Skip to content

Chapter 2 — Functions

2.4 — The Catalog of Prototype Functions and Asymptotic Behavior

Before transforming a function, one needs to know the function being transformed.


The vocabulary of §2.3, with its domain, image, zeros, sign, symmetry and monotonicity, was built for any function. This section puts it to work on a few concrete functions, chosen not for their difficulty but for the opposite reason: they are the simplest of each family, and for that reason they will serve as a base, in §2.5, for building many other functions by transformation without repeating the analysis from scratch. One of them, moreover, will display a phenomenon the book mentioned only in passing in §1.9: that of a function whose values grow without bound.

Six functions form the basic catalog of the chapter.

FunctionDomainImageParityMonotonicity
f(x)=xf(x)=xR\mathbb{R}R\mathbb{R}OddIncreasing on R\mathbb{R}
f(x)=x2f(x)=x^2R\mathbb{R}[0,∞)[0,\infty)EvenDecreasing on (−∞,0](-\infty,0], increasing on [0,∞)[0,\infty)
f(x)=x3f(x)=x^3R\mathbb{R}R\mathbb{R}OddIncreasing on R\mathbb{R}
f(x)=∣x∣f(x)=\vert x\vertR\mathbb{R}[0,∞)[0,\infty)EvenDecreasing on (−∞,0](-\infty,0], increasing on [0,∞)[0,\infty)
f(x)=xf(x)=\sqrt{x}[0,∞)[0,\infty)[0,∞)[0,\infty)NeitherIncreasing on [0,∞)[0,\infty)
f(x)=1xf(x)=\dfrac{1}{x}R∖{0}\mathbb{R}\setminus\{0\}R∖{0}\mathbb{R}\setminus\{0\}OddDecreasing on (−∞,0)(-\infty,0) and on (0,∞)(0,\infty), separately

The six prototypes of the catalog: identity, quadratic, cubic, reciprocal, square root and absolute value, each with its own window

Figure 2.2 — The six prototypes of the elementary gallery.

Each row of the table is a statement that can be proved with what we already have. The domains come from §2.2: x\sqrt{x} requires a nonnegative radicand and 1x\frac{1}{x} a nonzero denominator. The parities are checked as in §2.3: for example, 1−x=−1x\frac{1}{-x} = -\frac{1}{x} says that 1x\frac{1}{x} is odd, and ∣−x∣=∣x∣|-x| = |x| says that the absolute value is even. The monotonicity of x2x^2 was proved in §2.3, and that of ∣x∣|x| is immediate, because ∣x∣=−x|x| = -x for x≤0x \le 0 and ∣x∣=x|x| = x for x≥0x \ge 0. The images and the remaining monotonicities are left.

For the images, we must see that each value in the column is reached and that no other value is. Every real yy is the cube of some real, because by §1.2 y3\sqrt[3]{y} exists; the image of x3x^3 is R\mathbb{R}. Every y≥0y \ge 0 is the square root of y2y^2, and a square root is never negative; the image of x\sqrt{x} is [0,∞)[0, \infty). Every y≠0y \neq 0 is the reciprocal of 1y\frac{1}{y}, and a reciprocal is never zero; the image of 1x\frac{1}{x} is R∖{0}\mathbb{R} \setminus \{0\}.

For the monotonicities, the tool is again the lemma of §1.2: if 0≤u<v0 \le u < v, then un<vnu^n < v^n. The function x3x^3 is increasing on R\mathbb{R}: if 0≤x1<x20 \le x_1 < x_2, the lemma gives x13<x23x_1^3 < x_2^3; if x1<x2≤0x_1 < x_2 \le 0, the lemma applied to 0≤−x2<−x10 \le -x_2 < -x_1 gives −x23<−x13-x_2^3 < -x_1^3, that is, x13<x23x_1^3 < x_2^3; and if x1<0<x2x_1 < 0 < x_2, then x13<0<x23x_1^3 < 0 < x_2^3. The function x\sqrt{x} is increasing on [0,∞)[0, \infty): if 0≤x1<x20 \le x_1 < x_2 and we had x2≤x1\sqrt{x_2} \le \sqrt{x_1}, the lemma would give x2≤x1x_2 \le x_1, which is false. The function 1x\frac{1}{x} is decreasing on (0,∞)(0, \infty): if 0<x1<x20 < x_1 < x_2, multiplying by the positive number 1x1x2\frac{1}{x_1 x_2} gives 1x2<1x1\frac{1}{x_2} < \frac{1}{x_1}. On (−∞,0)(-\infty, 0) the argument is the same, because if x1x_1 and x2x_2 are both negative, x1x2x_1 x_2 is positive.

Two rows deserve a comment for what they anticipate.

The first is that of x\sqrt{x}, which is neither even nor odd for a structural reason: its domain, [0,∞)[0, \infty), is not symmetric with respect to the origin. The comparison between f(−x)f(-x) and f(x)f(x) that §2.3 requires is not even possible for x>0x > 0, and parity is ruled out beforehand.

The second is that of 1x\frac{1}{x}, whose monotonicity is declared separately on each branch, (−∞,0)(-\infty, 0) and (0,∞)(0, \infty), and not over its whole domain. This is no oversight. The function is decreasing on each branch, but it is not decreasing on R∖{0}\mathbb{R} \setminus \{0\}: with x1=−1x_1 = -1 and x2=1x_2 = 1 we have x1<x2x_1 < x_2 and yet f(x1)=−1<1=f(x2)f(x_1) = -1 < 1 = f(x_2). Nor is it increasing, because with x1=−1x_1 = -1 and x2=−12x_2 = -\frac{1}{2} we have f(x1)=−1>−2=f(x2)f(x_1) = -1 > -2 = f(x_2). The definition of monotonicity in §2.3 asks for an interval, and R∖{0}\mathbb{R} \setminus \{0\} is not one: it lacks precisely the point that separates the two branches. What happens near that point is the topic of the rest of the section.


§2. Unbounded growth: the vertical asymptote

Section titled “§2. Unbounded growth: the vertical asymptote”

Of the six functions in the catalog, five have no “forbidden” point in the middle of their domain near which something unusual might happen. The exception is 1x\frac{1}{x}, at the only point its domain excludes, x=0x = 0. Let us ask precisely what happens to 1x\frac{1}{x} as xx approaches that point without being able to reach it.

The intuitive answer, ”1x\frac{1}{x} gets larger and larger”, needs the same kind of rigor that the idea of xx being near aa received in §1.1: there, “near” was translated into the condition ∣x−a∣<ϵ|x - a| < \epsilon. Here we must translate what “becoming arbitrarily large” means.

Definition. Let ff be a function and a∈Ra \in \mathbb{R} such that, for every δ>0\delta > 0, there is some x∈Dom⁡(f)x \in \operatorname{Dom}(f) with a<x<a+δa < x < a + \delta. We say that f(x)f(x) grows without bound as xx tends to aa from the right, and we write x→a+x \to a^+, if

∀M>0,∃δ>0 such that, for all x∈Dom⁡(f),a<x<a+δ  ⟹  f(x)>M\forall M > 0, \exists \delta > 0 \text{ such that, for all } x \in \operatorname{Dom}(f), \quad a < x < a + \delta \implies f(x) > M

Likewise, f(x)f(x) decreases without bound as x→a+x \to a^+ if the conclusion is f(x)<−Mf(x) < -M, and the versions from the left, x→a−x \to a^-, are obtained by replacing the condition a<x<a+δa < x < a + \delta with a−δ<x<aa - \delta < x < a. If any of these four situations occurs, the vertical line x=ax = a is called a vertical asymptote of ff.

The definition parallels that of a neighborhood in §1.1. There the idea was “for every margin ϵ\epsilon, there is an interval where the distance to aa is less than that margin”; here it is “for every bound MM, however large, there is an interval to the right of aa where ff already exceeds it”. One measures closeness to a fixed value; the other says there is no possible fixed value, because ff exceeds any ceiling that is proposed.

Proposition. 1x\frac{1}{x} grows without bound as x→0+x \to 0^+.

Proof. Let M>0M > 0 be arbitrary, and choose δ=1M\delta = \frac{1}{M}. If 0<x<1M0 < x < \frac{1}{M}, we multiply this inequality by the positive number Mx\frac{M}{x}, which preserves its direction (§1.7), and obtain M<1xM < \frac{1}{x}. We have proved that 0<x<δ0 < x < \delta implies 1x>M\frac{1}{x} > M. Since MM was arbitrary, the definition is satisfied. ■\blacksquare

The value δ=1M\delta = \frac{1}{M} does not come out of nowhere: it is what one obtains by solving the inequality 1x>M\frac{1}{x} > M for x>0x > 0, with the methods of §1.7, and choosing δ\delta at the exact point where the inequality starts to hold.


§3. Approach at infinity: the horizontal asymptote

Section titled “§3. Approach at infinity: the horizontal asymptote”

There is a second phenomenon at the other end of the domain of 1x\frac{1}{x}: what happens when xx itself grows without bound. The question is reversed with respect to the previous one, because it is no longer the variable that approaches a forbidden point, but the variable that moves away indefinitely. And the answer is reversed too: f(x)f(x) does not grow without bound, but comes closer and closer to a fixed value.

Definition. Let ff be a function defined for every xx greater than some number. We say that f(x)f(x) approaches a real LL as xx grows without bound, and we write x→+∞x \to +\infty, if

∀ϵ>0,∃N>0 such that x>N  ⟹  ∣f(x)−L∣<ϵ\forall \epsilon > 0, \exists N > 0 \text{ such that } x > N \implies |f(x) - L| < \epsilon

If this occurs, the horizontal line y=Ly = L is called a horizontal asymptote of ff. The version for x→−∞x \to -\infty is obtained by replacing x>Nx > N with x<−Nx < -N.

The symbol +∞+\infty is not a number: the notation x→+∞x \to +\infty is an abbreviation of ”xx grows without bound”, and the whole definition speaks only of real numbers NN and ϵ\epsilon. This definition reuses unchanged the distance ∣f(x)−L∣<ϵ|f(x) - L| < \epsilon of the neighborhood of §1.1; the only new thing is the hypothesis that triggers it, x>Nx > N, instead of xx lying in a bounded interval.

Proposition. 1x\frac{1}{x} approaches 00 as x→+∞x \to +\infty.

Proof. Let ϵ>0\epsilon > 0 be arbitrary, and choose N=1ϵN = \frac{1}{\epsilon}. If x>1ϵx > \frac{1}{\epsilon}, then x>0x > 0, and multiplying by the positive number ϵx\frac{\epsilon}{x} gives ϵ>1x\epsilon > \frac{1}{x}. Moreover 1x>0\frac{1}{x} > 0, so that

∣1x−0∣=1x<ϵ\left|\frac{1}{x} - 0\right| = \frac{1}{x} < \epsilon

We have proved that x>Nx > N implies ∣1x−0∣<ϵ\left|\frac{1}{x} - 0\right| < \epsilon, for any ϵ>0\epsilon > 0. ■\blacksquare


Two cases of the behavior of 1x\frac{1}{x} remain: what happens as x→0−x \to 0^- and as x→−∞x \to -\infty. There is no need to repeat the arguments from scratch, because 1x\frac{1}{x} is odd, and oddness translates each case on the right into the corresponding case on the left.

If −δ<x<0-\delta < x < 0, the number t=−xt = -x satisfies 0<t<δ0 < t < \delta. With δ=1M\delta = \frac{1}{M}, the proposition of §2 gives 1t>M\frac{1}{t} > M, and by oddness 1x=−1t<−M\frac{1}{x} = -\frac{1}{t} < -M. Therefore 1x\frac{1}{x} decreases without bound as x→0−x \to 0^-, and the line x=0x = 0 is a vertical asymptote from the left as well, with the sign reversed.

Likewise, if x<−Nx < -N with N=1ϵN = \frac{1}{\epsilon}, the number t=−xt = -x satisfies t>Nt > N, and the proposition of §3 gives ∣1t∣<ϵ\left|\frac{1}{t}\right| < \epsilon; since ∣1x∣=∣−1t∣=∣1t∣\left|\frac{1}{x}\right| = \left|-\frac{1}{t}\right| = \left|\frac{1}{t}\right|, also ∣1x−0∣<ϵ\left|\frac{1}{x} - 0\right| < \epsilon. The line y=0y = 0 is a horizontal asymptote toward the left as well, with the same value L=0L = 0. This is no accident: if an odd function approaches LL to the right, it approaches −L-L to the left, and here L=−L=0L = -L = 0.

None of the other five functions in the catalog has asymptotes. Their domains exclude no interior point, and as xx grows or decreases without bound, their values do not settle down either: xx and x3x^3 grow without bound to the right and decrease without bound to the left, x2x^2 and ∣x∣|x| grow without bound in both directions, and x\sqrt{x} grows without bound to the right, its only unbounded end. Within this catalog, 1x\frac{1}{x} is the only one that displays both phenomena, that of an excluded point near which the function shoots off and that of a function that settles around a value as the variable moves away.


What this catalog lets us glimpse, beyond the case of 1x\frac{1}{x}, is that infinity appears in two forms that should not be confused. One is infinity as a destination: what happens to f(x)f(x) when xx itself moves away without bound, a question about the long run. The other is infinity as a local symptom: a function defined near a point, except at the point itself, whose values shoot off as xx approaches it. In neither does ff “reach” infinity: no MM and no NN, however large, exhaust the definition. Both are statements about every finite threshold.

And at bottom they are the same phenomenon. Both are particular cases of a single notion, that of limit, which this section built twice without naming it, once with MM and δ\delta, once with ϵ\epsilon and NN. Naming it now changes no definition and no proof: what we defined are, exactly, the limits ”f(x)f(x) tends to +∞+\infty as x→a+x \to a^+” and ”f(x)f(x) tends to LL as x→+∞x \to +\infty”, together with the variants we obtained by symmetry. What lies outside this book, and properly belongs to Calculus, is the general theory: the limit as xx approaches aa from both sides at once, the rules that allow one to compute the limit of a sum, a product or a quotient from the limits of the parts without repeating each time an argument like those of this section, and the notion of continuity that rests on all of it. This book built the concept in the cases it needed; systematizing it is a different work.

With the gallery of prototypes analyzed, including its behavior at the edges of the domain, the chapter has the raw material for the next step: transforming these six functions, shifting, stretching and reflecting them, to obtain a much larger family without repeating the analysis from scratch. That is the topic of the next section.