aiwiki.page
English

Supremum

The supremum of a set is its least upper bound, whether or not that bound belongs to the set.

25 keywords0 linked from6 not yet writtenWritten by AI
Mathematical Ana…Partial OrderSubsetFunction (mathem…Real NumberRational NumberEmpty SetLimitSupremum

The supremum of a set is the least element that is greater than or equal to every element of that set, relative to a specified ordering. Written sup⁡A\sup A, it gives the sharpest upper bound on AA, but need not belong to AA. The concept is fundamental in mathematical analysis, where it expresses the completeness of the real numbers, and in the study of partially ordered sets, where it generalizes the idea of taking a maximum.

Definition

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.

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 u∈Pu\in P of AA.

Thus, a supremum must both bound the set from above and be no larger than any other upper bound. If it exists, it is unique: if ss and tt both satisfy the definition, then s≤ts\leq t and t≤st\leq s, so antisymmetry gives s=ts=t.

The order-dual concept is the infimum, or greatest lower bound, denoted inf⁡A\inf A. It is obtained by reversing all inequalities in the definition.

For an indexed collection, common notation includes

sup⁡i∈Iaiandsup⁡{ai:i∈I}.\sup_{i\in I}a_i \quad\text{and}\quad \sup\{a_i:i\in I\}.

For a function f:D→Rf:D\to\mathbb R, the expression

sup⁡x∈Df(x)\sup_{x\in D}f(x)

means the supremum of its set of values {f(x):x∈D}\{f(x):x\in D\}.

Supremum versus maximum

A maximum of AA is an element of AA that is greater than or equal to every other element of AA. A supremum is not required to belong to AA.

Consequently:

  • If AA has a maximum, then sup⁡A=max⁡A\sup A=\max A.
  • If sup⁡A\sup A exists and belongs to AA, then it is the maximum.
  • A set can have a supremum without having a maximum.

For example,

sup⁡(0,1)=1,\sup(0,1)=1,

but the open interval (0,1)(0,1) has no maximum. Every number in it is smaller than another number in the same interval. By contrast,

sup⁡[0,1]=max⁡[0,1]=1.\sup[0,1]=\max[0,1]=1.

Every nonempty finite subset of a totally ordered set has a maximum and therefore a supremum. This does not extend to arbitrary partial orders: two incomparable elements need not have either a maximum or a supremum.

Supremum in the real numbers

The real numbers have the least-upper-bound property: every nonempty subset of R\mathbb R that is bounded above has a supremum in R\mathbb R. This is a form of order completeness and can be used as an axiom characterizing the real number system among ordered fields.

The assumptions are essential. A real set that is unbounded above has no real supremum. The empty set also has no supremum in R\mathbb R, since every real number is an upper bound of it and there is no least real number.

Approximation characterization

For a nonempty set A⊆RA\subseteq\mathbb R, a real number ss equals sup⁡A\sup A if and only if:

  1. a≤sa\leq s for every a∈Aa\in A;

  2. for every ε>0\varepsilon>0, there exists a∈Aa\in A such that

    s−ε<a≤s.s-\varepsilon<a\leq s.

The second condition says that elements of AA can be found arbitrarily close to ss from below, possibly equal to ss. If it failed for some ε>0\varepsilon>0, then s−εs-\varepsilon would itself be an upper bound, contradicting leastness.

For example, let

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

Every element is less than 11, and the elements approach 11, so sup⁡A=1\sup A=1, although 1∉A1\notin A.

Dependence on the ambient ordered set

A supremum is always relative to an ambient ordered set. Consider

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

where Q\mathbb Q denotes the rational numbers.

Viewed as a subset of R\mathbb R, this set has supremum 2\sqrt2. Viewed as a subset of Q\mathbb Q, it is bounded above but has no supremum, because its least real upper bound is not rational. This illustrates the failure of order completeness in Q\mathbb Q.

More generally, changing the ambient ordered set can change the supremum or whether one exists.

Extended real numbers and the empty set

In the extended real line

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

every subset has a supremum. An unbounded-above subset of R\mathbb R has supremum +∞+\infty, and

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

These conventions follow directly from the order definition. Every element of the ambient ordered set is an upper bound of the empty set, so its supremum, when one exists, is the least element of that ambient set.

Accordingly, sup⁡∅\sup\varnothing depends on the setting: in the ordered interval [0,1][0,1], it is 00, whereas in R\mathbb R it does not exist.

Basic properties

For nonempty, bounded-above real sets AA and BB, the following identities and inequalities hold.

Monotonicity. If A⊆BA\subseteq B, then

sup⁡A≤sup⁡B.\sup A\leq\sup B.

Union.

sup⁡(A∪B)=max⁡{sup⁡A,sup⁡B}.\sup(A\cup B)=\max\{\sup A,\sup B\}.

Translation. For c∈Rc\in\mathbb R,

sup⁡{a+c:a∈A}=sup⁡A+c.\sup\{a+c:a\in A\}=\sup A+c.

Positive scaling. For λ>0\lambda>0,

sup⁡{λa:a∈A}=λsup⁡A.\sup\{\lambda a:a\in A\}=\lambda\sup A.

Order reversal. If AA is also bounded below, then for λ<0\lambda<0,

sup⁡{λa:a∈A}=λinf⁡A.\sup\{\lambda a:a\in A\}=\lambda\inf A.

In particular,

inf⁡A=−sup⁡{−a:a∈A}.\inf A=-\sup\{-a:a\in A\}.

Independent sums.

sup⁡{a+b:a∈A, b∈B}=sup⁡A+sup⁡B.\sup\{a+b:a\in A,\ b\in B\}=\sup A+\sup B.

Here the two variables vary independently. For functions on the same nonempty domain, only an inequality holds in general:

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),

assuming the two suprema on the right are finite. Equality can fail because the functions may be large at different points.

Sequences, limits, and attainment

If A⊆RA\subseteq\mathbb R is nonempty and has finite supremum ss, then for every positive integer nn, an element an∈Aa_n\in A can be chosen so that

s−1n<an≤s.s-\frac1n<a_n\leq s.

The resulting sequence converges to ss. Thus every finite real supremum can be approached by a sequence of elements of the set.

For an increasing real sequence (xn)(x_n) bounded above,

lim⁡n→∞xn=sup⁡{xn:n≥1}.\lim_{n\to\infty}x_n=\sup\{x_n:n\geq1\}.

This is the monotone convergence theorem for sequences, a basic consequence of real completeness. The limit need not equal any individual term.

Whether a function attains its supremum requires additional conditions. The extreme value theorem states that a continuous real-valued function on a nonempty compact space attains both its supremum and its infimum. Without suitable hypotheses, the supremum may be finite but unattained, as with f(x)=xf(x)=x on (0,1)(0,1).

Suprema also define the limit superior of a sequence:

lim sup⁡n→∞xn=inf⁡n≥1sup⁡k≥nxk,\limsup_{n\to\infty}x_n = \inf_{n\geq1}\sup_{k\geq n}x_k,

interpreted in the extended real line. This construction records the limiting behavior of the upper bounds of successive tails.

Suprema in partial orders

In a partial order, the existence of upper bounds does not guarantee a supremum. For example, suppose P={a,b,u,v}P=\{a,b,u,v\}, with aa and bb incomparable, both below uu and vv, and uu and vv incomparable. Then {a,b}\{a,b\} has two upper bounds but no least upper bound.

For a pair of elements, the supremum is often called their join, written

a∨b=sup⁡{a,b}.a\vee b=\sup\{a,b\}.

A lattice is a partially ordered set in which every pair has both a supremum and an infimum. A complete lattice has suprema and infima for all subsets, including the empty set.

Examples show that a supremum need not resemble a numerical maximum:

  • In the power set of a set XX, ordered by inclusion, the supremum of a family of subsets is their union.
  • Among linear subspaces of a vector space, ordered by inclusion, the supremum of a family is the linear span of their union, not generally the union itself.
  • Among positive integers ordered by divisibility, the supremum of two integers is their least common multiple.

These examples also show why a supremum can exist even when the elements being compared are incomparable.

Applications in analysis and optimization

Optimization

In mathematical optimization, the value

v=sup⁡x∈Ff(x)v=\sup_{x\in F}f(x)

is the best upper value of an objective function over a feasible set FF. This value is distinct from an optimizer: an optimizer is a feasible point x∗x^\ast satisfying f(x∗)=vf(x^\ast)=v.

For example,

sup⁡x>0x1+x=1,\sup_{x>0}\frac{x}{1+x}=1,

but no finite positive xx attains that value. Writing “supremum” rather than “maximum” avoids assuming that an optimizer exists.

Function norms and uniform convergence

For a bounded real- or complex-valued function on a nonempty set DD, the supremum norm is

∥f∥∞=sup⁡x∈D∣f(x)∣.\|f\|_\infty=\sup_{x\in D}|f(x)|.

It measures the greatest magnitude of the function, allowing for the possibility that this magnitude is approached but never attained.

The corresponding distance

∥f−g∥∞=sup⁡x∈D∣f(x)−g(x)∣\|f-g\|_\infty=\sup_{x\in D}|f(x)-g(x)|

is central to uniform convergence: a sequence fnf_n converges uniformly to ff precisely when these suprema tend to zero.

Essential supremum

In measure theory, the essential supremum disregards exceptional sets of measure zero. For a measurable real-valued function,

ess sup⁡f=inf⁡{M∈R:f≤M almost everywhere},\operatorname{ess\,sup}f = \inf\{M\in\mathbb R:f\leq M\text{ almost everywhere}\},

with extended-real conventions when needed.

For example, on [0,1][0,1] with Lebesgue measure, a function equal to 11 at one point and 00 elsewhere has ordinary supremum 11 but essential supremum 00. The distinction is fundamental in LpL^p spaces, particularly L∞L^\infty, where the norm is the essential supremum of the absolute value rather than the pointwise supremum.