ohms, for t>0, and we found what it approaches as t approaches 6. Now replace the second
resistor by larger and larger ones. What happens to the combined resistance? Rounded, in ohms:
t=10 gives R(10)≈2.3077;
t=100 gives R(100)≈2.9126;
t=1000 gives R(1000)≈2.9910;
t=1000000 gives R(1000000)≈2.999991.
The values creep up towards 3 and never reach it. Indeed, for every t>0,
which is positive, and as small as we like when t is large enough: it is less than 0.001 as
soon as t>8997. A huge resistor in parallel lets almost no current through, so the pair
behaves almost like the 3-ohm resistor alone.
This is a new kind of limit. The variable t does not approach a number a: it grows without
bound. We will write limt→∞R(t)=3, and say that the line y=3 is a
horizontal asymptote of the graph of R. We also want to describe x→−∞, and values
that grow without bound themselves, as x2 does when x grows. This page defines all of
these precisely, carries the limit laws over to them, and uses the laws to find the limits of
every rational function and of some expressions with square roots.
Informally, limx→∞f(x)=L means that f(x) is as close to L as we like for
allx large enough. The ε–δ game of the definition of the limit changes
in one place: the challenger still names a tolerance ε>0, but instead of a window
0<∣x−a∣<δ around a point a you reply with a thresholdN, and you win if
every x>N has f(x) within ε of L. The window is now the ray of all x>N. For
values that grow without bound, such as x2, the challenger names a height B instead, and
you win if f(x)>B for every x>N. Values of f at small x play no part; f only has to
be defined from some point on.
In words. Parts (a) and (b) are the game of the definition of the limit
with the punctured window 0<∣x−a∣<δ replaced by the ray x>N, or x<N.
Parts (c) and (d) replace the tolerance band around L by a height B that the values must
stay above, or below, from the threshold on. Four points to notice:
If a threshold N wins a round, so does every larger one when x→∞ (every smaller
one when x→−∞): if N′≥N and x>N′, then x>N by transitivity
(property 2 of the order rules). So N may be taken as large as we
like, and it depends on ε (or B), never on x.
Since f is defined on (c,∞) in (a), every N≥c meets the requirement that f
be defined at every x>N. That requirement only rules out thresholds that are too small,
as the window’s domain clause in the definition of the limit only rules
out windows that are too wide.
The symbols ∞ and −∞ are not real numbers. “limx→∞f(x)=∞”
does not say that a limit exists: it says how the limit fails to exist as a real number,
because the values eventually exceed every real number (property 1 of the remark
below proves this). This is the convention that the page
Infinite Limits and Vertical Asymptotes uses for infinite limits at a point a (the
proposition there proves that such a limit is not a
real number).
When we say that a limit at infinity exists, we mean that it is a real number.
On the page Infinite Limits and Vertical Asymptotes the heights are M>0; that is the same
condition, since a threshold that wins the height max(B,1) also wins B: if
f(x)>max(B,1), then f(x)>B, by transitivity. (For −∞, the heights there are
−M with M>0, and the height −max(−B,1) plays the same role.)
Example.limx→∞x1=0, by part (a). The function is defined at
every x>0, so on (0,∞). Let ε>0 and take N=ε1, which is
positive by property 6 of the order rules. If x>N, then
0<ε1<x, and property 6 gives x1<1/ε1=ε and
x1>0. So ∣∣x1−0∣∣=x1<ε, by
the definition of the absolute value.
Example.limx→∞x2=∞, by part (c). Let B be a real number and take
N to be the larger of 1 and B. If x>N, then x>1 and x>B. Multiplying x>1 by
the positive number x keeps the inequality (property 5(a) of the order
rules): x2>x. So x2>x>B, and x2>B by transitivity.
Non-example. No real number L is the limit of x2 as x→∞. Take ε=1.
Whatever N is offered, let x be 1 more than the larger of N and ∣L∣+1. Then
x>N, and x>∣L∣+1≥1, so as in the example x2>x>∣L∣+1. Adding −L
to both sides (property 3 of the order rules), and using
∣L∣−L≥0 (property 3 of the absolute value),
gives x2−L>∣L∣−L+1≥1. So x2−L is positive, and
∣∣x2−L∣∣=x2−L>1=ε: the threshold N loses the round ε=1.
The figure shows one round of the game of part (a).
Figure 1:The graph of f(x)=x2+12x2+x, with the band 1.9<y<2.1 of half-width
ε=0.1 around y=2; the vertical axis shows only the heights from 1.7 to 2.15. The
graph crosses the line y=2 at x=2, leaves the band at x=3 and returns to it at
x=7. To the right of the threshold N=7 (shaded) it stays inside the band.
For this ε, the threshold N=7 wins the round, and no smaller threshold does
(Exercise 5 proves both). A smaller ε needs a larger
N. The graph meets the line y=2 and leaves the band before it settles: the definition does
not care what happens before the threshold.
Before we use the definition, three facts that make it easier to use. The first says that we may
speak of the limit at infinity.
A limit at infinity describes the far ends of a graph. When the limit is a real number L, the
graph runs ever closer to the horizontal line y=L.
In words. Far to the right, or far to the left, the graph comes as close to the line
y=L as we like and stays that close.
Example. The line y=3 is a horizontal asymptote of the graph of the combined resistance
R of Why this matters, which is defined only for t>0, so only the direction t→∞
is available (Example 5 proves the limit). The line y=2 is a
horizontal asymptote of the graph in Figure 1: as the figure shows,
a graph may cross its horizontal asymptote.
Non-example. The graph of x2 has no horizontal asymptote. As x→∞ its limit is
∞, so by property 1 of the remark it is no real
number L. As x→−∞ the same holds, by property 3 of the remark: g(t)=(−t)2=t2.
By property 1 of the remark, a graph has at most two horizontal asymptotes, one for each
direction. Each of the two directions may give the same line, a different one, or none.
Horizontal and vertical asymptotes. The page Infinite Limits and Vertical Asymptotes defines
a vertical asymptotex=a: a vertical line near which
f(x) tends to ∞ or to −∞ as x approaches the number a from at least one side.
The roles of x and y are exchanged. At a vertical asymptote, x approaches a number and
f(x) grows or falls without bound; at a horizontal asymptote, x grows without bound and f(x)
approaches a number. A graph may have many vertical asymptotes, but at most two horizontal ones;
and it meets a vertical line x=a at most once, because f has at most one value at a, while
it may cross a horizontal asymptote. The graph of x1 has both kinds:
the line y=0, as x→∞ by the example above, and the line x=0, since
limx→0+x1=∞ by part (a) of the proposition on reciprocal
powers, with a=0 and n=1.
The limit laws of Limit Laws hold at infinity too, with the same hypotheses.
Part (a) differs: the function x has no real limit as x→∞, so the identity law
becomes a statement about ∞.
(If f and g are defined on two different intervals (α1,∞) and
(α2,∞), both are defined on the one whose left end is the larger of α1 and
α2, so one α is no restriction.)
When the laws do not apply. As on the Limit Laws page, the laws need the limits L and M of
the parts to be real numbers. If both parts tend to ∞, the laws say nothing about the
difference, “∞−∞”, or the quotient, “∞∞”, and such limits can
be anything. For every real number c, the functions x+c and x both tend to ∞, yet
their difference is the constant c, whose limit is c. The quotients x2x and
xx2 are both “∞∞”; they equal 2 and x at every x>0, so by
property 2 of the remark their limits are 2 and ∞,
by part (a) of the theorem. Such forms are a signal to rewrite the function first, as the next
proof and the examples do.
For a rational function, the leading terms decide everything: far out, q(x)p(x)
behaves like the quotient of the leading terms. We first need that the reciprocal powers
xj1 tend to 0.
In words. Compare the degrees. A higher degree in the denominator pulls the quotient to 0;
equal degrees give the quotient ba of the leading coefficients; a higher degree in the
numerator sends the quotient to ∞ or −∞, with the sign of
bxkaxm=baxm−k far out on that side. So the graph of r has the
horizontal asymptote y=0 in case (b), y=ba in case (c), in both directions, and
none in case (d).
Example.2x2+53x2−x+2 has m=k=2, so its limit in both directions is
23. Non-example. For xx2+1, m=2>k=1 and ba=1:
the limit is ∞ as x→∞, and, since (−1)1⋅1<0, −∞ as
x→−∞. There is no horizontal asymptote.
Limits at infinity let us say which of two functions grows faster. Informally, fgrows faster
thang as x→∞ if f(x)g(x)→0: eventually g(x) is a tiny fraction of
f(x).
Higher powers grow faster. If m>k≥0, then xmxk=xm−k1
for x=0, which tends to 0 by part (a) of the
proposition. So x3 grows faster than 1000x2, although
1000x2 is far larger for a long time: at x=10 it is 100000 against 1000, at
x=1000 the two are equal, and at x=10000, x3 is ten times larger.
A polynomial grows like its leading term. By part (c) of the
proposition, applied to p(x) over its leading term
axm, axmp(x)→1: the other terms become negligible in comparison.
The square root grows, but more slowly than x.x eventually exceeds every
height (Exercise 13), yet
x+1x→0 (Exercise 8).
From left to right. Let c′=max(c,1)>0. For 0<t<c′1, t1>c′,
so g is defined on (0,c′1). If N wins the round ε for f, so does
N′=max(N,1), which is positive; then δ=N′1 wins it for g: if
0<t<δ, then t1>N′, so g(t)=f(t1) is defined and
within ε of L.
From right to left. If g is defined on (0,ρ), then f(x)=g(x1) for
x>ρ1, since 0<x1<ρ there. If δ wins the round ε for
g, then N=δ1 wins it for f: if x>N, then 0<x1<δ, so
f(x)=g(x1) is defined and within ε of L.
Example. For f(x)=x2+12x2+x of Figure 1,
multiplying the numerator and the denominator by t2>0 gives
g(t)=1+t22+t for t>0. This rational function has the limit 12=2
at 0, by direct substitution, so its right-hand limit is 2
too, by the theorem on one-sided limits; the right-hand limit
uses only t>0, where it agrees with g. So
limx→∞f(x)=2, as part (c) of Proposition 1 says.
The same substitution turns limx→−∞ into limt→0−. It shows why the
two kinds of limit behave alike, and it is how some books define limits at infinity.
limx→∞f(x)=L means: for every ε>0 there is a threshold N with
∣f(x)−L∣<ε for every x>N (Definition 1). For x→−∞
the ray is x<N; for the value ∞, a height B replaces the tolerance.
“=∞” describes how a limit fails to exist as a real number; it is not a limit that
exists.
The limit laws hold at infinity, with max(N1,N2) in place of min(δ1,δ2)
(Theorem 1); they need the limits of the parts to be real, so
“∞−∞” and “∞∞” call for rewriting first.
For a rational function, compare degrees: lower over higher gives 0, equal gives the quotient
of the leading coefficients, higher over lower gives ±∞
(Proposition 1).
With square roots, multiply by the conjugate for “∞−∞”, and remember
x2=∣x∣=−x for x<0.
y=L is a horizontal asymptote if f(x)→L as x→∞ or as x→−∞
(Definition 2); a graph has at most two, and may cross them.
The numerator x4−x has degree m=4 and leading coefficient a=1. The denominator
−2x3+3 has degree k=3 and leading coefficient b=−2. So m>k and
ba=−21. As x→−∞, part (d) of Proposition 1
looks at the sign of
which is positive. So the limit is ∞. (Far to the left the function behaves like
−2x3x4=−2x, which is large and positive there; at x=−10 it is about
5.0.)
The numerator −4x2+1 has degree 2 and leading coefficient −4; the denominator has
degree 2 and leading coefficient 1. (The denominator is never 0:
x2+x+1=(x+21)2+43, where the square is ≥0 by part (c) of
the sign rules, so it is at least 43, and f is defined on R.) By part (c) of Proposition 1, both limits at infinity
are 1−4=−4, so y=−4 is the horizontal asymptote, in both directions, and there
is no other (property 1 of the remark).
Every x>7 works. Let x>7. Then x−2>0, so f(x)−2 is a quotient of two
positive numbers, positive by part (c) of the sign rules, and
∣f(x)−2∣=f(x)−2. Multiplying by the positive number 10(x2+1), the inequality
f(x)−2<0.1 holds exactly when 10(x−2)<x2+1, that is, when
For x>0, x1+x6 is non-negative and its square is x2+6x, so it equals
x2+6x by the uniqueness in the square-root remark.
Dividing the numerator and the denominator by x>0,
By part (a) of Proposition 1 and the laws of
Theorem 1, 1+x6→1>0, so by part (f)
1+x6→1, and by parts (b) and (d) the quotient tends to
1+16=3. The quotient equals the original function at every x>0, so by
property 2 of the remark the limit is 3.
The function is defined on R, since x2+3>0. For x<0, the number
−x1+x23 is non-negative, and its square is x2+3, so it is
x2+3 (the square-root remark). Dividing the numerator and
the denominator by −x>0,
As x→−∞, x1→0 and x21→0 by part (a) of
Proposition 1. By Theorem 1, as
x→−∞, the numerator 2−x1 tends to 2, and 1+x23→1>0,
so the root tends to 1 by part (f). By the quotient law and the “in particular” of part (c),
with c=−1, the limit is −12=−2; by property 2 of the
remark, so is that of the original function.
(As x→∞ the limit is 2: the graph has two horizontal asymptotes, y=2 and
y=−2.)
For x>0, x>0 (it is ≥0, and not 0, since 02=0=x), and
x1=x1: the right-hand side is non-negative, and its square is
x1 (the square-root remark). Since x⋅x=x,
xx=x1, and dividing the numerator and the denominator by x,
By part (a) of Proposition 1, x1→0, and x1≥0
for every x>0, so part (f) of Theorem 1, with L=0 and K=0,
gives 1/x→0. The denominator tends to 1=0, so the quotient law gives
10=0; by property 2 of the remark, so does the
original function. So x grows more slowly than x+1.
For x>0, A(x) is the value of the rational function x45x+2000, whose numerator
and denominator have degree 1 and leading coefficients 45 and 1. By part (c) of
Proposition 1 and property 2 of the
remark, limx→∞A(x)=45.
The fixed cost is shared by more and more chairs, so the average cost approaches the cost of the
materials and labour for one chair, £45. Indeed A(x)−45=x2000, which is less than
5 exactly when x>400 (multiplying by the positive number 5x): more than 400
chairs bring the average below £50.
False. Take f(x)=x+1 and g(x)=x, defined on R. By part (a) of
Theorem 1, limx→∞g(x)=∞. For f, given a real
number B, the threshold N=B−1 wins: if x>B−1, then x+1>B, adding 1
(property 3 of the order rules). So limx→∞f(x)=∞.
But f(x)−g(x)=1 at every x, and the constant 1 has the limit 1 (part (a)), not 0: by
property 1 of the remark, a limit is unique.
With m=1 and c=−1, the difference is x+12 at every x>−1. This rational
function has a numerator of degree 0 and a denominator of degree 1, so its limit is 0 by
part (b) of Proposition 1, and by property 2 of the
remark so is that of the difference.
Uniqueness. Suppose m′ and c′ work too. By the difference law, part (b) of
Theorem 1, applied to the two differences,
If m′=m, then (m′−m)x+(c′−c) is a polynomial of degree 1, and by part (d) of the
proposition (with the denominator 1, of degree 0) its limit is ∞ or −∞, which
contradicts the limit 0 by property 1 of the remark. So
m′=m, and then the function is the constant c′−c, whose limit is c′−c (part (a) of the
laws); uniqueness gives c′−c=0. So m=1 and c=−1 is the only pair.
The line y=x−1 is not horizontal, but the graph approaches it far to the right.
The domain. With the height B=1, there is a threshold N0 such that f is defined and
f(x)>1>0 at every x>N0. So f(x)=0 there, and f1 is defined at every
point of (N0,∞), as part (a) of the definition requires.
The rounds. Let ε>0. Then ε1 is a real number, so there is a threshold
N such that f is defined and f(x)>ε1 at every x>N. Let x>N. Since
ε1>0 (property 6 of the order rules), we have
0<ε1<f(x), and property 6 gives
So f(x)1 is defined, and ∣∣f(x)1−0∣∣=f(x)1<ε, by
the definition of the absolute value. So N wins the round ε,
and since ε>0 was arbitrary, limx→∞f(x)1=0.
The function x is defined at every x≥0 (the square-root
remark), so on (0,∞). Let B be a real number.
If B≤0. Take N=0. If x>0, then x is defined, x≥0, and
x=0, because 02=0=x. So x>0≥B, and x>B by
transitivity (property 2 of the order rules).
If B>0. Take N=B2, which is ≥0 by the “in particular” of part (c) of the
sign rules. If x>B2, then x>0, by transitivity from
0≤B2<x (property 2 of the order rules), so x is
defined. The numbers B and x are non-negative, and B2<x=(x)2,
so the order part of the square-root remark, with s=B and t=x, gives
B<x.
In both cases N wins the round B. Since B was arbitrary, limx→∞x=∞.
Indeterminate Forms and L'Hôpital's Rule (coming soon)
Area and Riemann Sums (coming soon)
Found an error on this page? Report it (the form opens with this page filled in).
Limits at infinity return wherever “in the long run” matters: in curve sketching, where horizontal
and oblique asymptotes describe the far ends of a graph; in L’Hôpital’s rule, whose general form
covers “∞∞” and x→±∞; in improper integrals over unbounded
intervals; and in sequences, where n→∞ runs through the integers only.