WhatsApp us
University Maths18 Nov 2026

The Epsilon-Delta Definition of a Limit: How to Write the Proof

What the epsilon-delta definition of a limit actually says, a two-stage method for writing the proof, five graded worked examples (linear to non-existence), the δ-choice rules for each function type and practice with answers.

The Epsilon-Delta Definition of a Limit: How to Write the Proof

The epsilon-delta definition says that lim (x→a) f(x) = L means: for every ε > 0 there is a δ > 0 such that whenever 0 < |x − a| < δ, we have |f(x) − L| < ε. An epsilon-delta proof is a recipe for δ: you are handed any ε, and you must produce a δ (usually a formula in ε) that forces f(x) to within ε of L. Below: how to find that recipe and write it up for full marks.

Reading the definition one phrase at a time

Break the sentence into four parts.

  • "For every ε > 0": someone else chooses how close to L they want f(x) to be. Your argument must work for every positive value, however tiny.
  • "there is a δ > 0": your job. δ may depend on ε (and on a), but never on x.
  • "whenever 0 < |x − a| < δ": x lies within δ of a, but x ≠ a. The value f(a), or whether f is even defined at a, plays no part.
  • "|f(x) − L| < ε": the output lands inside the target band.

The definition is the formal backbone of Calculus I limits and returns in proof-based courses. Our online college maths tutoring for US students page explains how sessions with Ajay Vatsyayan work for students meeting these proofs for the first time.

The two-stage method: scratch work, then the proof

Every epsilon-delta proof has two stages, and only the second is handed in.

Stage 1, scratch work (backwards). Start from |f(x) − L|, simplify it, and try to show it is at most (some constant) × |x − a|. Once you have |f(x) − L| ≤ K|x − a|, the choice δ = ε/K will work. If the "constant" depends on x, restrict δ ≤ 1 first so that you can bound it.

Stage 2, the written proof (forwards). Begin with "Let ε > 0. Choose δ = ...". Then assume 0 < |x − a| < δ and show, line by line, that |f(x) − L| < ε. End with a sentence that names the conclusion.

Markers are looking for the forward direction. Scratch work that starts from |f(x) − L| < ε and ends at "so δ = ε/4" is discovery, not proof.

Worked example 1: a linear function

Prove that lim (x→3) (4x − 5) = 7.

Scratch work:

Distance from L: |4x − 5 − 7| = |4x − 12| = 4|x − 3|

We need 4|x − 3| < ε, i.e. |x − 3| < ε/4. So take δ = ε/4.

Proof. Let ε > 0 and choose δ = ε/4. If 0 < |x − 3| < δ, then |(4x − 5) − 7| = 4|x − 3| < 4δ = ε. Hence lim (x→3) (4x − 5) = 7. ∎

For any linear function mx + c with m ≠ 0, the same steps give δ = ε/|m|: steeper lines need smaller δ.

Worked example 2: a quadratic at a negative point

Prove that lim (x→−1) (2x² − 3x) = 5.

Scratch work. Factor the difference so that (x − a) appears:

2x² − 3x − 5 = (x + 1)(2x − 5), so |f(x) − 5| = |x + 1| · |2x − 5|

The factor |2x − 5| depends on x, so bound it. Agree in advance that δ ≤ 1. Then |x + 1| < 1 means −2 < x < 0, so −9 < 2x − 5 < −5 and |2x − 5| < 9. Now |f(x) − 5| < 9|x + 1|, which is less than ε when |x + 1| < ε/9.

Both restrictions must hold, so δ = min(1, ε/9).

Proof. Let ε > 0 and set δ = min(1, ε/9). Suppose 0 < |x + 1| < δ. Since δ ≤ 1, we have −2 < x < 0, so |2x − 5| < 9. Since δ ≤ ε/9,

Distance: |2x² − 3x − 5| = |x + 1| · |2x − 5| < (ε/9) · 9 = ε. ∎

Testing the δ numerically

A numerical check catches algebra slips before you write up. For each ε we took δ from the formula and found the largest |f(x) − 5| over 0 < |x + 1| < δ in Python:

εδ = min(1, ε/9)largest distance of f(x) from 5 in the δ-windowinside the band?
0.50.05560.395yes
0.10.01110.0780yes
0.010.001110.00778yes

The δ works with room to spare because 9 is a generous bound. That is fine: you need some δ that works, not the largest.

Worked example 3: a rational function

Prove that lim (x→4) 1/(x − 2) = 1/2.

Scratch work. Combine the fractions:

Distance: |1/(x − 2) − 1/2| = |2 − (x − 2)| / (2|x − 2|) = |x − 4| / (2|x − 2|)

The danger is a small denominator, so keep x well away from 2. With δ ≤ 1 we get 3 < x < 5, so |x − 2| > 1 and the distance is less than |x − 4| / 2, which is below ε when |x − 4| < 2ε. So δ = min(1, 2ε).

Proof. Let ε > 0 and δ = min(1, 2ε). If 0 < |x − 4| < δ, then 3 < x < 5, so |x − 2| > 1 and

Distance: |x − 4| / (2|x − 2|) < |x − 4| / 2 < 2ε/2 = ε. ∎

Here the restriction δ ≤ 1 does two jobs: it bounds the denominator below and keeps x away from 2, where f is undefined.

Worked example 4: a square root

Prove that lim (x→9) √x = 3.

Scratch work. Rationalise:

Distance: |√x − 3| = |x − 9| / (√x + 3) ≤ |x − 9| / 3

since √x ≥ 0. So |x − 9| < 3ε is enough. We also need x ≥ 0 for √x to exist; δ ≤ 9 keeps x inside (0, 18). Choose δ = min(9, 3ε).

Proof. Let ε > 0 and δ = min(9, 3ε). If 0 < |x − 9| < δ, then x > 0 and |√x − 3| = |x − 9| / (√x + 3) ≤ |x − 9| / 3 < ε. ∎

Worked example 5: proving a limit does not exist

Let f(x) = x² for x < 1 and f(x) = 3x − 1 for x ≥ 1. The left values approach 1 and the right values approach 2, so we expect no limit at x = 1. To prove it you need the negation of the definition:

lim (x→a) f(x) ≠ L means: there is some ε > 0 such that for every δ > 0 there is an x with 0 < |x − a| < δ but |f(x) − L| ≥ ε.

Proof. Suppose the limit were some number L. Take ε = 1/2 and let δ be the δ that is supposed to work, shrunk if necessary so that δ < 1. Pick x₁ = 1 − δ/2 and x₂ = 1 + δ/2. Both are within δ of 1, so both values would lie within 1/2 of L, giving |f(x₂) − f(x₁)| < 1 by the triangle inequality. But

f(x₁) = (1 − δ/2)² < 1 and f(x₂) = 2 + 3δ/2 > 2

so f(x₂) − f(x₁) > 1, a contradiction. No L works, and the limit does not exist. ∎

The key move is fixing ε at most half the jump.

How to choose δ: a rule for each function type

function typescratch-work movetypical δ
linear, mx + cfactor out the slope mε divided by the size of m
polynomialfactor out (x − a), bound the rest using δ ≤ 1min(1, ε/K)
rationalcombine fractions, bound the denominator below using δ ≤ 1min(1, cε)
square rootmultiply by the conjugate, use √x + √a ≥ √amin(a, √a · ε)
piecewise with a jumpnegate the definition, fix ε below half the jumpnone (the limit fails)

K and c come from your bounds, so they change from question to question. If the restriction δ ≤ 1 lets x reach a bad point (a zero of the denominator, or negative values under a root), use a smaller restriction such as δ ≤ 1/2.

Continuity is the same definition with L = f(a)

The epsilon-delta definition of continuity says f is continuous at a when, for every ε > 0, there is a δ > 0 with |x − a| < δ implying |f(x) − f(a)| < ε. Two changes from the limit version: the limit value must be f(a), and the condition 0 < |x − a| relaxes to |x − a| because x = a now does no harm.

Quick example: f(x) = x³ is continuous at 1. Here |x³ − 1| = |x − 1| · |x² + x + 1|. With δ ≤ 1 we have 0 < x < 2, so x² + x + 1 < 7, and δ = min(1, ε/7) works. When students reach uniform continuity in analysis, the question becomes whether one δ works for every a at once.

Where students lose marks

  • δ written in terms of x. "Choose δ = ε/|x + 3|" is not allowed. Bound the x-factor first, then δ depends on ε alone.
  • Forgetting the min. In example 2, δ = ε/9 by itself fails when ε is large: ε = 18 gives δ = 2, and near x = −3 the distance |f(x) − 5| reaches almost 22. Write min(1, ε/9) and say where each part is used.
  • Submitting the scratch work as the proof. Starting from |f(x) − L| < ε and "solving for x" runs the logic backwards. Rewrite it forwards.
  • Bounding in the wrong direction. For the denominator you need a lower bound (|x − 2| > 1), because a smaller denominator makes the fraction larger. Students often write an upper bound there by habit.
  • Mixing up < and ≤. If you end with |f(x) − L| ≤ ε rather than < ε, some markers deduct. Strict inequality from |x − a| < δ usually carries through; keep it.
  • Wrong negation. "For every ε there is a δ that fails" is not the negation. The correct form is "there exists an ε such that every δ fails", as in example 5.
  • Plugging in. Showing f(3) = 7 proves nothing about a limit; f(a) may not exist or may differ from L.

Practice set

Write a full proof for each. Answers give a δ that works (others are possible).

  1. lim (x→−2) (5 − 3x) = 11
  2. lim (x→1) (x² + 4x) = 5
  3. lim (x→3) 6/x = 2
  4. lim (x→4) √x = 2
  5. Show that lim (x→2) ⌊x⌋ does not exist, where ⌊x⌋ is the greatest integer ≤ x.

Answers:

  1. |5 − 3x − 11| = 3|x + 2|, so δ = ε/3.
  2. |x² + 4x − 5| = |x − 1| · |x + 5|; with δ ≤ 1, |x + 5| < 7, so δ = min(1, ε/7).
  3. |6/x − 2| = 2|x − 3| / |x|; with δ ≤ 1, x > 2, so the expression is less than |x − 3|. δ = min(1, ε).
  4. |√x − 2| = |x − 4| / (√x + 2) ≤ |x − 4| / 2, so δ = min(4, 2ε).
  5. Left of 2 the value is 1, right of 2 it is 2. Take ε = 1/2; for any δ < 1 the points 2 − δ/2 and 2 + δ/2 give values 1 and 2, which cannot both be within 1/2 of one number L.

We checked each δ numerically for several values of ε, in the same way as the table above.

What this definition leads to

The derivative is a limit of difference quotients, so its rules rest on this definition. Series convergence uses the sequence version (for every ε there is an N), which sits behind our series convergence tests flowchart. Heading into proof-based courses? Read how to start real analysis proofs; our real analysis tutoring page explains how we support those courses.

In Calculus I the applied side follows: related rates problems worked step by step, then integration by parts and the tabular method in Calculus II. Our guide to the US college math sequence shows how these courses fit together at different colleges.

How Maths Bodhi helps

Epsilon-delta proofs improve fastest when someone reads your written proof line by line and shows where the logic runs backwards or a bound points the wrong way. In online sessions we work through your own attempts, set fresh limits to prove, and build the habit of the two-stage method. We explain and review; we don't complete graded work. The university maths tutoring hub lists the other courses we support.

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

What is the epsilon-delta definition of a limit in simple words?
However close (ε) you want f(x) to be to L, you can find a distance (δ) so that every x within δ of a, except a itself, gives f(x) that close to L.
How do you find delta in an epsilon-delta proof?
Simplify |f(x) − L| into a constant times |x − a|, restricting δ ≤ 1 to bound any leftover x-terms, then take δ = min(1, ε/constant).
Can delta depend on x?
No. δ may depend on ε and the point a, but never on x.
Why do we use min(1, ε/K) for delta?
The 1 keeps x close enough to a for your bound K to be true, and ε/K makes the final inequality work; δ must satisfy both.
What is the difference between the epsilon-delta definition of a limit and of continuity?
For continuity the limit value must equal f(a), and x = a is allowed in the condition |x − a| < δ.
Are epsilon-delta proofs on Calculus 1 exams?
It depends on the course and instructor, so check your syllabus and past exams; proof-based and honors courses use them most.
epsilon delta prooflimitsCalculus Icontinuitycollege calculus

Related guides

Keep exploring this maths topic

University Maths18 Nov 2026

Series Convergence Tests: A Flowchart for Choosing the Right Test

A step-by-step flowchart for deciding which series convergence test to use, a cheat sheet table with each test's blind spot, seven worked examples, absolute versus conditional convergence, error bounds and practice with answers.

series convergence testsinfinite seriesCalculus IIratio test

By Maths Bodhi Team

Read guide

University Maths18 Nov 2026

How to Start Real Analysis Proofs: A Guide for First-Time Students

Real analysis asks you to prove what calculus let you assume. This guide shows how to read the definitions, which proof templates to learn first, and four short complete proofs with the scratch work shown separately.

real analysisproofsepsilon Nsupremum

By Maths Bodhi Team

Read guide

University Maths2 Nov 2026

The US College Math Sequence: Calculus I to III, Linear Algebra and Differential Equations

How the usual US college math sequence runs from Calculus I to Calculus III, linear algebra and differential equations, how four real universities number and bundle it, what AP credit changes, and the maths that trips students up at each step, with worked examples and practice.

college math sequenceCalculus IIIlinear algebradifferential equations

By Maths Bodhi Team

Read guide

University Maths18 Nov 2026

Integration by Parts and the Tabular Method, With Worked Examples

The integration by parts formula, how to choose u (LIATE and when to break it), the tabular method for repeated parts and for exponential-times-trig integrals, six graded worked examples, common errors and practice with answers.

integration by partstabular methodCalculus IIintegration techniques

By Maths Bodhi Team

Read guide

University Maths18 Nov 2026

Related Rates Problems, Step by Step (With Solutions)

A six-step method for related rates problems in Calculus I, six graded worked solutions (balloon, ladder, cone, cars, shadow, rocket camera), a table of the equations that link the variables, common traps and practice with answers.

related ratesCalculus Iimplicit differentiationchain rule

By Maths Bodhi Team

Read guide

University Maths2 Nov 2026

From JEE to IIT First-Year Maths: What Changes, with Worked Examples

Why strong JEE students often struggle in first-year maths at an IIT, what the current IIT courses contain, six worked examples showing exactly what changes (ε-δ, proofs, abstract linear algebra, several variables), a first-semester plan, common mistakes and practice with answers.

IIT first year mathsJEE to IIT mathsB.Tech mathslinear algebra

By Maths Bodhi Team

Read guide