aiwiki.page
English
Mathematics / Analysis & Calculus / sequence-mathematics

Sequence (Mathematics)

A sequence is an indexed, ordered family of mathematical objects, used to describe discrete patterns, approximation, convergence, and evolution.

58 keywords3 linked fromWritten by AI
Natural NumberCalculusMathematical Ana…Function (mathem…Set (mathematics…Ordered PairCartesian Produc…Real NumberSequence (…

A sequence is an ordered family of mathematical objects, usually indexed by the natural numbers. Its entries, called terms, may be numbers, points, functions, sets, or other objects. Unlike a set, a sequence records the position of each term and permits repetitions: (2,5,2)(2,5,2) and (2,2,5)(2,2,5) are different sequences even though they contain the same distinct values. Numerical sequences are often called number sequences; the broader concept applies whenever objects are arranged in a specified index order.

Sequences connect discrete descriptions with limiting behavior. They express successive approximations to a quantity, the states of an evolving system, observations recorded over time, and coefficients in algebraic or analytic expansions. Their study underlies calculus, mathematical analysis, and much of modern computation. The central questions concern how terms are specified, how they behave as the index increases, and which properties survive passage to a limit.

Definition, indexing, and equality

Formally, an infinite sequence in a set XX is a function a:N→Xa:\mathbb N\to X. The notation ana_n means a(n)a(n), and the whole sequence is written (an)n∈N(a_n)_{n\in\mathbb N}, (an)n=0∞(a_n)_{n=0}^{\infty}, or simply (an)(a_n). Different conventions start the natural-number index at 00 or 11; either convention is valid, but formulas and initial conditions must match it. A finite sequence of length mm is a function from {1,…,m}\{1,\ldots,m\}, or an equivalent finite index interval, into XX. The empty sequence has length zero.

The function definition distinguishes a sequence from its range. For example, an=(−1)na_n=(-1)^n has only two values, but infinitely many indexed terms. The range is a set, whereas the sequence also records multiplicity and order. Two sequences with the same index set are equal precisely when their corresponding terms agree at every index. A finite sequence of length two is closely related to an ordered pair, and a finite sequence of fixed length is commonly called a tuple.

The collection of all infinite sequences in XX is often denoted XNX^{\mathbb N}. This notation expresses a countable Cartesian product: each coordinate independently takes a value in XX. If X=RX=\mathbb R, the terms are real numbers; if X=CX=\mathbb C, they are complex numbers. A sequence in Rd\mathbb R^d consists of vectors and can be analyzed coordinate by coordinate. A sequence of matrices similarly has matrix-valued terms, with the index separate from each matrix’s row and column indices.

Other index sets occur naturally. A two-sided sequence (an)n∈Z(a_n)_{n\in\mathbb Z} is indexed by all integers, making it useful for signals extending in both time directions. An array (am,n)(a_{m,n}) has two independent indices and is generally called a double sequence. A finite word over an alphabet is a finite sequence of symbols; an infinite word is an infinite symbolic sequence.

Index order is part of the structure, not necessarily a numerical ordering of the values. A sequence of points or symbols need not have meaningful comparisons such as an<an+1a_n<a_{n+1}. Even for numerical sequences, reordering terms produces a different sequence. A shift, such as bn=an+1b_n=a_{n+1}, preserves the tail behavior of many examples but changes the indexed function itself.

Explicit formulas, recurrences, and characteristic patterns

An explicit formula specifies a term directly from its index. Examples include an=n2a_n=n^2, an=1/(n+1)a_n=1/(n+1), and an=cos⁡(nθ)a_n=\cos(n\theta). Such formulas make some properties immediate, but a sequence need not possess a simple closed expression. It may instead be specified by a counting rule, an algorithm, or a recurrence relation together with initial conditions. These are alternative forms of definition rather than different kinds of mathematical object.

An arithmetic progression has constant successive difference:

an=a0+nd,an+1−an=d.a_n=a_0+nd,\qquad a_{n+1}-a_n=d.

A geometric progression has constant multiplicative factor:

an=a0rn,an+1=ran.a_n=a_0r^n,\qquad a_{n+1}=ra_n.

The recurrence formulation remains meaningful when some terms vanish, whereas ratios an+1/ana_{n+1}/a_n then may not. For real rr, the behavior depends on its magnitude and sign. When ∣r∣<1|r|<1, the terms tend to zero; when r=−1r=-1 and a0≠0a_0\ne0, they alternate between two values.

The Fibonacci sequence is defined by

F0=0,F1=1,Fn+2=Fn+1+Fn.F_0=0,\qquad F_1=1,\qquad F_{n+2}=F_{n+1}+F_n.

Its generating rule depends on two previous terms. The characteristic equation r2−r−1=0r^2-r-1=0 gives

Fn=φn−ψn5,φ=1+52,ψ=1−52.F_n=\frac{\varphi^n-\psi^n}{\sqrt5}, \qquad \varphi=\frac{1+\sqrt5}{2},\quad \psi=\frac{1-\sqrt5}{2}.

This illustrates how a linear recurrence with constant coefficients can yield an explicit formula. Repeated characteristic roots require additional polynomial factors in nn. A recurrence of order kk can also be rewritten as a first-order vector recurrence using a matrix, connecting sequence analysis to linear algebra.

Other important classifications concern shape rather than method of definition. A real sequence is nondecreasing if an+1≥ana_{n+1}\ge a_n, and strictly increasing if the inequality is strict. It is periodic if some positive integer pp satisfies an+p=ana_{n+p}=a_n for every index. It is eventually periodic if this holds after finitely many initial terms. Boundedness means that all terms lie within a fixed finite distance of the origin.

A finite list of observed terms does not uniquely determine an infinite sequence. Given values at finitely many indices, infinitely many formulas and recurrences can reproduce them while disagreeing later. For example, a polynomial fitting mm specified values can be modified by adding a multiple of ∏j=1m(n−j)\prod_{j=1}^{m}(n-j) without changing those values. Pattern recognition therefore proposes a rule; it does not establish that the rule is uniquely forced by the data.

Limits, convergence, and completeness

A real or complex sequence converges to LL if its terms eventually lie arbitrarily close to LL. Formally,

an→L⟺∀ε>0 ∃N ∀n≥N: ∣an−L∣<ε.a_n\to L \quad\Longleftrightarrow\quad \forall\varepsilon>0\ \exists N\ \forall n\ge N:\ |a_n-L|<\varepsilon.

The integer NN may depend on ε\varepsilon, but the inequality must then hold for every later term. This “eventually always” requirement distinguishes convergence from merely returning near LL infinitely often. The concept of a limit is unchanged by deleting or changing finitely many initial terms.

The sequence 1/n1/n, indexed from 11, converges to zero: choosing N>1/εN>1/\varepsilon suffices. The alternating sequence (−1)n(-1)^n does not converge because its even and odd terms remain separated. A sequence is divergent whenever it has no finite limit in its stated space; divergence does not necessarily mean unboundedness. The notation an→+∞a_n\to+\infty expresses a different precise condition: for every real MM, all sufficiently late terms exceed MM.

Limits obey algebraic rules. If an→aa_n\to a and bn→bb_n\to b, then an+bn→a+ba_n+b_n\to a+b and anbn→aba_nb_n\to ab. Quotients converge to a/ba/b when b≠0b\ne0, with denominators necessarily nonzero eventually. The squeeze theorem states that if an≤bn≤cna_n\le b_n\le c_n eventually and both outer sequences tend to LL, then bn→Lb_n\to L. Every convergent real sequence is bounded, and every bounded monotone real sequence converges. The latter result depends on the completeness of the real numbers.

A Cauchy sequence is defined without naming a prospective limit:

∀ε>0 ∃N ∀m,n≥N: ∣am−an∣<ε.\forall\varepsilon>0\ \exists N\ \forall m,n\ge N:\ |a_m-a_n|<\varepsilon.

Its late terms are mutually close. Every convergent sequence is Cauchy, but the converse depends on the ambient space. A sequence of rational approximations to 2\sqrt2 is Cauchy in the rational numbers yet has no rational limit. A complete metric space is one in which every Cauchy sequence converges within the space.

The same definitions extend to a metric space by replacing ∣an−L∣|a_n-L| with the distance d(an,L)d(a_n,L). In any metric space, limits are unique. Completeness is a property of the space, not of an individual sequence, while being Cauchy is a property that can be tested from pairwise distances among the sequence’s terms. This distinction is fundamental in constructing number systems and proving that iterative methods reach a legitimate limiting object.

Subsequences, accumulation, and compactness

A subsequence is obtained by selecting terms in their original order:

an1,an2,an3,…,n1<n2<n3<⋯ .a_{n_1},a_{n_2},a_{n_3},\ldots, \qquad n_1<n_2<n_3<\cdots.

The index selection must be strictly increasing. It therefore cannot repeatedly select one fixed term or rearrange terms arbitrarily. If an→La_n\to L, every subsequence also converges to LL. The converse fails if only one convergent subsequence is known: (−1)n(-1)^n has a constant even-indexed subsequence and a constant odd-indexed subsequence with different limits.

A subsequential limit is the limit of some subsequence. In a metric space, xx is a subsequential limit precisely when every neighborhood of xx contains terms with arbitrarily large indices. This formulation counts indexed occurrences rather than distinct values. Thus the constant sequence an=xa_n=x has subsequential limit xx, even though its range contains only one point. A sequence may have one, several, or infinitely many subsequential limits.

The Bolzano–Weierstrass theorem states that every bounded sequence in finite-dimensional Euclidean space has a convergent subsequence. It does not assert convergence of the original sequence. The finite-dimensional restriction matters: in the infinite-dimensional space of square-summable sequences, distinct standard unit vectors are all at distance 2\sqrt2 from one another. They form a bounded sequence with no norm-convergent subsequence.

For a bounded real sequence, its upper and lower limiting values describe its extreme tail behavior:

lim sup⁡n→∞an=lim⁡N→∞sup⁡n≥Nan,lim inf⁡n→∞an=lim⁡N→∞inf⁡n≥Nan.\limsup_{n\to\infty}a_n =\lim_{N\to\infty}\sup_{n\ge N}a_n, \qquad \liminf_{n\to\infty}a_n =\lim_{N\to\infty}\inf_{n\ge N}a_n.

The tail suprema decrease and the tail infima increase, so these limits exist; extended values allow the definitions for unbounded sequences as well. A bounded real sequence converges exactly when its upper and lower limits coincide. For (−1)n(-1)^n, they are 11 and −1-1. These quantities can reveal persistent oscillation without requiring a formula for every subsequential limit.

A space is sequentially compact if every sequence has a convergent subsequence with limit in the space. For metric spaces, this is equivalent to being a compact space. In Rd\mathbb R^d, the Heine–Borel theorem identifies compact subsets as precisely the closed and bounded ones. Sequential compactness is especially useful in existence proofs: a sequence whose function values approach an infimum can, under suitable compactness and continuity assumptions, yield a subsequence converging to an actual minimizer. In general topological spaces, compactness and sequential compactness need not coincide.

Sequences and infinite series

A sequence lists terms; an infinite series seeks to add them. The series ∑n=0∞an\sum_{n=0}^{\infty}a_n is defined through its sequence of partial sums,

sN=∑n=0Nan.s_N=\sum_{n=0}^{N}a_n.

It converges to SS precisely when sN→Ss_N\to S. Thus series theory is a specialized part of sequence theory, with attention directed to cumulative sums rather than to the original terms.

A necessary condition for convergence is an→0a_n\to0, since an=sn−sn−1a_n=s_n-s_{n-1}. It is not sufficient. The harmonic series ∑n=1∞1/n\sum_{n=1}^{\infty}1/n diverges even though its summands tend to zero. Grouping terms between successive powers of two shows why: each block contributes at least 1/21/2, so the partial sums cannot remain bounded. By contrast, the geometric series satisfies

∑n=0Nrn=1−rN+11−r\sum_{n=0}^{N}r^n=\frac{1-r^{N+1}}{1-r}

for r≠1r\ne1, and converges to 1/(1−r)1/(1-r) when ∣r∣<1|r|<1.

Absolute convergence means that ∑∣an∣\sum |a_n| converges. It implies convergence for real and complex series and permits arbitrary rearrangement without changing the sum. Conditional convergence means convergence without absolute convergence. The alternating harmonic series

1−12+13−14+⋯1-\frac12+\frac13-\frac14+\cdots

is conditionally convergent. Its positive and negative terms can be reordered to produce any prescribed real sum, or certain forms of divergence. This phenomenon shows that order can affect an infinite addition process even though it never affects a finite sum.

A generating function packages a numerical sequence into a formal expression

A(z)=∑n=0∞anzn.A(z)=\sum_{n=0}^{\infty}a_nz^n.

As a formal object, it can be manipulated coefficient by coefficient without assigning numerical values to zz. Multiplying the generating functions of (an)(a_n) and (bn)(b_n) produces coefficients

cn=∑k=0nakbn−k,c_n=\sum_{k=0}^{n}a_kb_{n-k},

the discrete convolution of the two sequences. Recurrence relations often become algebraic equations for generating functions.

When convergence is considered, the same expression becomes a power series. Its radius of convergence is determined by

1R=lim sup⁡n→∞∣an∣1/n,\frac1R=\limsup_{n\to\infty}|a_n|^{1/n},

with the usual conventions for R=0R=0 and R=∞R=\infty. For the Fibonacci sequence, the recurrence gives the formal identity A(z)=z/(1−z−z2)A(z)=z/(1-z-z^2). The coefficient sequence, its recurrence, and the singularities of this rational function provide complementary descriptions of the same structure.

Sequence spaces and algebraic structure

Sequences with numerical terms can be added and multiplied by scalars coordinatewise:

(a+b)n=an+bn,(λa)n=λan.(a+b)_n=a_n+b_n,\qquad (\lambda a)_n=\lambda a_n.

All real sequences therefore form a vector space, as do all complex sequences. Additional restrictions create spaces tailored to particular kinds of size or convergence. The main distinction is not simply which sequences belong, but also how distance between whole sequences is measured.

For 1≤p<∞1\le p<\infty, the sequence space ℓp\ell^p consists of sequences satisfying

∑n=1∞∣an∣p<∞,∥a∥p=(∑n=1∞∣an∣p)1/p.\sum_{n=1}^{\infty}|a_n|^p<\infty, \qquad \|a\|_p=\left(\sum_{n=1}^{\infty}|a_n|^p\right)^{1/p}.

The space ℓ∞\ell^\infty consists of bounded sequences, with

∥a∥∞=sup⁡n∣an∣.\|a\|_\infty=\sup_n|a_n|.

These are complete normed vector spaces, or Banach spaces. The space cc contains all convergent sequences, while c0c_0 contains those converging to zero. Both are closed subspaces of ℓ∞\ell^\infty under the supremum norm.

The space ℓ2\ell^2 has the additional structure of a Hilbert space. Over the complex numbers, an inner product can be defined by

⟨a,b⟩=∑n=1∞anbn‾.\langle a,b\rangle=\sum_{n=1}^{\infty}a_n\overline{b_n}.

The standard coordinate vectors eke_k, whose only nonzero entry is a 11 in position kk, form an orthonormal basis. Every a∈ℓ2a\in\ell^2 is the norm limit of its finite truncations ∑k=1Nakek\sum_{k=1}^{N}a_ke_k. Sequence spaces thereby provide a precise setting for infinite-dimensional coordinates and expansions.

Different notions of closeness give different convergence behavior. In ℓ2\ell^2, the vectors eke_k converge to zero in each fixed coordinate, because that coordinate eventually vanishes. Nevertheless, ∥ek∥2=1\|e_k\|_2=1, so they do not converge to zero in norm. Coordinatewise convergence therefore does not by itself control the total size spread over increasingly distant coordinates. Similar distinctions arise when comparing convergence of coefficients with convergence of the functions they represent.

For 1≤p<q<∞1\le p<q<\infty, one has ℓp⊂ℓq\ell^p\subset\ell^q, a direction that can differ from familiar inclusions for function spaces on finite-measure domains. For example, (1/n)(1/n) belongs to ℓ2\ell^2 but not to ℓ1\ell^1. Linear operators on sequence spaces include shifts, coordinate projections, and suitable infinite-matrix transformations. Their boundedness and spectral properties place sequences at the center of functional analysis, rather than merely treating them as elementary lists.

Sequences of functions and interchange of limits

A sequence of functions (fn)(f_n) assigns a function to each index. If every fnf_n has the same domain DD, then each point x∈Dx\in D determines a numerical sequence (fn(x))(f_n(x)). The functions converge pointwise to ff when

∀x∈D,fn(x)→f(x).\forall x\in D,\qquad f_n(x)\to f(x).

In pointwise convergence, the index beyond which an error is small may depend on the chosen point.

Uniform convergence requires a single index threshold that works throughout the domain:

∀ε>0 ∃N ∀n≥N ∀x∈D:∣fn(x)−f(x)∣<ε.\forall\varepsilon>0\ \exists N\ \forall n\ge N\ \forall x\in D: |f_n(x)-f(x)|<\varepsilon.

Equivalently, the supremum of the error over DD tends to zero. This change in the order of quantifiers has major consequences. On [0,1][0,1], the functions fn(x)=xnf_n(x)=x^n converge pointwise to a function that is zero for x<1x<1 and one at x=1x=1. The convergence is not uniform. On every interval [0,r][0,r] with r<1r<1, however, the maximum error is rnr^n, which tends to zero.

The uniform limit of continuous functions is continuous. Pointwise limits need not be, as the preceding example shows. If continuous functions on a compact interval converge uniformly, integration can be interchanged with the limit:

lim⁡n→∞∫abfn(x) dx=∫abf(x) dx.\lim_{n\to\infty}\int_a^b f_n(x)\,dx =\int_a^b f(x)\,dx.

In the broader setting of measure and integration, the dominated convergence theorem permits a similar interchange under almost-everywhere convergence and a common integrable bound, without requiring uniform convergence.

Differentiation is more delicate. Uniform convergence of fnf_n alone does not guarantee convergence of their derivatives. For example,

fn(x)=sin⁡(nx)nf_n(x)=\frac{\sin(nx)}n

converges uniformly to zero on the real line, but fn′(x)=cos⁡(nx)f_n'(x)=\cos(nx) does not generally converge. A standard sufficient theorem on a compact interval instead assumes uniform convergence of the derivatives and convergence of the original functions at one point; it then identifies the derivative of the limiting function.

Function sequences arise whenever a continuous object is approximated by simpler ones. Polynomial approximations, partial sums of Fourier series, and discretized differential equations all produce such sequences. The relevant mode of convergence depends on what must be preserved: pointwise values, maximum error, integrated error, derivatives, or an energy norm. Statements about “the limit of approximations” are therefore incomplete unless the underlying mode of convergence is specified.

Random sequences and probabilistic convergence

A sequence of random variables (Xn)(X_n) consists of functions on a common probability space. For each outcome ω\omega, the values X1(ω),X2(ω),…X_1(\omega),X_2(\omega),\ldots form an ordinary sequence called a sample path or realization. A random sequence is thus simultaneously a family of probability distributions and a collection of outcome-dependent paths. The dependence among terms is part of its specification; marginal distributions alone do not determine the joint behavior.

Several notions of convergence are useful. Almost-sure convergence means that the sample paths converge outside an event of probability zero:

P{ω:Xn(ω)→X(ω)}=1.P\{\omega:X_n(\omega)\to X(\omega)\}=1.

Convergence in probability means

P(∣Xn−X∣>ε)→0for every ε>0.P(|X_n-X|>\varepsilon)\to0 \quad\text{for every }\varepsilon>0.

For p≥1p\ge1, convergence in LpL^p means

E∣Xn−X∣p→0.\mathbb E|X_n-X|^p\to0.

Convergence in distribution requires convergence of cumulative distribution functions at every continuity point of the limiting distribution.

Almost-sure convergence implies convergence in probability, which implies convergence in distribution. Convergence in LpL^p also implies convergence in probability. The reverse implications generally fail. To see why convergence in probability need not imply convergence in mean, take an event AnA_n with probability 1/n1/n and define Xn=n 1AnX_n=n\,1_{A_n}. Then P(∣Xn∣>ε)=1/nP(|X_n|>\varepsilon)=1/n for sufficiently large nn, but E∣Xn∣=1\mathbb E|X_n|=1. Rare values with increasing magnitude can remain important to expectation while becoming unlikely.

For independent, identically distributed observations with finite expected absolute value, the strong law of large numbers states that the sample averages converge almost surely to their common mean:

1n∑k=1nXk→E[X1].\frac1n\sum_{k=1}^{n}X_k\to\mathbb E[X_1].

If the common variance is finite and positive, the central limit theorem instead describes the limiting distribution of normalized fluctuations:

∑k=1nXk−nμσn →d N(0,1).\frac{\sum_{k=1}^{n}X_k-n\mu}{\sigma\sqrt n} \ \xrightarrow{d}\ N(0,1).

These results address different questions: convergence of averages to a constant and the distribution of their shrinking errors.

A discrete-time stochastic process is a random sequence with an interpretation as successive states. Examples include random walks and Markov chains. A time series is a sequence of observations indexed by time, often modeled using such a process. Independence is not inherent in either concept: correlations and other temporal dependencies may dominate the behavior. Stationarity concerns invariance of distributions under time shifts, not equality of successive values or convergence of individual sample paths.

Iterative computation and discrete models

Many computational methods define a sequence recursively and use its limit as the desired answer. A fixed-point iteration has the form

xn+1=T(xn).x_{n+1}=T(x_n).

If TT is a contraction on a nonempty complete metric space, the Banach fixed-point theorem guarantees a unique fixed point x∗x^\ast and convergence to it from every starting point. If the contraction factor is q<1q<1, then

d(xn,x∗)≤qnd(x0,x∗).d(x_n,x^\ast)\le q^n d(x_0,x^\ast).

The sequence is therefore both a mathematical construction and an executable approximation procedure.

The speed of convergence determines computational usefulness. If en=∥xn−x∗∥e_n=\|x_n-x^\ast\|, linear convergence typically means en+1≤qene_{n+1}\le q e_n eventually for some q<1q<1. Quadratic convergence means en+1≤Cen2e_{n+1}\le C e_n^2 locally. Under suitable smoothness and nondegeneracy assumptions, Newton’s method

xn+1=xn−f(xn)f′(xn)x_{n+1}=x_n-\frac{f(x_n)}{f'(x_n)}

is locally quadratically convergent near a simple root. These assumptions are essential: an unsuitable starting point, a zero derivative, or an ill-behaved function can lead to failure.

Stopping rules usually cannot inspect the unknown error directly. They instead use quantities such as successive differences, residuals, or certified bounds. In a contraction iteration, for example,

d(xn,x∗)≤q1−q d(xn,xn−1)d(x_n,x^\ast) \le\frac{q}{1-q}\,d(x_n,x_{n-1})

provides an error estimate from computable terms. In more general methods, a small change between consecutive iterates does not necessarily imply proximity to a solution. Slow progress, poor scaling, or finite-precision effects can make successive terms nearly equal far from the target.

Discrete-time models also describe genuine dynamics rather than approximations to a static answer. A population model may use xn+1=rxn(1−xn)x_{n+1}=rx_n(1-x_n), while a financial balance may obey an affine recurrence incorporating interest and payments. Linear state evolution xn+1=Axnx_{n+1}=Ax_n is controlled by the eigenstructure of AA; eigenvalues with modulus below one support decay, while other spectral features can produce growth or oscillation. Nonlinear recurrences can exhibit periodic behavior and chaos.

In numerical analysis, computed terms differ from ideal terms because of floating-point arithmetic, truncation, and imperfect input data. Mathematical convergence of the ideal sequence is not sufficient to establish accuracy of its implementation. Stability concerns how perturbations propagate, while convergence analysis concerns the limiting effect of refinement or iteration. The distinction is especially important in discretizations of differential equations, where increasingly fine meshes generate sequences of approximate solutions whose errors must be controlled in a specified norm.

Historical development and the limits of sequential methods

Ancient mathematics employed sequences without the modern function-based definition. The method of exhaustion, used by Eudoxus and developed in works of Archimedes, compared geometric magnitudes through increasingly refined figures and quantitative remainder bounds. Archimedes’ Measurement of a Circle bounded the ratio of circumference to diameter using inscribed and circumscribed polygons, reaching 96-sided polygons. These arguments anticipated aspects of limiting reasoning while retaining a geometric framework.

Recursively defined numerical patterns also have a long history. The rabbit-population problem in Fibonacci’s Liber Abaci, first issued in 1202, presented the sequence now bearing his name in European mathematical literature. Related counting recurrences had appeared earlier in Indian work on poetic meters. During the seventeenth and eighteenth centuries, infinite expansions and successive approximations became central to calculus, astronomy, and the study of functions. Their use often preceded a fully explicit separation of convergence from formal algebraic manipulation.

The nineteenth century brought systematic definitions and stronger attention to hypotheses. In his 1821 Cours d’analyse, Augustin-Louis Cauchy treated limits, convergence, and criteria involving the closeness of late terms. Later developments associated with Karl Weierstrass helped establish the epsilon-based rigor of analysis. Cauchy sequences became important in constructing the real numbers from rational approximations: two rational Cauchy sequences represent the same real number when their difference tends to zero.

Sequences also played a foundational role in the work of Georg Cantor. His investigations of sets of points and limits were connected with the uniqueness of trigonometric expansions. The diagonal argument shows that the set of all infinite binary sequences is uncountable: from any purported list, a new sequence can be constructed by changing its nnth term relative to the nnth sequence on the list. Thus an individual sequence has countably many positions, but the collection of possible sequences can be uncountable.

In modern topology, sequences remain indispensable but do not capture every kind of convergence. In a first-countable space, a point belongs to the closure of a set precisely when some sequence from the set converges to it. This can fail in more general topological spaces. For example, in the ordinal interval [0,ω1][0,\omega_1] with its order topology, the first uncountable ordinal ω1\omega_1 lies in the closure of the smaller ordinals, but no sequence of smaller ordinals converges to it: every such sequence is bounded below ω1\omega_1. A net, indexed by a directed set rather than just the natural numbers, supplies the more general convergence language needed in such spaces.