Primality Testing Patrick Lee 12 July 2003 (updated on 13 July 2003) Finding a Prime Number Finding a prime number is critical for publickey cryptosystems, such as RSA and DiffieHellman. Nave approach: Randomly pick a number n. Try if n is divided by 2, 3, 5, 7, ., p, where p is the largest prime number less than or equal to the square root of n. Computationally expensive. You need to pre-obtain all small prime numbers. 2 Introduction to Number Theory

Number theory: modular arithmetic on a finite set of integers Most of the randomized algorithms starts by choosing a random number from some domain and then works deterministically from there on. We hope that with high probability the chosen number has some desirable properties. Goal: Given a number n, the desired complexity is O(logn), i.e., polynomial in the length of n. 3 Computing GCD gcd(a, b): greatest common divisor of (a,b)

Euclids algorithm: a and b are co-prime iff gcd(a,b) = 1 Finding gcd(a,b) for a>b, gcd(a,b) = gcd(b, a mod b) Extended Euclids: Finding gcd d and numbers x and y such that d=ax+by 4 Groups Additive Group:

Zn = {0, 1, , n-1} forms a group under addition modulo n. Multiplicative Group: Zn* = {x | 1 <= x < n and gcd(x,n) = 1} forms a group under multiplication modulo n. For prime p, Zp* includes all elements [1,p-1]. E.g., Z6* = {1, 5} E.g., Z7* = {1, 2, 3, 4, 5, 6} 5 Chinese Remainder Theorem (CRT)

Given n1, n2,, nk are pairwise co-prime. There exists a unique r, r in [0, n = n1n2nk), satisfying r = ri mod ni for any sequence {r1,..,rk}, where ri in [0, ni). E.g., r = 2 (mod 3) r = 3 (mod 5) r = 2 (mod 7) We have r = 23, unique in [0,105). 6 Euler phi Function: phi(.) phi(n) = |Zn*| Theorem: if n= p1e1p2e2pkek, phi(n) = (p1-1)p1e1 - 1...(pk-1)pkek 1

e.g., phi(p) = p1 for prime p e.g., if n = pq, phi(n) = (p-1)(q-1) If we know phi(n), we can factorize n. Eulers Theorem: for all n and x in Zn* xphi(n) = 1 (mod n) For any prime p, xp-1 = 1 (mod p) for all x in [1, p-1]. (Fermats Little Theorem). If xn-1 <> 1, n is not prime (e.g., 45 mod 6 = 4). 7 Order and Generator ord(x): smallest t such that xt = 1 mod n

Generator: an element whose order = group size. E.g., in Z11*, ord(3) = 5, ord(2) = 10 E.g., 3 is the generator of Z7* Subgroup: generated from an element of order t < phi(n) {1,3,32=9,33=5,34=4} = {1,3,4,5,9} is a subgroup of Z11* A group is cyclic if it has a generator. For any prime p, the group Zp* is cyclic, i.e, every Zp* has a generator, say g. Zp* = {1, g, g2, g3, , gp-2}

8 Group Size Subgroup size divides group size (for all n) Group size = phi(n) We use an element of order t < phi(n) as the generator of the subgroup, (say 2 in Z7*). The subgroup spans t elements. For x in subgroup, we observe t has to divide phi(n) so that xtk = xphi(n) = 1, for some integer k. You can prove it by contradiction by assuming t does not divide phi(n). E.g., H = {1, 3, 4, 5, 9} is a subgroup of Z11*, |H| dividies | Z11*|. This proposition applies to all n (prime / composite).

9 Quadratic Residue y is a quadratic residue (mod n) if there exists x in Zn* such that x2 = y (mod n) i.e., y has a square root in Zn* Claim: For any prime p, every quadratic residue has exactly two square roots x, -x mod p. Proof: if x2 = u2 (mod p), then (x-u)(x+u) = 0 (mod p), so either p divides x-u (i.e., x=u), or p divides x+u (i.e., x=-u). It implies if x2 = 1 (mod p), x = 1 or -1. 10

Quadratic Residue (contd) Theorem: For any prime p, and g is generator, gk is a quadratic residue iff k is even. Given Zp* = {1, g, g2, g3, , gp-2} Even powers of g are quadratic residues Odd powers of g are not quadratic residues Legendre symbol: [a/p] = 1 if a is a quadratic residue mod p, and 1 if a is not a quadratic residue mod p. 11 Quadratic Residue (contd)

Theorem: For prime p and a in Zp*, [a/p] = a(p-1)/2 (mod p). Zp* is cyclic, a = gk for some k. If k is even, let k = 2m, a(p-1)/2 = g(p-1)m = 1. If k is odd, let k = 2m+1, a(p-1)/2 = g(p-1)/2 = -1. Reasons: This is a square root of 1. g(p-1)/2 <> 1 since ord(g) <> (p-1)/2. But 1 has two square roots. Thus, the only solution is -1. If n is prime, a(n-1)/2= 1 or -1. If we find a(n-1)/2 is not 1 and -1, n is composite. 12 Ideas of Primality Testing

Idea 1: If xn-1 mod n <> 1, n is definitely composite. If xn-1 mod n = 1, n is probably prime. Idea 2: If x(n-1)/2 mod n <> {1,-1}, n is definitely composite. If x(n-1)/2 mod n = {1,-1}, n is probably prime. 13 Simple Primality Testing Alg. Repeat k times:

Pick a in {2,...,n-1} at random. If gcd(a,n) != 1, then output COMPOSITE. [this is actually unnecessary but conceptually helps] If a(n-1)/2 is not congruent to +1 or -1 (mod n), then output COMPOSITE. Now, if we ever got a "-1" above output "PROBABLY PRIME" else output "PROBABLY COMPOSITE". 14 Error of the Simple Alg. The alg is BPP with error probability 1/2k. If n is prime, half of them makes a(n-1)/2 = 1. Prob. error in

each iteration is . If n is composite, error occurs if n is claimed to be PROBABLY PRIME. We use the key lemma. Key Lemma: Let n be an odd composite, not a prime power, and let t=(n-1)/2. If there exists a in Zn* such that at = -1 (mod n), then at most half of the x's in Zn* have xt = {-1,+1} (mod n). 15 Error of the Simple Alg. (contd) Let S = {x in Zn* | xt = 1 or -1} (let t = (n-1)/2). Wed like to show S is a proper subgroup of Zn*. S is a subgroup of Zn* since it's closed under multiplication (xt)(yt) = (xy)t.

Find b in Zn* but not in S. Let n = qr, where q and r are co-prime. Using the CRT notation, let b = (a,1), denoting b=a (mod q), b=1 (mod r). CRT assures the existence of b. Thus, bt = (at, 1t) = (-1, 1), implying b <> 1 and -1, since 1 = (1, 1) and -1 = (-1,-1). S is a proper subgroup. Since the subgroup size divides the group size, |S| <= |Zn*|. 16 Case of Prime-Power Composites

Key Lemma doesnt apply if n is a prime-power. However, it doesnt matter since it cannot pass the test of step (3), i.e., we are sure that a(n-1)/2 <> 1,-1 mod n for all a. Proof (assume all operations are mod n): Write n = pe, where p is prime. Consider an-1, which is equal to ape-1. Note that phi(n) = pe-1(p-1) = pe-pe-1, according to the theorem in slide 7. ape-1 = aphi(n)+pe-1-1 = ape-1-1 (by Eulers Theorem) Recursively, we get ape-1 = a-1. Since a<>1, a-1 <> 1. We have an-1 <> 1, and its square root is not 1 and -1.

Thus, if n is prime-power, it does not pass the test case in step (3). We can safely ignore the case of prime-powers in the Key Lemma. 17 Miller-Rabin Algorithm 1) pick a in {2,...,n-1} at random. 2) If an-1 != 1 (mod n), then output COMPOSITE 3) Let n-1 = 2r * B, where B is odd. 4) Compute aB, a2B, ..., an-1 (mod n). 5) 6) 7)

If we found a non {-1,+1} root of 1 in the above list, then output COMPOSITE. else output POSSIBLY PRIME. 18 Error of MR Algorithm It is RP. For prime n, the algorithm always returns prime. For non-Carmichael composite n, the algorithm returns prime with probability at most in each iteration (i.e., step 2 detects compositeness with probability at least ). Carmichael number: a composite n such that for all a in Zn*, an-1 = 1 mod n. (e.g., 561, 1729)

19 Error of MR Algorithm (Proof) Let Fn = {x in Zn* | xn-1 = 1 mod n}, the set of elements that do not violate Fermats theorem. Lemma: Let n be a composite non-Carmichael number. Then |Fn| <= |Zn*|. Clearly, Fn <> Zn* . Fn forms a group. There exists a such that an-1 <> 1 mod n. It is closed under multiplication (trivial proof!)

Fn is a proper subgroup of Zn*. |Fn| divides |Zn*|, and |Fn| is strictly less than |Zn*|. 20 Detecting Carmichael Numbers Computing aB, a2B, ..., a2rB (mod n), where B =(n-1)/2r, detects Carmichael numbers. Idea: a(n-1)/2 = {1,-1}, how about a(n-1)/4? If a(n-1)/4 = {1,-1}, how about a(n-1)/8? Prove by contradiction.

Assume n is Carmichael, for all a, aB = 1 mod n. Property: Carmichael number is the product of distinct prime. Thus, let n = p1p2..pk. Let g is a generator of Zp1*. Let a = (g, 1), i.e., a = g (mod p1), a = 1 (mod p2..pr), by CRT By assumption, aB = 1 (mod n). It implies gB = 1 (mod p1) (why?). Since g is the generator, B = p-1, which contradicts B is odd. Thus, for some a, aB <> 1. The probability is > . 21 How to Find a Prime Number? Algorithm: Randomly pick a number from [1,n-1]. Plug it into the primality testing algorithm. If fails, repeat the test with another number.

Are prime numbers rare? No. Prime number theorem: No. of prime numbers less than n ~ n/ln(n). 22 References R. Motwani and P. Raghavan, Randomized Algorithms, Ch. 14. CMU, Randomized algorithms, http://www-2.cs.cmu.edu/afs/cs/usr/avrim/w ww/Randalgs98/home.html

CLRS, Introduction to Algorithms, 2nd edition. Ch. 31. 23