A courier charges £3 for a parcel of up to 2 kg, and £5 for a parcel over 2 kg and up to
10 kg. For a parcel of w kilograms, where 0<w≤10, the price in pounds is
What does the price do near 2 kg? A parcel of 1.999 kg costs £3 and one of 2.001 kg costs £5,
and however close to 2 kg the two weights are, the two prices stay £2 apart. So P(w)
approaches no single number as w approaches 2: limw→2P(w) does not exist, as we
prove in Example 2. But “does not exist” throws away what we
know. Just below 2 kg the price is £3, and just above it is £5. We would like to say that
P(w) approaches 3 as w approaches 2 from the left, approaches 5 from the right, and so
jumps by £2 at 2 kg.
A second situation has the same cure. The definition of the limit asks for
f to be defined on an open interval around a, except possibly at a. The square root
x is defined for no x<0, because a negative number has no real square root
(the square-root remark), so that definition says nothing about
x at 0. Yet x is as small as we like when x>0 is small enough:
0.0001=0.01.
A one-sided limit looks at one side of a only. It describes the jump of the price, and
the square root at 0, and it gives a practical test for whether a limit exists: look at the
two sides separately, then compare.
Informally, limx→a+f(x)=L means that f(x) is as close to L as we like for
all x close enough to a and greater thana. The window of the ε–δ game becomes the
right half of the window of the definition of the limit: the punctured
neighbourhood 0<∣x−a∣<δ is the set (a−δ,a)∪(a,a+δ), by
part (e) of the proposition on distance inequalities, and a
one-sided limit keeps only one of the two intervals.
In words. The right-hand limit is the definition of the limit with its
window 0<∣x−a∣<δ replaced by the right half, a<x<a+δ; the
left-hand limit keeps the left half, a−δ<x<a. The game is the same: a challenger
names ε>0, you reply with δ>0, and you win the round if every x in your
half-window has f(x) within ε of L. Three points to notice:
The values of f on the other side of a play no part either. For the right-hand limit, f
need not even be defined to the left of a: that is what makes limx→0+x
meaningful.
As before, if a δ wins a round, so does every smaller δ>0. Since f is
defined on (a,a+r) (in part (a)), every δ≤r gives a half-window inside that
interval, so the requirement that f be defined in the window only rules out windows that
are too wide.
Example. Let g(x)=x for x≤0 and g(x)=x+1 for x>0. Then
limx→0+g(x)=1. Given ε>0, take δ=ε. If 0<x<δ, then
g(x)=x+1, so ∣g(x)−1∣=∣x∣=x<ε, where ∣x∣=x because x>0
(the definition of the absolute value).
Non-example.limx→0+g(x) is not 0, although g(0)=0. Take
ε=21. Whatever δ>0 is offered, let d=21min(δ,1),
which is positive. Since min(δ,1)≤δ, multiplying by the positive number
21 keeps the inequality, so d≤2δ<δ, and the half-window
0<x<δ contains x=d. There g(d)=d+1, and adding 1 to d>0 keeps the
inequality, d+1>1; so d+1 is positive, and
∣g(d)−0∣=d+1>1>21 by
the definition of the absolute value. So no δ wins the round
ε=21. (From
the left, g(x)=x, and limx→0−g(x)=0, with δ=ε as in the
example.)
The figure plays the game of the definition of the limit for the parcel
price of Why this matters, with the two-sided claim limw→2P(w)=5. It reports on
the two sides of 2 separately, the left half-window and the right half-window, which is
exactly what the one-sided limits do.
Figure 1:The ε–δ game for the claim that the parcel price has the limit 5 at 2 kg. Graph of the
price for weights from 1 kg to 3 kg: height 3 to the left of 2 and height 5 to the
right, with an open circle at (2,5), the claimed limit. Around it are a horizontal band of
half-width ε around y=5 (translucent, dashed edges) and a vertical window
0<∣x−2∣<δ (translucent, solid edges; the dotted line x=2 itself is left
out, so the widget never uses the price at exactly 2 kg). A slider sets ε from 0.1 to
2.5 in steps of 0.1, starting at 0.5.
For each ε the widget states, separately for the left and the right of 2, the largest
δ it found or that it observed none. For ε=0.5 it reports on the right that the
values stayed in the band at every point it checked, out to the edge of the view (δ=1),
and on the left that it observed no δ: the values leave the band at points as close to
2 as it checks. A second slider sets your own δ; points of the graph inside the window
but outside the band are marked with crosses, and a sentence says whether your δ passed
the widget’s check at the points it tested, or names a point where it fails.
Try this:
At ε=0.5, read what the widget reports on each side of 2. Which side’s values are
within ε of 5?
Select the ε slider and raise ε one step at a time. At which ε does the widget
first find a δ on the left? Why there, and not one step earlier?
Now the widget finds a δ on both sides. Does that make 5 the limit of P(w) as
w→2? (The definition asks for a δ for everyε>0.)
The widget checks finitely many points, so what it reports are observations. The proofs are
on this page: Example 2 proves that the right-hand limit is 5 and
the left-hand limit is 3, so that limw→2P(w) does not exist, and exercise
Exercise 8 finds exactly the ε for which some δ
works on the left.
Could a function have two different right-hand limits at the same point? No, by the argument
of the uniqueness of limits, now with a point taken in the
half-window.
So we may speak of the right-hand limit and the left-hand limit, when they exist.
The two half-windows together make up the punctured neighbourhood 0<∣x−a∣<δ.
So a limit should exist exactly when f(x) settles near the same number on both sides.
By part (e) of the proposition on distance inequalities, the window
0<∣x−a∣<δ is made of the two half-windows a−δ<x<a and
a<x<a+δ. So a δ that works for the limit works on each side, and
conversely the smaller of a δ for each side works for the limit.
The one-sided limits make sense. By hypothesis, f is defined at every point of an open
interval I containing a, except possibly at a. By
the definition of an interval, I is (c,d), (c,∞),
(−∞,d) or R, and since a∈I, we have c<a whenever I has a left endpoint
c, and a<d whenever it has a right endpoint d. Let r be the smaller of the numbers
a−c and d−a that occur (one of them, if I has only one endpoint), and r=1 if
I=R. Then r>0: adding −c to c<a keeps the inequality, 0<a−c, and adding
−a to a<d gives 0<d−a. Every x with a−r<x<a+r lies in I. Adding
c−r to r≤a−c keeps the inequality, c≤a−r, so c<x by transitivity; and
adding a to r≤d−a gives a+r≤d, so x<d by transitivity. So f is
defined at every point of (a−r,a) and of (a,a+r), as parts (b) and (a) of
Definition 1 require.
If the limit is L, so are both one-sided limits. Let ε>0. Since
limx→af(x)=L, there is a δ>0 such that f is defined at every x with
0<∣x−a∣<δ, and ∣f(x)−L∣<ε for each such x. If
a<x<a+δ, then 0<∣x−a∣<δ, by part (e) of the proposition, read from
right to left. So f(x) is defined and ∣f(x)−L∣<ε: this δ wins the round
ε for the right-hand limit. In the same way, a−δ<x<a implies
0<∣x−a∣<δ, so the same δ wins the round for the left-hand limit. Since
ε>0 was arbitrary, limx→a+f(x)=L and limx→a−f(x)=L.
If both one-sided limits are L, so is the limit. Let ε>0. By part (b) of the
definition there is a δ1>0 such that f is defined and ∣f(x)−L∣<ε at
every x with a−δ1<x<a; by part (a) there is a δ2>0 such that the same
holds at every x with a<x<a+δ2. Let δ=min(δ1,δ2), which
is positive, and let 0<∣x−a∣<δ. By part (e) of the proposition, read from left
to right, either a−δ<x<a or a<x<a+δ.
In the first case, multiplying δ≤δ1 by the negative number −1 reverses
it, −δ1≤−δ, and adding a keeps it: a−δ1≤a−δ<x<a.
So x is in the half-window of δ1.
In the second case, adding a to δ≤δ2 gives
a<x<a+δ≤a+δ2, so x is in the half-window of δ2.
Either way, f(x) is defined and ∣f(x)−L∣<ε. So δ wins the round ε
for the limit; since ε>0 was arbitrary, limx→af(x)=L.
The theorem gives the limit when the two one-sided limits agree. Read the other way, it is the
usual way to show that a limit does not exist.
How to use them. To decide whether limx→af(x) exists, find both one-sided limits
and compare them. If they are equal, the limit is their common value, by
Theorem 1. If they are different, or one of them does not exist,
there is no limit, by part (a) or part (b) of Corollary 1.
A function given by different formulas on the two sides of a is made for one-sided limits:
each half-window sees only one formula. The value at a itself, whichever formula or special
value gives it, plays no part.
The jump of x∣x∣ at 0, the first of the three ways a limit can fail to exist
in The Limit of a Function, fits the same pattern: its one-sided limits are
−1 and 1 (exercise Exercise 1), so part (a) of
Corollary 1 gives again what
the jump example proved there with the triangle inequality.
Here domf is the domain of f. To negate, each ∀
becomes ∃ and each ∃ becomes ∀, the implication P⟹Q becomes
“P and not Q”, “not (A and B)” becomes “(not A) or (not B)”, and “not
(u<v)” becomes “u≥v”, by trichotomy (the order rules of
Absolute Value and Inequalities, “Working with
inequalities”). So L is not the right-hand limit when
This is the pattern of the non-example after Definition 1: ε=21,
and for each δ the point x=21min(δ,1). For the left-hand limit, replace
a<x<a+δ by a−δ<x<a.
Why f must be defined on one side of a
Part (a) of Definition 1 asks for f to be defined on an interval
(a,a+r), and, in its domain clause, at every point of the half-window. The reason in
Remark 1 rests on that clause: it picks the point
x=a+2δ of the half-window and uses f(x).
Suppose the definition dropped both requirements, and only asked that “x∈domf and
a<x<a+δ imply ∣f(x)−L∣<ε”. Take f(x)=−x, which is defined
exactly for x≤0, and a=0. No x in its domain satisfies 0<x<δ, so the
condition holds for everyL, and every real number would be “the right-hand limit” of
−x at 0. Our definition does not apply there: −x is defined on no interval
(0,r). It does apply from the left, where limx→0−−x=0, by the argument
of Example 3 with −x in place of x.
f is defined at every point of (−1,0) and of (0,1), so both parts of
Definition 1 apply.
(a) For x<0, ∣x∣=−x (the definition of the absolute value),
so f(x)=x−x=−1. Given ε>0, let δ=1: if −1<x<0, then
∣f(x)−(−1)∣=0<ε. So limx→0−f(x)=−1.
(b) For x>0, ∣x∣=x, so f(x)=xx=1, and δ=1 works in the same
way: limx→0+f(x)=1.
The two one-sided limits differ, so by part (a) of Corollary 1,
limx→0f(x) does not exist, as the jump example proved
directly from the definition. The jump (Remark 2) is
1−(−1)=2.
f is defined at every real number, and f(1)=0 plays no part in (a), (b) or (d), because
no window contains 1.
(a) For x<1, ∣(2x+1)−3∣=∣2(x−1)∣=2∣x−1∣, by
property 4 of the absolute value and ∣2∣=2.
Given ε>0, let δ=2ε, and let 1−δ<x<1. Since
x−1<0, ∣x−1∣=−(x−1)=1−x by
the definition of the absolute value. Adding δ−x to
1−δ<x keeps the inequality, 1−x<δ, and adding −x to x<1 gives
0<1−x. So 0<∣x−1∣<2ε, and multiplying by the positive number 2
keeps the inequality: ∣f(x)−3∣<ε. So limx→1−f(x)=3.
(b) For x>1, ∣(4−x)−3∣=∣1−x∣=∣x−1∣=x−1, by
property 2 of the absolute value and because x−1>0.
Given ε>0, let δ=ε: if 1<x<1+ε, then
∣f(x)−3∣=x−1<ε. So limx→1+f(x)=3.
(c) f(1)=0, by the definition of f.
(d) Both one-sided limits are 3, so limx→1f(x)=3 by
Theorem 1, although f(1)=0.
For x>3, x−3 is defined and non-negative. By the order part of
the square-root remark, for the non-negative numbers x−3
and 0.1, x−3<0.1 exactly when (x−3)2<0.12, that is,
x−3<0.01. So every δ≤0.01 works. A larger δ does not: its half-window
contains x=3.01, where x−3=0.1, which is not less than 0.1. The largest
δ is 0.01=1001.
This is the round ε=0.1 of limx→3+x−3=0. Only the right-hand
limit is defined here: x−3 is undefined for every x<3.
It is false. Let H(x)=0 for x<0 and H(x)=1 for x≥0, and a=0, so H is
defined at every real number and H(0)=1.
For x>0, H(x)=1, so ∣H(x)−1∣=0<ε for every ε>0 (with δ=1,
say): limx→0+H(x)=1=H(0). For x<0, H(x)=0, and in the same way
limx→0−H(x)=0. The one-sided limits differ, so by part (a) of
Corollary 1, limx→0H(x) does not exist; in particular it
is not H(0). A right-hand limit says nothing about the left of a.
C is defined at every point of (0,3), an open interval containing 1. The value
C(1)=2, from the first line, plays no part in either one-sided limit, because neither
half-window contains 1.
(a) For 0<t<1, C(t)=2. Given ε>0, δ=1 works: if 0<t<1, then
∣C(t)−2∣=0<ε. So limt→1−C(t)=2.
(b) For 1<t<2, C(t)=4, and δ=1 works in the same way:
limt→1+C(t)=4.
(c) By Remark 2, the jump is 4−2=2: a stay just over an
hour costs £2 more than one just under. Since the one-sided limits differ,
limt→1C(t) does not exist, by part (a) of Corollary 1.
The value g(2)=3⋅2−1=5, from the second formula, plays no part: none of the three
windows contains x=2, so that g(2) equals 5 helps no δ.
(a) For x>2, ∣g(x)−5∣=∣3(x−2)∣=3∣x−2∣ by
property 4 of the absolute value, and
∣x−2∣=x−2 because x−2>0
(the definition of the absolute value). Dividing by the positive
number 3 keeps the inequality, so 3(x−2)<0.03 exactly when x−2<0.01.
So every δ≤0.01 works, and a larger one does not: its half-window contains
x=2.01, where ∣g(x)−5∣=0.03. The largest is 0.01=1001.
(b) For x<2, ∣g(x)−5∣=∣x−2∣=−(x−2)=2−x, because x−2<0
(the definition of the absolute value). Adding x−0.03 to both
sides keeps an inequality, so 2−x<0.03 exactly when x>1.97. So every
δ≤0.03 works, and a larger one does not: its half-window contains x=1.97,
where ∣g(x)−5∣=0.03. The largest is 0.03=1003.
(c) By part (e) of the proposition on distance inequalities,
0<∣x−2∣<δ means 2−δ<x<2 or 2<x<2+δ. So δ works
for the window exactly when it works for both half-windows, that is, when δ≤0.01 and
δ≤0.03. The largest is the smaller number, 1001: the choice
δ=min(δ1,δ2) of the proof of Theorem 1.
f is defined at every real number. The value f(1)=7 plays no part, because neither
half-window contains 1.
From the right. For x>1, ∣(5−2x)−3∣=∣−2(x−1)∣=2∣x−1∣, by
property 4 of the absolute value and ∣−2∣=2. Given
ε>0, let δ=2ε: if 1<x<1+δ, then
0<∣x−1∣<2ε, and multiplying by the positive number 2 keeps the
inequality, so ∣f(x)−3∣<ε. So limx→1+f(x)=3.
From the left. For x<1, ∣(cx+1)−(c+1)∣=∣c(x−1)∣=∣c∣∣x−1∣,
by property 4 of the absolute value. If c=0, this is
0<ε for every x, and any δ works. If c=0, then ∣c∣>0; let
δ=∣c∣ε. If 1−δ<x<1, then ∣x−1∣<∣c∣ε,
and multiplying by the positive number ∣c∣ gives ∣f(x)−(c+1)∣<ε. Either
way, limx→1−f(x)=c+1.
The limit. If c+1=3, that is, c=2, both one-sided limits are 3, and
limx→1f(x)=3 by Theorem 1. If c=2, the one-sided
limits c+1 and 3 differ, and the limit does not exist, by part (a) of
Corollary 1. So
the limit exists exactly when c=2.
For 0<w<2, P(w)=3, so ∣P(w)−5∣=2 at every point of every half-window
2−δ<w<2 with δ≤2.
If ε>2, then δ=1 works, because ∣P(w)−5∣=2<ε for 1<w<2.
If 0<ε≤2, no δ works: every half-window contains the point
w=2−21min(δ,1), which lies in (1,2), and there
∣P(w)−5∣=2≥ε.
So the set is (2,∞). On the slider, whose values are 0.1,0.2,…,2.5, the widget
observes no δ on the left up to ε=2 and finds one from ε=2.1 on. Since the
definition needs a δ for every ε>0, and already ε=2 has none,
limw→2−P(w)=5: from the left, the limit is 3
(Example 2).
It is false. The function x1 is defined at every point of (0,1), so part (a) of
Definition 1 applies. Let L be any real number. We show that no δ wins
the round ε=1.
Let δ>0, and let x=min(2δ,∣L∣+21). Both
numbers are positive, so 0<x≤2δ<δ: the point x is in the
half-window. Also 0<x≤∣L∣+21. Multiplying this inequality by the
positive number x∣L∣+2 keeps it: ∣L∣+2≤x1. By
property 3 of the absolute value, L≤∣L∣;
multiplying by −1 reverses this, −L≥−∣L∣, and adding x1 keeps it.
Adding −∣L∣ to ∣L∣+2≤x1 gives x1−∣L∣≥2. So
So x1−L is positive and, by the definition of the absolute
value, ∣∣x1−L∣∣=x1−L≥2>1. The
point x fails the round, so no δ works for ε=1, and L is not the right-hand
limit. Since L was arbitrary, the right-hand limit does not exist.
The function x1 is defined at every point of (0,1), so part (b) of
Definition 1 applies at a=1.
Scratch work. For 0<x<1, x1−1=x1−x, a quotient of two
positive numbers, so it is positive and ∣∣x1−1∣∣=x1−x. If
21<x, multiplying by the positive number x2 keeps the inequality:
x1<2. Then multiplying by the positive number 1−x gives
x1−x<2(1−x), which is less than ε when 1−x<2ε.
Proof. Let ε>0 and δ=min(21,2ε), and let
1−δ<x<1. Since δ≤21, multiplying by −1 reverses the inequality
and adding 1 keeps it, so 1−δ≥21 and 21<x<1. Then x1 is
defined, x1<2, and 0<1−x<δ≤2ε, so, by the scratch work,