Start real analysis proofs by writing out the exact definition you have to satisfy, doing rough scratch work to find the number that makes the inequality work (usually an N or a δ), and only then writing a clean proof that runs forwards from "Let ε > 0". Most first-course proofs follow a handful of templates, and learning those templates matters more than any single trick.
This guide is for students starting a first analysis course after calculus. It explains what changes, how to read the definitions, the templates to learn first, and four short proofs with commentary.
What changes after calculus
In calculus you were mostly asked to compute: a limit, a derivative, an integral. The theorems behind those computations were stated and used. Real analysis goes back and proves them, starting from the properties of the real numbers. Four things change at once.
- The answer is an argument, not a number. A correct final line with a gap in the middle earns little.
- Definitions are exact and you must use them literally. "aₙ gets close to L" is not the definition; the ε-N statement is.
- Inequalities replace equalities. Most of the work is bounding a quantity above by something simpler, and accepting that you lose a little on the way.
- Examples are not proofs. Checking n = 1, 10 and 1000 shows a pattern; it does not prove a statement about every n.
Students we teach often find the first three weeks harder than anything in calculus, then find that the course settles once the definitions feel familiar. That is normal. If you want guided practice with your own course's problem sheets, our real analysis tutoring online page explains how sessions with Ajay Vatsyayan work.
How to read a definition
Take the definition of a convergent sequence:
aₙ → L means: for every ε > 0 there exists N ∈ ℕ such that for all n ≥ N, |aₙ − L| < ε.
Read it as a game. Your opponent picks any tolerance ε > 0, however small. You must reply with a point N in the sequence. After that point, every term must lie within ε of L. Three things follow from reading it this way.
- Order matters. ε comes first, so N is allowed to depend on ε. N may not depend on n, because n is chosen after N.
- "For every ε" means you cannot pick ε. Your proof starts "Let ε > 0 be arbitrary", and everything after it must work for that unknown ε.
- "There exists N" means you must produce one. A formula such as N > 5/ε is the usual way.
To prove a statement is false, negate it carefully: swap "for every" with "there exists" and reverse the final inequality.
| Statement | Its negation |
|---|---|
| for every ε > 0 there is N such that every aₙ with n ≥ N lies within ε of L | there is ε > 0 such that for every N some aₙ with n ≥ N is at distance ε or more from L |
| S is bounded above: there is M with s ≤ M for all s ∈ S | for every M there is s ∈ S with s > M |
| f is increasing: x < y implies f(x) ≤ f(y) | there are x < y with f(x) > f(y) |
The templates to learn first
| To prove | Start with | Usual tools |
|---|---|---|
| aₙ → L | Let ε > 0. Choose N … | simplify the distance from L, bound it by c/n, Archimedean property |
| sup S = M | show M is an upper bound, then that nothing smaller is | for each ε > 0 find s ∈ S with s > M − ε |
| a sequence diverges | suppose aₙ → L, aim for a contradiction | the negation above, a specific ε such as 1 |
| a property holds for all n | induction, or a direct argument | base case, inductive step |
| a convergent sequence has a property | fix a convenient ε such as 1 | definition, triangle inequality |
The triangle inequality |x + y| ≤ |x| + |y| appears in almost every early proof. So does the Archimedean property: for any real number r there is a natural number N > r.
If induction or proof by contradiction is new, our posts on proof by induction and on why proof by contradiction builds deeper thinking are a gentle place to start before the analysis versions.
Proof 1: a limit from the definition
Claim: (3n + 1)/(n + 2) → 3.
Scratch work (not part of the proof). Simplify the distance from the limit:
Distance from 3: |(3n + 1)/(n + 2) − 3| = |(3n + 1 − 3n − 6)/(n + 2)| = 5/(n + 2)
We want this below ε. Solving 5/(n + 2) < ε exactly is possible, but it is easier to use 5/(n + 2) < 5/n and ask for 5/n < ε, which holds when n > 5/ε.
Proof. Let ε > 0. By the Archimedean property choose N ∈ ℕ with N > 5/ε. Then for every n ≥ N the distance from 3 satisfies
Distance: |(3n + 1)/(n + 2) − 3| = 5/(n + 2) < 5/n ≤ 5/N < ε.
Hence (3n + 1)/(n + 2) → 3. ∎
Commentary. The scratch work runs backwards from what we want. The proof runs forwards from ε, and the chain of inequalities is written in one line so a reader can check each step. Throwing away the 2 in the denominator made N simpler at no cost. The epsilon-delta definition of a limit uses the same game for functions, with δ in place of N.
Proof 2: a supremum
Claim: for S = {2 − 1/n : n ∈ ℕ}, sup S = 2.
Scratch work. Two parts. First, every element is below 2, which is clear. Second, nothing smaller than 2 is an upper bound: for any ε > 0 we need some element above 2 − ε. Now 2 − 1/n > 2 − ε exactly when 1/n < ε, that is n > 1/ε.
Proof. For every n ∈ ℕ, 2 − 1/n < 2, so 2 is an upper bound of S. Let ε > 0. Choose n ∈ ℕ with n > 1/ε. Then 1/n < ε, so 2 − 1/n > 2 − ε, and 2 − ε is not an upper bound. Since this holds for every ε > 0, no number below 2 is an upper bound, so sup S = 2. ∎
Commentary. Notice that 2 is not in S, so S has no maximum. The supremum need not belong to the set, and that is exactly why the two-part template exists.
Proof 3: a convergent sequence is bounded
Claim: if aₙ → L, then there is M with |aₙ| ≤ M for every n.
Scratch work. The definition only controls the tail of the sequence, after some N. Before N there are finitely many terms, and finitely many numbers always have a largest. So: use the definition once with a convenient ε, say ε = 1, then deal with the first few terms separately.
Proof. Apply the definition with ε = 1: there is N such that |aₙ − L| < 1 for all n ≥ N. For those n, by the triangle inequality,
Size of the term: |aₙ| = |(aₙ − L) + L| ≤ |aₙ − L| + |L| < 1 + |L|.
Let M be the largest of the numbers |aₙ| for n = 1, 2, …, N − 1, together with 1 + |L|. This is a maximum of finitely many numbers, so it exists. Every term with n < N is bounded by M because it appears in that list, and every term with n ≥ N is bounded by 1 + |L| ≤ M. ∎
Commentary. Here you get to choose ε, because the definition is a hypothesis you are using rather than a goal you are proving. Students often mix up these two roles. When convergence is given, you may pick any ε that helps.
Proof 4: showing a sequence diverges
Claim: the sequence aₙ = (−1)ⁿ does not converge.
Scratch work. Consecutive terms are 1 and −1, which are 2 apart. If every term after N were within 1 of some L, two consecutive terms would be less than 2 apart. That gives the contradiction, and suggests ε = 1.
Proof. Suppose, for contradiction, that aₙ → L for some real L. Taking ε = 1, there is N with |aₙ − L| < 1 for all n ≥ N. Then aₙ and aₙ₊₁ with n = N are one 1 and one −1, so
2 = |aₙ − aₙ₊₁| ≤ |aₙ − L| + |L − aₙ₊₁| < 1 + 1 = 2,
which is impossible. So (−1)ⁿ does not converge. ∎
Commentary. This is the negation from the table in action: there is an ε (namely 1) for which no N works.
Where students lose marks
- N depends on n. "Choose N = n + 1" makes no sense; N must be fixed before n is chosen.
- Presenting the scratch work as the proof. Working backwards from |aₙ − L| < ε and ending with "n > 5/ε, true" proves the wrong direction. Rewrite it forwards.
- "For large n" with no N. Say how large, with a formula.
- Choosing ε in a proof of convergence. You may only choose ε when convergence is assumed, as in proof 3.
- Assuming what you want. Using the limit's value inside the argument for that limit is circular.
- Unjustified inequality steps. Each "<" or "≤" in the chain should be obvious or have a reason next to it.
Practice set
- Prove (2n + 3)/(n + 1) → 2.
- Prove 1/√n → 0.
- Find sup and inf of {n/(n + 1) : n ∈ ℕ}, with proof.
- Prove (n + 5)/(n² + 1) → 0.
- Prove (n² + 1)/n does not converge.
Answers (key step). (1) The distance is 1/(n + 1) < 1/n, so take N > 1/ε. (2) Take N > 1/ε². (3) sup = 1 (not attained, use n > 1/ε − 1), inf = ½ (attained at n = 1). (4) For n ≥ 1, n + 5 ≤ 6n and n² + 1 > n², so the term is below 6/n; take N > 6/ε. (5) The terms exceed n, so the sequence is unbounded, and by proof 3 it cannot converge.
A weekly routine that works
Keep one sheet with every definition from the course, written exactly, each with an example and a non-example. Before each problem sheet, rewrite the relevant definitions from memory. For each proof, write the scratch work on the left of the page and the final proof on the right. When you read a proof in lectures, cover it, try the first line yourself, then compare.
The same shift from computing to proving happens in linear algebra, which is why our refreshed piece on why linear algebra feels hard gives similar advice. UK students moving up from A Level can also read about the step from A Level to first-year university maths. Later in an analysis course these templates carry series convergence (see the series convergence tests flowchart), continuity, the derivative and the Riemann integral.
How Maths Bodhi helps
We read your attempted proofs line by line, show where a quantifier or inequality slipped, and set new problems on the templates you find hardest. Other courses we cover are on our university maths hub. We explain, practise and review your own work; we don't write graded assignments or sit exams for anyone. Home tuition is available in Gurgaon (Gurugram). Students elsewhere in India and abroad learn with us online. You can book a free demo class.
Written and checked by the Maths Bodhi team (November 2026).
Frequently asked questions
- How do I start a real analysis proof?
- Write the exact definition you must satisfy, do scratch work to find N or δ, then write the proof forwards from "Let ε > 0".
- Why is real analysis so hard?
- It asks you to prove results calculus let you assume, using exact definitions, quantifiers and inequalities rather than computation.
- What should I know before taking real analysis?
- Calculus, comfort with inequalities and absolute values, and basic proof methods such as induction and contradiction.
- What is the difference between the scratch work and the proof?
- Scratch work runs backwards to find N or δ; the proof runs forwards from ε and is what you hand in.
- Can the supremum of a set not be in the set?
- Yes. The set {2 − 1/n : n ∈ ℕ} has supremum 2, which is not an element.
- How many hours a week should I spend on real analysis?
- Enough to rewrite the definitions from memory and attempt every problem sheet question yourself before the class discusses it.
