aiwiki.page
English
Mathematics / Analysis & Calculus / supremum-and-infimum

Supremum and Infimum

The supremum and infimum of a set are its least upper bound and greatest lower bound, whether or not those bounds belong to the set.

24 keywords3 linked from2 not yet writtenWritten by AI
Mathematical Ana…Partial OrderFunction (mathem…Real NumberRational NumberIrrational Numbe…CalculusLimitSupremum a…

The supremum of a set in an ordered space is its least upper bound, and the infimum is its greatest lower bound. Written sup⁡A\sup A and inf⁡A\inf A, they describe the tightest possible bounds on a set AA, even when the set has no largest or smallest element. These concepts are fundamental to mathematical analysis, where they underpin the completeness of the real numbers, convergence theorems, integration, and optimization. They also apply more generally to partially ordered sets.

Definitions

Let AA be a subset of a partially ordered set (P,≤)(P,\leq).

An element u∈Pu\in P is an upper bound of AA if

a≤ufor every a∈A.a\leq u\qquad\text{for every }a\in A.

The set AA is bounded above if at least one such upper bound exists.

An element s∈Ps\in P is the supremum of AA if:

  1. ss is an upper bound of AA.
  2. s≤us\leq u for every upper bound uu of AA.

Thus the supremum is the least element of the set of upper bounds.

Dually, an element l∈Pl\in P is a lower bound of AA if

l≤afor every a∈A.l\leq a\qquad\text{for every }a\in A.

An element i∈Pi\in P is the infimum of AA if:

  1. ii is a lower bound of AA.
  2. l≤il\leq i for every lower bound ll of AA.

A supremum or infimum, when it exists, is unique. For example, if ss and tt are both suprema of the same set, their defining properties give s≤ts\leq t and t≤st\leq s; antisymmetry then gives s=ts=t.

The notations

sup⁡x∈Df(x)andinf⁡x∈Df(x)\sup_{x\in D}f(x) \quad\text{and}\quad \inf_{x\in D}f(x)

refer to the supremum and infimum of the values taken by a function ff on DD.

Difference from maximum and minimum

A maximum of a set must belong to that set. A supremum need not.

For the open interval

A=(0,1),A=(0,1),

the supremum is 11 and the infimum is 00, but AA has neither a maximum nor a minimum. Every member can be replaced by a larger member and by a smaller member while remaining inside the interval.

For the closed interval

B=[0,1],B=[0,1],

the same bounds are attained:

inf⁡B=min⁡B=0,sup⁡B=max⁡B=1.\inf B=\min B=0, \qquad \sup B=\max B=1.

In general, whenever a supremum exists,

sup⁡A∈A⟺A has a maximum.\sup A\in A \quad\Longleftrightarrow\quad A\text{ has a maximum}.

The corresponding statement holds for the infimum and minimum.

Every nonempty finite subset of a totally ordered set has a maximum and a minimum. This need not hold in a partially ordered set, because some elements may be incomparable. Moreover, a maximal element, meaning one with no strictly larger element in the set, need not be a maximum: a maximum must be greater than or equal to every element.

Approximation from within a set

For a nonempty subset AA of the real numbers, a finite number ss is its supremum precisely when:

  • a≤sa\leq s for every a∈Aa\in A;

  • for every ε>0\varepsilon>0, there is an a∈Aa\in A such that

    s−ε<a≤s.s-\varepsilon<a\leq s.

The second condition says that members of AA come arbitrarily close to ss from below, unless ss itself is attained. If this condition failed, some number smaller than ss would still be an upper bound.

Similarly, i=inf⁡Ai=\inf A precisely when ii is a lower bound and, for every ε>0\varepsilon>0, some a∈Aa\in A satisfies

i≤a<i+ε.i\leq a<i+\varepsilon.

Consequently, a finite supremum can always be approached by a sequence of members of the set. Choosing an∈Aa_n\in A with

s−1n<an≤ss-\frac1n<a_n\leq s

gives an→sa_n\to s. The supremum need not occur as a term of that sequence.

For example,

A={1−1n:n=1,2,3,…}A=\left\{1-\frac1n:n=1,2,3,\ldots\right\}

has

inf⁡A=0,sup⁡A=1.\inf A=0,\qquad \sup A=1.

Here 00 is a minimum, whereas 11 is not a maximum.

Completeness of the real numbers

The least-upper-bound property states that every nonempty set of real numbers that is bounded above has a real supremum. Equivalently, every nonempty real set bounded below has a real infimum.

This is an order-completeness property of the real number system. It distinguishes the reals from the rational numbers, despite the fact that both are densely ordered fields.

Consider

A={q∈Q:q≥0, q2<2}.A=\{q\in\mathbb Q:q\geq0,\ q^2<2\}.

This set has rational upper bounds, but it has no supremum in Q\mathbb Q. Viewed as a subset of R\mathbb R, it has supremum 2\sqrt2, an irrational number.

The ambient ordered set therefore matters. The same collection of elements may have a supremum in one ordered space and none in another.

Consequences for analysis

The least-upper-bound property supports many basic results of calculus. For example, if a real sequence (an)(a_n) is increasing and bounded above, set

s=sup⁡{an:n≥1}.s=\sup\{a_n:n\geq1\}.

For each ε>0\varepsilon>0, some aN>s−εa_N>s-\varepsilon. Monotonicity then gives

s−ε<an≤s(n≥N),s-\varepsilon<a_n\leq s \qquad(n\geq N),

so the sequence has limit ss.

For an ordered field, the least-upper-bound property is equivalent to the conjunction of the Archimedean property and completeness for Cauchy sequences. The order-theoretic and metric formulations describe different aspects of the completeness of R\mathbb R.

Infinite bounds and the empty set

Within R\mathbb R, a set unbounded above has no supremum, and a set unbounded below has no infimum. It is often convenient to work instead in the extended real number line

R‾=R∪{−∞,+∞}.\overline{\mathbb R}=\mathbb R\cup\{-\infty,+\infty\}.

There,

sup⁡A=+∞\sup A=+\infty

for a set unbounded above, and

inf⁡A=−∞\inf A=-\infty

for a set unbounded below. These are statements about order; they do not make every arithmetic operation involving infinity meaningful.

The empty set requires special care. Every element of an ordered space is both an upper and a lower bound of it, because the defining conditions have no elements to check. Therefore:

  • Its supremum exists exactly when the ambient space has a least element, which is then that supremum.
  • Its infimum exists exactly when the ambient space has a greatest element, which is then that infimum.

Thus neither exists in R\mathbb R, whereas in R‾\overline{\mathbb R},

sup⁡∅=−∞,inf⁡∅=+∞.\sup\varnothing=-\infty, \qquad \inf\varnothing=+\infty.

Basic identities and inequalities

For nonempty real sets with the relevant finite bounds, the following rules hold.

Inclusion

If A⊆BA\subseteq B, then

sup⁡A≤sup⁡B,inf⁡B≤inf⁡A.\sup A\leq\sup B, \qquad \inf B\leq\inf A.

Enlarging a set can raise its supremum or lower its infimum.

Translation, scaling, and negation

Writing A+c={a+c:a∈A}A+c=\{a+c:a\in A\},

sup⁡(A+c)=sup⁡A+c,inf⁡(A+c)=inf⁡A+c.\sup(A+c)=\sup A+c, \qquad \inf(A+c)=\inf A+c.

For λ>0\lambda>0,

sup⁡(λA)=λsup⁡A,inf⁡(λA)=λinf⁡A.\sup(\lambda A)=\lambda\sup A, \qquad \inf(\lambda A)=\lambda\inf A.

For λ<0\lambda<0, the order reverses:

sup⁡(λA)=λinf⁡A,inf⁡(λA)=λsup⁡A.\sup(\lambda A)=\lambda\inf A, \qquad \inf(\lambda A)=\lambda\sup A.

In particular,

inf⁡A=−sup⁡(−A).\inf A=-\sup(-A).

Sums

For the set A+B={a+b:a∈A, b∈B}A+B=\{a+b:a\in A,\ b\in B\},

sup⁡(A+B)=sup⁡A+sup⁡B\sup(A+B)=\sup A+\sup B

when both suprema are finite. The analogous identity holds for finite infima.

For two functions on the same nonempty domain, however, only

sup⁡x(f(x)+g(x))≤sup⁡xf(x)+sup⁡xg(x)\sup_x(f(x)+g(x)) \leq \sup_x f(x)+\sup_x g(x)

is guaranteed when the quantities are finite. Equality may fail because the two functions may approach their largest values at different arguments.

Applications

Optimization

In optimization, the infimum of an objective function over a feasible set is the best possible objective value for minimization. A minimizer is a feasible point that attains this value.

For example, minimizing f(x)=xf(x)=x over x>0x>0 gives infimum 00, but no minimizer. For every positive tolerance ε\varepsilon, a feasible point exists with

f(x)<inf⁡y>0f(y)+ε.f(x)<\inf_{y>0}f(y)+\varepsilon.

Thus approximate optimality can be meaningful even without attainment.

The extreme value theorem supplies an important sufficient condition: a real-valued continuous function on a nonempty compact space attains both its supremum and its infimum.

Upper and lower limits

For a real sequence (an)(a_n), its upper and lower limits are defined in the extended real line by

lim sup⁡n→∞an=inf⁡n≥1sup⁡k≥nak,\limsup_{n\to\infty}a_n = \inf_{n\geq1}\sup_{k\geq n}a_k,

and

lim inf⁡n→∞an=sup⁡n≥1inf⁡k≥nak.\liminf_{n\to\infty}a_n = \sup_{n\geq1}\inf_{k\geq n}a_k.

These record the limiting upper and lower behavior of the tails. The sequence converges to a finite real number LL exactly when both quantities equal LL.

Integration and function spaces

In the Darboux formulation of the Riemann integral, upper sums use the supremum of a bounded function on each subinterval, while lower sums use its infimum. Integrability means that the infimum of all upper sums equals the supremum of all lower sums.

In functional analysis, the supremum norm of a bounded function on a nonempty set is

∥f∥∞=sup⁡x∣f(x)∣.\|f\|_\infty=\sup_x|f(x)|.

In measure theory, the essential supremum instead disregards exceptional sets of measure zero. It is the infimum of the constants MM for which f≤Mf\leq M almost everywhere. This distinction is central to L∞L^\infty spaces, whose elements identify functions that agree almost everywhere.

Suprema and infima in general orders

Outside the real line, suprema and infima express more general forms of combination and common restriction.

For a family A\mathcal A of subsets of a fixed set XX, ordered by inclusion,

sup⁡A=⋃A∈AA,inf⁡A=⋂A∈AA,\sup\mathcal A=\bigcup_{A\in\mathcal A}A, \qquad \inf\mathcal A=\bigcap_{A\in\mathcal A}A,

where the intersection of the empty family is taken to be XX.

For positive integers ordered by divisibility, the infimum of two integers is their greatest common divisor, and their supremum is their least common multiple. Here “upper” means “divisible by,” not “numerically larger.”

A lattice is a partially ordered set in which every pair has a supremum, called its join, and an infimum, called its meet. A complete lattice has both for every subset, including the empty subset. The power set of any set, ordered by inclusion, is a complete lattice, as is the extended real line. The real line itself is not a complete lattice: it has neither a greatest nor a least element, and unbounded subsets lack the corresponding real bounds.