Skip to content

Chapter 3 — Polynomials and Rational Functions

3.1 — Polynomial Functions and Their Graphs

Adding and multiplying is all it takes.


In §2.4 we presented xx, x2x^2 and x3x^3 as isolated entries of a table, without saying what they have in common. What they have in common is this: all three are obtained by applying to the variable xx, finitely many times, the two operations that R\mathbb{R} provides by axiom, addition and multiplication. In this section we formalize that observation and study three consequences: the domain, the behavior as xx grows without bound, and the behavior near a point where the function vanishes.

§1. Definition and elements of a polynomial

Section titled “§1. Definition and elements of a polynomial”

§1.3 introduced the polynomial as an algebraic expression. Here we look at it as what interests us in this chapter: a function from R\mathbb{R} to R\mathbb{R}.

Definition. A polynomial function is a function p:R→Rp:\mathbb{R}\to\mathbb{R} of the form

p(x)=anxn+an−1xn−1+⋯+a1x+a0p(x) = a_n x^n + a_{n-1}x^{n-1} + \cdots + a_1 x + a_0

where nn is a nonnegative integer and a0,a1,…,an∈Ra_0,a_1,\ldots,a_n \in \mathbb{R}. The numbers aka_k are called coefficients, and the subscript kk indicates which power of xx the coefficient belongs to. If an≠0a_n \neq 0, we say that pp has degree nn, written deg⁡(p)=n\deg(p)=n, and that ana_n is its leading coefficient. The zero polynomial, p(x)=0p(x)=0 for all xx, is left out of this classification: no coefficient singles out in it a term of greatest weight, so we assign it no degree.

The definition speaks of coefficients, and what interests us is the function. A legitimate question remains: if two different expressions defined the same function, could they have different degrees? The answer is no, and we prove it below, in the corollary of §2.

No structural restriction limits the domain. Unlike x\sqrt{x} or 1/x1/x in §2.2, adding and multiplying real numbers never produces a forbidden operation, so Dom(p)=R\text{Dom}(p)=\mathbb{R} for every polynomial pp, without exception. This is the first substantive difference with respect to the catalog of §2.4: where 1/x1/x had to exclude a point, no polynomial ever needs to.

The three prototypes we already know are, under this definition, the simplest cases. The function xx is the polynomial of degree 11 with a1=1a_1=1 and a0=0a_0=0. The functions x2x^2 and x3x^3 are those of degree 22 and 33 with a single nonzero coefficient. A constant function p(x)=cp(x)=c, including the constant 00 of §2.4, is a polynomial of degree 00 if c≠0c\neq0, and is the zero polynomial if c=0c=0.


§2.4 §2 defined what it means for f(x)f(x) to grow without bound as xx approaches a point aa from the right. The question that interests us now is different. No polynomial has a forbidden point to approach; what moves away is xx itself. §2.4 §3 studied a similar situation with 1/x1/x as x→+∞x\to+\infty, and there the result was an approach to a fixed value. For a nonconstant polynomial the opposite happens, and to say so with the same rigor we need a new definition.

Definition. Let ff be a function defined for all xx greater than some value. We say that f(x)f(x) grows without bound as x→+∞x\to+\infty if

∀M>0,∃N>0such thatx>N  ⟹  f(x)>M\forall M>0,\quad \exists N>0 \quad \text{such that} \quad x>N \implies f(x)>M

We read it like this: for every bound MM, however large, there is a threshold NN such that, past that threshold, ff exceeds MM. We say that f(x)f(x) decreases without bound as x→+∞x\to+\infty if the analogous condition holds with f(x)<−Mf(x)<-M in place of f(x)>Mf(x)>M. The definitions for x→−∞x\to-\infty are obtained by replacing the hypothesis x>Nx>N with x<−Nx<-N.

This definition has the same structure as those of §2.4 §2 and §3. Only the hypothesis that triggers the conclusion changes: here, that xx exceeds a threshold NN, instead of xx falling inside an interval (a,a+δ)(a,a+\delta). With it we can prove the central result of this section.

Proposition. Let p(x)=anxn+⋯+a0p(x)=a_nx^n+\cdots+a_0 with an≠0a_n\neq0 and n≥1n\geq1. The behavior of pp as x→+∞x\to+\infty depends only on the sign of ana_n: if an>0a_n>0, pp grows without bound; if an<0a_n<0, pp decreases without bound. The behavior as x→−∞x\to-\infty depends in addition on the parity of nn: it coincides with that of x→+∞x\to+\infty if nn is even, and is reversed if nn is odd.

Proof. We prove something stronger: for every M>0M>0 there is N>0N>0 such that, if ∣x∣>N|x|>N, then ∣p(x)∣>M|p(x)|>M and p(x)p(x) has the same sign as anxna_nx^n. The four conclusions of the proposition can be read off from this.

For x≠0x\neq0 we factor out the term of highest degree:

p(x)=anxn⋅r(x),r(x)=1+∑k=0n−1akanxn−kp(x) = a_nx^n\cdot r(x), \qquad r(x)=1+\sum_{k=0}^{n-1}\frac{a_k}{a_n x^{n-k}}

Let AA be the largest of the nn numbers ∣ak/an∣|a_k/a_n| with 0≤k≤n−10\le k\le n-1. If ∣x∣≥1|x|\geq1, since n−k≥1n-k\geq1 we have ∣x∣n−k≥∣x∣|x|^{n-k}\geq|x|, and each summand of r(x)r(x) other than 11 satisfies ∣akanxn−k∣≤A∣x∣\left|\dfrac{a_k}{a_nx^{n-k}}\right|\le\dfrac{A}{|x|}. There are nn summands of that kind, so

∣r(x)−1∣≤nA∣x∣|r(x)-1|\le\frac{nA}{|x|}

Let N0=max⁡{1,2nA}N_0=\max\{1, 2nA\}. If ∣x∣>N0|x|>N_0, then ∣x∣>2nA|x|>2nA, that is, nA∣x∣<12\dfrac{nA}{|x|}<\dfrac12, and therefore ∣r(x)−1∣<12|r(x)-1|<\dfrac12. In particular r(x)>12>0r(x)>\dfrac12>0 for every ∣x∣>N0|x|>N_0.

Now let M>0M>0 be arbitrary and take N=max⁡{N0,2M/∣an∣}N=\max\{N_0, 2M/|a_n|\}. If ∣x∣>N|x|>N, since ∣x∣>1|x|>1 we have ∣x∣n≥∣x∣>2M/∣an∣|x|^n\geq|x|>2M/|a_n|, and since r(x)>1/2r(x)>1/2,

∣p(x)∣=∣an∣⋅∣x∣n⋅r(x)>∣an∣⋅2M∣an∣⋅12=M|p(x)| = |a_n|\cdot|x|^n\cdot r(x) > |a_n|\cdot\frac{2M}{|a_n|}\cdot\frac12 = M

Moreover r(x)>0r(x)>0, so p(x)p(x) has the sign of anxna_nx^n. For x>Nx>N that sign is the sign of ana_n, so p(x)>Mp(x)>M if an>0a_n>0 and p(x)<−Mp(x)<-M if an<0a_n<0. For x<−Nx<-N, the sign of xnx^n is (−1)n(-1)^n, and the sign of p(x)p(x) is that of an(−1)na_n(-1)^n: it coincides with the case x→+∞x\to+\infty if nn is even and is the opposite if nn is odd. ■\blacksquare

The proposition reduces the behavior at the ends, which has four cases according to the sign of ana_n and the parity of nn, to an immediate reading of the term of highest degree. The rest of the polynomial plays no part in the conclusion: the lower-degree terms, however many they are or however large their coefficients, end up dominated by anxna_nx^n as soon as ∣x∣|x| is large enough.

Four cases of behavior at the ends according to degree and leading coefficient Figure 3.1 — The four possible cases, determined entirely by the sign of ana_n and the parity of nn.

The proposition also has a consequence that settles the question we left open in §1.

Corollary. If two polynomial expressions define the same function, they have the same coefficients. In particular, the degree and the leading coefficient of a polynomial function depend only on the function.

Proof. Let ∑kakxk\sum_k a_kx^k and ∑kbkxk\sum_k b_kx^k be two expressions that give the same value for every xx, and let ck=ak−bkc_k=a_k-b_k (padding with zeros any missing coefficients). The function d(x)=∑kckxkd(x)=\sum_k c_kx^k is then identically zero. Suppose some ckc_k is nonzero and let mm be the largest index with cm≠0c_m\neq0. If m=0m=0, dd is the constant c0≠0c_0\neq0, which is not the zero function. If m≥1m\geq1, the proposition applied to dd gives ∣d(x)∣>1|d(x)|>1 for ∣x∣|x| sufficiently large, and again dd is not the zero function. In both cases we reach a contradiction; hence all the ckc_k are zero, that is, ak=bka_k=b_k for all kk. ■\blacksquare


§3. Multiplicity of a root and its graphical effect

Section titled “§3. Multiplicity of a root and its graphical effect”

We say that rr is a zero or root of pp if p(r)=0p(r)=0. Let us compare xx, x2x^2 and x3x^3 at r=0r=0. All three vanish there, but the graph of the first crosses the axis, that of the second touches it and turns back, and that of the third crosses it flattened against it. What distinguishes the three is how many times the factor xx appears. To say this precisely, we define multiplicity.

Definition. Let pp be a nonzero polynomial and let r∈Rr\in\mathbb{R}. We say that rr is a zero of pp of multiplicity m≥1m\geq1 if there is a polynomial qq with q(r)≠0q(r)\neq0 such that

p(x)=(x−r)mq(x)p(x)=(x-r)^mq(x)

For it to make sense to speak of “the” multiplicity we need two things: that the number mm be determined by pp and rr, and that every zero have some multiplicity. We prove the first now.

Proposition. If p(x)=(x−r)mq(x)=(x−r)m′q′(x)p(x)=(x-r)^m q(x)=(x-r)^{m'}q'(x) with q(r)≠0q(r)\neq0 and q′(r)≠0q'(r)\neq0, then m=m′m=m'.

Proof. Suppose m<m′m<m'. Subtracting and factoring out (x−r)m(x-r)^m gives (x−r)m[q(x)−(x−r)m′−mq′(x)]=0(x-r)^m\big[q(x)-(x-r)^{m'-m}q'(x)\big]=0. By §1.3 §2, deg⁡(PQ)=deg⁡(P)+deg⁡(Q)\deg(PQ)=\deg(P)+\deg(Q) for nonzero polynomials; in particular, a product of nonzero polynomials is not zero. Since (x−r)m(x-r)^m is not zero, the bracket is: q(x)=(x−r)m′−mq′(x)q(x)=(x-r)^{m'-m}q'(x). Evaluating at x=rx=r we obtain q(r)=0⋅q′(r)=0q(r)=0\cdot q'(r)=0, which contradicts q(r)≠0q(r)\neq0. The case m>m′m>m' is symmetric, so m=m′m=m'. ■\blacksquare

The second thing, that every zero of a nonzero polynomial has a multiplicity, is a consequence of the Factor Theorem: each time we extract a factor (x−r)(x-r) the degree drops by one, so the process cannot continue indefinitely [Proved in §3.2].

The multiplicity determines how the graph of pp behaves near rr. To prove it we first need a result about polynomials that do not vanish at a point: they keep their sign in a neighborhood of that point.

Lemma (sign preservation). Let qq be a polynomial and r∈Rr\in\mathbb{R} with q(r)≠0q(r)\neq0. There exists δ>0\delta>0 such that, for every xx with ∣x−r∣<δ|x-r|<\delta, q(x)q(x) has the same sign as q(r)q(r).

Proof. We write q(x)=∑k=0dbkxkq(x)=\sum_{k=0}^{d}b_kx^k. For k≥1k\geq1 the following identity holds:

xk−rk=(x−r)∑j=0k−1xk−1−jrjx^k-r^k=(x-r)\sum_{j=0}^{k-1}x^{k-1-j}r^j

which generalizes the difference of squares and the difference of cubes of §1.3. It is verified by distributing: the product is ∑j=0k−1xk−jrj−∑j=0k−1xk−1−jrj+1\sum_{j=0}^{k-1}x^{k-j}r^j-\sum_{j=0}^{k-1}x^{k-1-j}r^{j+1}, and the terms of the two sums cancel in pairs, except the first of the first sum, xkx^k, and the last of the second, −rk-r^k.

Suppose ∣x−r∣<1|x-r|<1. Then ∣x∣<∣r∣+1|x|<|r|+1, and each of the kk terms of the preceding sum satisfies ∣x∣k−1−j∣r∣j≤(∣r∣+1)k−1|x|^{k-1-j}|r|^j\le(|r|+1)^{k-1}. Hence ∣xk−rk∣≤k(∣r∣+1)k−1∣x−r∣|x^k-r^k|\le k(|r|+1)^{k-1}|x-r|, and summing over kk,

∣q(x)−q(r)∣=∣∑k=1dbk(xk−rk)∣≤C∣x−r∣,C=∑k=1d∣bk∣k(∣r∣+1)k−1|q(x)-q(r)|=\left|\sum_{k=1}^{d}b_k(x^k-r^k)\right|\le C|x-r|,\qquad C=\sum_{k=1}^{d}|b_k|k(|r|+1)^{k-1}

Take δ=min⁡{1,∣q(r)∣2(C+1)}\delta=\min\left\{1, \dfrac{|q(r)|}{2(C+1)}\right\}. If ∣x−r∣<δ|x-r|<\delta, then ∣q(x)−q(r)∣≤C∣x−r∣<∣q(r)∣2|q(x)-q(r)|\le C|x-r|<\dfrac{|q(r)|}{2}. If q(r)>0q(r)>0, this gives q(x)>q(r)−q(r)2>0q(x)>q(r)-\dfrac{q(r)}{2}>0; if q(r)<0q(r)<0, it gives q(x)<q(r)+∣q(r)∣2<0q(x)<q(r)+\dfrac{|q(r)|}{2}<0. In both cases q(x)q(x) has the sign of q(r)q(r). ■\blacksquare

Proposition. Let rr be a zero of pp with multiplicity mm. If mm is odd, pp changes sign at rr: the graph crosses the xx-axis. If mm is even, pp does not change sign at rr: the graph touches the xx-axis and returns to the same side.

Proof. Let p(x)=(x−r)mq(x)p(x)=(x-r)^mq(x) with q(r)≠0q(r)\neq0. By the lemma, there is δ>0\delta>0 such that qq has on (r−δ,r+δ)(r-\delta,r+\delta) the constant sign of q(r)q(r). For xx in that interval with x≠rx\neq r, the sign of p(x)p(x) is the product of the sign of q(r)q(r) and that of (x−r)m(x-r)^m. If mm is even, (x−r)m=((x−r)m/2)2>0(x-r)^m=\big((x-r)^{m/2}\big)^2>0, so pp keeps the sign of q(r)q(r) on both sides of rr. If mm is odd, (x−r)m=(x−r)((x−r)(m−1)/2)2(x-r)^m=(x-r)\big((x-r)^{(m-1)/2}\big)^2 has the sign of x−rx-r: it is negative to the left of rr and positive to the right, so pp changes sign. ■\blacksquare

The case m=1m=1, the simple root, is the most frequent and the easiest to visualize: the graph crosses the axis cleanly. The case m=2m=2 is the one that already appeared, unnamed, in x2x^2 of §2.4: the graph touches the axis at x=0x=0 and returns to the same sign. Higher multiplicities repeat one of these two patterns, crossing if mm is odd and tangency if mm is even. To the eye, the graph flattens against the axis the more, the larger mm is; that is a visual impression, and what we have proved is only the behavior of the sign.

Local behavior of a polynomial near a root according to its multiplicity Figure 3.2 — Local behavior near rr for m=1m=1, m=2m=2 and m=3m=3: crossing, tangency, flattened crossing.


In this way we have gathered, for the polynomial in general, the three elements that already organized the study of each prototype in §2.4: the domain, which here has no restriction at all; the behavior at the ends, governed entirely by the term of highest degree; and the local behavior at the points where the function vanishes, governed by the multiplicity. One problem remains, the most elementary of all, which this section postponed on purpose: given a concrete polynomial, how do we effectively determine what its real zeros are? That is the subject of the next section.