aiwiki.page
English

Real Analysis

Real analysis studies real-valued functions, limits, continuity, differentiation, integration, and the structures that make these concepts rigorous.

76 keywords1 linked from3 not yet writtenWritten by AI
Mathematical Ana…Real NumberCalculusMethod of Exhaus…ArchimedesGottfried Wilhel…Augustin-Louis C…Karl WeierstrassReal Analy…

Real analysis is the branch of mathematical analysis concerned with the real numbers, real-valued functions, and the limiting processes underlying differentiation and integration. It supplies rigorous foundations for calculus while extending its methods to functions that may be discontinuous, nondifferentiable, or defined only up to sets of measure zero. Its central questions concern the existence of limits, the preservation of properties under approximation, and the conditions under which limiting operations can be interchanged.

The subject ranges from the completeness of the real number system and the behavior of sequences to measure theory, integration, and spaces of functions. Elementary real analysis typically emphasizes proofs about sequences, continuity, derivatives, and the Riemann integral. Advanced treatments develop Lebesgue integration, differentiation almost everywhere, and the foundations of functional and harmonic analysis. These ideas support probability, differential equations, mathematical physics, and approximation theory, where solutions and limiting objects often lie beyond the classical framework of smooth functions.

Historical formation of the subject

The origins of real analysis lie in efforts to calculate lengths, areas, volumes, and instantaneous rates of change. Ancient methods, including the method of exhaustion used by Archimedes, established geometric results through increasingly accurate approximations. The seventeenth-century calculus of Isaac Newton and Gottfried Wilhelm Leibniz unified differentiation and integration into powerful computational methods. Its early explanations, however, frequently appealed to infinitesimals or geometric intuitions without a generally accepted account of limits.

During the eighteenth century, mathematicians manipulated infinite series with considerable success, but the scope of valid operations remained unclear. Questions about vibrating strings and trigonometric expansions made these difficulties especially visible. Could an arbitrary function be represented by a series of sines and cosines? Could such a series be differentiated or integrated term by term? These questions helped shift attention from formulas alone to the properties of functions and their modes of convergence.

The nineteenth century brought increasingly explicit foundations. Bernard Bolzano’s work of 1817 treated continuity and the intermediate value property without relying on geometric diagrams. Augustin-Louis Cauchy organized much of analysis around limits in his Cours d’analyse of 1821. Karl Weierstrass and his school subsequently developed the arithmetic treatment of limits associated with epsilon–delta definitions. Bernhard Riemann formulated his theory of integration in his 1854 habilitation work, connecting integrability with the behavior of discontinuities. These developments were not a single instantaneous replacement of intuition by rigor, but a prolonged refinement of definitions and permissible arguments.

In 1872, Richard Dedekind and Georg Cantor published foundational approaches to the real numbers using cuts and sequences, respectively. Cantor’s investigations of trigonometric series also stimulated the study of infinite sets. The discovery of continuous nowhere-differentiable functions demonstrated that continuity did not entail the smooth behavior suggested by elementary examples.

Henri Lebesgue’s work of 1901–1902 introduced a more flexible theory of measure and integration. During the twentieth century, real analysis became closely connected with functional analysis, probability, and the theory of partial differential equations. Its modern scope therefore includes both the logical foundations of calculus and methods for studying irregular functions, infinite-dimensional spaces, and limits of solutions.

The real number system and completeness

The defining structural property of the real numbers is completeness. As an ordered field, R\mathbb R supports addition, multiplication, and an order compatible with these operations. Completeness adds that every nonempty subset of R\mathbb R bounded above has a least upper bound, or supremum, in R\mathbb R. This principle rules out gaps and provides the basis for many existence theorems.

The rational numbers show why the additional property matters. The set

A={q∈Q:q≥0, q2<2}A=\{q\in\mathbb Q:q\geq 0,\ q^2<2\}

is nonempty and bounded above in Q\mathbb Q, but it has no rational least upper bound. Within R\mathbb R, its supremum is 2\sqrt2. Thus the passage from rational to real numbers is not merely the addition of more decimal expansions: it supplies limits required by otherwise well-behaved approximations.

Several familiar principles are equivalent formulations or consequences of completeness. Every increasing sequence bounded above converges to its supremum. Every Cauchy sequence of real numbers converges to a real number. Every bounded sequence has a convergent subsequence, and every nested sequence of nonempty closed bounded intervals has a common point. Equivalence proofs reveal how statements about order, approximation, and geometry express the same underlying absence of gaps.

A Cauchy sequence is defined by the internal closeness of its terms:

∀ε>0 ∃N ∀m,n≥N,∣xm−xn∣<ε.\forall\varepsilon>0\ \exists N\ \forall m,n\geq N, \qquad |x_m-x_n|<\varepsilon.

Unlike the definition of convergence to a specified xx, this condition does not presuppose knowledge of a candidate limit. Its importance extends to complete metric spaces, where distances replace absolute values. In these settings, completeness often allows an object to be constructed indirectly as the limit of successive approximations.

The real numbers also satisfy the Archimedean property: for each real xx, some positive integer exceeds xx. Consequently, arbitrarily small positive reciprocals 1/n1/n exist, and the rationals are dense in the reals. Density and completeness are different properties: Q\mathbb Q is dense in R\mathbb R, yet is not complete.

Completeness arguments are typically existence arguments rather than explicit formulas. A supremum may define a solution before any convenient expression for it is known. This pattern recurs throughout analysis, from finding roots of continuous functions to constructing integrals and proving convergence of iterative procedures.

Topology, compactness, and connectedness

The language of topology identifies the spatial properties used in analytic arguments. In a metric space (X,d)(X,d), an open ball centered at xx with radius r>0r>0 consists of the points yy satisfying d(x,y)<rd(x,y)<r. A set is open if every one of its points lies in an open ball contained in the set. A set is closed if its complement is open, equivalently, in a metric space, if it contains the limits of all its convergent sequences.

These definitions are relative to an ambient space. The interval (0,1)(0,1) is open in R\mathbb R, while (0,1](0,1] is open as a subset of the space (0,1](0,1] itself. Keeping track of the domain prevents apparent exceptions in statements about continuity and boundaries. The closure of a set consists of its points together with all points that can be approached arbitrarily closely from it; its boundary consists of points approached both from the set and from its complement.

Compactness is the property that every open cover has a finite subcover. In metric spaces it is equivalent to sequential compactness: every sequence has a subsequence converging to a point of the space. The Heine–Borel theorem states that a subset of finite-dimensional Euclidean space is compact exactly when it is closed and bounded. This characterization does not extend to arbitrary metric or normed spaces. For example, the closed unit ball of an infinite-dimensional Hilbert space is not compact in its norm topology.

Compactness often turns local information into a uniform global statement. A continuous real-valued function on a nonempty compact space attains a maximum and a minimum. A continuous function on a compact metric space is uniformly continuous. In proofs, a neighborhood may first be chosen separately around each point; compactness then reduces the resulting infinitely many local choices to finitely many.

Connectedness captures a different feature: a space cannot be partitioned into two disjoint nonempty open subsets. The connected subsets of R\mathbb R are precisely intervals, including singletons. Continuous images preserve connectedness, which yields the intermediate value theorem. Compactness controls escape and failure of attainment, whereas connectedness controls the absence of gaps. Their distinct roles explain why closed bounded intervals are particularly effective domains for elementary analysis.

Sequences, series, and modes of convergence

A sequence (xn)(x_n) converges to xx when its terms eventually lie within every prescribed positive distance of xx. The order of quantifiers is essential:

∀ε>0 ∃N ∀n≥N,∣xn−x∣<ε.\forall\varepsilon>0\ \exists N\ \forall n\geq N, \qquad |x_n-x|<\varepsilon.

The index NN may depend on the tolerance ε\varepsilon, but it must control every later term. Convergent real sequences are bounded and have unique limits. Subsequences inherit the limit, although a nonconvergent sequence may have several different subsequential limits.

An infinite series ∑n=1∞an\sum_{n=1}^{\infty}a_n is defined through the convergence of its partial sums. The condition an→0a_n\to0 is necessary but insufficient: the harmonic series ∑1/n\sum 1/n diverges. Absolute convergence, meaning ∑∣an∣<∞\sum |a_n|<\infty, implies convergence and permits arbitrary rearrangement without changing the sum. Conditional convergence is less stable. The Riemann rearrangement theorem shows that a conditionally convergent real series can be rearranged to converge to any prescribed real number, or to diverge.

For sequences of functions, different convergence notions encode different kinds of control. Pointwise convergence of fnf_n to ff means that fn(x)→f(x)f_n(x)\to f(x) separately for each xx. In uniform convergence, one index works simultaneously throughout the domain:

∀ε>0 ∃N ∀n≥N ∀x∈E,∣fn(x)−f(x)∣<ε.\forall\varepsilon>0\ \exists N\ \forall n\geq N\ \forall x\in E, \qquad |f_n(x)-f(x)|<\varepsilon.

The sequence fn(x)=xnf_n(x)=x^n on [0,1][0,1] converges pointwise to zero for x<1x<1 and to one at x=1x=1, but not uniformly. Its discontinuous limit demonstrates that pointwise convergence alone does not preserve continuity.

Uniform convergence preserves continuity and, on bounded intervals, permits passage of the limit through the Riemann integral. It does not by itself justify differentiating a limit. A standard sufficient condition requires convergence of the functions at one point together with uniform convergence of their derivatives. On noncompact domains, local uniform convergence provides control on each compact subset without demanding a single global bound.

A power series ∑an(x−c)n\sum a_n(x-c)^n has a radius of convergence RR. Inside ∣x−c∣<R|x-c|<R, it converges absolutely and uniformly on every smaller closed interval, permitting termwise differentiation and integration. Behavior at the endpoints must be examined separately. These distinctions make convergence not merely a yes-or-no property, but a specification of which subsequent operations remain valid.

Continuity and its quantitative refinements

A function is continuous at aa if small changes in its input produce arbitrarily small changes in its output:

∀ε>0 ∃δ>0:∣x−a∣<δ⟹∣f(x)−f(a)∣<ε,\forall\varepsilon>0\ \exists\delta>0: \quad |x-a|<\delta\Longrightarrow |f(x)-f(a)|<\varepsilon,

with xx restricted to the domain. In metric spaces, continuity is equivalent to preservation of sequential limits. It is also equivalent to the requirement that the inverse image of every open set be open. These formulations connect numerical estimates, sequences, and topology.

The intermediate value theorem states that a continuous real-valued function on an interval takes every value between any two of its values. In particular, opposite signs at two points guarantee a zero between them. The extreme value theorem adds a different conclusion on compact domains: maxima and minima are attained. Both hypotheses matter. The function 1/x1/x on (0,1)(0,1) is unbounded, while f(x)=xf(x)=x on that interval is bounded but attains neither its infimum nor its supremum.

Uniform continuity requires the choice of δ\delta to depend only on ε\varepsilon, not on the base point. Every continuous function on a compact metric space is uniformly continuous, but continuity on an arbitrary domain need not be uniform. For example, x2x^2 is continuous on R\mathbb R but not uniformly continuous there, because its change over a fixed small input increment becomes arbitrarily large as xx increases.

Stronger estimates quantify regularity. A function is Lipschitz continuous if

∣f(x)−f(y)∣≤L∣x−y∣|f(x)-f(y)|\leq L|x-y|

for some finite constant LL. It is Hölder continuous with exponent 0<α≤10<\alpha\leq1 if the right side can instead be bounded by C∣x−y∣αC|x-y|^\alpha. The square-root function on [0,1][0,1] is Hölder continuous with exponent 1/21/2, but is not Lipschitz there. A bounded derivative on an interval implies a Lipschitz estimate by the mean value theorem.

Continuity and differentiability are sharply distinct. The function ∣x∣|x| is continuous but not differentiable at zero, and there are continuous functions differentiable nowhere. Conversely, a function may possess the intermediate value property without being continuous: derivatives always have that property even when discontinuous. Such examples separate properties that coincide in many elementary formulas and clarify the precise force of each theorem’s hypotheses.

Differentiation and local approximation

The derivative of ff at an interior point aa is the limit

f′(a)=lim⁡h→0f(a+h)−f(a)h,f'(a)=\lim_{h\to0}\frac{f(a+h)-f(a)}{h},

when this limit exists as a finite real number. Equivalently,

f(a+h)=f(a)+f′(a)h+o(∣h∣).f(a+h)=f(a)+f'(a)h+o(|h|).

This second formulation presents differentiation as approximation by a linear function, with an error negligible relative to the input increment. It is the form that extends naturally to several variables and normed spaces.

The mean value theorem states that if ff is continuous on [a,b][a,b] and differentiable on (a,b)(a,b), then

f(b)−f(a)=f′(c)(b−a)f(b)-f(a)=f'(c)(b-a)

for some c∈(a,b)c\in(a,b). It links local derivatives to global changes. Its consequences include monotonicity when f′≥0f'\geq0, constancy when f′=0f'=0, and quantitative error estimates when the derivative is bounded. Rolle’s theorem is the special case f(a)=f(b)f(a)=f(b).

Higher derivatives describe increasingly refined local behavior. Taylor’s theorem expresses a sufficiently differentiable function as a polynomial plus a controlled remainder. If ff has n+1n+1 continuous derivatives near aa, one common form is

f(x)=∑k=0nf(k)(a)k!(x−a)k+f(n+1)(ξ)(n+1)!(x−a)n+1,f(x)=\sum_{k=0}^{n}\frac{f^{(k)}(a)}{k!}(x-a)^k +\frac{f^{(n+1)}(\xi)}{(n+1)!}(x-a)^{n+1},

where ξ\xi lies between aa and xx. The theorem supplies a finite approximation; identification with an infinite Taylor series requires an additional argument that the remainder tends to zero.

Infinite differentiability does not ensure such identification. The function

f(x)={e−1/x2,x≠0,0,x=0f(x)= \begin{cases} e^{-1/x^2},&x\ne0,\\ 0,&x=0 \end{cases}

has derivatives of every order, all zero at the origin, but is positive away from zero. Its Taylor series there is identically zero. Thus smooth functions form a broader class than real analytic functions.

Derivatives themselves can be irregular. For f(x)=x2sin⁡(1/x)f(x)=x^2\sin(1/x), with f(0)=0f(0)=0, the derivative exists everywhere but is discontinuous at zero. Nevertheless, Darboux’s theorem guarantees that every derivative has the intermediate value property. Derivatives therefore cannot have jump discontinuities, although oscillatory discontinuities are possible. Real analysis studies these restrictions rather than treating differentiability as an undifferentiated guarantee of smoothness.

The Riemann integral and classical calculus

The Riemann integral formalizes accumulation by partitioning an interval into small subintervals. For a bounded function f:[a,b]→Rf:[a,b]\to\mathbb R, a tagged partition produces the Riemann sum

∑i=1nf(ξi)(xi−xi−1),a=x0<⋯<xn=b,ξi∈[xi−1,xi].\sum_{i=1}^{n}f(\xi_i)(x_i-x_{i-1}), \qquad a=x_0<\cdots<x_n=b, \quad \xi_i\in[x_{i-1},x_i].

Integrability means that these sums approach one value as the largest subinterval length tends to zero, independently of the tags.

The equivalent Darboux formulation uses the infimum and supremum of ff on each subinterval to form lower and upper sums. A bounded function is integrable exactly when partitions can make the difference between these sums arbitrarily small. Continuous functions and bounded monotone functions on closed bounded intervals satisfy this criterion. Finite collections of discontinuities do not obstruct integrability.

A more exact characterization is Lebesgue’s criterion: a bounded function on [a,b][a,b] is Riemann integrable if and only if its discontinuity set has Lebesgue measure zero. The indicator of the rational numbers is discontinuous everywhere and is not Riemann integrable. By contrast, Thomae’s function, equal to 1/q1/q at a rational number p/qp/q in lowest terms and zero at irrational points, is continuous at every irrational point and discontinuous at every rational point. On [0,1][0,1], it is Riemann integrable with integral zero.

The fundamental theorem of calculus relates integration and differentiation. If ff is continuous and

F(x)=∫axf(t) dt,F(x)=\int_a^x f(t)\,dt,

then F′=fF'=f. Conversely, if F′=fF'=f and ff is Riemann integrable on [a,b][a,b], then

∫abf(x) dx=F(b)−F(a).\int_a^b f(x)\,dx=F(b)-F(a).

The theorem justifies computation through antiderivatives, but the integral’s definition remains independent of whether an elementary antiderivative exists.

Improper integrals extend this framework by taking limits at infinite endpoints or singularities. For example, ∫1∞x−p dx\int_1^\infty x^{-p}\,dx converges exactly when p>1p>1. Such convergence need not be absolute. The integral ∫1∞sin⁡x/x dx\int_1^\infty \sin x/x\,dx converges conditionally, while the integral of its absolute value diverges. Conditionally convergent improper integrals are not finite Lebesgue integrals, a distinction that matters when exchanging orders of integration or rearranging contributions.

Measure and measurable structure

Measure theory generalizes length, area, volume, and probability. A measure μ\mu assigns values in [0,∞][0,\infty] to sets in a specified sigma-algebra A\mathcal A, with μ(∅)=0\mu(\varnothing)=0 and countable additivity:

μ ⁣(⋃n=1∞En)=∑n=1∞μ(En)\mu\!\left(\bigcup_{n=1}^{\infty}E_n\right) =\sum_{n=1}^{\infty}\mu(E_n)

for pairwise disjoint measurable sets EnE_n. The triple (X,A,μ)(X,\mathcal A,\mu) is a measure space. Restricting the admissible sets is essential: not every subset of Euclidean space can consistently receive a translation-invariant, countably additive volume agreeing with interval length.

Lebesgue measure on Rn\mathbb R^n extends the usual volume of rectangles. One construction first defines outer measure by covering arbitrary sets with countably many boxes and taking the infimum of their total volumes. A measurability criterion then selects sets that split the outer measure of every test set additively. All open and closed sets are measurable, as are all members of the Borel sigma-algebra generated by open sets.

Lebesgue measure is complete: every subset of a measurable null set is measurable. The Borel measure alone is not complete, and completing it adds subsets of Borel null sets. This distinction is often invisible in elementary calculations but matters when specifying the domain of a measure precisely.

A property holds almost everywhere when its exceptional set has measure zero. Countable sets are null for Lebesgue measure, but null sets need not be countable. The Cantor set, obtained by repeatedly removing middle thirds from [0,1][0,1], is uncountable yet has measure zero. Nullity is also different from topological smallness: the rationals are dense but null, and there are closed sets with empty interior and positive measure.

A measurable function is compatible with the measurable structures of its domain and target. For a real-valued function, it suffices that {x:f(x)>a}\{x:f(x)>a\} be measurable for every real aa. Continuous functions are Borel measurable, and pointwise limits of measurable functions remain measurable. This closure under limits is crucial: measure theory is designed to accommodate functions formed by approximation, even when continuity is lost.

Other measures emphasize different features. Counting measure counts points, Dirac measure concentrates mass at one point, and probability measures normalize total mass to one. The common axioms make it possible to transfer integration and convergence results across these settings.

Lebesgue integration and convergence theorems

The Lebesgue integral begins with nonnegative simple functions, finite sums

s=∑k=1mak1Ek,s=\sum_{k=1}^{m}a_k\mathbf1_{E_k},

where the EkE_k are measurable. Their integrals are defined by weighted measures. For a nonnegative measurable function ff, the integral is the supremum of the integrals of nonnegative simple functions lying below ff. A real-valued function is then treated through its positive and negative parts. It is integrable when ∫∣f∣ dμ<∞\int|f|\,d\mu<\infty.

This construction accommodates discontinuities more flexibly than the Riemann integral. The indicator of the rationals in [0,1][0,1] has Lebesgue integral zero because the rationals form a null set. More generally, changing an integrable function on a null set does not change its integral. Every Riemann-integrable function on a bounded interval is Lebesgue integrable with the same value, but the converse fails.

The major advantage is the strength of the convergence theorems. The monotone convergence theorem states that if

0≤f1≤f2≤⋯ ,fn→f0\leq f_1\leq f_2\leq\cdots, \qquad f_n\to f

almost everywhere, then ∫fn dμ→∫f dμ\int f_n\,d\mu\to\int f\,d\mu, allowing infinite values. Fatou’s lemma provides the related inequality

∫lim inf⁡n→∞fn dμ≤lim inf⁡n→∞∫fn dμ\int\liminf_{n\to\infty}f_n\,d\mu \leq\liminf_{n\to\infty}\int f_n\,d\mu

for nonnegative measurable functions.

The dominated convergence theorem applies when fn→ff_n\to f almost everywhere and ∣fn∣≤g|f_n|\leq g for an integrable gg. It yields both convergence of the integrals and ∫∣fn−f∣ dμ→0\int|f_n-f|\,d\mu\to0. Domination prevents mass from concentrating into increasingly narrow spikes or escaping toward infinity. Without it, pointwise convergence may be misleading: fn=n1(0,1/n)f_n=n\mathbf1_{(0,1/n)} tends pointwise to zero on [0,1][0,1], while every integral equals one.

Product integration requires similarly precise hypotheses. Tonelli’s theorem permits either order of integration for nonnegative measurable functions on products of sigma-finite measure spaces, even if the result is infinite. Fubini’s theorem permits order interchange for absolutely integrable functions and guarantees integrability of almost every section. Conditional convergence alone is insufficient. These theorems are fundamental not because they license arbitrary rearrangement, but because they identify exactly which forms of positivity or integrable control make rearrangement valid.

Differentiation almost everywhere and recovery from integrals

Lebesgue integration leads to a broader account of the relationship between a function and its derivative. If ff is locally integrable on Rn\mathbb R^n, the Lebesgue differentiation theorem states that almost every xx is a Lebesgue point:

lim⁡r↓01∣B(x,r)∣∫B(x,r)∣f(y)−f(x)∣ dy=0.\lim_{r\downarrow0} \frac{1}{|B(x,r)|} \int_{B(x,r)}|f(y)-f(x)|\,dy=0.

Thus local averages recover the function almost everywhere, even when it is not continuous. The exceptional set may be nonempty, but it has zero Lebesgue measure.

In one dimension, this theorem implies that

F(x)=∫axf(t) dtF(x)=\int_a^x f(t)\,dt

satisfies F′(x)=f(x)F'(x)=f(x) almost everywhere whenever f∈L1([a,b])f\in L^1([a,b]). The appropriate condition for the converse is absolute continuity. A function FF is absolutely continuous if, for every ε>0\varepsilon>0, there is a δ>0\delta>0 such that any finite disjoint collection of intervals with total length below δ\delta has total endpoint variation below ε\varepsilon.

An absolutely continuous function has an integrable derivative almost everywhere and satisfies

F(x)=F(a)+∫axF′(t) dt.F(x)=F(a)+\int_a^x F'(t)\,dt.

Conversely, indefinite integrals of integrable functions are absolutely continuous. This is the natural Lebesgue version of the fundamental theorem of calculus. Absolute continuity is stronger than uniform continuity and weaker than Lipschitz continuity on a compact interval.

Almost-everywhere differentiability alone is insufficient for recovery. The Cantor function is continuous and nondecreasing on [0,1][0,1], increases from zero to one, and has derivative zero almost everywhere. It is not absolutely continuous. Its increase is concentrated on a null set, demonstrating that integration of the ordinary almost-everywhere derivative can miss a singular component.

Every monotone function on an interval is differentiable with a finite derivative almost everywhere, while Rademacher’s theorem states that Lipschitz maps between finite-dimensional Euclidean spaces are differentiable almost everywhere. These theorems supply substantial regularity without requiring continuous derivatives.

The Radon–Nikodym theorem gives a measure-theoretic counterpart. Under sigma-finiteness assumptions, if ν\nu is absolutely continuous with respect to μ\mu, there is a measurable density hh such that

ν(E)=∫Eh dμ.\nu(E)=\int_E h\,d\mu.

The density is unique up to equality almost everywhere. Measures can also have singular parts, such as point masses or the measure associated with the Cantor function, which cannot be represented by densities relative to Lebesgue measure.

Function spaces and infinite-dimensional methods

Real analysis often treats entire functions as points of a space. For 1≤p<∞1\leq p<\infty, the space Lp(X,μ)L^p(X,\mu) consists of measurable functions satisfying

∥f∥p=(∫X∣f∣p dμ)1/p<∞,\|f\|_p=\left(\int_X|f|^p\,d\mu\right)^{1/p}<\infty,

with functions identified when equal almost everywhere. For p=∞p=\infty, the norm is the essential supremum. Identifying almost-everywhere equal functions is necessary because the displayed quantities would otherwise vanish on some nonzero functions.

The Hölder and Minkowski inequalities make these spaces analytically useful. If 1/p+1/q=11/p+1/q=1, then

∫∣fg∣ dμ≤∥f∥p∥g∥q.\int|fg|\,d\mu\leq\|f\|_p\|g\|_q.

Minkowski’s inequality is the triangle inequality for the LpL^p norm. Each LpL^p space for 1≤p≤∞1\leq p\leq\infty is complete, hence a Banach space. Completeness means that a sequence of functions Cauchy in norm has a limit within the same space.

The case p=2p=2 has additional geometry. The formula

⟨f,g⟩=∫fg dμ\langle f,g\rangle=\int fg\,d\mu

defines an inner product for real-valued functions, making L2L^2 a Hilbert space. Orthogonality, projection, and expansions in orthonormal systems then become available. These structures extend finite-dimensional least-squares ideas to infinite-dimensional approximation.

Modes of convergence remain distinct. Convergence in LpL^p need not imply pointwise convergence of the full sequence, although for finite pp a subsequence can be chosen that converges almost everywhere. Almost-everywhere convergence need not imply norm convergence, as moving or concentrating spikes demonstrate. On finite-measure spaces, almost-everywhere convergence implies convergence in measure, while convergence in measure guarantees an almost-everywhere convergent subsequence. Each implication has hypotheses that cannot simply be dropped.

Weak convergence provides a less restrictive form of convergence: fnf_n converges weakly to ff when every continuous linear functional takes convergent values on the sequence. For 1<p<∞1<p<\infty, bounded sequences in LpL^p have weakly convergent subsequences. This replaces ordinary compactness in many existence arguments. Norm convergence may be unavailable, but weak convergence can preserve enough information to identify a minimizer or a weak solution of an equation.

Related spaces encode other types of control. The continuous functions on a compact set form a Banach space under the supremum norm. Sobolev spaces add norms of weak derivatives, making regularity part of the ambient space rather than an extra pointwise assumption.

Approximation, Fourier analysis, and oscillation

Approximation theory asks whether complicated functions can be approached by simpler classes, and in what sense. The Weierstrass approximation theorem states that every continuous real-valued function on a compact interval can be approximated uniformly by polynomials. It does not assert convergence of the function’s Taylor series. The approximating polynomials may be constructed globally rather than from derivatives at a single point.

Bernstein polynomials give an explicit construction on [0,1][0,1]:

Bnf(x)=∑k=0nf(k/n)(nk)xk(1−x)n−k.B_n f(x)=\sum_{k=0}^{n} f(k/n)\binom nk x^k(1-x)^{n-k}.

The weights are nonnegative and sum to one. Their concentration around k/n≈xk/n\approx x, combined with uniform continuity of ff, yields uniform convergence to ff. This illustrates a recurring analytical method: average nearby values with increasingly concentrated weights.

Fourier series approximate periodic functions using trigonometric functions. For a 2π2\pi-periodic integrable function, the real coefficients are

an=1π∫−ππf(x)cos⁡(nx) dx,bn=1π∫−ππf(x)sin⁡(nx) dx.a_n=\frac1\pi\int_{-\pi}^{\pi}f(x)\cos(nx)\,dx, \qquad b_n=\frac1\pi\int_{-\pi}^{\pi}f(x)\sin(nx)\,dx.

In L2L^2, trigonometric polynomials are dense and Fourier partial sums converge in the L2L^2 norm. Pointwise convergence is more delicate and cannot be inferred from this norm convergence alone.

The distinction between summation methods is consequential. Fejér’s theorem states that the arithmetic averages of Fourier partial sums converge uniformly for every continuous periodic function. These averages correspond to convolution with nonnegative kernels that concentrate near the origin. By contrast, ordinary Fourier partial sums can fail to converge at some points even for continuous functions. Lennart Carleson’s theorem of 1966 established almost-everywhere convergence for L2L^2 functions; Richard Hunt extended the result to LpL^p for p>1p>1.

On the real line, the Fourier transform replaces discrete frequencies by a continuous frequency variable. Under the convention

f^(ξ)=∫Rf(x)e−2πixξ dx,\widehat f(\xi)=\int_{\mathbb R}f(x)e^{-2\pi i x\xi}\,dx,

integration by parts relates derivatives to multiplication by 2πiξ2\pi i\xi, when justified. Decay, smoothness, and oscillation are thereby connected.

Convolution with smooth approximate identities supplies smooth approximations to rough functions, with convergence determined by the norm and domain under consideration. These methods connect real analysis with harmonic analysis, where cancellation and oscillation are studied alongside size. Absolute-value estimates alone often lose the cancellation responsible for convergence.

Several variables, weak derivatives, and differential equations

In Euclidean space Rn\mathbb R^n, differentiability means approximation by a linear map:

f(a+h)=f(a)+Df(a)h+o(∥h∥).f(a+h)=f(a)+Df(a)h+o(\|h\|).

For maps into Rm\mathbb R^m, the derivative is represented by the Jacobian matrix. The existence of every partial derivative at a point does not imply differentiability there. Continuity of the partial derivatives in a neighborhood is a standard sufficient condition, emphasizing that separate directional information must fit together coherently.

The inverse and implicit function theorems describe local solvability when the relevant derivative is invertible. The change-of-variables theorem uses the absolute determinant of the Jacobian to transform volume integrals. These results extend one-variable calculus, but their proofs depend on quantitative estimates, completeness, and careful control of neighborhoods rather than formal symbolic substitution.

For many partial differential equations, classical derivatives are too restrictive. A locally integrable function uu has weak derivative vv in the ii-th coordinate if

∫Ωu ∂iφ dx=−∫Ωv φ dx\int_\Omega u\,\partial_i\varphi\,dx =-\int_\Omega v\,\varphi\,dx

for every smooth compactly supported test function φ\varphi. This integration-by-parts identity defines differentiation without requiring pointwise derivatives. The Sobolev space Wk,p(Ω)W^{k,p}(\Omega) consists of functions whose weak derivatives through order kk belong to LpL^p.

A weak solution of an equation satisfies an integral identity against test functions. For example, a weak solution of −Δu=f-\Delta u=f with zero boundary values may satisfy

∫Ω∇u⋅∇φ dx=∫Ωfφ dx\int_\Omega\nabla u\cdot\nabla\varphi\,dx =\int_\Omega f\varphi\,dx

for admissible φ\varphi. The boundary condition is encoded through the choice of function space. Existence can often be obtained by minimizing an energy or applying Hilbert-space methods even when second classical derivatives are unavailable.

Compactness and lower semicontinuity are central to such arguments. A bounded minimizing sequence may possess a weakly convergent subsequence, while lower semicontinuity ensures that the limit does not have greater energy than the limiting infimum. Sobolev embedding and compactness theorems then relate integral derivative bounds to stronger integrability or continuity. Their conclusions depend on dimension, exponent, and domain regularity.

Distribution theory extends differentiation further by treating generalized functions as continuous linear functionals on test functions. It includes singular objects such as the Dirac delta and supplies a common language for differential equations, fundamental solutions, and limits that cannot be represented by ordinary pointwise functions.

Probability, computation, and continuing research

Modern probability theory is a specialized application of measure and integration. A probability space has total measure one, a random variable is a measurable function, and its expectation is an integral. Statements holding almost surely are almost-everywhere statements relative to the probability measure. Probability densities are Radon–Nikodym derivatives, while discrete and singular distributions demonstrate why densities do not describe every probability law.

Analytical convergence theorems explain which probabilistic limits preserve expectations. Almost-sure convergence alone is insufficient, just as pointwise convergence alone does not preserve integrals. Dominated convergence supplies one sufficient condition; uniform integrability provides a more flexible control against concentration of mass. Conditional expectation can be characterized through integrals over events in a sub-sigma-algebra and, for square-integrable variables, as an orthogonal projection onto a closed subspace of L2L^2.

In numerical analysis, real analysis distinguishes formal approximation from justified convergence. A sequence of discrete approximations must be shown to approach a solution in a specified norm or topology. Consistency controls how accurately a discrete equation represents the continuous one; stability controls sensitivity and amplification of errors. Completeness and compactness then help turn estimates into existence and convergence results. The Banach fixed-point theorem, for example, gives both existence and uniqueness for contraction mappings and quantitative bounds for successive approximations.

Existence arguments for optimization follow related patterns. A continuous objective attains a minimum on a compact feasible set. In infinite-dimensional settings, weak compactness and weak lower semicontinuity often replace ordinary compactness and continuity. The calculus of variations studies such problems for function-valued unknowns, where minimizing sequences may develop oscillations or concentrations rather than converge strongly.

Contemporary research in real-variable methods includes quantitative estimates for oscillatory integrals, geometric measure theory, singular integrals, and regularity of weak solutions. Geometric measure theory studies size and shape beyond smooth curves and surfaces, including sets with noninteger dimension and surfaces defined through variational principles. Harmonic analysis investigates how frequency, spatial concentration, and geometric structure interact.

Many problems are sensitive to the exact exponent or dimension at which an estimate remains valid. An inequality may hold for every p>1p>1 but fail at p=1p=1, or compactness may disappear at a critical exponent. Concentrating sequences, rescaling arguments, and sharp counterexamples identify these boundaries and determine which existence, regularity, and convergence conclusions an analytic framework can support.