Can a supremum equal a maximum?

Yes. Whenever a nonempty set has a maximum, that value is also its supremum.

What is the corresponding lower-bound concept?

The infimum is the greatest lower bound; it need not be an attained minimum.

Can a bounded set have several suprema?

No. A least upper bound, when it exists, is unique.

AI Glossary / Mathematics and optimization / Intermediate

Supremum (or least upper bound)

Definition

The supremum of a set of real numbers is its least upper bound: the smallest number that is at least as large as every member of the set. Unlike a maximum, a supremum does not have to belong to the set.

What it means

Consider all numbers strictly between zero and one. Every member is below one, and no smaller number bounds them all from above. The supremum is therefore one, even though one is excluded. There is no largest member: any proposed member can be replaced by a slightly larger one still below one.

For a finite nonempty set, the supremum equals its maximum. The distinction becomes especially useful for infinite sets, limits and optimization problems with open constraints.

A supremum is an exact property of the specified set. It is not automatically the largest value observed in a sample, the largest number found by a search or a practical performance estimate. Those may only provide evidence about the underlying set.

How it works

To show that s is the supremum of a nonempty real set A, establish two facts. First, every a in A satisfies a ≤ s. Second, every smaller proposed bound fails: for any positive ε, some member of A exceeds s − ε.

The second condition says that members can approach the bound arbitrarily closely from below, unless the bound is already attained. Together the conditions rule out both an underestimate and an unnecessarily loose upper bound.

Completeness of the real numbers guarantees a finite real supremum for every nonempty set bounded above. An unbounded-above set has no finite real supremum; extended-real notation often writes its supremum as +∞. Empty-set conventions must be stated separately.

In optimization, supx f(x) means the supremum of attainable objective values. Writing “max” additionally indicates that an allowed x attains the greatest value.

Three practical examples

A theoretical performance analysis describes a limiting score that can be approached but not achieved under a strict constraint.
An optimization report distinguishes the best feasible solution found from an upper bound on all feasible objective values.
A learning-theory expression takes a supremum over a family of functions to represent a worst-case quantity, without assuming one function attains it.

Think of it this way

Think of a height ceiling that objects can approach as closely as desired. The ceiling is the tightest height bound even if no object actually reaches it.

The analogy concerns order and attainability. A supremum is not a physical barrier or an estimate with measurement uncertainty; it follows from the exact mathematical set being considered.

Worked example

An upper bound that is never reached

Let A = {1 − 1/n : n is a positive integer}. Its first values are 0, 1/2, 2/3, 3/4 and 4/5.

Every member is less than one, so one is an upper bound. To show that it is the least upper bound, take any ε > 0 and choose an integer n > 1/ε. Then 1/n < ε, so 1 − 1/n > 1 − ε. No number below one can bound the entire set. Thus sup A = 1.

There is no maximum because increasing n gives a larger member, while no finite n makes 1/n equal zero. If a computation searches only n = 1 through 100, it finds maximum 0.99 for that finite search set. That is not the maximum of the original infinite set, which has none.

Now add the number one to A. The supremum stays one, but it becomes an attained maximum. The bound and the question of attainment are therefore separate pieces of information.

Illustrative case study

A fictional optimization team reports that a parameterized score “has a maximum of one.” Its allowed parameter range actually excludes the limiting setting that would produce one.

The mathematical review replaces that wording with a supremum statement and reports the score attained by the selected feasible parameter. It also documents a tolerance-based target for practical stopping, rather than asking the numerical search to reach an impossible exact endpoint.

The lesson is that a tight theoretical limit and a deployable solution answer different questions. The corrected report remains useful: it explains how close the chosen solution is to the bound without inventing an attaining configuration.

Common misconceptions and mistakes

Supremum is another name for the largest member. A largest member is a maximum, which may not exist.
Any upper bound is the supremum. The supremum must be the smallest upper bound.
A long numerical search proves the supremum. Unsearched values can exceed the best observed value.
A supremum must be finite. Unbounded sets require a different statement or extended-real convention.
An unattained supremum makes optimization meaningless. Approximate feasible solutions can still be valuable when their gap is controlled.

Business use

Supremum language helps make technical claims precise. It is useful when a report describes worst-case exposure, theoretical capacity, limiting accuracy or an objective under strict feasibility conditions. Always identify the set over which the bound is taken.

For practical decisions, report an attainable solution alongside a valid bound and a clearly defined tolerance. Avoid presenting a model's theoretical limit as measured operating performance. If the bound comes from assumptions, state those assumptions so that readers can judge whether the comparison applies to their situation.

Check your understanding

For the interval [2, 5), what are its supremum and maximum?

Answer: Its supremum is 5. It has no maximum because 5 is excluded and every included number can be exceeded by another included number.

Related terms

Infimum (or greatest lower bound)
Upper bound
maximum
infimum
limit
feasible set
optimization gap

Learn next

Read infimum and compactness. Compare maximizing f(x) = x over [0, 1], over [0, 1) and over the entire real line. For each case, identify the upper-bound behavior and whether a feasible point attains a maximum.

Sources

Aalto Dictionary of Machine Learning, supremum or least upper bound: https://aaltodictionaryofml.github.io/ADictML_Main.pdf. Definition and real-number completeness statement. The sequence proof, finite-search comparison and fictional optimization report are original teaching examples.

Previous: Suppressor variable Next: SUPS