#A18 INTEGERS 9 (2009), 191-197

CHARACTERIZATIONS OF MIDYS PROPERTY

Gilberto Garca-Pulgarn
Universidad de Antioquia, Medelln, Colombia, Grupo de

Algebra, Teora de
N umeros y Aplicaciones, ERM
[email protected]
Hernan Giraldo
1
Universidad de Antioquia, Medelln, Colombia, Grupo de

Algebra, Teora de
N umeros y Aplicaciones, ERM
[email protected]
Received: 5/30/08, Revised: 3/4/09, Accepted: 3/3/09
Abstract
In 1836 E. Midy published in France an article where he showed that if p is a prime
number, such that the smallest repeating sequence of digits in the decimal expansion of
1
p
has an even length, when this sequence is broken into two halves of equal length if these
parts are added then the result is a string of 9s. Later, J. Lewittes and H. W. Martin
generalized this statement when the length of the smallest repeating sequence of digits
is e = kd and the sequence is broken into d blocks of equal length and the expansion is
over any number base; that fact was named Midys property. We will give necessary and
sucient conditions (that are easy to check) for the integer N to satisfy Midys property.
1. Introduction
By period we mean the smallest repeating sequence of digits in the decimal expansion
of
1
N
, with N a positive integer. According to [2], E. Midy published in France, in
1836, a paper where he showed that if p is a prime number, such that the decimal
expansion of
1
p
has a period of even length, when the period is broken into two
halves of equal length if these parts are added then we get a string of 9

s; this result
is called the property of nines. For instance, 1/7 = 0.142857, the bar indicating the
period, if the period is broken into halves of equal length and then added we get
142 + 857 = 999. We have a similar situation for 1/19 = 0.052631578947368421
with period length 18 and the halves are 052631578 and 947368421.
The property of nines has gained interest over the last few years, as it is shown
by the papers of Gupta and Sury [1], J. Lewittes [2] and the work of H. W. Martin
[3]. These last two works generalized this property when the period length is any
positive integer e, the period is broken into d blocks, where d is a divisor of e and
the expansion of 1/N is over any number base.
Let us x some notation. Let B be an integer greater than 1, B is the number
base, and N a positive integer relatively prime to B. We denote by e = o
N
(B) the
order of B in the multiplicative group U
N
, where this group is the set of positive
integers less than N and relatively prime to it, and the multiplication is the product
modulo N.
We know that e is the period length of x/N, where x U
N
(see [2]). Assume
e = kd for some integers k and d. Then
x
N
= 0.a
1
a
2
a
e
and the period a
1
a
2
a
e
1
Partial support from COLCIENCIAS and CODI (Universidad de Antioquia).
INTEGERS: 9 (2009) 192
can be broken into d blocks of equal length k. Let A
j
be the number represented
by the base B numeral consisting of the jth block of k B-digits in the period. We
denote by S
d
(x) =

d
j=1
A
j
.
With the above notation, if for all x U
N
the sum S
d
(x) is a multiple of B
k
1
then we say that N has the Midys property for the base B and the divisor d of e,
the period length; in that case we write N M
d
(B).
We denote by D
B
(N) the number represented by the base B numeral consisting
of the e B-digits of the period of the fraction
1
N
. It is easy to prove that ND
B
(N) =
B
e
1.
Using similar methods to those in [1], we shall prove the following characteriza-
tions of Midys property.
Theorem 1. With the above notation: N M
d
(B) if and only if D
B
(N) 0
(mod B
k
1). Furthermore, if N M
d
(B) and
B
e
1
B
k
1
= Nt for some integer t,
then D
B
(N) = (B
k
1)t.
Theorem 2. Let N, B, e = kd = o
N
(B) be as above and p
t
be the highest power of
p that divide N. Then, N M
d
(B) if and only if for every prime factor p of N,
the following is true: if o
p
(B) | k then p
t
| d.
Furthermore, if N M
d
(B) and

d
i=1
(B
ik
mod N) = rN for some integer r,
then

d
j=1
A
j
= (B
k
1)r.
Theorem 3. Under the same conditions of Theorem 2, N M
d
(B) if and only if
for every prime divisor p of gcd(B
k
1, N), we have p
t
| d.
2. Previous Results
Theorem 2 of [2] or Theorem 1 of [3], can be written with our notation in the
following way.
Theorem 4. The following are equivalent:
1. N M
d
(B).
2. For some x U
N
, S
d
(x) 0 (mod B
k
1).
3.
B
e
1
B
k
1
0 ( mod N).
Furthermore, if B
k
1 and N are relatively prime, then N M
d
(B).
It is clear from part three of this theorem that d cannot be 1, so we always assume
e = kd with d > 1.
The following result extends Theorem 3 of [3].
INTEGERS: 9 (2009) 193
Theorem 5. Let N be a positive integer relatively prime to the number base B and
e = kd the period length of 1/N. If for every prime factor p of N the integer k is
not a multiple of the period length of 1/p then N M
d
(B).
Proof. From our assumption, if p is a prime factor of N, then gcd(p, B
k
1) = 1
and therefore gcd(p
s
, B
k
1) = 1, for every integer s. Thus, gcd(N, B
k
1) = 1
and the result is a consequence of Theorem 4.
3. Main Results
Proof of Theorem 1. According to Theorem 4, N M
d
(B) if and only if
B
e
1
B
k
1

0 (mod N). But ND
B
(N) = B
e
1 and we get the result of the theorem.
From Theorem 1 we conclude immediately that if B1 is not a divisor of D
B
(N)
then N has not the Midys property for any factor d of the order e of B modulo N.
Now, we will give some previous results for the proof of Theorem 2.
Let
1
N
= 0.a
1
a
2
a
e
be a rational number, where the period is a
1
a
2
a
e
. We
denote this number by
1
N
= 0.A
1
A
2
A
d
, where the period was broken
into d blocks of equal length k, that is, e = kd. So, as we said above, A
j
be the
number represented by the base B numeral consisting of the jth block of k B-digits
of the period. And A
i
A
j
denotes the integer represented by the base B numeral
formed by the juxtaposition of the digits of A
i
and A
j
, for 1 i, j d.
Theorem 6. Let p and t be positive numbers with p a prime number such that B 1
mod p. Suppose that e = o
p
t (B) = kd. If
1
p
t
= 0.A
1
A
2
A
d
,
then
d

j=1
A
j
=
_

_
(B
k
1)(p
t
p
i
+2)
2p
i
if there is i, 1 i t 1, such that d = p
ti
,
0 mod B
k
1 in other case.
Proof. We have the following:
1
p
t
= 0.A
1
A
2
A
d
,
B
k
p
t
= A
1
.A
2
A
3
A
d
A
1
,
B
2k
p
t
= A
1
A
2
.A
3
A
4
A
1
A
2
, (1)
.
.
.
B
(d1)k
p
t
= A
1
A
2
A
d1
.A
d
A
1
A
d1
.
INTEGERS: 9 (2009) 194
Adding all these equations we get:
B
e
1
p
t
(B
k
1)
= A
1
+ A
1
A
2
+ + A
1
A
2
A
d1
+ 0.A
1
A
2
A
d
+ 0.A
2
A
3
A
d
A
1
(2)
+ + 0.A
d
A
1
A
d1
.
Notice that on the right side of the last equation, the sum of the values smaller than
1 is,
A
1
+ A
2
+ + A
d
B
k
1
.
On the other hand,
A
1
A
2
A
i
=
_
B
ik
p
t
_
,
where x is the floor function of x. But B
ik
= p
t
_
B
ik
p
t
_
+ r
i
with r
i
B
ik
mod
p
t
, from Equation 2 we have
B
e
1
p
t
(B
k
1)
=
d1

i=0
B
ik
r
i
p
t
+
A
1
+ A
2
+ + A
d
B
k
1
and therefore
A
1
+ A
2
+ + A
d
B
k
1
=

d1
i=0
r
i
p
t
,
so that
d

j=1
A
j
=
(B
k
1)

d1
i=0
r
i
p
t
. (3)
Hence r
i
B
ik
mod p
t
, so the set {r
i
| i = 0, 1, d 1} is a subgroup of U
p
t ,
and the result of the theorem follows from the next lemma.
Lemma 7. Let p and t be positive integers with p an odd prime. Let d be a factor
of p
t1
(p 1) and let U(p, t, d) be the unique subgroup of order d of U
p
t . Then d is
a power of p, d = p
ti
, if and only if, for all a U(p, t, d) we have a 1 mod p.
Furthermore

gU(p,t,d)
g =
_
_
_
p
ti
(p
t
p
i
+2)
2
if d = p
ti
0 mod p
t
otherwise
.
Proof. If d = p
ti
, we have
U(p, t, p
ti
) =
_
1, 1 + p
i
, 1 + 2p
i
, . . . , 1 + (p
ti
1)p
i
_
.
INTEGERS: 9 (2009) 195
Because this set is closed under the product modulo p
t
, we are done. If d p
t1
, then
there exists a U(p, t, d) such that a = 1 and o
p
t (a) | p 1 so that a / U(p, t, p
t1
)
and hence a 1 mod p. We have {ag | g U(p, t, d)} = U(p, t, d), so in the group
U
p
t ,

gU(p,t,d)
ag =

gU(p,t,d)
g.
Therefore, (a1)

gU(p,t,d)
g = 0, but a 1 mod p, and then

gU(p,t,d)
g 0 mod p
t
.
When p = 2, it is known that the group U
2
t with t > 2 has two subgroups of
order 2
ti
for i = 1, 2, , t 1. These are,
U
1
(2, t, 2
ti
) =
_
1, 1 + 1 2
i
, 1 + 2 2
i
, 1 + 3 2
i
, . . . , 1 + (2
ti
1)2
i
_
and
U
2
(2, t, 2
ti
) =
_
1, 1 2
i
1, 2 2
i
1, 3 2
i
1, . . . ,
(2
ti
2)2
i
1, (2
ti
1)2
i
1
_
.
With this, we get the following proposition.
Proposition 8. With the above notation,

gU(2,t,2
ti
)
g =
_
_
_
2
ti1
(2
t
2
i
+ 2) if U(2, t, 2
ti
) = U
1
(2, t, 2
ti
),
2
ti1
(2
t
2
i
) if U(2, t, 2
ti
) = U
2
(2, t, 2
ti
).
Note that the sum of the elements of the subgroup U
1
(2, t, 2
ti
) is the same as
the rst sum of Lemma 7.
From Theorem 6, we get the following statement.
Theorem 9. Let p and t be positive numbers with p prime and not a divisor of
B. Let N = p
t
and e = o
N
(B) = kd. Then N M
d
(B) if and only if d
p
t1
. Furthermore, if N M
d
(B) and

gU(p,t,d)
g = rp
t
for some integer r then

d
j=1
A
j
= (B
k
1)r.
In a similar way as we get Equation 3, we can prove that if gcd(N, B) = 1,
o
N
(B) = kd and r
i
B
ik
mod N, then
d

j=1
A
j
=
(B
k
1)

d
i=1
r
i
N
.
Therefore
INTEGERS: 9 (2009) 196

d
j=1
A
j
(B
k
1)
=

d
i=1
r
i
N
. (4)
In others words, the last equation says that to prove that N M
d
(B) is equivalent
to asking if the sum

d
i=1
r
i
is a multiple of N. If we have the decomposition of
N in prime factors N = p
t1
1
p
t2
2
p
ts
s
, it is well known from the Chinese Remainder
Theorem that there is an isomorphism between the groups U
N
and U
p
t
1
1
U
p
t
2
2

U
p
ts
s
; for this reason,

d
i=1
r
i
is a multiple of N if and only if it is a multiple
of p
tj
j
for all j = 1, 2, . . . , s.
Proof of Theorem 2. Let p be a prime divisor of N and p
t
the highest power of p
that divides N, and let be the canonical projection of U
N

= U
p
t
1
1
U
p
t
2
2
U
p
ts
s
over U
p
t . We have o
N
(B) = e = kd; hence p | B
e
1 and therefore o
p
(B) | e.
Suppose that o
p
(B) k. We know o
p
(B) | p 1, o
p
(B) | e and e = kd, so
we have that d cannot be a power of p and by Lemma 7, we get that

d
i=1
r
i
=

d
i=1
(B
ik
mod p
t
) is divisible by p
t
.
If o
p
(B) | k then p
t
| d, and therefore r
i
B
ik
1 mod p for i = 1, 2, . . . , d.
Now, the set B =
_
B
ik
| i = 1, 2, , d
_
is a subgroup of U
N
so we have that
(B) is a subgroup of U
p
t . By Lemma 7 it has order p
ti
, so (B) = U(p, t, p
ti
).
But,
B =
_
gU(p,t,p
ti
)

1
(g),
again by Lemma 7, we have that
d

i=1
r
i
=
d
p
ti

p
ti
(p
t
p
i
+ 2)
2
is a multiple of d and therefore of p
t
.
For any prime p that divides N, we conclude that the sum

d
i=1
r
i
is a multiple
of p
t
. So we can apply the isomorphism (given by the Chinese Remainder Theorem)
between the groups U
N
and U
p
t
1
1
U
p
t
2
2
U
p
ts
s
, to prove that this sum is
divisible by N; from Equation 4 we may conclude that

d
j=1
A
j
is a multiple of
B
k
1 and nally that N M
d
(B).
Reciprocally, let N M
d
(B), and p be a prime divisor of N. If o
p
(B) k the
result is immediate.
Suppose that o
p
(B) | k. Then r
i
B
ik
1 mod p for i = 1, 2, . . . , d. Again,
the set
__
B
ik
| i = 1, 2, . . . , d
__
is a subgroup of U
p
t of the type U(p, t, p
ti
).
Therefore, by Lemma 7,
INTEGERS: 9 (2009) 197
d

i=1
r
i
=
d
p
ti

p
ti
(p
t
p
i
+ 2)
2
= d
_
p
t
p
i
+ 2
2
_
.
We know that N M
d
(B) so that p
t
|

d
i=1
r
i
and therefore, p
t
| d.
Finally, Theorem 3 is immediate from Theorem 2 and it gives an easy character-
ization of Midys Property.
References
[1] A. Gupta and B. Sury, Decimal expansion of
1
p
and subgroup sums, Integers: Electronic
Journal Of Combinatorial Number Theory 5 (2005), # A19, 5 pp. (electronic). MR 2192238
[2] J. Lewittes, Midys theorem for periodic decimals, Integers: Electronic Journal Of Combina-
torial Number Theory 7 (2007), #A02, 11 pp. (electronic). MR 2282184
[3] H. W. Martin, Generalizations of Midys theorem on repeating decimals, Integers: Electronic
Journal Of Combinatorial Number Theory 7 (2007), #A03, 7 pp. (electronic). MR 2282186