Skip to content

Chapter 3 — Polynomials and Rational Functions

3.5 — Rational Functions

Where the polynomial never forbade anything, the quotient forbids exactly where the denominator vanishes.

§3.1 opened the chapter by noting that adding and multiplying exhausts the construction of the polynomial. The fourth operation of R\mathbb{R}, division, is still missing, and applying it to two polynomials produces a rational function: f(x)=p(x)/q(x)f(x)=p(x)/q(x), with p,qp,q polynomials and qq not the zero polynomial. Unlike pp or qq separately, ff inherits from division the same restriction that 1/x1/x already imposed in §2.2: there are points where ff is, quite simply, not defined.

Definition. A rational function is f(x)=p(x)q(x)f(x)=\frac{p(x)}{q(x)}, where pp and qq are polynomials and qq is not the zero polynomial. Its domain is

Dom(f)={x∈R:q(x)≠0}\text{Dom}(f) = \{x\in\mathbb{R} : q(x)\neq0\}

By §3.2, qq has at most finitely many real zeros, at most deg⁡(q)\deg(q), so that Dom(f)\text{Dom}(f) excludes, at most, finitely many points of R\mathbb{R}: never a whole interval, unlike what would happen if qq could vanish over a continuous range. This observation, trivial in appearance, is what guarantees that it is worth asking, as the rest of this section does, what happens near each of those excluded points: they are isolated points, surrounded by values where ff is defined.

§2. Near a zero of the denominator: holes and vertical asymptotes

Section titled “§2. Near a zero of the denominator: holes and vertical asymptotes”

A zero of qq does not always produce the same phenomenon. The difference depends on whether that same value also annuls pp, and with what multiplicity, the concept already formalized in §3.1 and extended in §3.4.

§2.4 formalized what it means for a function to grow without bound near an excluded point, but it did not formalize the case in which, instead, it stabilizes at a finite value near that point; it only did so for x→+∞x\to+\infty. We need that counterpart, now for a finite point.

Definition. Let ff be a function defined in a neighborhood of cc, except possibly at cc. We say that f(x)f(x) approaches a real number LL as x→cx\to c if

∀ϵ>0,∃δ>0such that0<∣x−c∣<δ  ⟹  ∣f(x)−L∣<ϵ\forall\epsilon>0,\quad\exists\delta>0\quad\text{such that}\quad 0<|x-c|<\delta\implies|f(x)-L|<\epsilon

This definition reuses, unaltered, the distance inequality ∣f(x)−L∣<ϵ|f(x)-L|<\epsilon already fixed in §2.4 §3; the only thing that changes is the hypothesis that activates it, 0<∣x−c∣<δ0<|x-c|<\delta, a neighborhood of a finite point, with that point explicitly excluded, in place of x>Nx>N.

Proposition. Let cc be a zero of qq of multiplicity mm, and let k≥0k\geq0 be the multiplicity of cc as a root of pp (k=0k=0 if p(c)≠0p(c)\neq0). Then:

(a) if k≥mk\geq m, f(x)f(x) approaches a real number LL as x→cx\to c; cc is a hole of ff;

(b) if k<mk<m, the line x=cx=c is a vertical asymptote of ff.

Proof. By the complete factorization of §3.4, write p(x)=(x−c)kp1(x)p(x)=(x-c)^kp_1(x) and q(x)=(x−c)mq1(x)q(x)=(x-c)^mq_1(x), with p1(c)≠0p_1(c)\neq0 and q1(c)≠0q_1(c)\neq0. For x≠cx\neq c,

f(x)=(x−c)kp1(x)(x−c)mq1(x)=(x−c)k−m⋅p1(x)q1(x)f(x) = \frac{(x-c)^kp_1(x)}{(x-c)^mq_1(x)} = (x-c)^{k-m}\cdot\frac{p_1(x)}{q_1(x)}

Case k≥mk\geq m. The exponent k−m≥0k-m\geq0, so that (x−c)k−m(x-c)^{k-m} is a polynomial well defined at cc, and p1(x)/q1(x)p_1(x)/q_1(x) is defined in a neighborhood of cc, since q1(c)≠0q_1(c)\neq0. Let L=0L=0 if k>mk>m, or L=p1(c)/q1(c)L=p_1(c)/q_1(c) if k=mk=m; in both cases, the continuity of the algebraic operations on real numbers (sum, product and quotient by something nonzero, already used without naming it in §3.1 §3) guarantees that, given ϵ>0\epsilon>0, there exists δ>0\delta>0 such that 0<∣x−c∣<δ0<|x-c|<\delta forces ∣f(x)−L∣<ϵ|f(x)-L|<\epsilon. The explicit proof repeats, coordinate by coordinate, the same kind of bounding already practiced in §3.1 §2, and it is omitted because it is mechanical.

Case k<mk<m. The exponent k−m<0k-m<0, so that (x−c)k−m=1/(x−c)m−k(x-c)^{k-m}=1/(x-c)^{m-k} with m−k≥1m-k\geq1. Since p1(c)/q1(c)≠0p_1(c)/q_1(c)\neq0 is a fixed nonzero real number, f(x)f(x) behaves, near cc, like a nonzero constant divided by (x−c)m−k(x-c)^{m-k}, exactly the situation of 1/x1/x near 00 already resolved in §2.4 §2, except for a multiplicative constant and, possibly, a power greater than one: the same argument as in that proposition, adapted by that constant, shows that ∣f(x)∣|f(x)| grows without bound as x→cx\to c, and therefore so does f(x)f(x) or −f(x)-f(x) on each side, according to the signs of p1(c)/q1(c)p_1(c)/q_1(c) and of (x−c)m−k(x-c)^{m-k}. The line x=cx=c is, by definition, a vertical asymptote. ■\blacksquare

The case k≥mk\geq m deserves a final clarification: although f(x)f(x) approaches LL, the point cc remains excluded from Dom(f)\text{Dom}(f), since the definition of ff as a quotient excludes it by construction, regardless of the fact that p1/q1p_1/q_1 was indeed defined there. That is, precisely, the phenomenon that informal vocabulary calls a hole: a single missing point in a graph that would otherwise be a continuous curve.

Comparison between a hole and a vertical asymptote in a rational function

Left: k≥mk\geq m, hole. Right: k<mk\lt m, vertical asymptote.

§3. Behavior at the ends: the horizontal asymptote

Section titled “§3. Behavior at the ends: the horizontal asymptote”

The question of interest now is the same one that organized §3.1 §2 for polynomials: what happens when xx itself grows without bound. But the result, for a quotient, is different in nature: it does not always grow without bound as every nonconstant polynomial did.

Proposition. Let f(x)=p(x)/q(x)f(x)=p(x)/q(x) with deg⁡(p)=n\deg(p)=n and deg⁡(q)=m\deg(q)=m.

(a) If n<mn<m, f(x)f(x) approaches 00 as x→+∞x\to+\infty, and also as x→−∞x\to-\infty.

(b) If n=mn=m, f(x)f(x) approaches an/bma_n/b_m, the quotient of the leading coefficients, as x→±∞x\to\pm\infty.

(c) If n>mn>m, ff has no horizontal asymptote: ∣f(x)∣|f(x)| grows without bound.

Proof. Factor, as in §3.1 §2, p(x)=anxn(1+rp(x))p(x)=a_nx^n(1+r_p(x)) and q(x)=bmxm(1+rq(x))q(x)=b_mx^m(1+r_q(x)), where rp(x)r_p(x) and rq(x)r_q(x) are sums of terms that can be bounded arbitrarily close to 00 as ∣x∣→∞|x|\to\infty, exactly as was proved in that section. Then, for xx sufficiently large in absolute value,

f(x)=anbmxn−m⋅1+rp(x)1+rq(x)f(x) = \frac{a_n}{b_m}x^{n-m}\cdot\frac{1+r_p(x)}{1+r_q(x)}

and the second factor approaches 11 as ∣x∣→∞|x|\to\infty, by the same bounding argument of §3.1 §2 applied to rpr_p and rqr_q separately. All the behavior of ff is then governed by anbmxn−m\frac{a_n}{b_m}x^{n-m}: if n<mn<m, the exponent is negative and xn−m→0x^{n-m}\to0, giving case (a); if n=mn=m, the exponent is 00 and the factor is the constant 11, giving case (b) directly; if n>mn>m, the exponent is positive and xn−mx^{n-m} grows without bound, by the same proposition of §3.1 §2, giving case (c). In each case, translating “the second factor approaches 11” into the corresponding ϵ\epsilon-δ\delta or ϵ\epsilon-NN definition completes the formal proof, with the same kind of bounding already practiced repeatedly. ■\blacksquare

One case remains that the proposition of §3 declared, without developing it, “without horizontal asymptote”: n>mn>m. Within that case there is one that is particularly orderly, n=m+1n=m+1, where ff does not stabilize at a horizontal line, but does stabilize at a line with slope.

Definition. The line y=cx+dy=cx+d is an oblique asymptote of ff if f(x)−(cx+d)f(x)-(cx+d) approaches 00 as x→+∞x\to+\infty, in the sense of §2 of this section, with L=0L=0.

Proposition. If deg⁡(p)=deg⁡(q)+1\deg(p)=\deg(q)+1, ff has an oblique asymptote.

Proof. By the division algorithm of §3.2, there exist unique c1(x)c_1(x) and r(x)r(x) such that p(x)=q(x)c1(x)+r(x)p(x)=q(x)c_1(x)+r(x), with deg⁡(r)<deg⁡(q)\deg(r)<\deg(q). Since deg⁡(p)=deg⁡(q)+1\deg(p)=\deg(q)+1, the quotient c1(x)c_1(x) has degree exactly 11: write c1(x)=cx+dc_1(x)=cx+d. Dividing the identity by q(x)q(x),

f(x)=p(x)q(x)=cx+d+r(x)q(x)f(x) = \frac{p(x)}{q(x)} = cx+d+\frac{r(x)}{q(x)}

so that f(x)−(cx+d)=r(x)/q(x)f(x)-(cx+d)=r(x)/q(x). Since deg⁡(r)<deg⁡(q)\deg(r)<\deg(q), case (a) of the proposition of §3, applied to r/qr/q in place of p/qp/q, guarantees that r(x)/q(x)r(x)/q(x) approaches 00 as x→+∞x\to+\infty. Hence f(x)−(cx+d)f(x)-(cx+d) approaches 00, and y=cx+dy=cx+d is an oblique asymptote by definition. ■\blacksquare

The construction is, in essence, the same division that already organized §3.2 completely: the oblique asymptote is not a new phenomenon in its mechanics, but the reappearance of the division algorithm proved there, now read from the perspective of behavior at infinity instead of the search for roots. And it is, by construction, mutually exclusive with the horizontal asymptote: a rational function has one or the other, never both, because the case n=m+1n=m+1 of §4 and the cases n≤mn\leq m of §3 are disjoint.

Rational function with oblique asymptote y=x

f(x)=x2−1x=x−1xf(x)=\frac{x^2-1}{x}=x-\frac{1}{x}: the term −1/x-1/x vanishes, leaving the line y=xy=x.

With domain, holes, and vertical, horizontal and oblique asymptotes already completely characterized, the catalog of possible behavior of a rational function is closed. That closure is not merely technical: it is also, in a broader sense, the closure of the question that opened this chapter in its index, about what it means to construct, and not simply to enumerate, the mathematical objects one works with. That reflection, more philosophical than algebraic, is the subject of the final section of the chapter.