathbb{R}^n$, $\ell^p$, $\ell^\infty$ and $(C[a,b],\|\cdot\|_\infty)$; $(C[0,1],\|\cdot\|_1)$ is not one.4.5 7.1 7.4
Bolzano–Weierstrass propertyEvery infinite subset of the space has a limit point in the space; for metric spaces it is equivalent to compactness.7.2
bounded linear operatorA linear map $T$ with $\|Tx\|\le M\|x\|$ for some $M$; for linear maps between normed spaces, bounded is the same as continuous.7.4
clopen setA subset that is both open and closed. A space is connected exactly when its only clopen subsets are $\varnothing$ and the whole space.7.1 7.3
closed setA set whose complement is open; equivalently, a set that contains the limit of every convergent sequence of its points.1.1 3.1 4.1 4.4
closureThe smallest closed set containing $A$: all points $x$ with $d(x,A)=0$, i.e. $A$ together with its limit points.3.1 3.2 5.2 7.1
complete metric spaceA metric space in which every Cauchy sequence converges to a point of the space, such as $\mathbb{R}^n$, $\ell^p$ and $C[a,b]$ with the sup metric; $\mathbb{Q}$ and $(0,1)$ are not complete.2.1 6.2 7.1 7.4
connected componentA maximal connected subset (component) or maximal path-connected subset (path component). Components partition the space and are closed; in an open subset of $\mathbb{R}^n$ they are open and countably many.7.3
connected setA set that cannot be written as the union of two disjoint nonempty relatively open subsets; equivalently, every continuous map from it to $\{0,1\}$ is constant. In $\mathbb{R}$ the connected sets are exactly the intervals.3.1 4.4 6.1 7.3
cut pointA point $p$ of a connected space $X$ such that $X\setminus\{p\}$ is disconnected. Homeomorphisms map cut points of each order to cut points of the same order, which is used to prove spaces are not homeomorphic.7.3
discrete metricThe metric with $d(x,y)=1$ for $x\ne y$ and $d(x,x)=0$. Every subset is open and closed, and a sequence converges only if it is eventually constant.3.1 7.1 7.2 7.3
Edelstein's fixed point theoremA map $T$ of a nonempty compact metric space with $d(Tx,Ty)\lt d(x,y)$ for $x\ne y$ has exactly one fixed point, and all its orbits converge to it.7.1 7.2
equicontinuousA family $\mathcal{F}$ of functions is equicontinuous at $x_0$ if one $\delta$ works for all $f\in\mathcal{F}$: $d(x,x_0)\lt\delta\Rightarrow|f(x)-f(x_0)|\lt\varepsilon$.7.2 7.4
equivalent metricsTwo metrics on one set are topologically equivalent if they have the same open sets, and strongly equivalent if $\alpha d\le\rho\le\beta d$ for constants $0\lt\alpha\le\beta$. Strong equivalence preserves Cauchy sequences and completeness; topological equivalence alone need not.7.1 7.2
equivalent normsNorms with $c\|x\|\le\|x\|'\le C\|x\|$ for constants $0\lt c\le C$; they have the same open sets and convergent sequences. All norms on a finite-dimensional space are equivalent.7.4
finite intersection propertyA family of sets has it if every finite subfamily has nonempty intersection. A metric space is compact iff every family of closed sets with this property has nonempty intersection.7.2
Hilbert cubeThe set of $x\in\ell^2$ with $|x_n|\le1/n$ for all $n$: an infinite-dimensional compact set, because its tails are uniformly small.7.2 7.4
Lebesgue numberFor an open cover $\mathcal U$, a number $\lambda\gt0$ such that every ball $B(x,\lambda)$ lies inside a single member of $\mathcal U$. Every open cover of a compact metric space has one.3.2 4.1 7.2
metric spaceA set $X$ with a distance $d(x,y)\ge0$ that vanishes only when $x=y$, is symmetric, and satisfies the triangle inequality $d(x,z)\le d(x,y)+d(y,z)$.2.1 3.1 5.1 6.2
normed linear spaceA vector space $X$ with a norm $\|\cdot\|$ (definite, homogeneous, satisfying the triangle inequality); it is a metric space with $d(x,y)=\|x-y\|$.7.1 7.2 7.3 7.4
nowhere denseA set whose closure has empty interior, such as $\mathbb{Z}$ or the Cantor set in $\mathbb{R}$, or a line in $\mathbb{R}^2$. A countable union of nowhere dense sets is called meagre (first category).4.1 4.4 4.5 7.1
open setA set $U$ such that every point $x\in U$ has a ball $B(x,r)=\{y:d(x,y)\lt r\}$ inside $U$. Arbitrary unions and finite intersections of open sets are open.1.1 3.1 4.1 4.4
operator norm$\|T\|=\sup_{\|x\|\le1}\|Tx\|$, the smallest $M$ with $\|Tx\|\le M\|x\|$; for matrices it is the largest column sum ($\|\cdot\|_1$), row sum ($\|\cdot\|_\infty$) or singular value ($\|\cdot\|_2$).6.1 6.2 7.4 10.2
path-connectedAny two points can be joined by a continuous path $\gamma:[0,1]\to X$. Path-connected implies connected, but not conversely (topologist's sine curve); for open subsets of $\mathbb{R}^n$ the two notions agree.7.3
Riesz's lemmaIf $Y$ is a closed proper subspace of a normed space and $0\lt\theta\lt1$, some unit vector is at distance at least $\theta$ from $Y$; it implies that the closed unit ball is compact only in finite dimensions.7.2 7.4
sequentially compactEvery sequence in the space has a subsequence converging to a point of the space. In a metric space this is equivalent to compactness.2.2 3.1 3.2 7.2
Stone–Weierstrass theoremA subalgebra of $C(K,\mathbb{R})$ ($K$ compact) that contains the constants and separates points is dense in the sup norm; in the complex case it must also be closed under conjugation.7.4 10.1
topologist's sine curveThe set $\{(x,\sin\frac1x):0\lt x\le1\}\cup(\{0\}\times[-1,1])$: compact and connected, but not path-connected.7.2 7.3
totally boundedFor every $\varepsilon\gt0$ the set is covered by finitely many balls of radius $\varepsilon$. Bounded sets in $\mathbb{R}^n$ are totally bounded; the unit ball of $\ell^2$ is not.7.2 7.4
totally disconnectedA space whose only nonempty connected subsets are single points, e.g. $\mathbb{Q}$, $\mathbb{R}\setminus\mathbb{Q}$ and the Cantor set. A subset of $\mathbb{R}$ is totally disconnected iff it contains no nondegenerate interval.4.4 7.3
ε-netA finite set of points such that every point of the space lies within $\varepsilon$ of one of them.
Ch 8 · Vector Spaces and Linear Transformations
TermMeaningUsed in
adjugateThe transpose of the cofactor matrix, $(\operatorname{adj}A)_{ij}=C_{ji}$. It satisfies $A\operatorname{adj}A=\operatorname{adj}A\,A=(\det A)I$, so $A^{-1}=\operatorname{adj}A/\det A$ when $\det A\ne0$.6.2 8.3 9.1 10.1
basisA linearly independent spanning set of a vector space. Every vector is a combination of basis vectors in exactly one way; the coefficients are its coordinates.3.1 4.4 6.1 7.1
cofactor$C_{ij}=(-1)^{i+j}M_{ij}$, where the minor $M_{ij}$ is the determinant of $A$ with row $i$ and column $j$ deleted; $\det A=\sum_ka_{ik}C_{ik}$ along any row (or column).8.3
Cramer's ruleIf $\det A\ne0$, the unique solution of $Ax=b$ is $x_i=\det A_i/\det A$, where $A_i$ is $A$ with column $i$ replaced by $b$. It gives no information when $\det A=0$.8.3
determinant$\det A=\sum_{\sigma\in S_n}\operatorname{sgn}\sigma\,a_{1\sigma(1)}\cdots a_{n\sigma(n)}$ for a square matrix: the unique function of the rows that is linear in each row, vanishes when two rows are equal and equals $1$ at $I$. $A$ is invertible iff $\det A\ne0$.6.1 6.2 7.2 7.3
dimensionThe number of vectors in any basis of a finite-dimensional vector space (all bases have the same size, by the Steinitz exchange theorem). It depends on the field: $\dim_{\mathbb{R}}\mathbb{C}=2$, $\dim_{\mathbb{C}}\mathbb{C}=1$.3.1 4.4 6.2 7.2
direct sum$W_1+\cdots+W_k$ is a direct sum, written $W_1\oplus\cdots\oplus W_k$, if every vector of the sum is $w_1+\cdots+w_k$ ($w_i\in W_i$) in only one way; for two subspaces this means $U\cap W=\{0\}$.8.1 8.2 9.1 9.2
dual space$V^*$, the space of linear functionals $V\to\mathbb{F}$. A basis $b_1,\dots,b_n$ of $V$ has a dual basis $f_1,\dots,f_n$ of $V^*$ with $f_i(b_j)=\delta_{ij}$, so $\dim V^*=\dim V$.7.4 8.2 10.1
elementary matrixA matrix obtained from the identity by one elementary row operation (swap two rows, scale a row by $c\ne0$, add a multiple of one row to another). Multiplying $A$ on the left by it performs that operation on $A$; every elementary matrix is invertible.8.3
Gaussian binomial coefficient$\binom nk_q$, the number of $k$-dimensional subspaces of an $n$-dimensional space over the field with $q$ elements: $\prod_{i=0}^{k-1}\frac{q^n-q^i}{q^k-q^i}$.8.1
idempotent matrixA square matrix with $P^2=P$: the projection onto $C(P)$ along $N(P)$. It is similar to $\operatorname{diag}(I_r,0)$, so $\operatorname{rank}P=\operatorname{tr}P$ over fields of characteristic $0$.7.2 7.3 8.3 10.1
index of nilpotencyFor a nilpotent $N$ (some power is $0$), the least $k$ with $N^k=0$. On an $n$-dimensional space $k\le n$, and $k=n$ whenever $\operatorname{rank}N=n-1$.8.2 9.3
isomorphismA bijective linear map. Finite-dimensional spaces over the same field are isomorphic exactly when they have the same dimension; $v\mapsto[v]_{\mathcal{B}}$ is an isomorphism onto $\mathbb{F}^n$.1.2 8.2
kernel of a linear mapThe kernel of a linear map $T$ is $\ker T=\{v:Tv=0\}$, a subspace of the domain; $T$ is injective exactly when $\ker T=\{0\}$. Its dimension is the nullity.8.2 8.3 10.1
linear mapA map $T:V\to W$ between vector spaces with $T(u+v)=Tu+Tv$ and $T(av)=aTv$. It is determined by its values on a basis, and those values can be chosen freely.6.1 6.2 7.4 8.2
linear spanThe span of a set $S$ is the set of all finite linear combinations $a_1v_1+\cdots+a_kv_k$ of vectors of $S$ — the smallest subspace containing $S$. $S$ is a spanning set of $V$ if its span is $V$.8.1
linearly independentVectors $v_1,\dots,v_k$ are linearly independent if $a_1v_1+\cdots+a_kv_k=0$ forces all $a_i=0$; otherwise they are linearly dependent, and one of them is a combination of the others.6.2 7.4 8.1 9.1
matrix of a linear map$[T]_{\mathcal{B}}^{\mathcal{C}}$: the matrix whose $j$-th column is the coordinate vector of $Tb_j$ in the basis $\mathcal{C}$. Then $[Tv]_{\mathcal{C}}=[T]_{\mathcal{B}}^{\mathcal{C}}[v]_{\mathcal{B}}$ and $[ST]=[S][T]$.8.2
quotient spaceFor a subspace $W\le V$, the set $V/W$ of cosets $v+W$ with $(u+W)+(v+W)=(u+v)+W$ and $a(v+W)=av+W$; it has dimension $\dim V-\dim W$.8.1
rank of a matrixThe dimension of the column space of $A$, which always equals the dimension of its row space; it is the number of pivots in an echelon form and the size of the largest nonzero minor.8.3 9.1 10.1
rank–nullity theoremFor a linear map $T:V\to W$ with $V$ finite-dimensional, $\dim V=\dim\operatorname{im}T+\dim\ker T$ (rank plus nullity).6.2 8.2 8.3 9.1
reduced row echelon formA matrix whose nonzero rows have leading entries $1$ moving strictly to the right, with zero rows at the bottom and each leading $1$ alone in its column. Every matrix is row equivalent to exactly one such matrix.8.3
Rouché–Capelli theorem$Ax=b$ is consistent iff $\operatorname{rank}A=\operatorname{rank}[A\mid b]$; the solutions then form $x_0+N(A)$, an affine subspace of dimension $n-\operatorname{rank}A$.8.3
similar matrices$A$ and $B$ are similar if $B=P^{-1}AP$ for an invertible $P$ — equivalently, they represent one operator in two bases. Similar matrices share rank, trace, determinant and characteristic polynomial (the converse fails).8.2 8.3 9.1 9.3
Steinitz exchange theoremIf $m$ vectors span $V$ and $k$ vectors of $V$ are linearly independent, then $k\le m$, and the independent vectors can be exchanged one by one into the spanning list.8.1
Sylvester's rank inequalityFor $A$ ($m\times n$) and $C$ ($n\times q$): $\operatorname{rank}A+\operatorname{rank}C-n\le\operatorname{rank}(AC)$. It is the case $B=I$ of Frobenius' inequality $\operatorname{rank}(AB)+\operatorname{rank}(BC)\le\operatorname{rank}B+\operatorname{rank}(ABC)$.8.3
transition matrixFor two ordered bases $\mathcal{B},\mathcal{B}'$ of $V$, the matrix $P$ whose columns are the vectors of $\mathcal{B}'$ in $\mathcal{B}$-coordinates; $[v]_{\mathcal{B}}=P[v]_{\mathcal{B}'}$ and $[T]_{\mathcal{B}'}=P^{-1}[T]_{\mathcal{B}}P$.8.2 10.3
vector spaceA set $V$ with an addition and a multiplication by scalars from a field $\mathbb{F}$ satisfying eight axioms (commutativity and associativity of $+$, a zero vector, negatives, $a(bv)=(ab)v$, $1v=v$ and the two distributive laws).4.3 7.1 7.4 8.1
Ch 9 · Eigenvalues and Canonical Forms
TermMeaningUsed in
algebraic multiplicityThe multiplicity $a(\lambda)$ of $\lambda$ as a root of the characteristic polynomial. Always $g(\lambda)\le a(\lambda)$.9.1 9.2 9.3 10.1
Cayley–Hamilton theoremEvery square matrix satisfies its own characteristic polynomial: $p_A(A)=0$. Hence the minimal polynomial divides the characteristic polynomial.9.1 9.2 9.3
characteristic polynomial$p_A(x)=\det(xI-A)$, monic of degree $n$: $x^n-(\operatorname{tr}A)x^{n-1}+\cdots+(-1)^n\det A$. Its roots in the field are the eigenvalues; similar matrices have the same one.8.2 9.1 9.2 9.3
companion matrixFor $f=x^n+c_{n-1}x^{n-1}+\cdots+c_0$, the matrix with $1$s on the subdiagonal and last column $(-c_0,\dots,-c_{n-1})^T$; its characteristic and minimal polynomials both equal $f$.9.1 9.2 9.3
cyclic vectorA vector $v$ such that $v,Av,\dots,A^{n-1}v$ is a basis. $A$ has one iff $m_A=p_A$ iff $A$ is similar to the companion matrix of $p_A$.9.1
diagonalisable$A$ is diagonalisable over $\mathbb{F}$ if $P^{-1}AP$ is diagonal for some invertible $P$ over $\mathbb{F}$; equivalently there is a basis of eigenvectors, or the minimal polynomial is a product of distinct linear factors over $\mathbb{F}$.7.3 8.3 9.1 9.2
eigenspace$E_\lambda=\ker(T-\lambda I)$: the eigenvectors belonging to $\lambda$ together with $0$. Its dimension is the geometric multiplicity of $\lambda$.9.1 9.2 9.3 10.2
eigenvalueA scalar $\lambda$ in the field such that $Tv=\lambda v$ for some nonzero vector $v$; equivalently a root of the characteristic polynomial, or a $\lambda$ for which $T-\lambda I$ is not invertible (in finite dimension).6.2 7.1 7.2 7.3
eigenvectorA nonzero vector $v$ with $Tv=\lambda v$ for some scalar $\lambda$: a direction that $T$ only stretches (or reverses). Eigenvectors for distinct eigenvalues are linearly independent.6.2 7.4 9.1 9.2
Fitting decompositionFor $N=T-\lambda I$ and $s$ the step where the kernels of $N^k$ stop growing, $V=\ker N^s\oplus\operatorname{im}N^s$: $N$ is nilpotent on the first summand and invertible on the second.9.3
generalised eigenspace$G_\lambda(T)=\ker(T-\lambda I)^n$ ($n=\dim V$), the set of vectors killed by some power of $T-\lambda I$. Its dimension is the algebraic multiplicity of $\lambda$.9.2 9.3
generalised eigenvectorA nonzero vector $v$ with $(T-\lambda I)^kv=0$ for some $k\ge1$; the least such $k$ is its height. Height $1$ means an ordinary eigenvector.9.3
geometric multiplicity$g(\lambda)=\dim\ker(T-\lambda I)=n-\operatorname{rank}(T-\lambda I)$, the number of independent eigenvectors for $\lambda$.9.1 9.3
invariant factorsThe monic polynomials $f_1\mid\cdots\mid f_r$ of the rational canonical form; $f_r$ is the minimal polynomial and $f_1\cdots f_r$ the characteristic polynomial. Two matrices are similar iff they have the same invariant factors.9.3
invariant subspaceA subspace $W$ with $T(W)\subseteq W$. In a basis extending one of $W$, the matrix of $T$ is block upper triangular; kernels and images of polynomials in $T$ are invariant.9.2 9.3
Jordan blockThe $k\times k$ matrix $J_k(\lambda)$ with $\lambda$ on the diagonal, $1$ on the superdiagonal and $0$ elsewhere; its characteristic and minimal polynomials are both $(x-\lambda)^k$ and it has a single eigenvector.8.2 9.3
Jordan canonical formA block-diagonal matrix of Jordan blocks similar to a given matrix. It exists when the characteristic polynomial splits (always over $\mathbb{C}$) and is unique up to the order of the blocks.8.2 9.2 9.3
Jordan chainVectors $v_1,\dots,v_k$ with $(T-\lambda I)v_1=0$ and $(T-\lambda I)v_j=v_{j-1}$; in the basis $v_1,\dots,v_k$ of their span, $T$ acts as the Jordan block $J_k(\lambda)$.9.3
Jordan–Chevalley decompositionThe unique way to write $A=D+N$ with $D$ diagonalisable, $N$ nilpotent and $DN=ND$; both $D$ and $N$ are polynomials in $A$.9.3
minimal polynomialThe monic polynomial $m_A$ of least degree with $m_A(A)=0$. A polynomial annihilates $A$ iff $m_A$ divides it; $m_A$ and $p_A$ have the same roots.7.3 9.1 9.2 9.3
primary decomposition theoremIf $m_T=q_1^{e_1}\cdots q_k^{e_k}$ with distinct irreducible $q_i$, then $V=\bigoplus\ker q_i(T)^{e_i}$, each piece is invariant, $T$ restricted to it has minimal polynomial $q_i^{e_i}$, and the projections are polynomials in $T$.9.2 9.3
rational canonical formA block-diagonal matrix $\operatorname{diag}(C(f_1),\dots,C(f_r))$ of companion matrices with $f_1\mid f_2\mid\cdots\mid f_r$. Every matrix over any field is similar to exactly one such matrix.9.1 9.3
Schur's theoremEvery complex square matrix is unitarily similar to an upper triangular matrix, $A=UTU^*$. Consequence (Schur's inequality): $\sum|\lambda_i|^2\le\sum|a_{ij}|^2$, with equality iff $A$ is unitarily diagonalisable.9.2 9.3 10.1
simultaneously diagonalisableA family of operators is simultaneously diagonalisable if one basis diagonalises all of them. Diagonalisable operators are simultaneously diagonalisable iff they commute; commuting complex matrices are always simultaneously triangularisable.9.2 9.3
triangularisable$T$ is triangularisable if some basis makes its matrix upper triangular, i.e. there is a flag of invariant subspaces of every dimension. This happens iff the characteristic polynomial splits over the field.9.2 9.3 10.1 10.2
Ch 10 · Inner Product Spaces and Quadratic Forms
TermMeaningUsed in
Bessel's inequalityFor an orthonormal family $(e_k)$ and any $x$: $\sum_k|\langle x,e_k\rangle|^2\le\|x\|^2$. In particular the Fourier coefficients $\langle x,e_k\rangle$ tend to $0$.10.1
bilinear formA map $B:V\times V\to\mathbb{F}$ linear in each variable. In a basis $v_i$ it is $B(x,y)=[x]^TM[y]$ with $M=(B(v_i,v_j))$; it is symmetric iff $M$ is.10.3
Cauchy–Schwarz inequality$|\langle x,y\rangle|\le\|x\|\,\|y\|$ in every inner product space, with equality exactly when $x$ and $y$ are linearly dependent.2.3 4.1 4.3 6.1
congruent matrices$A$ and $B$ are congruent if $B=P^TAP$ for an invertible $P$ — they represent the same bilinear or quadratic form in two bases. Congruence preserves rank and, over $\mathbb{R}$, the inertia, but not the eigenvalues.10.3
Gershgorin discFor $A=(a_{ij})$, the disc $|z-a_{ii}|\le\sum_{j\ne i}|a_{ij}|$. Every eigenvalue lies in the union of these discs, and an isolated group of $k$ discs contains exactly $k$ eigenvalues.10.2
Gram–Schmidt processThe recursion $u_k=v_k-\sum_{j\lt k}\langle v_k,e_j\rangle e_j$, $e_k=u_k/\|u_k\|$, which turns linearly independent vectors into orthonormal ones with the same successive spans; in matrix form it is the $QR$ factorisation.7.3 9.2 10.1 10.2
index of a quadratic formThe index is $n_+$, the number of positive squares in a diagonal form; the inertia is the triple $(n_+,n_-,n_0)$. $n_+$ is the largest dimension of a subspace on which the form is positive definite.10.3
inner productA function $\langle x,y\rangle$ on a real or complex vector space that is linear in the first slot, conjugate symmetric ($\langle y,x\rangle=\overline{\langle x,y\rangle}$) and positive definite ($\langle x,x\rangle\gt0$ for $x\ne0$). It defines lengths $\|x\|=\sqrt{\langle x,x\rangle}$ and orthogonality.10.1 10.2
Lagrange's reductionReduction of a quadratic form to $\sum d_ky_k^2$ by completing squares one variable at a time (substituting $x_i=u+v$, $x_j=u-v$ when no square term is left); each step is a congruence.10.3
least-squares solutionA vector $\hat x$ minimising $\|Ax-b\|$; equivalently a solution of the normal equations $A^*A\hat x=A^*b$. It always exists, and is unique iff $A$ has independent columns.10.1
normal operatorAn operator that commutes with its adjoint, $T^*T=TT^*$; equivalently $\|Tx\|=\|T^*x\|$ for all $x$. Over $\mathbb{C}$, exactly the operators with an orthonormal basis of eigenvectors.7.3 7.4 9.2 10.2
orthogonal complementFor a subset $S$ of an inner product space, $S^\perp=\{x:\langle x,s\rangle=0\text{ for all }s\in S\}$, a subspace. For a finite-dimensional subspace $W$, $V=W\oplus W^\perp$.6.2 10.1 10.2
orthogonal projectionThe map $P_W$ sending $x$ to the unique $w\in W$ with $x-w\perp W$; it is the nearest point of $W$ to $x$. An idempotent operator is an orthogonal projection iff it is self-adjoint.8.2 10.1 10.2
orthonormal basisA basis $e_1,\dots,e_n$ with $\langle e_i,e_j\rangle=\delta_{ij}$. In it every vector is $x=\sum\langle x,e_j\rangle e_j$ and $\|x\|^2=\sum|\langle x,e_j\rangle|^2$.8.1 9.2 10.1 10.2
parallelogram law$\|x+y\|^2+\|x-y\|^2=2\|x\|^2+2\|y\|^2$. A norm comes from an inner product if and only if it satisfies this identity (Jordan–von Neumann).10.1
Parseval's identity$\|x\|^2=\sum_k|\langle x,e_k\rangle|^2$; it holds for all $x$ exactly when the orthonormal family is complete (a basis in finite dimension; dense span in general).10.1
polar decompositionThe factorisation $A=WP$ of a square matrix with $W$ unitary and $P=\sqrt{A^*A}$ positive semidefinite — the matrix analogue of $z=e^{i\theta}|z|$.10.2
positive definiteA self-adjoint $T$ (Hermitian matrix $A$) is positive definite if $\langle Tx,x\rangle\gt0$ ($x^*Ax\gt0$) for all $x\ne0$, and positive semidefinite if $\ge0$; equivalently all eigenvalues are $\gt0$ (resp. $\ge0$), or $A=B^*B$ with $B$ invertible (resp. any $B$).6.2 7.3 7.4 10.1
quadratic formA function $q(x)=B(x,x)$ with $B$ symmetric bilinear; on $\mathbb{R}^n$, $q(x)=x^TAx$ with $A$ symmetric, the coefficient of $x_ix_j$ ($i\ne j$) being $2a_{ij}$.6.2 10.1 10.2 10.3
Rayleigh quotient$R(x)=\langle Tx,x\rangle/\langle x,x\rangle$ for a self-adjoint $T$ and $x\ne0$. Its maximum and minimum are the largest and smallest eigenvalues, attained at eigenvectors (Courant–Fischer gives the others).10.2
self-adjoint operatorAn operator with $T^*=T$; for matrices $A^*=A$ (real case: symmetric). Its eigenvalues are real, eigenvectors for distinct eigenvalues are orthogonal, and it has an orthonormal eigenbasis.10.1 10.2 10.3
signatureFor a real quadratic form with $n_+$ positive and $n_-$ negative squares, the signature is $n_+-n_-$ (some texts call the pair $(n_+,n_-)$ the signature). With the rank it determines the form up to congruence.7.3 10.3
singular value decompositionEvery $m\times n$ matrix factors as $A=U\Sigma V^*$ with $U,V$ unitary and $\Sigma$ diagonal with entries $\sigma_1\ge\sigma_2\ge\dots\ge0$, the singular values: the square roots of the eigenvalues of $A^*A$. $\|A\|=\sigma_1$.7.4 8.3 10.1 10.2
spectral theoremA complex operator is normal iff it is unitarily diagonalisable, $A=UDU^*$; a real operator is self-adjoint (symmetric) iff it is orthogonally diagonalisable, $A=QDQ^T$. Equivalently $T=\sum\lambda_jP_j$ with orthogonal eigenprojections $P_j$.6.2 7.3 7.4 10.1
Sylvester's criterionA real symmetric (or Hermitian) matrix is positive definite iff all its leading principal minors $D_1,\dots,D_n$ are positive; negative definite iff $(-1)^kD_k\gt0$. For semidefiniteness all principal minors must be checked.6.2 10.1 10.3
Sylvester's law of inertiaIn every diagonal form $\sum d_ky_k^2$ of a real quadratic form, the numbers of positive, negative and zero coefficients are the same. They are the numbers of positive, negative and zero eigenvalues of its matrix.10.3
unitary matrixA square matrix with $U^*U=I$ (real case $Q^TQ=I$: orthogonal). It preserves inner products and lengths, its columns are orthonormal, and its eigenvalues have modulus $1$.7.3 8.3 9.2 10.2