A WORD ABOUT PRIMITIVE ROOTS

PETE L. CLARK

Let N be a positive integer. An integer g is said to be a primitive root modulo N if every element x of (Z/N Z) is of the form g i for some positive integer i. Equivalently, the nite group (Z/N Z) is cyclic and g (mod N ) is a generator. Wed like to nd primitive roots mod N , if possible. There are really two problems: Question 1. For which N does there exist a primitive root modulo N ? Question 2. Assuming there does exist a primitive root modulo N , how do we nd one? How do we nd all of them? We can and shall give a complete answer to Question 1. We already know that the group of units of a nite eld is nite, and we know that Z/N Z is a eld if (and only if) N is prime. Thus primitive roots exist modulo N when N is prime. When N is not prime we might as well ask a more general question: what is the structure of the unit group (Z/N Z) ? From our work on the Chinese Remainder ar 1 theorem, we know that if N = pa 1 pr , there is an isomorphism of unit groups
r

(Z/N Z) =

1 Z/(pa 1

r pa r Z)

=
i=1

i (Z/pa i Z) .

Thus it is enough to gure out the group structure when N = pa is a prime power. Theorem 1. The nite abelian group (Z/pa Z) is cyclic whenever p is an odd prime, or when p = 2 and a is 1 or 2. For a 3, we have (Z/2a Z) = Z2 Z2a2 . Before proving Theorem 1, let us nail down the answer it gives to Question 1. Corollary 2. Primitive roots exist modulo N in precisely the following cases: (i) N = 1, 2 or 4. (ii) N = pa is an odd prime power. (iii) N = 2pa is twice an odd prime power. Proof: Theorem 1 gives primitive roots in cases (i) and (ii). If p is odd, then (Z/2pa Z) = (Z/2Z) (Z/pa Z) = (Z/pa Z) since (Z/2Z) is the trivial group. Conversely, if N is not of the form (i), (ii) or (iii) then N is divisible either by 8 or by two distinct odd primes p and q . In the rst case, write N = 2a M with (2, M ) = 1 and a 3. Then (Z/N Z) = (Z/2a Z) (Z/M Z) ,
1

PETE L. CLARK

and (Z/N Z) , having the noncylic subgroup (Z/2a Z) , cannot itself be cyclic [Handout A2.5, Corollary 6]. In the second case write N = pa q b M ; then (Z/pa Z) (Z/q b Z) (Z/M Z) . (Z/N Z) = Both (Z/pa Z) and (Z/q a Z) have even order, hence their orders are not relatively prime and the product group cannot be cyclic [Handout A2.5, Corollary 10]. Proof of Theorem 1: The idea for odd p is as follows: if g is a primitive root mod p, then [Handout A2.5, Corollary 2] the order of g mod pa is divisible by k p 1, hence of the form pk (p 1) for some k a 1. Therefore g = g p has order p 1 [Handout A2.5, Proposition 7]. We claim z = 1 + p has order pa1 ; since gcd(pa1 , p 1) = 1, g z has order pa1 (p 1) [Handout A2.5, Example 4]. Lemma 3. Let p be an odd prime and z an integer which is congruent to 1 (mod p). a) ordp (z p 1) = ordp (z 1) + 1. k b) For all k Z+ , ordp (z p 1) = ordp (z 1) + k . Proof: We may write z = 1 + xp for some integer x, so ordp (z 1) = 1 + ordp (x). Then p p p (1) z p 1 = (1 + xp)p 1 = (xp) + (xp)2 + . . . + (xp)p1 + (xp)p . 1 2 p1 For the rst term on the right hand side of (1), we have ordp ( p xp) = 2 + ordp (x) = ordp (z 1) + 1. 1

The remaining terms have larger p-orders, so the p-order of z p 1 is ordp (z 1) + 1, k k 1 whence part a). Since z p 1 = (z p )p 1, part b) follows by induction. Applying Lemma 3 to z = 1 + p gives ordp (z p 1) = k for all k Z+ . So a 2 a 1 zp = 1 (mod pa ) and z p 1 (mod pa ): z has order exactly pa1 in (Z/pa Z) . Therefore, with notation as above, g z has order pa1 (p 1) = #(Z/pa Z) , so is a primitive root mod pa . Now for p = 2. Note that (Z/2Z) and (Z/4Z) have orders 1 and 2 respectively so are certainly cyclic, and we may take a 3. We claim that the subgroup of (Z/2a Z) generated by 5 has order 2a2 and is disjoint from the subgroup generated by 1, of order 2. It follows that the group is isomorphic to Z2 Z2a2 . When p = 2 Lemma 3 breaks down because the right hand side of (1) becomes just 4x + 4x2 = 4x(x + 1), whose 2-order is at least 3 + ord2 (x) if x is odd. So instead we take x even. In fact we may just take x = 2, so z = 1 + 2x = 5, ord2 (z 2 1) = ord2 (z 1) + ord2 (z + 1) = ord2 (z 1) + ord2 (6) = ord2 (z 1) + 1. Again, inductively, we get ord2 (z 2 1) = ord2 (z 1) + k, or ord2 (52 1) = k + 2. Thus for a 2, 5 has order 2a2 in (Z/2a Z) . Moreover 5k + 1 2 (mod 4) for all k , so 5k = 1 (mod 2a ), so the subgroups generated by the classes of 5 and of 1 are disjoint. This completes the proof of Theorem 1.
k k k 1

A WORD ABOUT PRIMITIVE ROOTS

Question 2 remains: when there is a primitive root, then (ZN Z) is a cyclic group, so has (n) generators, where n is its order. Since the order of (Z/N Z) is n = (N ), if there is one primitive root there are in fact exactly ((N )) of them, which is interesting. When N = p is a prime, we get that there are (p 1) primitive roots. But how many is that?? We will turn to questions like this shortly. Suppose now that N = p is prime, so we know that there are a fair number of primitive roots modulo p, but how do we nd one? This is a much deeper question. Suppose for instance we ask whether 2 is a primitive root modulo p. Well, it depends on p. Among odd primes less than 100, 2 is a primitive root modulo 3, 5, 11, 13, 19, 29, 37, 53, 59, 61, 67, 83 and is not a primitive root modulo 7, 17, 23, 31, 41, 43, 47, 71, 73, 79, 89, 97; each list has 12 members. If you extend the list you will nd that the chance 1 that 2 is a primitive root modulo p seems to dip below 2 and approach a number closer to 37%. In fact Emil Artin conjectured that with 2 replaced by any prime number a, a is a primitive root modulo (100C )% of the primes, with C=
p

1 p(p 1)

= 0.3739558136 . . . ,

and in particular that a is a primitive root modulo innitely many primes. This is yet another example of a classical problem which has seen dramatic progress in our own lifetime. Following work of Gupta and Murty in 1984 and Heath-Brown in 1986, it is now known that there are at most two bad prime numbers a such that a is a primitive root modulo only nitely many primes p. So, for instance, if 2 is not a primitive root modulo innitely many primes and 3 is not either, than we can be sure that 5 is a primitive root modulo innitely many primes! There are further concrete questions of great interest: for instance, what can be said about the smallest primitive root mod p? Or, suppose we are given p and want to nd a primitive root of p very quickly: what do we do? An extremely large literature exists on such matters.