Skip to content

Chapter 1 — Fundamental Concepts

1.2 — Exponents and Radicals

Algebra becomes powerful when we abbreviate repetition.


In §1.1 we established the rules of addition and multiplication in R\mathbb{R}. Now we take one more step. Just as, among natural numbers, multiplication abbreviates a repeated sum, exponentiation abbreviates a repeated product. This section tells how that simple idea extends, without losing coherence at any step, from natural exponents to integer and rational ones. Along the way appear the radicals, which undo exponentiation, and a question that §1.1 left open: why 2\sqrt{2} exists.

Let a∈Ra \in \mathbb{R} and n∈Nn \in \mathbb{N}. The power ana^n is the product of nn factors equal to aa:

an=a⋅a⋅⋯⋅a⏟n factorsa^n = \underbrace{a \cdot a \cdot \dots \cdot a}_{n \text{ factors}}

The number aa is called the base and nn the exponent. With a single factor, a1=aa^1 = a. The ellipsis in the formula has the same character as the one in N\mathbb{N} in §1.1. Rigorously, ana^n is defined step by step, with a1=aa^1 = a and an+1=an⋅aa^{n+1} = a^n \cdot a, and §10.2 proves that this definition is legitimate. This definition counts factors and says nothing, as yet, about zero or negative exponents.

Those cases are not defined at whim. We define them in the only way that preserves a rule that natural powers already satisfy: if m>nm > n and a≠0a \neq 0, then am/an=am−na^m / a^n = a^{m-n}, because the nn factors of the denominator cancel with nn of the mm in the numerator. If we want the rule to hold also when m=nm = n, the left side is an/an=1a^n / a^n = 1 and the right side is a0a^0. There is no choice but to define

a0=1(a≠0)a^0 = 1 \qquad (a \neq 0)

If we now want the rule to hold with m=0m = 0, the left side is a0/an=1/ana^0 / a^n = 1/a^n and the right side is a−na^{-n}. This forces us to define the negative exponent as

a−n=1an(a≠0, n∈N)a^{-n} = \frac{1}{a^n} \qquad (a \neq 0,\ n \in \mathbb{N})

Thus a−na^{-n} is exactly the multiplicative inverse of ana^n, in the sense of §1.1. Since the inverse of a product is the product of the inverses, also a−n=(1/a)na^{-n} = (1/a)^n.

The expression 000^0 is left out of these definitions, because the argument that led to a0=1a^0 = 1 divides by aa. In Calculus, "000^0" appears as an indeterminate form: expressions that approach that form can approach different values. In algebraic contexts, on the other hand, the convention 00=10^0 = 1 is convenient, and that is how the binomial theorem of §10.5 uses it. In this section we do not need it, and we leave 000^0 undefined.

For real bases aa and bb, nonzero when some exponent is zero or negative, and integer exponents mm and nn, five laws hold:

LawFormula
Product of equal basesam⋅an=am+na^m \cdot a^n = a^{m+n}
Quotient of equal basesam/an=am−na^m / a^n = a^{m-n}
Power of a power(am)n=amn(a^m)^n = a^{mn}
Power of a product(ab)n=anbn(ab)^n = a^n b^n
Power of a quotient(a/b)n=an/bn(a/b)^n = a^n/b^n

With natural exponents, each law is a matter of counting factors. Multiplying ama^m by ana^n means writing mm copies of aa followed by nn more copies, that is, m+nm + n copies. Raising ama^m to the nn means repeating the block of mm copies nn times, which gives mnmn copies. And in (ab)n(ab)^n, commutativity lets us rearrange the nn factors abab into nn factors aa followed by nn factors bb.

For arbitrary integer exponents we must check the cases that counting does not cover. We do so for the first law, and the others reduce to it.

Product of equal bases. If m,n≥0m, n \ge 0, it is the counting above, with a0=1a^0 = 1. If m≥0>nm \ge 0 > n and m+n≥0m + n \ge 0, then m=(m+n)+(−n)m = (m+n) + (-n) is a sum of two naturals or zeros, and by the previous case am=am+na−na^m = a^{m+n} a^{-n}. Multiplying by ana^n, which is the inverse of a−na^{-n}, gives aman=am+na^m a^n = a^{m+n}. If m≥0>nm \ge 0 > n and m+n<0m + n < 0, the previous case applied to −n=m+(−(m+n))-n = m + (-(m+n)) gives a−n=ama−(m+n)a^{-n} = a^m a^{-(m+n)}, and taking inverses, an=a−mam+na^n = a^{-m} a^{m+n}, whence aman=am+na^m a^n = a^{m+n}. If m,n<0m, n < 0, then aman=1a−ma−n=1a−m−n=am+na^m a^n = \dfrac{1}{a^{-m}a^{-n}} = \dfrac{1}{a^{-m-n}} = a^{m+n}. The case m<0≤nm < 0 \le n is symmetric.

The other four. The quotient is the product with a−na^{-n} in place of ana^n: am/an=ama−n=am−na^m / a^n = a^m a^{-n} = a^{m-n}. For the power of a power with n≥0n \ge 0, we apply the product law nn times: (am)n=am+⋯+m=amn(a^m)^n = a^{m + \dots + m} = a^{mn}. If n=−k<0n = -k < 0, then (am)−k=1/(am)k=1/amk=a−mk(a^m)^{-k} = 1/(a^m)^k = 1/a^{mk} = a^{-mk}. For the power of a product with n=−k<0n = -k < 0, we have (ab)−k=1akbk=a−kb−k(ab)^{-k} = \dfrac{1}{a^k b^k} = a^{-k} b^{-k}. The quotient law is the product law applied to aa and 1/b1/b.

There is a constant temptation that is best disarmed from the start: exponentiation does not distribute over addition or subtraction. The equality (a+b)n=an+bn(a+b)^n = a^n + b^n is not an identity when n≥2n \ge 2: it may hold for particular values, such as b=0b = 0, but it fails in general. The case n=2n = 2 is the most useful to remember:

(a+b)2=(a+b)(a+b)=a2+ab+ba+b2=a2+2ab+b2(a+b)^2 = (a+b)(a+b) = a^2 + ab + ba + b^2 = a^2 + 2ab + b^2

The term 2ab2ab is precisely what is lost by “distributing” the exponent. With a=b=1a = b = 1, the left side is 44 and a2+b2a^2 + b^2 is 22.


Exponentiation answers the question “what is bb raised to the nn?”. The inverse question, “what number, raised to the nn, gives aa?”, leads to root extraction. Before giving the answer a name, we must make sure that it exists and that it is unique. We begin with uniqueness, which follows from a property of order.

Lemma. Let n∈Nn \in \mathbb{N}. If 0≤x<y0 \le x < y, then xn<ynx^n < y^n.

Proof. Multiplying x<yx < y by x≥0x \ge 0 and by y>0y > 0 gives x2≤xy<y2x^2 \le xy < y^2. Repeating the argument, multiplying xk<ykx^k < y^k by xx and by yy, we pass from each exponent to the next, up to nn. ■\blacksquare

Consequently, two nonnegative numbers with the same nnth power are equal: if they were different, the smaller would have the smaller power.

Existence requires something that algebra alone does not provide: the supremum axiom of §1.1. This is the moment to fulfill the promise made there.

Theorem. Let n∈Nn \in \mathbb{N} with n≥2n \ge 2 and a≥0a \ge 0. There is a unique real number b≥0b \ge 0 such that bn=ab^n = a.

Proof. Uniqueness is the lemma. For existence, if a=0a = 0 then b=0b = 0 works, so we assume a>0a > 0. We will use an inequality: if 0≤x≤y0 \le x \le y, then

yn−xn≤nyn−1(y−x)y^n - x^n \le ny^{n-1}(y - x)

To see it, note that yn−xn=(y−x)(yn−1+yn−2x+⋯+xn−1)y^n - x^n = (y - x)(y^{n-1} + y^{n-2}x + \dots + x^{n-1}). When the product on the right is expanded, every term appears once with a plus sign and once with a minus sign, except yny^n and −xn-x^n. The second factor has nn terms, and each is at most yn−1y^{n-1} because x≤yx \le y.

Let T={x≥0∣xn≤a}T = \{x \ge 0 \mid x^n \le a\}. It is nonempty, because 0∈T0 \in T, and it is bounded above by 1+a1 + a: if x>1+ax > 1 + a, then x>1x > 1, so xn≥x>ax^n \ge x > a. By the supremum axiom, TT has a supremum; call it ss. We will prove that sn=as^n = a by ruling out the other two possibilities.

If we had sn<as^n < a, take h=min⁡{12, a−sn2n(s+1)n−1}h = \min\left\{\dfrac{1}{2},\ \dfrac{a - s^n}{2n(s+1)^{n-1}}\right\}, which is positive and less than 11. By the inequality with x=sx = s and y=s+h≤s+1y = s + h \le s + 1,

(s+h)n−sn≤n(s+h)n−1h≤n(s+1)n−1h≤a−sn2<a−sn(s+h)^n - s^n \le n(s+h)^{n-1}h \le n(s+1)^{n-1}h \le \frac{a - s^n}{2} < a - s^n

Hence (s+h)n<a(s+h)^n < a, so s+h∈Ts + h \in T and it is greater than ss, contradicting the fact that ss is an upper bound.

If we had sn>as^n > a, then s>0s > 0. Take h=min⁡{s2, sn−a2nsn−1}h = \min\left\{\dfrac{s}{2},\ \dfrac{s^n - a}{2n s^{n-1}}\right\}, which is positive and less than ss. By the inequality with x=s−hx = s - h and y=sy = s,

sn−(s−h)n≤nsn−1h≤sn−a2<sn−as^n - (s-h)^n \le n s^{n-1} h \le \frac{s^n - a}{2} < s^n - a

Hence (s−h)n>a(s-h)^n > a. Every x∈Tx \in T satisfies xn≤a<(s−h)nx^n \le a < (s-h)^n, and by the lemma x<s−hx < s - h. Thus s−hs - h is an upper bound of TT less than ss, contradicting the fact that ss is the least. With both possibilities ruled out, sn=as^n = a. ■\blacksquare

With n=2n = 2 and a=2a = 2, the theorem says that there is a unique positive real number whose square is 22: it is the number 2\sqrt{2} of §1.1, and its existence rests on the supremum axiom.

Definition. Let n∈Nn \in \mathbb{N} with n≥2n \ge 2 and a≥0a \ge 0. The nnth root of aa, written an\sqrt[n]{a}, is the unique real number b≥0b \ge 0 with bn=ab^n = a. In the expression an\sqrt[n]{a}, the sign  \sqrt{\ } is the radical, nn is the index and aa is the radicand. With index 22 the index is not written: a\sqrt{a} is the square root.

What if the radicand is negative? Here the index makes a difference. If nn is even, bn=(bn/2)2≥0b^n = (b^{n/2})^2 \ge 0 for every real bb, so no real number raised to nn gives a negative number. Negative numbers have no root of even index in R\mathbb{R}: −4\sqrt{-4} is not a real number. If nn is odd, on the other hand, (−b)n=−bn(-b)^n = -b^n, and the number −−an-\sqrt[n]{-a} satisfies (−−an)n=−(−a)=a\big(-\sqrt[n]{-a}\big)^n = -(-a) = a. For this reason, for odd nn and a<0a < 0, we define an=−−an\sqrt[n]{a} = -\sqrt[n]{-a}. For example, −83=−2\sqrt[3]{-8} = -2. This root is also unique, because for odd nn the function x↦xnx \mapsto x^n is strictly increasing on all of R\mathbb{R}: if x<0≤yx < 0 \le y, then xn<0≤ynx^n < 0 \le y^n, and if x<y<0x < y < 0, the lemma applied to 0<−y<−x0 < -y < -x gives −yn<−xn-y^n < -x^n.

The even-index restriction has a consequence that often goes unnoticed and that connects with the absolute value of §1.1.

Proposition. For every a∈Ra \in \mathbb{R} and every even nn, ann=∣a∣\sqrt[n]{a^n} = |a|.

Proof. The number ∣a∣|a| is nonnegative and, since nn is even, ∣a∣n=an|a|^n = a^n: if a≥0a \ge 0 this is immediate, and if a<0a < 0, then ∣a∣n=(−a)n=an|a|^n = (-a)^n = a^n. By the uniqueness of the root, ann=∣a∣\sqrt[n]{a^n} = |a|. ■\blacksquare

In particular, a2=∣a∣\sqrt{a^2} = |a|, and not aa. A numerical example exposes the trap: (−3)2=9=3\sqrt{(-3)^2} = \sqrt{9} = 3, which is ∣−3∣|-3| and not −3-3.

Roots allow us to extend exponentiation to rational exponents. Here we must proceed carefully, because the extension works without surprises only when the base is positive. That is why we begin with a>0a > 0.

Definition. Let a>0a > 0, m∈Zm \in \mathbb{Z} and n∈Nn \in \mathbb{N}. We define

a1/n=anam/n=(an)ma^{1/n} = \sqrt[n]{a} \qquad\qquad a^{m/n} = \left(\sqrt[n]{a}\right)^m

With n=1n = 1, understanding that a1=a\sqrt[1]{a} = a, we recover the integer power. The number am/na^{m/n} is positive, and it can also be computed as amn\sqrt[n]{a^m}. Indeed, ((an)m)n=((an)n)m=am\big((\sqrt[n]{a})^m\big)^n = \big((\sqrt[n]{a})^n\big)^m = a^m by the laws of part 2, and the uniqueness of the root gives (an)m=amn(\sqrt[n]{a})^m = \sqrt[n]{a^m}.

Before going on, we must solve a problem that the notation hides. The same rational number can be written in infinitely many ways: 12\frac{1}{2}, 24\frac{2}{4} and 50100\frac{50}{100} are the same number. The definition starts from one of those forms, mn\frac{m}{n}, without saying what happens if we use another. For the symbol am/na^{m/n} to make sense, two ways of writing the same exponent must give the same real number. This is not automatic; it must be proved.

Proposition (the rational exponent is well defined). Let a>0a > 0. If mn=pq\frac{m}{n} = \frac{p}{q}, that is, if mq=npmq = np, then am/n=ap/qa^{m/n} = a^{p/q}.

Proof. Let b=am/nb = a^{m/n} and c=ap/qc = a^{p/q}, both positive. We raise each to the integer power nqnq and use the laws of part 2:

bnq=((an)m)nq=((an)n)mq=amqcnq=((aq)p)nq=((aq)q)np=anpb^{nq} = \left(\left(\sqrt[n]{a}\right)^m\right)^{nq} = \left(\left(\sqrt[n]{a}\right)^n\right)^{mq} = a^{mq} \qquad\qquad c^{nq} = \left(\left(\sqrt[q]{a}\right)^p\right)^{nq} = \left(\left(\sqrt[q]{a}\right)^q\right)^{np} = a^{np}

Since mq=npmq = np, we get bnq=cnqb^{nq} = c^{nq}, and since bb and cc are positive, the lemma of part 3 gives b=cb = c. ■\blacksquare

With the definition secured, the five laws of part 2 hold for rational exponents when the bases are positive. For rational rr and ss we write both with a common denominator, r=mnr = \frac{m}{n} and s=pns = \frac{p}{n}. Then the product is the integer law applied to the base an\sqrt[n]{a}:

aras=(an)m(an)p=(an)m+p=ar+sa^r a^s = \left(\sqrt[n]{a}\right)^m \left(\sqrt[n]{a}\right)^p = \left(\sqrt[n]{a}\right)^{m+p} = a^{r+s}

For the power of a product, anbn\sqrt[n]{a}\sqrt[n]{b} is positive and its nnth power is abab, so abn=anbn\sqrt[n]{ab} = \sqrt[n]{a}\sqrt[n]{b} by uniqueness; raising to the mm gives (ab)m/n=am/nbm/n(ab)^{m/n} = a^{m/n} b^{m/n}. For the power of a power, let d=(am/n)p/qd = (a^{m/n})^{p/q}. Then dq=(am/n)p=amp/nd^q = (a^{m/n})^p = a^{mp/n}, so dqn=ampd^{qn} = a^{mp}, and since d>0d > 0, uniqueness gives d=amp/(nq)d = a^{mp/(nq)}. The quotient reduces to the product, as in the integer case.

Example. 82/3=(83)2=22=48^{2/3} = \left(\sqrt[3]{8}\right)^2 = 2^2 = 4, and also 823=643=4\sqrt[3]{8^2} = \sqrt[3]{64} = 4.

Why exclude negative bases? Because with them the definition ceases to be well defined. With odd index, (−8)1/3=−83=−2(-8)^{1/3} = \sqrt[3]{-8} = -2 makes sense. But 13=26\frac{1}{3} = \frac{2}{6}, and the form 26\frac{2}{6} would give (−8)26=646=2\sqrt[6]{(-8)^2} = \sqrt[6]{64} = 2. The same exponent would produce two different values. With negative bases we therefore use only radicals of odd index, an\sqrt[n]{a}, and not the fractional-exponent notation.


The properties of radicals are the laws of exponents written in another notation. With nonnegative radicands, and b>0b > 0 in the quotient,

abn=anbnabn=anbnanm=amn\sqrt[n]{ab} = \sqrt[n]{a}\sqrt[n]{b} \qquad\qquad \sqrt[n]{\frac{a}{b}} = \frac{\sqrt[n]{a}}{\sqrt[n]{b}} \qquad\qquad \sqrt[m]{\sqrt[n]{a}} = \sqrt[mn]{a}

All three are proved in the same way: the right side is nonnegative and, raised to the corresponding index, gives the radicand; uniqueness of the root does the rest. In the third, for example, (amn)mn=a\big(\sqrt[mn]{a}\big)^{mn} = a says that (amn)m\big(\sqrt[mn]{a}\big)^m is the nnth root of aa, and then amn\sqrt[mn]{a} is the mmth root of an\sqrt[n]{a}. Nesting two roots is thus equivalent to a single root whose index is the product of the indices.

The condition on the radicands matters. With even index and negative radicands, the first formula does not even make sense on the right side: (−2)(−8)=16=4\sqrt{(-2)(-8)} = \sqrt{16} = 4, but −2\sqrt{-2} and −8\sqrt{-8} are not real numbers.

We say that a radical is simplified when three conditions hold. The first is that all the factors appearing in the radicand with exponent greater than or equal to the index have been extracted. The second, that the index is as small as possible. The third, that no radicals remain in any denominator. For example, since 72=36⋅2=62⋅272 = 36 \cdot 2 = 6^2 \cdot 2, we get 72=622=62\sqrt{72} = \sqrt{6^2}\sqrt{2} = 6\sqrt{2}. Likewise, 543=33⋅23=323\sqrt[3]{54} = \sqrt[3]{3^3 \cdot 2} = 3\sqrt[3]{2}. And since 94=9=3\sqrt[4]{9} = \sqrt{\sqrt{9}} = \sqrt{3}, the index 44 reduces to 22. The third condition is the subject of the next part.

To rationalize a fraction is to transform it into an equal one whose denominator has no radicals. The procedure depends on whether the denominator has one term or two.

A single radical. Suppose the denominator is akn\sqrt[n]{a^k}, with a>0a > 0 and 0<k<n0 < k < n. For the radical to disappear, we must complete the nnth power of the radicand. That is why we multiply numerator and denominator by an−kn\sqrt[n]{a^{n-k}}:

Aakn=Aan−knaknan−kn=Aan−knann=Aan−kna\frac{A}{\sqrt[n]{a^k}} = \frac{A\sqrt[n]{a^{n-k}}}{\sqrt[n]{a^k}\sqrt[n]{a^{n-k}}} = \frac{A\sqrt[n]{a^{n-k}}}{\sqrt[n]{a^n}} = \frac{A\sqrt[n]{a^{n-k}}}{a}

Examples. 53=5333=533\dfrac{5}{\sqrt{3}} = \dfrac{5\sqrt{3}}{\sqrt{3}\sqrt{3}} = \dfrac{5\sqrt{3}}{3}, and 123=432343=432\dfrac{1}{\sqrt[3]{2}} = \dfrac{\sqrt[3]{4}}{\sqrt[3]{2}\sqrt[3]{4}} = \dfrac{\sqrt[3]{4}}{2}.

Two terms. Suppose the denominator is a+b\sqrt{a} + \sqrt{b} or a−b\sqrt{a} - \sqrt{b}, with a,b≥0a, b \ge 0 and a≠ba \neq b. Here the device is different: we multiply by the conjugate, the same expression with the middle sign changed. The difference of squares, (x+y)(x−y)=x2−y2(x+y)(x-y) = x^2 - y^2, obtained by expanding, makes the roots get squared and disappear:

Aa+b=A(a−b)(a+b)(a−b)=A(a−b)a−b\frac{A}{\sqrt{a}+\sqrt{b}} = \frac{A\left(\sqrt{a}-\sqrt{b}\right)}{\left(\sqrt{a}+\sqrt{b}\right)\left(\sqrt{a}-\sqrt{b}\right)} = \frac{A\left(\sqrt{a}-\sqrt{b}\right)}{a-b}

The condition a≠ba \neq b guarantees that the denominator a−ba - b is not zero. With a−b\sqrt{a} - \sqrt{b} in the denominator we proceed in the same way, multiplying by a+b\sqrt{a} + \sqrt{b}. For example, 15−2=5+25−2=5+23\dfrac{1}{\sqrt{5} - \sqrt{2}} = \dfrac{\sqrt{5} + \sqrt{2}}{5 - 2} = \dfrac{\sqrt{5} + \sqrt{2}}{3}.


This settles the handling of powers and roots in R\mathbb{R} and the existence of roots. The next section applies these tools to algebraic expressions: polynomials, special products and factoring.