An upper bound is not automatically a supremum. To identify the least upper bound, we must also show that elements of the set approach it from below. This second obligation is what makes supremum rules for sums, differences and products valid even when no maximum exists. It also explains why intersections behave differently from unions.
This chapter builds on sets and functions and real numbers and completeness, which establish the approximation criterion and the set-sum rule. Here we revisit those ideas and explore attainment, products and intersections.
Prerequisites and companion notes
The prerequisites are ordered real arithmetic, the completeness axiom, the Archimedean property, and set unions and intersections. Limits and continuity are not needed for the main proofs. Two companion notes apply the rules below: bounds, reflection and attainment treats a parity-defined set and difference sets, and products, unions and intersections treats product sets and intersections.
The two obligations
Definition 8.1.1
(Supremum, infimum, maximum and minimum)
Let . A number is an upper bound of if for every ; the supremum is the least upper bound. A number is a lower bound if for every ; the infimum is the greatest lower bound. A maximum (minimum) is a supremum (infimum) that belongs to .
All bounds in this chapter are finite real numbers. We therefore state nonemptiness and the required boundedness before using them. The empty set has no real supremum or infimum in this convention.
Theorem 8.1.2
(Approximation criterion)
Let be nonempty and bounded above, and let be an upper bound of . Then if and only if for every there is with . Dually, for nonempty and bounded below and a lower bound, if and only if for every there is with .
Proof
(Necessity and sufficiency of the second condition)
For necessity, suppose and . Then is smaller than the least upper bound, so it is not an upper bound, and some satisfies ; also .
For sufficiency, suppose the approximation condition holds and let be an upper bound with . Choose . The witness satisfies , contradicting that bounds . So no upper bound is smaller than , and is the least upper bound. This proves leastness, not just boundedness. The infimum version reverses every inequality.
Proof X-ray
Why one tolerance is enough
The proof of sufficiency does not test the condition for all against all rivals. It takes a single rival upper bound and turns it into a single tolerance : the gap between the claimed and the rival bound. The hypothesis is then used once, at exactly this , and produces an element above . The contradiction closes the argument because was assumed to bound every element. Positivity of is where is used; if no tolerance is available, which is as it should be, since is not smaller than itself. Without nonemptiness the witness could not exist.
If an upper bound already belongs to the set, the second condition is automatic; the criterion matters when no maximum exists.
Reflection and scaling must respect the sign
For nonempty bounded below, the set is bounded above. Reversing inequalities in the definition gives
Indeed is a lower bound of exactly when is an upper bound of , and greatest among the former becomes least among the latter. This also proves the equivalence of infimum completeness and supremum completeness.
Proposition 8.1.3
(Affine images)
Let be nonempty and bounded, , , and for constants put . If , then and . If , then and . If , the image is , whose maximum and minimum both equal .
Proof
(Order preserved or reversed)
For the map preserves order. The bounds and bound the image, and an input witness within of (or ) gives an image element within of (or ). For the map reverses order: the tolerance is , and witnesses near the lower bound produce elements near the upper bound of the image. Do not substitute into a proof that divides by the scale.
For nonzero , attainment travels with the relevant endpoint. With negative scale, an attained minimum becomes an attained maximum, while a missing supremum becomes a missing infimum.
A sum uses two independently chosen witnesses
For nonempty bounded sets , define
Proposition 8.1.4
(Sum rule and attainment)
Let be nonempty and bounded, and . Then , and has a maximum if and only if both and have maxima. Likewise , and has a minimum if and only if both have minima.
Proof
(Independent witnesses and two nonnegative gaps)
Every sum is at most . Given , choose and in their respective sets. Then , which proves . The elements can be selected independently because the definition of allows every pair.
Now suppose . Since and , the equation forces both terms to be zero. So and , and both suprema are attained. Conversely, attained suprema give the pair . The minimum statement is the mirror image. Reflection supplies the infimum rule, and applying the sum rule to gives the difference rules treated in the companion note on reflection.
An approximation to an endpoint is not an attained endpoint. Neither supremum has to be attained for the sum rule itself to hold.
Concept lensStructural
The set of upper bounds is a ray
For nonempty bounded above, let be its set of upper bounds. Any number above an upper bound is again one, and the supremum is the least, so , whether or not . The sum rule says exactly that , because . Attainment asks a different question: whether the left endpoint of the ray lies in the set itself.
A union chooses a set; an intersection must satisfy both
For nonempty bounded ,
The larger supremum bounds both sets, while approximation in the set with that supremum proves leastness. The infimum proof reverses the order.
If is nonempty, inclusion yields only
The separate extremal approximations may not be points belonging to both sets. Equality therefore needs its own justification, and the intersection may even be empty.
Counterexample mode
An intersection can lose the smaller supremum
The claim is false. Take and . Then , and . But , so and . The inclusion inequality holds strictly. Equality would need elements of approaching , and there are none.
For products, sign matters in a different way. Multiplication preserves order on nonnegative inputs; negative factors reverse it. The companion note on products proves the nonnegative product rule with explicit approximation estimates and then tests its failure when that hypothesis is removed. This is another use of the two obligations, not an instruction to multiply endpoint symbols without checking order.
Worked examples
Worked example 8.1.5
(An affine image with negative scale)
Let and consider . Here and , and . The proposition gives and . Directly, gives . The supremum is attained at , so it is a maximum; the infimum would need , so there is no minimum. The attained minimum of became an attained maximum of the image.
Worked example 8.1.6
(A sum with neither extremum attained)
Let and . The sum has infimum and supremum . It has neither a minimum nor a maximum: the lower endpoint is missing from , and the upper endpoint is missing from . The set-sum bound rules still apply, and every sum lies strictly between and .
Worked example 8.1.7
(A union and an intersection of adjacent intervals)
Let and . The union is with supremum and infimum , neither attained. The intersection is , with supremum and infimum : here the inclusion inequalities are equalities. Contrast the disjoint pair , : the intersection is empty and has no real supremum.
Common mistakes
Common mistake
(Stopping after the upper bound)
Showing for every proves only that is an upper bound. For the number bounds , but , because elements approach from below. The approximation step is what identifies the least upper bound.
Common mistake
(Confusing attainment with approximation)
A supremum can be approached by elements without being one of them. In the supremum is approached by for large , yet , so there is no maximum. Membership in the set is a separate requirement.
Practice
Quick check
(Let , and . Find the supremum and infimum of and say whether either is attained.)
The scale is negative, so the roles of and swap.
Solution · Answer
Here and . For , and . Directly, ranges over . Neither endpoint is attained, since .
Quick check
(If an upper bound of belongs to , why is ?)
Compare with an arbitrary upper bound .
Solution · Answer
Let be any upper bound of . Since , we have . So is smaller than or equal to every upper bound, i.e. it is the least one. The approximation condition holds trivially with the witness .
Quick check
(Let and . Find and , and decide whether has a maximum or a minimum.)
Check which of , has a maximum and which has a minimum.
Solution · Answer
The sum rule gives and . has a minimum but no maximum, and has a maximum but no minimum. A maximum of the sum needs maxima in both sets, and a minimum needs minima in both, so has neither. In fact .
Checklist
- Establish nonemptiness and boundedness before naming finite extrema.
- Prove a bound and then supply arbitrary-accuracy witnesses; neither step replaces the other.
- Keep maximum/minimum membership separate from supremum/infimum approximation.
- Track the sign when scaling or multiplying inequalities.
- For independent set sums, choose the two witnesses independently and divide the tolerance between them.
- For intersections, first check nonemptiness and do not promote inclusion inequalities into equalities.