Fermat & his Little Theorem
Fermat & his Little Theorem
Fermat & his Little Theorem
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
79<br />
<br />
<strong>Fermat</strong> & <strong>his</strong> <strong>Little</strong> <strong>Theorem</strong><br />
We have seen that our subject has a long <strong>his</strong>tory and tradition in many cultures and<br />
countries. But if we consider our culture as part of the culture born out of the<br />
Renaissance, then one of the people most responsible for bringing up our subject in our<br />
culture is Pierre de <strong>Fermat</strong>, the great 17 th century French mathematician whose name we<br />
encountered in the first section. In t<strong>his</strong> section we will discuss what is known as <strong>Fermat</strong>’s<br />
<strong>Little</strong> <strong>Theorem</strong>, a very useful fact and one that leads in many ways to the birth of abstract<br />
algebra later on. Before we prove it we need to investigate something interesting and<br />
entertaining.<br />
Let us recall the basic properties of the binomial coefficients:<br />
n<br />
k<br />
, which reads, n<br />
choose k , is the number of ways of choosing k objects out of a total of n objects, and it<br />
is referred to as a binomial coefficient since:<br />
n<br />
n n n<br />
2<br />
n<br />
n 2<br />
n<br />
n 1<br />
n<br />
n<br />
(1 x)<br />
x x x x x .<br />
0 1 2 n 2 n 1 n<br />
The first 12 powers are given by:<br />
1<br />
(1 x) 1 x<br />
2 2<br />
(1 x) 1 2x x<br />
3 2 3<br />
(1 x) 1 3x 3x x<br />
4 2 3 4<br />
(1 x) 1 4x 6x 4x x<br />
5 2 3 4 5<br />
(1 x) 1 5x 10x 10x 5x x<br />
6 2 3 4 5 6<br />
(1 x) 1 6x 15x 20x 15x 6x x<br />
7 2 3 4 5 6 7<br />
(1 x) 1 7x 21x 35x 35x 21x 7x x<br />
8 2 3 4 5 6 7 8<br />
(1 x) 1 8x 28x 56x 70x 56x 28x 8x x<br />
9 2 3 4 5 6 7 8 9<br />
(1 x) 1 9x 36x 84x 126x 126x 84x 36x 9x x<br />
10 2 3 4 5 6 7 8 9 10<br />
(1 x) 1 10x 45x 120x 210x 252x 210x 120x 45x 10x x<br />
11 2 3 4 5 6 7 8 9 10 11<br />
(1 x) 1 11x 55x 165x 330x 462x 462x 330x 165x 55x 11x x<br />
12 2 3 4 5 6 7 8 9 10 11 12<br />
(1 x) 1 12x 66x 220x 496x 792x 924x 792x 495x 220x 66x 12x x<br />
The following are basic facts:<br />
n<br />
0<br />
n<br />
n<br />
1, and<br />
n<br />
k<br />
n<br />
n k .<br />
They also satisfy Pascal’s Recursion:<br />
n<br />
k<br />
1<br />
n<br />
k<br />
k<br />
n<br />
1
80<br />
and are explicitly given by Newton’s Formula:<br />
n n!<br />
n( n 1)( n 2) ( n k 1)<br />
.<br />
k k!( n k)!<br />
k( k 1)( k 2) 1<br />
Suppose now we look at the expansion of 1<br />
n<br />
x mod n :<br />
1<br />
(1 x ) 0<br />
mod 1<br />
2 2<br />
(1 x) 1 x mod 2<br />
3 3<br />
(1 x) 1 x mod 3<br />
4 2 4<br />
(1 x) 1 2x x mod 4<br />
5 5<br />
(1 x) 1 x mod 5<br />
6 2 3 4 6<br />
(1 x) 1 3x 2x 3x x mod 6<br />
7 7<br />
(1 x) 1 x mod 7<br />
8 2 4 6 8<br />
(1 x) 1 4x 6x 4x x mod 8<br />
9 3 6 9<br />
(1 x) 1 3x 3x x mod 9<br />
10 2 5 8 10<br />
(1 x) 1 5x 2x 5x x mod 10<br />
11 11<br />
(1 x) 1 x mod 11<br />
12 2 3 4 8 9 10 12<br />
(1 x) 1 6x 4x 4x 4x 4x 6x x mod 12<br />
We could venture a conjecture in that the only exponents that have a nonzero<br />
coefficient are NOT relatively prime to n .<br />
Indeed,<br />
n<br />
Lemma. For any n and k , 0<br />
k<br />
n<br />
then 0<br />
k<br />
mod n .<br />
Proof. From Newton’s Formula, we get that<br />
so<br />
k<br />
n<br />
k<br />
n<br />
n<br />
k<br />
1<br />
1<br />
mod<br />
n<br />
k<br />
n<br />
n k<br />
. In particular, if n k 1 ,<br />
n! n ( n 1)! n<br />
k!( n k)! k ( k 1)!( n k)!<br />
k<br />
n<br />
, so k 0 mod n . But then we are done by cancellation. <br />
k<br />
n<br />
k<br />
1<br />
1<br />
,<br />
12<br />
Note that the converse of t<strong>his</strong> lemma is not true: 0<br />
6<br />
We could now venture another guess: that (1<br />
n<br />
x) 1<br />
that is exactly the theorem we prove now.<br />
mod 12.<br />
n<br />
x exactly when n is prime, and
81<br />
<strong>Theorem</strong>. Primeness. Let n 1. Then n is a prime number if and only if<br />
n<br />
n<br />
(1 x) 1 x mod n .<br />
Proof. One direction is clear from the previous lemma: if n is prime, then any<br />
k 1, , n 1 is relatively prime to n, and so we have our conclusion. Assume now that<br />
n is not prime. Let p be the smallest prime dividing n , so p n and ( p 1)! is<br />
n n<br />
relatively prime to n. We claim 0 mod n . T<strong>his</strong> follows from:<br />
p p<br />
n<br />
=<br />
n( n 1)( n 2) ( n p 1)<br />
,<br />
p p( p 1)( p 2) 1<br />
and so mod n , we have<br />
( p 1)! n n<br />
( n<br />
p p<br />
1)( n 2) ( n p 1) ,<br />
but then mod n, we have<br />
( p 1)!<br />
n n n 1<br />
( 1)( 2) ( p 1) ( 1) ( p<br />
p p p<br />
1)! ,<br />
and we can cancel the ( p 1)! since it is relatively prime to n, and we have arrived at<br />
n<br />
n<br />
p<br />
( 1) 1<br />
1<br />
( 1) p mod p p n . If p is odd, then 1<br />
and thus we are finished.<br />
, and if p 2 , then<br />
n<br />
2<br />
n<br />
2<br />
mod n<br />
<br />
10<br />
Example 1. 45 5<br />
2<br />
35<br />
324,632 7 mod 35.<br />
5<br />
12<br />
mod 10, 66 6<br />
2<br />
15<br />
mod 12, 455 5<br />
3<br />
mod 15 and<br />
And we are ready for a fundamental theorem:<br />
<strong>Theorem</strong>. <strong>Fermat</strong>'s <strong>Little</strong> <strong>Theorem</strong>. Let p be a prime. Then the<br />
following are true:<br />
p<br />
For any m , m m mod p .<br />
p 1<br />
If m is relatively prime to p , then m 1 mod p .<br />
Proof. We start by proving that the two statements are equivalent: If is true, and m is<br />
relatively prime to p , then by cancellation, follows. Conversely, assume . Then by<br />
simply multiplying we get when m is relatively prime p . And if m is not relatively<br />
prime to p , then m 0 and so m p m mod p is trivial. We proceed to prove ,<br />
namely<br />
p<br />
m m mod p by induction on m . If m 1, it is obvious. Suppose the theorem<br />
holds for m , so<br />
p<br />
m m mod p . We now have to show that it holds for m 1, that is,<br />
p<br />
we have to prove that mod p , ( m 1) m 1 . But by the previous theorem and by<br />
p p<br />
induction, we get: ( m 1) m 1 m 1, and we are done.
82<br />
Example 2. Let us consider p 7 , then since<br />
every m is congruent to a standard residue, we<br />
only need to consider those.<br />
Example 3. On the Number 341 Again.<br />
30<br />
2 1073741824 1 mod 31. But actually<br />
5<br />
2 32 1 mod 31, so although we know one power (namely p 1) that will give 1, it<br />
may not be the smallest power. We will pursue t<strong>his</strong> smallest power below. Once<br />
5<br />
10 5 2 2<br />
2 1 mod 31, then, of course, 2 (2 ) 1 1 mod 31. By the theorem,<br />
10<br />
10<br />
2 1 mod 11, and so since 11 and 31 are relatively prime, 2 1 mod 341, and so<br />
341 10 34<br />
2 (2 ) 2 2 mod341 , a fact we encountered in a previous section, except now we<br />
can argue it with almost no computational aides.<br />
One can wonder whether the property of the theorem characterizes primes. In other<br />
p<br />
words, suppose m m mod p for all m , is then p necessarily a prime? Or in the other<br />
version, if whenever m and p are relatively prime, we have that<br />
then necessarily a prime?<br />
p 1<br />
m 1 mod p , is p<br />
Example 4. On the Number 561, a Pseudoprime. Let us consider a very special<br />
number: p 561 3 11 17 . If we take an m relatively prime to 561, then we know<br />
2<br />
10<br />
16<br />
that m 1 mod 3, m 1 mod 11 m 1 mod 17 since we can apply the theorem to<br />
the three primes, and so we have that<br />
560 2 280<br />
560 10 56<br />
560 16 35<br />
m ( m ) 1 mod 3, m ( m ) 1 mod 11 and m ( m ) 1 mod 17,<br />
thus<br />
560<br />
m 1 mod 561, so holds for 561. But how about ?<br />
m m 7 mod( m 7 ,7) m 6 mod( m<br />
6 ,7)<br />
1 1 1 1 1<br />
2 128 2 64 1<br />
3 2187 3 729 1<br />
4 16384 4 4096 1<br />
5 78125 5 15625 1<br />
6 279936 6 46656 1<br />
Computing powers using the squaring algorithm from previously, we can see that mod<br />
561<br />
561, where the bolded columns are the ones relevant to the power 561, 3 3,<br />
561<br />
561<br />
11 11 and 17 17:<br />
1 2 4 8 16 32 64 128 256 512 561<br />
3 9 81 390 69 273 477 324 69 273 3<br />
11 121 55 220 154 154 154 154 154 154 11<br />
17 289 493 136 544 289 493 136 544 289 17<br />
e f g<br />
And thus if m is arbitrary, then m 3 11 17 k where k is relatively prime to 561, and<br />
the exponents e, f and g are nonnegative. Then mod 561,<br />
561 e 561 f 561 g 561 561 561 e 561 f 561 g 560 e f g<br />
m (3 ) (11 ) (17 ) k (3 ) (11 ) (17 ) k k 3 11 17 k m<br />
and thus <strong>Fermat</strong>’s <strong>Little</strong> <strong>Theorem</strong> also holds for 561. But 561 is not prime. So it is<br />
referred to as a pseudoprime. The existence of infinitely many such pseudoprimes was<br />
proven in 1992.
83<br />
Alas, 341 is not a pseudoprime since<br />
341<br />
3 168 mod 341.<br />
One of the reasons why statement of the theorem is more attractive is because it is<br />
more powerful algebraically—any power of 1 is also 1, and we used t<strong>his</strong> property several<br />
times in the argument above, for example claiming that since mod 17, m 16 1, m 560 1<br />
also. Euler and <strong>his</strong> successful pursuit of the generalization of <strong>Fermat</strong>’s <strong>Little</strong> <strong>Theorem</strong><br />
will be the topic of the next section, and it is statement that he will generalize.<br />
So the key idea becomes that of the smallest positive exponent that will make the<br />
power of a number equal to 1 in a given mod. T<strong>his</strong> smallest power is called the order<br />
of the number in that modulus.<br />
So <strong>Fermat</strong>’s <strong>Little</strong> <strong>Theorem</strong> claims that modulo a prime p , the order of anything<br />
(which is not a multiple of p ) is at most p 1. We will soon improve on t<strong>his</strong> once we<br />
understand order better.<br />
Example 5. On the Order of 2 mod 3, 5, 7, 9, 11 and 13. Let us consider the powers of<br />
2 in several different moduli:<br />
Exponent 1 2 3 4 5 6 7 8 9 10 11 12<br />
Power 2 4 8 16 32 64 128 256 512 1024 2048 4096<br />
mod 3 2 1 2 1 2 1 2 1 2 1 2 1<br />
mod 5 2 4 3 1 2 4 3 1 2 4 3 1<br />
mod 7 2 4 1 2 4 1 2 4 1 2 4 1<br />
mod 9 2 4 8 7 5 1 2 4 8 7 5 1<br />
mod 11 2 4 8 5 10 9 7 3 6 1 2 4<br />
mod 13 2 4 8 3 6 12 11 9 5 10 7 1<br />
So the order of 2 mod 3 is 2, mod 5 is 4, mod 7 is 3 (not 6), mod 9 is 6, mod 11 is 10 and<br />
mod 13, the order is 12. So the theorem predicted the order most of the time. It did not for<br />
3<br />
3 2 2<br />
7, but in any case 2 1, so (2 ) 1 1, so the theorem is satisfied. Mod 9 is not<br />
6<br />
12<br />
relevant because 9 is not a prime, but once we have 2 1, then certainly 2 1 also. Do<br />
1234<br />
10 1230<br />
1234 4<br />
we know 2 ? mod 11? Since 2 1, 2 1, so 2 2 16 5 mod 11.<br />
We should clarify that not everything has an order mod anything: for example mod 9,<br />
m<br />
3 0 for all m 2, or mod 12, the powers of 2 are, respectively, 2, 4, 8, 4, 8, … and so<br />
on, so not power of 2 will ever be 1.<br />
The next lemma explains it better.
84<br />
<strong>Theorem</strong>. Existence of Order. Let k be a positive integer. Then a has<br />
an order mod k if and only if they are relatively prime, a k 1.<br />
m<br />
Proof. Suppose a has an order mod k , so a 1 mod k for some positive integer m .<br />
m<br />
Let p be a common divisor of a and k . Then mod p , a 0 , but a 1, also, so p 1<br />
and we have a k 1. Conversely, suppose a and k are relatively prime. Since the<br />
2 3 4<br />
powers of a : a, a , a , a ,... go on indefinitely so two of them have to congruent mod k ,<br />
say for i j,<br />
a<br />
i<br />
j<br />
a mod k . Because of relative primeness, we can cancel a , so we get<br />
i 1 j 1<br />
i 2 j 2<br />
mod k , a a , and then a a , and so on until we obtain 1<br />
positive power of a is 1, and so order exists.<br />
j i<br />
a<br />
, so some<br />
<br />
Example 6. On the Order Mod 10. Thus mod<br />
10, only 1, 3, 7 and 9 will have an order.<br />
Indeed the table verifies that:<br />
Thus the order of 1 is 1 (always), the order of 3<br />
is 4, the order of 7 is 4 also and the order of 9<br />
is 2.<br />
By the fifth power we have a repetition in each<br />
row, so the table will just keep repeating.<br />
Exponent 1 2 3 4 5<br />
1 1 1 1 1 1<br />
2 2 4 8 6 2<br />
3 3 9 7 1 3<br />
4 4 6 4 6 4<br />
5 5 5 5 5 5<br />
6 6 6 6 6 6<br />
7 7 9 3 1 7<br />
8 8 4 2 6 8<br />
9 9 1 9 1 9<br />
Let us pursue properties of order further. Suppose for example, that, mod k , for some k ,<br />
m 44 1 and that m 43 1 also. Then without hesitation we can claim m 1, since<br />
44 43<br />
1 m m m m , all mod k .<br />
Suppose instead we had that m 44 1 and also m 32 1 (all t<strong>his</strong> mod k ). What can we<br />
44 32 12 12<br />
claim then? Easily, 1 m m m m and we conclude m 12 1. But t<strong>his</strong> is not the<br />
best conclusion we can make. Since 4 is the g.c.d. of 32 and 44, we know we can write it<br />
as a combination of them: 4 3 44 4 32, and so 3 44 4 4 32 , and thus<br />
44 3 4 4 32 4 32 4 4<br />
1 ( m ) m m ( m ) m , and we have a superior result. The superiority<br />
4<br />
stems from the fact that from m 1, we can readily conclude that m 12 1 easily—the<br />
cube of 1 is 1, but not the other way around—just because the cube is 1 does not mean<br />
the number is 1. In fact, we could not conclude that m 2 1 for the same reasons, since<br />
m 2<br />
4<br />
2<br />
1 is possible. E.g., 2 1 mod 5 and 2 1.<br />
Roughly a century and a half after <strong>Fermat</strong>, these ideas would be developed much further<br />
and gain much power in the hands of Lagrange, Gauss and Galois at the time abstract<br />
algebra is born.<br />
And we are ready with a fundamental fact about order.
85<br />
<strong>Theorem</strong>. Order. Let a be an integer. Let m, n,<br />
k be positive. If<br />
m<br />
a<br />
n<br />
1 mod k and a<br />
m n<br />
1 mod k , then also a 1 mod k .<br />
Proof. Without loss, by Kuttaka, if we let d m n, then we can assume there are<br />
positive integers rs , such that d rm sn , or equivalently d sn rm. But then mod<br />
k , of course,<br />
and the result is established.<br />
r m<br />
r<br />
d sn d sn d n<br />
s<br />
d s d<br />
1 1 a a a a a a a 1 a<br />
<br />
Corollary. Meaning of Order. Let a have order n modulo k . Then for<br />
m<br />
any m , a 1 mod k if and only if n|<br />
m.<br />
m n<br />
Proof. One direction is clear: if n|<br />
m, then mod k , we have<br />
n n<br />
a ( a ) 1 1.<br />
m<br />
m n<br />
Conversely, suppose a 1, but then a 1, and since n is smallest positive number<br />
for which t<strong>his</strong> happens, m n n and we are done. <br />
As a direct consequence of t<strong>his</strong> corollary and <strong>Fermat</strong>’s <strong>Little</strong> <strong>Theorem</strong>, we get<br />
Corollary. Order Mod a Prime. Let p be a prime and let m p 1. Then<br />
the order of m mod p is a divisor of p 1.<br />
Example 7. On the Order Mod 11. The order of anything mod 11 is thus a divisor of 10:<br />
Exponent 1 2 3 4 5 6 7 8 9 10<br />
1 1 1 1 1 1 1 1 1 1 1<br />
2 2 4 8 5 10 9 7 3 6 1<br />
3 3 9 5 4 1 3 9 5 4 1<br />
4 4 5 9 3 1 4 5 9 3 1<br />
5 5 3 4 9 1 5 3 4 9 1<br />
6 6 3 7 9 10 5 8 4 2 1<br />
7 7 5 2 3 10 4 6 9 8 1<br />
8 8 9 6 4 10 3 2 5 7 1<br />
9 9 4 3 5 1 9 4 3 5 1<br />
10 10 1 10 1 10 1 10 1 10 1<br />
so the respective orders of 1, 2, 3, 4, 5, 6, 7, 8 and 9 are: 1, 10, 5, 5, 5, 10, 10, 10, 5 and 2.<br />
m<br />
m<br />
For the remainder of the section we explore some classical results with our new found<br />
power based on the concept of order.<br />
Let a b 1 be relatively prime integers. Consider the sequence<br />
n n<br />
A ( , ) a b<br />
n<br />
a b An<br />
a b ,
86<br />
where we will use An<br />
( a, b ) for clarification when necessary.<br />
Some concrete examples follow:<br />
a b A<br />
1<br />
A<br />
2<br />
A<br />
3<br />
A<br />
4<br />
A<br />
5<br />
A<br />
6<br />
A<br />
7<br />
A<br />
8<br />
A<br />
9<br />
A<br />
10<br />
2 1 1 3 7 15 31 63 127 255 511 1023<br />
3 2 1 5 19 65 211 665 2059 6305 19171 58025<br />
4 3 1 7 37 175 781 3367 14197 58975 242461 989527<br />
5 2 1 7 39 203 1031 5187 25999 130123 650871 3254867<br />
5 3 1 8 49 272 1441 7448 37969 192032 966721 4853288<br />
6 1 1 7 43 259 1555 9331 55987 335923 2015539 12093235<br />
6 5 1 11 91 671 4651 31031 201811 1288991 8124571 50700551<br />
7 2 1 9 67 477 3355 23517 164683 1152909 8070619 56494845<br />
7 5 1 12 109 888 6841 51012 372709 2687088 19200241 136354812<br />
8 3 1 11 97 803 6505 52283 418993 3354131 26839609 214736555<br />
8 5 1 13 129 1157 9881 82173 673009 5462197 44088201 354658733<br />
9 2 1 11 103 935 8431 75911 683263 6149495 55345711 498111911<br />
9 5 1 14 151 1484 13981 128954 1176211 10664024 96366841 869254694<br />
9 7 1 16 193 2080 21121 206896 1979713 18640960 173533441 1602154576<br />
10 1 1 11 111 1111 11111 111111 1111111 11111111 111111111 1111111111<br />
10 3 1 13 139 1417 14251 142753 1428259 14284777 142854331 1428562993<br />
10 7 1 17 219 2533 27731 294117 3058819 31411733 319882131 3239174917<br />
Table of<br />
An<br />
s for various a s & b s<br />
As the table exhibits, the first claim we can make is that<br />
A is always an integer.<br />
n<br />
n n n n<br />
n n<br />
That is easy: mod a b, a b, so a b , so a b 0 , so ( a b) | ( a b ) and so<br />
A<br />
n<br />
is a (positive) integer. T<strong>his</strong> argument may look familiar, but it is both beautiful and<br />
useful so it is worth repeating. Also those among the readers that are algebraically<br />
1 2 3 2 2 1<br />
minded will readily recognize that A a n a n b a n b ab n b n .<br />
The next remark is as easy as the first one:<br />
if n|<br />
m, then A | A .<br />
Again, we do modular arithmetic: mod<br />
n<br />
n n<br />
a b ,<br />
n<br />
m<br />
m<br />
m<br />
n n m n<br />
a b , so ( ) n n n m<br />
( )<br />
m m<br />
so a b 0, and so by the Cancellation <strong>Theorem</strong>, mod<br />
A<br />
m<br />
m<br />
a<br />
a<br />
b<br />
b<br />
m<br />
0 .<br />
Next we need to show that<br />
a ( a b) 1 b ( a b ) and a A 1 b A .<br />
n<br />
a a b b ,<br />
n<br />
A<br />
n<br />
n n<br />
a b<br />
a b ,
87<br />
The first claims are immediate, since if for some mod prime p , a 0 and a b 0 ,<br />
then b 0 also, which is impossible; similarly for the second equality. Suppose now p<br />
n n n<br />
a b b<br />
is a prime and mod p , A<br />
n<br />
0 and a 0. But then An<br />
a b b<br />
necessarily b 0 which is impossible; ; similarly for the last equality.<br />
We are on our way to proving the following very elegant result<br />
<strong>Theorem</strong>. GCDs and<br />
Then for any ,<br />
b<br />
n 1<br />
A ' s. Let a b 1 be relatively prime integers.<br />
nm:<br />
n m n m<br />
n<br />
A A A .<br />
, and so<br />
Proof. Let d n m, then we know already by the previous discussion that A | A and<br />
Ad<br />
| A<br />
m. Now let q be a common divisor of A<br />
n<br />
and A<br />
m<br />
. But then we know that ( a b)<br />
q<br />
n n<br />
m m<br />
is a common divisor of a b and a b . We also know that since q and a are<br />
relatively prime, (and similarly for b ), hence a ( a b) q 1 b ( a b)<br />
q . Thus, mod<br />
( a b)<br />
q , we have that<br />
n<br />
n n a<br />
a b , so 1<br />
b<br />
m<br />
a<br />
, and similarly 1 , so by the Order<br />
b<br />
d<br />
a<br />
d d d d<br />
<strong>Theorem</strong>, 1. But then mod ( a b)<br />
q , we have a b , so a b 0, so by<br />
b<br />
Cancellation, A<br />
d<br />
0 mod q , and we are finished. <br />
Example 8. Let a 7 and b 2 . Then A<br />
6<br />
23517 and A<br />
9<br />
8070619 . And as readily<br />
verified by the Euclidean algorithm, A6 A9 67 A<br />
3<br />
. Similarly, if a 8 and b 5,<br />
then A<br />
10<br />
354658733 and A<br />
8<br />
5462197 and A10 A8 13 A<br />
2<br />
.<br />
The reader may suspect that the relative primeness of a and b was essential for t<strong>his</strong><br />
theorem, and indeed that is the case.<br />
d<br />
n<br />
Corollary. Primeness of<br />
A<br />
n<br />
. If<br />
A<br />
n<br />
is prime, then n is prime.<br />
Proof. If n has a proper prime divisor p , then A<br />
p<br />
1 is a proper divisor of A<br />
n<br />
.<br />
<br />
However, the converse is not true: e.g. if a 4 and b 3, then A<br />
5<br />
781 11 71. In<br />
fact, no one knows whether for any specific a and b , there is always a prime among the<br />
A ' s, or infinitely many primes, or whatever.<br />
n<br />
Some specific cases in the table are particularly interesting—we start with a very<br />
classical situation.
88<br />
Example 9. The Mersenne Primes. Let a 2 and b 1.<br />
Then 2 n<br />
A 1<br />
n<br />
and these are<br />
called the Mersenne Numbers after the cleric and letter writer of the 17 th century, a<br />
contemporary of <strong>Fermat</strong>—for <strong>his</strong>torical reasons, we will denote them by M .<br />
The primes among them are naturally the Mersenne Primes. As we saw above, if<br />
to be a prime, n has to be also. In fact,<br />
2<br />
3<br />
primes, but M 2047 23 89 is not prime.<br />
11<br />
M , 7<br />
M<br />
n<br />
is<br />
M , M 31 and M 127 are all<br />
And although one sees quite few Mersenne Primes in the table above, they are much<br />
scarcer than they seem at first.<br />
In fact, at present, there are only 47 known Mersenne Primes, those for the following<br />
exponents n :<br />
2 3 5 7 13 17 19 31<br />
61 89 107 127 521 607 1279 2203<br />
2281 3217 4253 4423 9689 9941 11213 19937<br />
21701 23209 44497 86243 110503 132049 216091 756839<br />
859433 1257787 1398269 2976221 3021377 6972593 13466917 20996011<br />
24036583 25964951 30402457 32582657 37156667 42643801 43112609<br />
The first prime ever found by a computer was<br />
known prime known today is<br />
was discovered in 2008!<br />
3<br />
5<br />
M in 1952, and the largest<br />
521<br />
521<br />
2 1<br />
43,112,609<br />
M<br />
43,112,609<br />
2 1 which has 12,978,189 digits and<br />
p<br />
Let us consider a prime divisor q of M 2 1 where p is a prime. Then we know that<br />
p<br />
p<br />
mod q , 2 1. Since the order of 2 mod q has to be a divisor of p , then we know that<br />
the order is p since it is not 1. But then by <strong>Fermat</strong>’s <strong>Little</strong> <strong>Theorem</strong>, p| q 1. Thus<br />
q kp 1 and since q is odd, it must be that k is even, and so we have proven that<br />
If q is a prime factor of M , then q 2kp 1.<br />
In the list of Mersenne Primes given above, the first five omissions were: 11, 23, 29, 37,<br />
and 41. The table below illustrates why:<br />
p M p<br />
Primes of the form 2kp 1 Factorization of M p<br />
11 2047 23, 67, 89, … 23 89<br />
23 8388607 47, 139, 277, 461,… 47 178481<br />
29 536870911 233, 349, 523, 639,… 233 1103 2089<br />
37 13743895347 149, 223, 593, 1259,… 223 616318177<br />
41 2199023255551 83, 329, 739, 821,… 13367 164511353<br />
Later on we will see a connection between Mersenne primes and the Classical Greece<br />
idea of a perfect number.<br />
p<br />
7<br />
n
89<br />
Example 10. Repunits Revisited. Consider the case a 10 and b 1, then we can<br />
recognize A<br />
n n, the repunit with n digits—all ones.<br />
Thus, when we apply the theorem to t<strong>his</strong> case, we get<br />
the g.c.d. of two repunits is the repunit of the g.c.d..<br />
For example,<br />
100 130 10 .<br />
And of course if<br />
n<br />
is to be a prime, then n will have to be also,<br />
2<br />
One can extend the idea of the An<br />
s as follows. Let p( x)<br />
x mx n be a polynomial<br />
with integer coefficients, so mn , , and assume m n 1. Let and be its roots<br />
(which may be complex numbers). Note that m and n . We define a<br />
sequence as follows:<br />
S ( m, n)<br />
S .<br />
k<br />
The case we discussed above corresponded to the case m a b and n ab since our<br />
roots were a and b . For example, the Mersenne case corresponds to the polynomial<br />
2<br />
p( x) x 3x 2.<br />
It is not difficult to show that S<br />
1<br />
1 and S2<br />
m, and from then on<br />
Sk 1<br />
mSk nS<br />
k 1,<br />
and thus we can conclude they are all integers.<br />
Although the proofs are a bit more technical and not necessarily appropriate for t<strong>his</strong><br />
course, we need to remark that it is still true that<br />
if d | k, then Sd<br />
| S<br />
k<br />
,<br />
and that<br />
Sk Sl S<br />
k l<br />
.<br />
One of the most famous cases of t<strong>his</strong> generalization is<br />
Example 11. The Fibonacci Numbers. Take the polynomial<br />
1 5<br />
m n 1. Then<br />
and<br />
2<br />
Sk 1<br />
Sk S<br />
k 1, we have that Sk F<br />
k, the<br />
k<br />
k<br />
k<br />
2<br />
p x x x 1, so<br />
1 5<br />
. Since S<br />
1<br />
1, S2 m 1 and<br />
2<br />
th<br />
k Fibonacci number. So we have that<br />
5<br />
and we also get that Fk Fl F<br />
k l<br />
as mentioned in a previous section.<br />
F<br />
k<br />
k<br />
k
90<br />
Similarly to the<br />
A ' s, we could consider<br />
n<br />
n n<br />
Bn<br />
a b ,<br />
where we still assume a and b are relatively prime. As before some examples are given<br />
below:<br />
a b B<br />
1<br />
B<br />
2<br />
B<br />
3<br />
B<br />
4<br />
B<br />
5<br />
B<br />
6<br />
B<br />
7<br />
B<br />
8<br />
B<br />
9<br />
B<br />
10<br />
2 1 3 5 9 17 33 65 129 257 513 1025<br />
3 2 5 13 35 97 275 793 2315 6817 20195 60073<br />
4 3 7 25 91 337 1267 4825 18571 72097 281827 1107625<br />
5 2 7 29 133 641 3157 15689 78253 390881 1953637 9766649<br />
5 3 8 34 152 706 3368 16354 80312 397186 1972808 9824674<br />
6 1 7 37 217 1297 7777 46657 279937 1679617 10077697 60466177<br />
6 5 11 61 341 1921 10901 62281 358061 2070241 12030821 70231801<br />
7 2 9 53 351 2417 16839 117713 823671 5765057 40354119 282476273<br />
7 5 12 74 468 3026 19932 133274 901668 6155426 42306732 292240874<br />
8 3 11 73 539 4177 33011 262873 2099339 16783777 134237411 1073800873<br />
8 5 13 89 637 4721 35893 277769 2175277 17167841 136170853 1083507449<br />
9 2 11 85 737 6577 59081 531505 4783097 43046977 387421001 3486785425<br />
9 5 14 106 854 7186 62174 547066 4861094 43437346 389373614 3496550026<br />
9 7 16 130 1072 8962 75856 649090 5606512 48811522 427774096 3769259650<br />
10 1 11 101 1001 10001 100001 1000001 10000001 100000001 1000000001 10000000001<br />
10 3 13 109 1027 10081 100243 1000729 10002187 100006561 1000019683 10000059049<br />
10 7 17 149 1343 12401 116807 1117649 10823543 105764801 1040353607 10282475249<br />
Table of<br />
The claims about the<br />
nevertheless. It is still true that<br />
Bn<br />
s for various a s & b s<br />
Bn<br />
s are not as elegant as those for the<br />
B a 1 B b<br />
in a similar fashion: if mod prime p , B<br />
n<br />
0 and a 0, then so is b 0<br />
Here the relevant theorem is a little weaker:<br />
<strong>Theorem</strong>. GCDs and<br />
and<br />
Bn<br />
m<br />
are both odd, then Bn Bm B<br />
n m<br />
.<br />
n m<br />
n<br />
n<br />
' s. Let n and m be positive integers. If<br />
An<br />
s, but they are interesting<br />
, and vice versa.<br />
n<br />
Proof. The proof is similar to the one before. Let d n m and k . We consider<br />
d<br />
d d<br />
d d n d k d k n<br />
computations mod a b . In that mod, a b , so a ( a ) ( b ) b since k<br />
is odd. Similarly,<br />
a<br />
m<br />
m<br />
b , and so we get that<br />
d d<br />
a b is a common divisor of<br />
m<br />
n<br />
n<br />
m<br />
B and<br />
B<br />
n<br />
. As before, we can assume there exists positive integers r and s such that<br />
d rm sn. But then 1<br />
r m<br />
s n<br />
d d<br />
. So mod 2, we get that 1 r s, or in other words, r
91<br />
and s have different parities. Let q be a common divisor of<br />
m<br />
a<br />
Then mod q , we have that 1<br />
b<br />
a<br />
b<br />
rm<br />
1<br />
B<br />
m<br />
and<br />
B<br />
n<br />
(so q b 1).<br />
n<br />
a<br />
and 1. But then (still mod q )<br />
b<br />
ns<br />
a<br />
depending on whether r is even or odd, but also 1<br />
b<br />
have different parity. Now since<br />
a<br />
b<br />
d<br />
1<br />
mod q , and so q divides<br />
since r and s<br />
d mr ns<br />
a a a<br />
, we get that in either case<br />
b b b<br />
d d<br />
a b . <br />
Example 12. Let a 7 and b 2 . Then B<br />
6<br />
117713 and B<br />
9<br />
40354119<br />
. One can<br />
verify that B6 B<br />
9<br />
1, and the conclusion of the theorem does not hold. But neither does<br />
6<br />
the hypothesis since 6 9 3 and<br />
3<br />
B<br />
10<br />
282476273 and B6 B10 53 B<br />
2<br />
.<br />
2 , which is even. On the other hand,<br />
As a corollary, we get the following statement (which also explains what occurred in the<br />
previous example):<br />
Corollary. Let d be a divisor of n . Then<br />
only if d<br />
n is odd. Moreover, if d<br />
n is even, then<br />
B<br />
d<br />
is a divisor of<br />
B<br />
d<br />
divides<br />
B<br />
n<br />
if and<br />
n d<br />
Proof. If is odd, since n d d and is odd, we get n<br />
Bn Bd B<br />
d<br />
. If is even,<br />
d d d<br />
then mod<br />
d d<br />
a b ,<br />
n n<br />
An<br />
a b , not<br />
a<br />
d<br />
d<br />
b , so<br />
n<br />
n d d n<br />
a b b , so<br />
A<br />
n<br />
.<br />
d d<br />
Bd<br />
a b is a factor of<br />
n n<br />
Bn<br />
a b . <br />
Example 13. Let a 7 and b 2 . Then B3<br />
351 and A6 23517 B<br />
3<br />
67 .<br />
In particular,<br />
n n<br />
if Bn<br />
a b is prime, then n cannot have any odd divisors<br />
—in other words,<br />
n has to be a power of 2 ,<br />
and of course, one of a and b has to be even.<br />
Here are a few values (the primes are bolded):
92<br />
a<br />
b<br />
B<br />
1<br />
B<br />
2<br />
B<br />
4<br />
B<br />
8<br />
B16<br />
2 1 3 5 17 257 65537<br />
3 2 5 13 97 6817 43112257<br />
4 1 5 17 257 65537 4294967297<br />
4 3 7 25 337 72097 4338014017<br />
5 2 7 29 641 390881 152587956161<br />
6 1 7 37 1297 1679617 2821109907457<br />
10 1 11 101 10001 100000001 10000000000000001<br />
Finally, we end the section with <strong>Fermat</strong>’s most famous mistake. The most interesting of<br />
all the B s is the case a 2 . and b 1 In that case they are named after <strong>Fermat</strong> himself.<br />
n<br />
n<br />
Example 14. The <strong>Fermat</strong> Primes. So we are concerned now with B 2 1. We know<br />
that if<br />
B is to be a prime, then n has to be a power of 2. So we will refer to these<br />
n<br />
k<br />
2<br />
2 1<br />
numbers: Fk as the <strong>Fermat</strong> numbers. We see that F<br />
0<br />
3, F<br />
1<br />
5, F<br />
2<br />
17 ,<br />
F<br />
3<br />
257 , and F<br />
4<br />
65537 are all primes indeed. <strong>Fermat</strong>’s error in prediction came when<br />
he conjectured that all of them would be primes.<br />
Certainly F k<br />
is odd, and hence relatively prime to 2. Let p be a divisor of F k<br />
. Then the<br />
order of 2 mod p has to divide p 1 as we saw before. But mod p , 0, so<br />
2<br />
so 2 k 1<br />
1. So the order of 2 mod p has to be a divisor of<br />
n<br />
F k<br />
k<br />
2<br />
2 1<br />
1<br />
2 k , but any proper divisor<br />
of t<strong>his</strong> number is a divisor of 2 k and t<strong>his</strong> exponent does not give 1, so the order is<br />
k 1<br />
1<br />
Thus, 2 | p 1, or p 1 mod 2 k 1<br />
, or equivalently, p 1 r 2 k for some r .<br />
,<br />
1<br />
2 k .<br />
3<br />
2<br />
We can easily prove now that F 3<br />
2 1 257 is prime, since if it were to have a prime<br />
divisor p , then p 1 16r, and so the smallest possible prime is 17, but it is already<br />
bigger than<br />
257 . Short sieve!<br />
4<br />
2<br />
Let us consider F 4<br />
2 1 65537 . Any prime divisor of it would have to satisfy<br />
p 1 mod 32 , which limits p to be in the list 33, 65, 97, 129, 161, 193 and 225 since we<br />
are stopping before 65537 256. Of these, if we look at the list of primes, we only<br />
have two: 97and 193. Since these are not factors of 65,537 we can conclude the latter is<br />
prime. Short sieve again!<br />
Unfortunately for <strong>Fermat</strong> who believed all the F k<br />
’s were primes, Euler (who was the<br />
person responsible for first proving <strong>Fermat</strong>’s own <strong>Little</strong> <strong>Theorem</strong>) proved that<br />
32<br />
F5<br />
2 1=4294967297= (641)(6700417)<br />
is not a prime. As usual, any prime divisor p of it would have to satisfy p 1 mod 64 .<br />
The first five primes that satisfy p 1 mod 64 are 193, 257, 449, 577 and 641. Thus, the
93<br />
factor 641 is not that difficult to find. Note that the other factor: 6700417 is of the same<br />
form: 6700417 64 104694 1, which is a little further on in the list to be tested.<br />
In fact, we know of no other <strong>Fermat</strong> primes besides those 5 on the list above: 3, 5, 17,<br />
257, and 65537. As of 2008, it is known that F k<br />
is composite for 5 k 32 . Note<br />
32<br />
2<br />
F 32<br />
2 1<br />
4294967296<br />
2 1.<br />
More than 100 years after <strong>Fermat</strong>, and shortly after Euler’s death, Gauss made an<br />
unexpected and remarkable connection between <strong>Fermat</strong> primes and constructability of a<br />
regular polygon by straightedge and compass alone. Namely, Gauss proved that<br />
a polygon can be built with straight edge and compass alone if and only the<br />
number of sides is a power of two times a product of distinct <strong>Fermat</strong> primes!<br />
So we can construct the 17-gon, 257-gon and the 65537-gon because they are <strong>Fermat</strong><br />
primes, and the Pentadecagon (15) because it is a product of distinct <strong>Fermat</strong> primes. In<br />
particular, we cannot construct using straightedge and compass alone a regular heptagon<br />
since 7 is not a <strong>Fermat</strong> prime. Isn't t<strong>his</strong> sort of amazing?<br />
Note that since the order of 2 mod any prime divisor of F k<br />
is<br />
1<br />
2 k , we cannot have any<br />
prime dividing F k<br />
and F m<br />
, and so they are always relatively primes. T<strong>his</strong> provides<br />
another reason why there are infinitely many primes.<br />
In the next section we will pursue the generalization of <strong>Fermat</strong>’s <strong>Little</strong> <strong>Theorem</strong> to other<br />
mods besides primes—a generalization Euler accomplished after much deliberation.