A set may have a supremum without having a maximum: the number is then approached from below but never reached. This note applies that idea to a set defined by a parity-dependent formula, to its image under a reflection, and to the difference of two sets, and it records exactly when the bounds are attained. The approximation criterion and the set-sum rule are proved in the companion note on suprema and set operations.
Criteria to keep in view
- For a nonempty set bounded above, an upper bound is its supremum exactly when elements approach it from below to arbitrary accuracy.
- For an infimum, use a lower bound and approximation from above.
- when the latter is finite; negative scaling swaps the roles of the two bounds.
- A maximum or minimum must belong to the set.
- For nonempty bounded sets, ; the summands may be chosen independently.
Difference sets
For nonempty sets put . Write , , and .
Proposition 8.2.1
(Bounds of a difference set)
Let be nonempty and bounded. Then
Proof
(Approximation from both sides)
Every difference satisfies , because and . Given , choose with and with . Then , so is the least upper bound.
Every difference also satisfies . Choose with and with . Then , which proves the infimum formula. Nonemptiness and boundedness make all four source bounds finite, so the witnesses exist.
Proposition 8.2.2
(Attainment for a difference set)
With the same hypotheses, holds exactly when and . Hence has a maximum if and only if has a maximum and has a minimum. Likewise holds exactly when and , so has a minimum if and only if has a minimum and has a maximum.
Proof
(Two nonnegative gaps)
Suppose . Then
Both terms are nonnegative, since bounds above and bounds below. A sum of two nonnegative numbers vanishes only if each vanishes, so and . In particular and . Conversely, if and , the pair gives .
For the minimum, use
which forces and ; the converse is the same pair argument. The argument describes every attaining pair, not only the existence of one.
Proof X-ray
Where the sign of each gap comes from
The proof hinges on writing the equation as a sum of two gaps. The gap is nonnegative because is an upper bound of ; the gap is nonnegative because is a lower bound of . The sign of the second gap is the reflection at work: enters the difference with a minus sign, so its infimum, not its supremum, is paired with . Nonemptiness and boundedness are used only to make and finite numbers. The argument closes because a sum of nonnegative terms can vanish in only one way.
Replace by and the pairing breaks: then the two gaps and enter as , a difference of nonnegative numbers, which can vanish while both are positive.
The pairing of with is easy to get wrong, as the next example shows.
Counterexample mode
The supremum of a difference is not a difference of suprema
A tempting claim is . Take . Every number in is a difference: equals and equals . So and (attained by ). But . The correct formula pairs with , giving .
Worked examples
Worked example 8.2.3
(A parity-defined set)
Let
Find , and every maximum or minimum, proving leastness and greatestness by approximation, and identify each index attaining an endpoint.
Write each member as
The numerator of the subtracted term is for even and for odd . It is always positive, so every member is strictly below . The subtracted term is at most , so every member is at least . At the value is exactly . For the bound is strict, so is the only index attaining the lower endpoint. Hence .
To prove , fix and choose an integer . The even index gives the value , which lies less than below because . These witnesses show that no number below is an upper bound. No member equals , so has no maximum.
Worked example 8.2.4
(A reflected image)
For as above, let . Find the supremum, infimum, maximum and minimum of .
Since , we get
The upper endpoint is attained exactly when , that is, at index . So . For the lower endpoint, given use to choose with . Then . Hence . It is not a minimum, because would require , which does not belong to .
The order reversal is visible in the answer: and , and the attained endpoint of (its minimum) becomes the attained endpoint of (its maximum).
Worked example 8.2.5
(A difference of two intervals)
Let and . Here , , and .
By the bounds proposition, and . By the attainment proposition, is attained: and , and the unique attaining pair is . So . The infimum would need and ; neither holds, so has no minimum. Differences come arbitrarily close to , but no pair reaches it.
Common mistakes
Common mistake
(An upper bound is not yet the supremum)
For the set above, every member is less than , and one may be tempted to conclude . That only proves that is an upper bound. The missing obligation is approximation from below: for the criterion would demand a member of greater than , but every member is less than . So is not the least upper bound; the supremum is .
Common mistake
(Scaling by a negative number swaps the bounds)
For it is wrong to write . That expression equals , which is the infimum of . A negative scale reverses inequalities, so .
Practice
Quick check
(Let . Find the supremum, infimum, maximum and minimum of .)
The scale is negative.
Solution · Answer
Since , we get . The value is attained at , so . The value would need , so and there is no minimum.
Quick check
(Let and . Does have a maximum? A minimum?)
Compare with and .
Solution · Answer
Here and , neither belonging to . By the attainment proposition, has no maximum (that would need ) and no minimum (that would need ). Its bounds are and .
Quick check
(In the proof of , why are the two witnesses chosen within of their bounds?)
Count the total error in .
Solution · Answer
The error of below and the error of above add up in . Each is less than , so the total is less than , which is the tolerance the criterion requires.