Skip to content

Chapter 3 — Polynomials and Rational Functions

3.2 — Real Zeros of Polynomials

Every root is, first of all, a factor that has not yet been extracted.


§1.3 taught us to factor by inspection: recognize a common factor, group terms, identify a special product, solve a quadratic trinomial. Those techniques suffice for the cases we treated there, but none of them says what to do in front of p(x)=x4−2x3−13x2+14x+24p(x)=x^4-2x^3-13x^2+14x+24. No common factor simplifies it, no special product recognizes it, and its degree exceeds what the general formula solves. We need a more general tool, and that tool is division.

Polynomial division answers the same question as integer division: given a dividend and a divisor, what quotient and remainder result? The answer has the same structure and the same guarantee of uniqueness.

Theorem (division algorithm). Let p(x)p(x) and d(x)d(x) be polynomials with real coefficients, with d(x)d(x) nonzero. There exist unique polynomials q(x)q(x) and r(x)r(x) such that

p(x)=d(x)q(x)+r(x),with r(x)=0 or deg⁡(r)<deg⁡(d)p(x) = d(x)q(x) + r(x), \qquad \text{with } r(x)=0 \text{ or } \deg(r) < \deg(d)

We call qq the quotient and rr the remainder.

Proof. Existence. We proceed by strong induction on the degree of pp. If p=0p=0 or deg⁡(p)<deg⁡(d)\deg(p)<\deg(d), we take q=0q=0 and r=pr=p, and the condition holds trivially. For the inductive step, suppose deg⁡(p)=n≥deg⁡(d)=k\deg(p)=n\geq\deg(d)=k and that the theorem holds for every polynomial of degree less than nn, and also for the zero polynomial. Let ana_n be the leading coefficient of pp and bkb_k that of dd. Consider

p1(x)=p(x)−anbkxn−kd(x)p_1(x) = p(x) - \frac{a_n}{b_k}x^{n-k}d(x)

which is constructed to cancel the term of degree nn: its coefficient of xnx^n is an−anbkbk=0a_n - \frac{a_n}{b_k}b_k = 0, and no term of pp or of xn−kd(x)x^{n-k}d(x) has degree greater than nn. Therefore p1p_1 is the zero polynomial or has degree less than nn. By the inductive hypothesis there exist q1(x)q_1(x) and r(x)r(x) with p1(x)=d(x)q1(x)+r(x)p_1(x)=d(x)q_1(x)+r(x) and r(x)=0r(x)=0 or deg⁡(r)<deg⁡(d)\deg(r)<\deg(d). Substituting,

p(x)=anbkxn−kd(x)+p1(x)=d(x)(anbkxn−k+q1(x))+r(x)p(x) = \frac{a_n}{b_k}x^{n-k}d(x) + p_1(x) = d(x)\left(\frac{a_n}{b_k}x^{n-k}+q_1(x)\right) + r(x)

which is the desired form, with q(x)=anbkxn−k+q1(x)q(x)=\frac{a_n}{b_k}x^{n-k}+q_1(x).

Uniqueness. Suppose that p(x)=d(x)q1(x)+r1(x)=d(x)q2(x)+r2(x)p(x)=d(x)q_1(x)+r_1(x)=d(x)q_2(x)+r_2(x), with both remainders zero or of degree less than deg⁡(d)\deg(d). Subtracting,

d(x)(q1(x)−q2(x))=r2(x)−r1(x)d(x)\big(q_1(x)-q_2(x)\big) = r_2(x)-r_1(x)

If q1≠q2q_1\neq q_2, the left side is a product of nonzero polynomials and, by §1.3 §2, has degree deg⁡(d)+deg⁡(q1−q2)≥deg⁡(d)\deg(d)+\deg(q_1-q_2)\geq\deg(d). The right side, on the other hand, is zero or has degree less than deg⁡(d)\deg(d), being the difference of two polynomials of that kind. This is a contradiction. Hence q1=q2q_1=q_2, and then r1=r2r_1=r_2. ■\blacksquare

The proof describes a procedure: cancel the term of highest degree, subtract, and repeat on the partial remainder until its degree falls below that of dd. This is long division of polynomials, the same mechanics as long division of integers.

When the divisor is d(x)=x−cd(x)=x-c, the procedure simplifies and is called synthetic division. Let us see why. If p(x)=∑k=0nakxkp(x)=\sum_{k=0}^{n}a_kx^k and we look for p(x)=(x−c)(bn−1xn−1+⋯+b0)+rp(x)=(x-c)(b_{n-1}x^{n-1}+\cdots+b_0)+r with rr constant, expanding the product and comparing the coefficients of each power we obtain

bn−1=an,bk−1=ak+cbk(1≤k≤n−1),r=a0+cb0b_{n-1}=a_n,\qquad b_{k-1}=a_k+cb_k\quad(1\le k\le n-1),\qquad r=a_0+cb_0

At each step, then, the previous coefficient of the quotient is multiplied by cc and added to the next coefficient of pp. It is the same algorithm as in the theorem, specialized to deg⁡(d)=1\deg(d)=1.


§2. The Remainder Theorem and the Factor Theorem

Section titled “§2. The Remainder Theorem and the Factor Theorem”

The final observation of §1, that dividing by x−cx-c leaves a constant remainder, has a consequence that is not evident at first sight: that remainder can be computed without dividing.

Theorem (Remainder). If p(x)p(x) is divided by x−cx-c, the remainder is p(c)p(c).

Proof. By §1, p(x)=(x−c)q(x)+rp(x)=(x-c)q(x)+r with rr constant. This equality between polynomials holds for every value of xx, because adding and multiplying polynomials is nothing more than applying distributivity, and those operations give the same values when evaluated. Evaluating at x=cx=c:

p(c)=(c−c)q(c)+r=0+r=r■p(c) = (c-c)q(c) + r = 0 + r = r \qquad \blacksquare

The theorem turns a question about division, what is the remainder?, into one of simple evaluation. And it immediately isolates the case in which that remainder vanishes.

Theorem (Factor). The number cc is a root of pp if and only if (x−c)(x-c) divides p(x)p(x) exactly, that is, with remainder 00.

Proof. By the Remainder Theorem, the remainder of dividing p(x)p(x) by (x−c)(x-c) is p(c)p(c). This remainder is 00 exactly when p(c)=0p(c)=0, that is, exactly when cc is a root of pp. ■\blacksquare

This is the result that §1.3 left pending. There, factoring a quadratic trinomial meant finding its two roots with the general formula and writing a(x−x1)(x−x2)a(x-x_1)(x-x_2). The Factor Theorem generalizes that procedure to any degree and does not need a general formula. In fact, for degree five or higher there is no general formula in radicals, a theorem we do not prove here. It is enough to locate a root cc by other means: division by (x−c)(x-c) yields the rest of the factorization, of degree one less, and on that quotient we can repeat the process.

That repetition lets us settle the pending matter of §3.1: that every zero has a multiplicity.

Corollary. Every zero rr of a nonzero polynomial pp has a multiplicity mm (in the sense of §3.1 §3), and m≤deg⁡(p)m\le\deg(p).

Proof. By the Factor Theorem, p(x)=(x−r)p1(x)p(x)=(x-r)p_1(x) with deg⁡(p1)=deg⁡(p)−1\deg(p_1)=\deg(p)-1. If p1(r)≠0p_1(r)\neq0, we are done with m=1m=1. If p1(r)=0p_1(r)=0, we apply the theorem again to p1p_1 and obtain p1(x)=(x−r)p2(x)p_1(x)=(x-r)p_2(x), of degree deg⁡(p)−2\deg(p)-2. Each step lowers the degree by one and a nonzero polynomial has degree at least 00, so the process stops at some step m≤deg⁡(p)m\le\deg(p), when the quotient pmp_m satisfies pm(r)≠0p_m(r)\neq0 (a nonzero constant polynomial satisfies this). Then p(x)=(x−r)mpm(x)p(x)=(x-r)^mp_m(x) with pm(r)≠0p_m(r)\neq0. ■\blacksquare


The Factor Theorem solves the problem of checking whether a given number is a root, but it does not say where to get that candidate. When the coefficients of pp are integers, we can bound in advance the set of rational candidates.

The proof uses a fact about integers that is worth making explicit. We say that two integers are coprime if they have no common divisors greater than 11.

Lemma. Let s,t,as,t,a be integers, with ss coprime to tt. Then ss is coprime to tnt^n for every n≥1n\geq1, and if ss divides atnat^n, then ss divides aa.

Proof. We use Bézout’s identity, a basic property of the integers that we accept without proof: two integers are coprime if and only if there exist integers u,vu,v with us+vt=1us+vt=1. If ss is coprime to bb and to cc, with us+vb=1us+vb=1 and u′s+v′c=1u's+v'c=1, multiplying both equalities we obtain

1=(us+vb)(u′s+v′c)=s(uu′s+uv′c+vbu′)+vv′bc1=(us+vb)(u's+v'c)=s(uu's+uv'c+vbu')+vv'bc

so ss is coprime to bcbc. Applying this repeatedly with b=c=tb=c=t, we obtain that ss is coprime to tnt^n. For the second claim, if us+vtn=1us+vt^n=1 and ss divides atnat^n, then a=usa+v(atn)a=usa+v(at^n) is a sum of two multiples of ss, and therefore ss divides aa. ■\blacksquare

Theorem (Rational Root). Let p(x)=anxn+⋯+a1x+a0p(x)=a_nx^n+\cdots+a_1x+a_0 with all ai∈Za_i\in\mathbb{Z} and a0,an≠0a_0,a_n\neq0. If st\frac{s}{t} is a rational root of pp, written in lowest terms (s,t∈Zs,t\in\mathbb{Z}, t≠0t\neq0, ss and tt coprime), then ss divides a0a_0 and tt divides ana_n.

Proof. Suppose p(s/t)=0p(s/t)=0:

an(st)n+an−1(st)n−1+⋯+a1st+a0=0a_n\left(\frac{s}{t}\right)^n + a_{n-1}\left(\frac{s}{t}\right)^{n-1}+\cdots+a_1\frac{s}{t}+a_0=0

Multiplying both sides by tnt^n and moving a0tna_0t^n to the other side,

a0tn=−s(ansn−1+an−1sn−2t+⋯+a1tn−1)a_0t^n = -s\big(a_ns^{n-1}+a_{n-1}s^{n-2}t+\cdots+a_1t^{n-1}\big)

The right side is an integer multiple of ss, so ss divides a0tna_0t^n. Since ss is coprime to tt, the lemma gives that ss divides a0a_0. From the same equality, moving now ansna_ns^n to the other side, we obtain ansn=−t(an−1sn−1+an−2sn−2t+⋯+a0tn−1)a_ns^n=-t\big(a_{n-1}s^{n-1}+a_{n-2}s^{n-2}t+\cdots+a_0t^{n-1}\big). This says that tt divides ansna_ns^n, and since tt is coprime to ss, the lemma gives that tt divides ana_n. ■\blacksquare

The theorem does not guarantee that any of those candidates is a root, nor even that pp has any rational root. What it provides is a finite, computable list of candidates, on which we apply the Factor Theorem by direct verification.

Let us do so with the polynomial from the beginning, p(x)=x4−2x3−13x2+14x+24p(x)=x^4-2x^3-13x^2+14x+24. Since the leading coefficient is 11, the candidates are the divisors of 2424, with both signs. We try c=1c=1: p(1)=1−2−13+14+24=24p(1)=1-2-13+14+24=24. By the Remainder Theorem, dividing by x−1x-1 leaves remainder 2424, so 11 is not a root. We try c=−1c=-1: p(−1)=1+2−13−14+24=0p(-1)=1+2-13-14+24=0, so x+1x+1 divides pp. Synthetic division by x+1x+1 (with c=−1c=-1) on the coefficients 1,−2,−13,14,241,-2,-13,14,24 gives successively 11; −2+(−1)(1)=−3-2+(-1)(1)=-3; −13+(−1)(−3)=−10-13+(-1)(-3)=-10; 14+(−1)(−10)=2414+(-1)(-10)=24; and the remainder 24+(−1)(24)=024+(-1)(24)=0. Therefore

p(x)=(x+1)(x3−3x2−10x+24)p(x)=(x+1)(x^3-3x^2-10x+24)

We repeat the process with the quotient. The candidates are still divisors of 2424, and c=2c=2 gives 8−12−20+24=08-12-20+24=0. Synthetic division with c=2c=2 on 1,−3,−10,241,-3,-10,24 gives 11; −3+2=−1-3+2=-1; −10+(−2)=−12-10+(-2)=-12; and the remainder 24+(−24)=024+(-24)=0. The quotient is x2−x−12x^2-x-12, which §1.3 factors by looking for two numbers with product −12-12 and sum −1-1: they are −4-4 and 33. In total,

p(x)=(x+1)(x−2)(x−4)(x+3)p(x)=(x+1)(x-2)(x-4)(x+3)

and the real zeros of pp are −3-3, −1-1, 22 and 44, all of multiplicity 11.


One last structural question remains, which §3.1 touched on when speaking of multiplicity without specifying how many zeros a polynomial can have in total. To count them with multiplicity we need to know how the multiplicity changes when we divide by a linear factor. We write μp(r)\mu_p(r) for the multiplicity of rr as a zero of pp, with μp(r)=0\mu_p(r)=0 if p(r)≠0p(r)\neq0.

Lemma. Let p(x)=(x−c)q(x)p(x)=(x-c)q(x) with pp nonzero. Then μp(c)=μq(c)+1\mu_p(c)=\mu_q(c)+1, and μp(r)=μq(r)\mu_p(r)=\mu_q(r) for every r≠cr\neq c.

Proof. Let m=μq(c)m=\mu_q(c), so that q(x)=(x−c)mg(x)q(x)=(x-c)^mg(x) with g(c)≠0g(c)\neq0. Then p(x)=(x−c)m+1g(x)p(x)=(x-c)^{m+1}g(x), and by the uniqueness of §3.1 §3, μp(c)=m+1\mu_p(c)=m+1. If r≠cr\neq c and j=μq(r)j=\mu_q(r), we write q(x)=(x−r)jh(x)q(x)=(x-r)^jh(x) with h(r)≠0h(r)\neq0. Then p(x)=(x−r)j[(x−c)h(x)]p(x)=(x-r)^j\big[(x-c)h(x)\big], and the bracket equals (r−c)h(r)≠0(r-c)h(r)\neq0 at x=rx=r, because it is a product of two nonzero numbers. By the same uniqueness, μp(r)=j\mu_p(r)=j. ■\blacksquare

Proposition. A polynomial pp of degree n≥1n\geq1 has finitely many real zeros, and the sum of their multiplicities is at most nn.

Proof. By induction on nn. If n=1n=1, p(x)=a1x+a0p(x)=a_1x+a_0 with a1≠0a_1\neq0 has exactly one zero, x=−a0/a1x=-a_0/a_1, of multiplicity 11: the proposition holds. Suppose the result valid for every degree less than nn and let deg⁡(p)=n\deg(p)=n. If pp has no real zeros, the sum is 0≤n0\le n. If cc is a zero, the Factor Theorem gives p(x)=(x−c)q(x)p(x)=(x-c)q(x) with deg⁡(q)=n−1\deg(q)=n-1. By the lemma, the zeros of pp are cc together with the zeros of qq, and the sum of multiplicities of pp is that of qq plus one. If n−1≥1n-1\geq1, the inductive hypothesis says that the sum for qq is at most n−1n-1; if n−1=0n-1=0, qq is a nonzero constant and has no zeros, with sum 00. In both cases the sum for pp is at most 1+(n−1)=n1+(n-1)=n. ■\blacksquare

This bound is not, in general, an equality: nothing guarantees that nn zeros exist, nor even a single one. The simplest case already shows it: p(x)=x2+1p(x)=x^2+1 has degree 22, but no real satisfies x2=−1x^2=-1, the same impossibility that §1.3 pointed out for the sum of squares and that §1.2 established for the even root of a negative number. The bound of this section says how many real zeros a polynomial can have at most; it does not say what happens with the ones that are missing.


The tools of this section, division, the Factor Theorem, the Rational Root Theorem and the bound by the degree, completely solve the problem of finding the real zeros of a polynomial when they exist. But the question with which we closed §4 was deliberately left unanswered: what are the roots that x2+1x^2+1 lacks in R\mathbb{R}? Answering it requires extending the number system itself, and that is the subject of the next section.