Working With Suprema? Always Ask: Least Upper Bound, or Just an Upper Bound?

Here is a habit that will save you points on JEE analysis questions, catch your own mistakes during proofs, and quietly separate you from the crowd of students who nod along when they hear the word "supremum" without actually holding it down. The habit is a single question, asked every time you see or write the symbol \sup:

Is this number the least upper bound, or just an upper bound?

They are not the same thing. A set can have thousands of upper bounds. It has exactly one least upper bound. Writing \sup S = M is a commitment to the stronger claim — and students routinely make that commitment while only doing the work for the weaker one.

The trap in one sentence

The definition of supremum has two parts, and the part most people forget is the second one. Here it is laid out explicitly, because you cannot ask the question correctly without seeing both halves.

A real number M is the supremum of a set S \subseteq \mathbb{R} when:

(Upper bound.) s \le M for every s \in S.
(Least.) For every \varepsilon > 0, there exists some s \in S with s > M - \varepsilon.

Part (i) says M sits above the set. Part (ii) says nothing below M can also sit above the set — because any candidate M - \varepsilon gets dethroned by some element of S that pokes above it.

Both parts matter. Checking only (i) tells you M is an upper bound. That is a weaker claim, true of infinitely many numbers. Checking (i) and (ii) together is what pins M down as the supremum.

The habit is: whenever you or a textbook writes "M = \sup S", pause and verify, mentally, that both parts hold. Not one. Both.

Why the habit exists — a toy that exposes the gap

Take S = [0, 1), the half-open interval from 0 (included) to 1 (excluded).

A careless student will glance at this and say: "the supremum is 2, since every element is less than 2." The glance is honest: 2 does satisfy part (i). Every s \in [0, 1) is below 2. But does 2 satisfy part (ii)? Take \varepsilon = 0.5. Is there some s \in S with s > 2 - 0.5 = 1.5? No — every element of [0, 1) is below 1, let alone below 1.5. Part (ii) fails. So 2 is an upper bound, not the supremum.

Try M = 1.5 instead. Part (i): still fine, every element of [0, 1) is below 1.5. Part (ii) with \varepsilon = 0.2: need some s \in S with s > 1.3. Same problem — no element of [0, 1) exceeds 1. Part (ii) fails again. So 1.5 is also just an upper bound.

Now try M = 1. Part (i): every s \in [0, 1) satisfies s < 1, so s \le 1. Fine. Part (ii): given \varepsilon > 0, can we find s \in S with s > 1 - \varepsilon? Yes — take s = 1 - \varepsilon/2, which is in [0, 1) as long as \varepsilon < 2, and which is strictly bigger than 1 - \varepsilon. Both parts pass. So \sup S = 1.

Notice what happens if you only ever ran check (i). You would have accepted 2 as an answer. Or 10. Or \pi^2. Each satisfies the weak condition. Only (ii) rules them out and forces you down to the exact number 1.

The number-line picture

Every tick above the line is a legitimate upper bound of $S = [0, 1)$. The supremum is the leftmost of them — the one that nothing below can replace. Asking only "is this an upper bound?" would green-light any of them. Asking also "is there nothing smaller?" forces you to the unique leftmost tick, $1$.

When the habit looks redundant — and when it saves you

Case where it looks redundant. Take S = \{1/n : n \in \mathbb{N}\} = \{1,\ 1/2,\ 1/3,\ 1/4,\ \ldots\}. A student glances and says, "sup is 1, because 1/1 = 1 is in the set." That is correct. But run the habit anyway. Part (i): yes, every 1/n \le 1. Part (ii): given \varepsilon > 0, is there s \in S with s > 1 - \varepsilon? Yes — take s = 1. Both parts pass trivially, because 1 is both in the set and above everything. Here \sup S = \max S = 1, and the habit cost you three seconds.

The habit looked redundant. That is fine. The point is not that it always catches something — it is that you never know in advance when it will.

Case where it saves you. Take S = \{x \in \mathbb{Q} : x^2 < 2\}, a set of rationals whose square is below 2.

A typical JEE-prep student reasons: "the biggest x with x^2 < 2 is about 1.414, so \sup S = 1.414." The habit catches this instantly. Check M = 1.414. Part (i): does every s \in S satisfy s \le 1.414? No. The rational 1.4142 has 1.4142^2 = 1.9999\ldots < 2, so 1.4142 \in S, and 1.4142 > 1.414. Part (i) already fails. 1.414 is not even an upper bound.

Push to M = 1.4143. Then 1.41421356 is still in S and exceeds 1.4143… Wait, no: 1.4143^2 = 1.99024 < 2, so 1.4143 \in S itself. Still not an upper bound.

The habit forces you to keep pushing M up until you stop finding rationals that beat it. Where do you stop? At exactly \sqrt{2}. The set of rationals whose square is less than 2 is bounded above by \sqrt{2} and by nothing smaller — so in \mathbb{R}, \sup S = \sqrt{2}. Crucially, this supremum is an irrational, and it does not belong to S.

Without the habit, you would have stopped at some decimal approximation, wrong by a little bit, and the grader would have circled it red. With the habit, you keep going until part (ii) is airtight, and you land on the exact answer — \sqrt{2}, a real number that exists only because \mathbb{R} is complete. See Supremum on a Leash for a draggable version of this phenomenon, and Bounded vs. Has a Maximum for the membership question lurking beneath.

Where the habit shows up in disguise

Once you know what to look for, you start spotting the same trap across analysis, calculus, and probability — dressed in different clothes each time.

In limits. When you write \lim_{n \to \infty} a_n = L for an increasing bounded sequence, that L is secretly \sup \{a_1, a_2, a_3, \ldots\}. Most students accept the symbol L without ever asking whether some smaller L' would also work as the limit. Run the habit. Part (i) says every term is \le L. Part (ii) says every smaller candidate gets overtaken. Both parts are exactly what the definition of limit (for monotone sequences) demands.

In integration. The definition of the Riemann integral uses \sup over lower sums and \inf over upper sums. Whenever a textbook writes \underline{\int} f = \sup_P L(f, P), the same two-part check governs it. Any single partition P_0 gives you a lower sum — but that lower sum is not automatically the supremum over all partitions. Asking "least, or just some?" is what forces you to consider finer and finer partitions until part (ii) closes.

In optimisation. JEE problems like "find the maximum value of f(x) = x(1 - x) on (0, 1)" look harmless. You take a derivative, get f'(x) = 1 - 2x = 0, find x = 1/2, compute f(1/2) = 1/4. That is the maximum — and also the sup, and they agree because 1/2 is in the open interval. Fine. But change the interval to (0, 1/3) and the critical point leaves the domain. Now there is no maximum at all; the supremum is f(1/3) = 2/9, and f approaches it without attaining it, because 1/3 is not in (0, 1/3). A student who does not run the habit writes "\max = 2/9" and loses the mark. A student who does writes "\sup = 2/9; max does not exist" and keeps it.

A tiny mental script

To make the habit automatic, rehearse this inner monologue exactly three times on problems you have already solved, and then it will fire on its own:

"I claim \sup S = M. Step one: is M above every element of S? Let me check. … Yes. Step two: can I shrink M by any positive \varepsilon and still have something above S? Let me try to find an element of S above M - \varepsilon. … Yes, here is one. So nothing smaller than M works. So M is the sup."

The script is short and boring. Boring is exactly what habits need to be — the interesting thinking is reserved for the problem itself, not for remembering which check to run. Once the script is automatic, you read "\sup" with both hands on the wheel.

The one-line diagnostic

If you take nothing else away, take this. Whenever you write M = \sup S, immediately write two check-lines next to it:

(i) For all s \in S, s \le M. — verify.
(ii) For all \varepsilon > 0, some s \in S satisfies s > M - \varepsilon. — verify.

If both lines check, you have earned the "=" sign. If only the first does, you have written something weaker than you meant, and you need to find a smaller M and try again.

That is the entire habit. Two lines, every time. Nothing fancy. The students who do this stop losing marks on supremum problems and start catching themselves mid-derivation when they were about to slip. The ones who skip it keep writing "\sup" when they mean "some upper bound I happened to spot," and the distinction quietly eats their scores.

Why the second check is the one that matters

Everything hard about suprema lives in part (ii). Part (i) — "M is above the set" — is the easy half, usually obvious from the definition of the set. For [0, 1), the bound 1 is visible at a glance. For \{1 - 1/n\}, the bound 1 follows from 1/n > 0. For almost any set a textbook will hand you, finding some upper bound is a one-line observation. So if part (i) were all there was, the word "supremum" would be unnecessary — you could just say "upper bound" and move on.

Part (ii) is the whole payload. It is what distinguishes "there exists a ceiling" (the Archimedean-adjacent statement) from "this specific ceiling is the tightest one" (the completeness statement). The real numbers were built precisely so that part (ii) always has an answer inside \mathbb{R} — that is the content of the least upper bound axiom. On the rationals, the answer can fail to exist, which is why \{x \in \mathbb{Q} : x^2 < 2\} has no rational supremum. When the habit asks you to verify part (ii), it is asking you to do the work that the completeness axiom does for you — and to notice, occasionally, that the answer lands outside your current number system. That noticing is half of why we built \mathbb{R} in the first place.