Femtobytes
Problem type

Limit of a function from the epsilon-delta definition

What it solves
Prove that a polynomial (a straight line, a quadratic or a cubic) has the limit you expect at a point, using only the epsilon-delta definition: for every output tolerance epsilon, find an input tolerance delta that guarantees it.
Subject
Real Analysis
Topic
Limits from the definition
Also called
epsilon delta, limit of a function, continuity from the definition
Your problem

Prove this limit using only the epsilon-delta definition, then try the proof out with the tolerance ε=110\varepsilon = \frac{1}{10}.

lim⁡x→2(x2−3x+5)=3\lim_{x \to 2} \left( x^{2} - 3 x + 5 \right) = 3
Invent a new one:
The problem

Prove this limit using only the epsilon-delta definition, then try the proof out with the tolerance ε=110\varepsilon = \frac{1}{10}.

lim⁡x→2(x2−3x+5)=3\lim_{x \to 2} \left( x^{2} - 3 x + 5 \right) = 3
The result
lim⁡x→2(x2−3x+5)=3\lim_{x \to 2} \left( x^{2} - 3 x + 5 \right) = 3

Proved. Given any tolerance ε>0\varepsilon > 0, the choice δ=min⁡{1,  ε2}\delta = \min\left\{ 1, \; \frac{\varepsilon}{2} \right\} works.

checked 18 lines of this solution were evaluated with exact arithmetic before it was shown to you. A solution with a false line is never served.
Picture 1 of 6Step 1: Find the candidate limit, then say exactly what must be proved
L + εL = 3L − εc − δc = 2c + δinput xoutput f(x)
The graph of the function near the point. The upright strip is the input window of half-width delta; the level band is the output tolerance of half-width epsilon. Inside the strip the graph never leaves the band. Here c=2c = 2, L=3L = 3, ε=110\varepsilon = \frac{1}{10} and δ=120\delta = \frac{1}{20}. The bright part of the graph is the part inside the strip. The hollow circle marks the point itself, which the definition leaves out.
1

Find the candidate limit, then say exactly what must be proved

Why this step

The definition can only CHECK a number we propose; it never hands us one. For a polynomial the natural candidate is the value at the point itself, because a polynomial is built from nothing but adding and multiplying, and neither of those can make a sudden jump. That is a belief. The proof turns it into a fact.

Put x=2x = 2 into the function:

22−3⋅2+5\displaystyle 2^{2} - 3 \cdot 2 + 5==4−3⋅2+5\displaystyle 4 - 3 \cdot 2 + 5
Work out each power of 22 first. Powers come before multiplying, and multiplying before adding. checked
==4−6+5\displaystyle 4 - 6 + 5
Do each multiplication. checked
==3\displaystyle 3
Add the numbers up. checked
Claim: lim⁡x→2(x2−3x+5)=3\text{Claim: } \lim_{x \to 2} \left( x^{2} - 3 x + 5 \right) = 3
This is what we will prove, with c=2c = 2 and L=3L = 3.
2

Rewrite everything in terms of the distance from c

Why this step

We control one thing only: how far xx is from 22. But ff is written in powers of xx, which hides that distance. So we give the distance a name, h=x−2h = x - 2, and rewrite the function in powers of hh. After this the quantity we control is the only letter on the page.

From h=x−2h = x - 2 we get x=2+hx = 2 + h. The condition 0<∣x−c∣<δ0 < \left| x - c \right| < \delta becomes simply 0<∣h∣<δ0 < \left| h \right| < \delta.

x2−3x+5−3\displaystyle x^{2} - 3 x + 5 - 3==(2+h)2−3(2+h)+5−3\displaystyle \left( 2 + h \right)^{2} - 3 \left( 2 + h \right) + 5 - 3
Replace every xx by 2+h2 + h. checked
==(4+4h+h2)−3(2+h)+5−3\displaystyle \left( 4 + 4 h + h^{2} \right) - 3 \left( 2 + h \right) + 5 - 3
Multiply out the square: (u+h)2=u2+2uh+h2(u + h)^2 = u^2 + 2uh + h^2 with u=2u = 2. checked
==h2+4h+4−3h−6+5−3\displaystyle h^{2} + 4 h + 4 - 3 h - 6 + 5 - 3
Multiply each bracket by the number in front of it. A minus sign in front reaches every term inside. checked
==h2+h\displaystyle h^{2} + h
Collect the terms with the same power of hh. The plain numbers add up to exactly 00, and that is no accident: they are f(2)−Lf(2) - L, and we CHOSE L=f(2)L = f(2). A wrong candidate for LL would leave a number here that no δ\delta could ever shrink. checked
==h(h+1)\displaystyle h \left( h + 1 \right)
Every remaining term contains hh, so take hh out as a common factor. checked
∣x2−3x+5−3∣=∣h∣⋅∣h+1∣\left| x^{2} - 3 x + 5 - 3 \right| = \left| h \right| \cdot \left| h + 1 \right|
Take absolute values. The absolute value of a product is the product of the absolute values. checked
3

Cage the second factor

Why this step

∣h+1∣\left| h + 1 \right| depends on hh, so it has no single value. But we do not need its value, only a ceiling it can never pass. A ceiling exists only if hh is kept inside some fixed region, so we make a promise now: whatever δ\delta we finally choose, it will be at most 11. Then ∣h∣<1\left| h \right| < 1.

∣h+1∣\displaystyle \left| h + 1 \right|≤\le∣h∣+∣1∣\displaystyle \left| h \right| + \left| 1 \right|
The triangle inequality. checked
==∣h∣+1\displaystyle \left| h \right| + 1
In each term, the absolute value of a product is the product of the absolute values. checked
≤\le1+1\displaystyle 1 + 1
Because ∣h∣<1\left| h \right| < 1: a number below 11 only gets smaller when raised to a power, so every ∣h∣\left| h \right| and ∣h∣2\left| h \right|^2 is at most 11. checked
==2\displaystyle 2
Add the numbers up. checked
4

Rough work: choose delta

Why this step

Inside the cage the output distance is at most 2∣h∣2 \left| h \right|. That is now a straight-line problem: a fixed number times the input distance. We ask how small ∣h∣\left| h \right| must be to push it under ε\varepsilon.

2∣h∣<ε2 \left| h \right| < \varepsilon
What would be enough.
∣h∣<ε2\left| h \right| < \frac{\varepsilon}{2}
Divide both sides by 22, which is positive, so the sign stays. checked
Choose δ=min⁡{1,  ε2}\text{Choose } \delta = \min\left\{ 1, \; \frac{\varepsilon}{2} \right\}
min⁡\min means the smaller of the two numbers. It is positive because both 11 and ε2\frac{\varepsilon}{2} are.
5

The proof, written forwards

Why this step

Rough work convinces us; a proof must convince the sceptic, in the order the definition demands: ε\varepsilon arrives first, we name δ\delta, then ANY permitted xx must obey. Every relation below is justified.

Let ε>0\varepsilon > 0 be given. Choose δ=min⁡{1,  ε2}\delta = \min\left\{ 1, \; \frac{\varepsilon}{2} \right\}. Suppose 0<∣x−2∣<δ0 < \left| x - 2 \right| < \delta, and write h=x−2h = x - 2.

Since δ≤1\delta \le 1 we have ∣h∣<1\left| h \right| < 1, so the cage holds. Since δ≤ε2\delta \le \frac{\varepsilon}{2} we have ∣h∣<ε2\left| h \right| < \frac{\varepsilon}{2}. Now the distance itself:

∣x2−3x+5−3∣\displaystyle \left| x^{2} - 3 x + 5 - 3 \right|==∣h∣⋅∣h+1∣\displaystyle \left| h \right| \cdot \left| h + 1 \right|
The rewritten distance. checked
≤\le∣h∣⋅2\displaystyle \left| h \right| \cdot 2
The cage: the second factor is at most 22. Multiplying the non-negative number ∣h∣\left| h \right| by something larger cannot make the product smaller. checked
<<ε2⋅2\displaystyle \frac{\varepsilon}{2} \cdot 2
Because ∣h∣<ε2\left| h \right| < \frac{\varepsilon}{2}, and multiplying both sides by the positive number 22 keeps the sign. checked
==ε\displaystyle \varepsilon
The 22 on top cancels the 22 underneath. checked
6

Test the rule with a real tolerance

Why this step

A rule for δ\delta that has never been run on a number is only a claim. We play the sceptic ourselves, pick ε=110\varepsilon = \frac{1}{10}, and check inputs spread across the whole permitted window.

With ε=110\varepsilon = \frac{1}{10} the rule δ=min⁡{1,  ε2}\delta = \min\left\{ 1, \; \frac{\varepsilon}{2} \right\} gives δ=120\delta = \frac{1}{20}, which is 0.050.05. Every xx below is inside the window 0<∣x−2∣<δ0 < \left| x - 2 \right| < \delta, and the outputs are computed exactly.

xxf(x)f(x)∣f(x)−L∣\left| f(x) - L \right|under ε\varepsilon?
1.95051.95052.952950252.952950250.047049750.04704975yes
1.9751.9752.9756252.9756250.0243750.024375yes
2.0052.0053.0050253.0050250.0050250.005025yes
2.0252.0253.0256253.0256250.0256250.025625yes
2.04952.04953.051950253.051950250.051950250.05195025yes
Inputs across the window, tolerance 1/10. Decimals are rounded for display; the comparison uses exact fractions.