Professional Documents
Culture Documents
Wely's Teorem PDF
Wely's Teorem PDF
Wely's Teorem PDF
We will give some preliminary material before introducing Weyls Theorem and its proof. In
particular, the following result is fairly straightforward, but nevertheless it leads to some nice
examples.
Equivalently, (
1
6 0
Z
i2kx 0 if k =
e dx =
0 1 if k = 0.
Also, if n > |k|, then (
n1
X 0 6 0
if k =
ei2kj/n =
j=0
n if k = 0.
We give a few examples of the usefulness of Theorem 23. We do this by posing problems and
demonstrating a solution to each based on the above theorem. It should be kept in mind that we do
not mean to imply the the following solutions are the most elegant.
From Theorem 23, we deduce that each of the integrals in the sum is 0 unless 1 +22 + +mm =
0. Also, if an integral in the sum is not zero, then it is positive. Therefore, Im 6= 0 if and only if there
exist 1 , . . . , m {1, 1} such that 1 +22 + +mm = 0. Note that if 1 +22 + +mm = 0,
then
m(m + 1)
0 1 + 22 + + mm 1 + 2 + + m (mod 2)
2
so that
m 0 or 3 (mod 4).
34
Thus, Im = 0 for m {1, 2, 5, 6, 9, 10}. To see that the answer is the remaining m, that is
m {3, 4, 7, 8}, observe that
1+23=0 and 1 2 3 + 4 = 0
and if
1 + 22 + + mm = 0,
then
1 + 22 + + mm (m + 1) + (m + 2) + (m + 3) (m + 4) = 0.
Note that in general Im 6= 0 if and only if m 0 or 3 (mod 4).
Thus,
where g(x) = (3x + 1)(2x + 1) = 6x2 + 5x + 1. Since g(0) = 1, this completes the solution.
35
Solution: We do not use Theorem 23 directly here but rather the following which is of a similar
flavor:
Suppose and u are real numbers. Then is an integer if and only if
Z u+
i2x
e dx = 0.
u
Position the rectangles so that their sides are parallel to the x and y-axes and so that the lower
left corner of R is (0, 0). Suppose j and j are the horizontal and vertical dimensions of Rj ,
respectively, for j = 1, 2, . . . , n. Let (uj , vj ) be the lower left corner of Rj . Then since either j is
an integer or j is an integer, we get from the above that
Z Z Z
vj +j Z uj +j uj +j vj +j
i2(x+y) i2x i2y
e dx dy = e dx e dy = 0.
vj uj uj vj
On the other hand, if and are the horizontal and vertical dimensions of R, respectively, then
Z Z Z Z
i2x
i2y
i2(x+y)
e dx
e dy =
e dx
0 0 R
n ZZ n Z
X X vj +j Z uj +j
i2(x+y) i2(x+y)
= e dx = e dx dy = 0.
Rj
j=1
vj uj
j=1
R R
Hence, either 0 ei2x dx = 0 or 0 ei2y dy = 0, which implies that either or is an integer,
completing the proof.
We turn now to the main topic of this section. For real, let {} denote the fractional part
of . Note that {} [0, 1) for all . Thus, for example, {2.341} = 0.341, {22/7} = 1/7, and
{22/7} = 6/7. Let {1 , 2 , . . . } be a sequence of real numbers. We say that the sequence is
uniformly distributed modulo one if for every a and b with 0 a b 1,
|{r n : {r } [a, b]}|
lim = b a.
n n
The following result is due to Weyl.
Theorem 24. If is a real irrational number, then the sequence {r}
r=1 is uniformly distributed
modulo one.
Before proving Weyls Theorem, we discuss some preliminaries. First, we observe that de-
termining whether a sequence is uniformly distributed may not even be intuitively clear. For
example, suppose R is irrational. Does it follow that {r }
r=1 is uniformly distributed modulo
one? One can see that this is not the case by considering = 2 since
r n1
|{r n : { 2 } [0, 1/3]}| .
2
36
Is this an unfair example since = 2 has the property thatsome power of is an integer? To
partially answer this question, we show that for = (1 + 5)/2, the sequence {r } r=1 is not
uniformly distributed modulo one. Define
!r !r
1+ 5 1 5
ur = + for r Z+ {0}.
2 2
Note that u0 = 2, u1 = 1, and for r 2,
!r !r
1+ 5 1 5
ur = +
2 2
!r2 !2 !r2 !2
1+ 5 1+ 5 1 5 1 5
= +
2 2 2 2
!r2 ! !r2 !
1+ 5 3+ 5 1 5 3 5
= +
2 2 2 2
!r2 ! !r2 !
1+ 5 1+ 5 1 5 1 5
= 1+ + 1+
2 2 2 2
= ur2 + ur1 .
r
Therefore, ur Z+ for all r 0. Note that (1 5)/2 (1, 0) so that (1 5)/2 tends to
zero as r tends to infinity. Therefore,
1 + 5 2r+1
!
lim =0
r 2
and
1 + 5 2r
!
lim = 1.
r 2
37
Pr
Theorem 25. Let {aj } j=0 be a sequence of complex numbers, and set sr = j=0 aj for r
+
Z {0}. Then
Pn
sr
(i) if limr sr = S, then limr r=0 = S, and
n+1
Pn
sr
(ii) if aj = (1) j(j+1)/2
, then limr sr does not exist and limr r=0 = 0.
n+1
Proof. (i) Suppose limr P sr = S. Let > 0. Then there is an R such that if r R, then
|sr S| < /2. Let A = R r=0 |sr S|. Let N max{R, 2A/}. Then for n N ,
Pn n
r=0 sr 1 X
n + 1 S = n + 1 (sr S)
r=0
R n
1 X 1 X
= (sr S) + (sr S)
n+1 n+1
r=0 r=R+1
1 1
A + (n + 1) n + (n + 1)
n+1 2 n+1 2 2
< + = .
2 2
Thus, Pn
sr r=0
lim = S.
r n + 1
38
Hence, Pn Pr
r=0 j=0 aj
(n + 2) log(n + 2)
.
n+1 n+1
Since this last expression tends to infinity with n, we deduce that the Cesaro sum of 1 + 12 + 13 +
is infinite. P
As another
Pr example, we show that it is possible to have a series j=0 aj with all of its partial
sums sr = j=0 aj bounded and with no Cesaro sum. We take
1
if j = 22k where k Z+ {0}
aj = 1 if j = 22k1 where k Z+ .
0 otherwise
3
22k3 + 22k2 = (22k ).
8
Also,
22k+1
X1 k
X 1
sj = 22j 22k = (22k+1 ).
j=0 j=0
2
P P
n n
Thus, infinitely often j=0 a j /(n + 1) 3/8 and infinitely often j=0 a j /(n + 1) 1/2.
This easily implies that the Cesaro sum cannot exist.
To prove Weyls Theorem (Theorem 24), we will use a little material from Fourier Analysis
which we introduce here. The basic idea is to write a function f (x), which maps the real numbers
to the complex numbers, in the form
X
f (x) = f(r)eirx . (11)
r=
Here, we wish to find numbers f(r) for which the above holds. Since eirx = cos(rx) + i sin(rx),
the series on the right in (11) is often referred to as a trignometric series. It is not always possible
to obtain a trignometric series representation for f (x), but lets suppose for the moment that f (x)
can be expressed in the form (11). We temporarily ignore rigor. Using Theorem 23, observe that
for n |r|,
n Z 2 Z 2 X n
!
1 X 1
f(r) = f(j)ei(jr)x dx = f(j)eijx eirx dx.
2 j=n 0 2 0 j=n
39
Letting n tend to infinity, we deduce that
Z 2
1
f(r) = f (x)eirx dx. (12)
2 0
Thus, using (12), it is seemingly easy to write a function in the form (11). However, we still need
to discuss when the series in (11) converges. We will require certian conditions on f (x). They are:
(i) f (x) is continuous over the reals.
(ii) f (x + 2) = f (x) for all real x.
The latter condition shouldnt be surprising since the values of f(r) determine f (x) by (11) and
they only depend on the values of f (x) for x [0, 2] by (12). Before continuing, it is worth
noting that (i) and (ii) are not sufficient to imply the convergence of the series in (11). On the other
hand, we will want to consider f (x) in this generality. To deal with this difficulty, we prove a
result of Fejer that with f (x) satisfying (i) and (ii), the series in (11) has Cesaro sum f (x) (so that
convergence of the series will not be necessary).
Define f(r) by (12), and set
j
n
!
1 X X
n (f, x) = f (r)eirx
n + 1 j=0 r=j
1
= (n + 1)f(0) + nf(1)eix + nf(1)eix +
n+1
n
X n + 1 |r|
= f (r)eirx .
r=n
n+1
We will now prove
Theorem 26. If f (x) satisfies (i) and (ii), then n (f, x) converges to f uniformly on R.
We begin with a relationship between n (f, x) and
n
X n + 1 |r| irx
Kn (x) = e .
r=n
n + 1
Observe that
n
X n + 1 |r|
n (f, x) = f (r)eirx
r=n
n + 1
n Z 2
X n + 1 |r| 1 irt
= f (t)e dt eirx
r=n
n + 1 2 0
2 n
n + 1 |r| ir(xt)
Z
1 X
= f (t) e dt
2 0 r=n
n+1
Z 2
1
= f (t)Kn (x t) dt.
2 0
40
Note that in addition to (ii), we have that Kn (x + 2) = Kn (x) for all real x. Letting y = x t
gives
1 x2
Z
n (f, x) = f (x y)Kn (y) dy
2 x
Z x Z 2
1 1
= f (x y)Kn (y) dy = f (x y)Kn (y) dy.
2 x2 2 0
We now proceed with two lemmas.
Lemma 1. Let x [0, 2). Then Kn (0) = n + 1, and if x 6= 0, then
2
1 sin((n + 1)x/2)
Kn (x) = .
n+1 sin(x/2)
Proof. Clearly,
n n
!
X n + 1 |r| 1 X
Kn (0) = = n+1+2 r = n + 1.
r=n
n+1 n+1 r=1
Note that
n 2n
X
i(r+n)x
X ei(2n+1)x 1
e = eirx = .
r=n r=0
eix 1
Also,
n
X ei(n+1)x 1
eirx = 1
r=1
eix 1
so that by taking derivatives, we obtain
n
X (eix 1)i(n + 1)ei(n+1)x (ei(n+1)x 1)ieix
i reirx = .
r=1
(eix 1)2
Thus,
n
X nei(n+2)x (n + 1)ei(n+1)x + eix
reirx =
r=1
(eix 1)2
and n
X nei(n+2)x (n + 1)ei(n+1)x + eix
reirx = .
r=1
(eix 1)2
41
Hence,
1
Kn (x) = (n + 1)einx (eix 1)(ei(2n+1)x 1)
(n + 1)(eix 1)2
nei(n+2)x (n + 1)ei(n+1)x + eix neinx (n + 1)ei(n1)x + eix
1 i(n+2)x inx ix
= e + e 2e
(n + 1)(eix 1)2
eix i(n+1)x i(n+1)x
= e 2 + e
(n + 1)(eix 1)2
eix i(n+1)x/2 i(n+1)x/2 2
= e e
(n + 1)(eix 1)2
1 2
= (eix/2 eix/2 )2 ei(n+1)x/2 ei(n+1)x/2
n+1
2 i(n+1)x/2 2
eix/2 ei(n+1)x/2
ix/2
1 e e
=
n+1 2i 2i
2
1 sin((n + 1)x/2)
= .
n+1 sin(x/2)
This completes the proof.
Lemma 2. Kn (x) has the following properties:
(a) Kn (x) 0 for all real numbers x.
(b) For every > 0 and > 0, there is an N = N (, ) such that if n N and x (, 2 ),
then |Kn (x)| < .
1 R 2
(c) Kn (x) dx = 1.
2 0
Comment: (b) is simply asserting that for all > 0, Kn (x) approaches 0 uniformly on (, 2 ).
Proof. Since Kn (x+2) = Kn (x), (a) follows from Lemma 1. For (b), note that for x (, 2)
and for n sufficiently large (independent of x), Lemma 1 implies
1 1
Kn (x) 2 < .
(n + 1) sin (x/2) (n + 1) sin2 (/2)
Finally, we deduce (c) from
Z 2 n
Z 2 X
1 1 n + 1 |r| irx
Kn (x) dx = e dx
2 0 2 0 r=n n + 1
n Z 2 Z 2
X n + 1 |r| 1 irx 1
= e dx = dx = 1,
r=n
n + 1 2 0 2 0
completing the proof.
42
We now give the basic idea behind the proof of Theorem 26. Note that Lemma 2 implies that
for > 0 fixed and small and for n large,
Z 2 Z 2
1 1
Kn (y) dy 0 and Kn (y) dy = 1
2 2 0
so that Z
1
Kn (y) dy 1.
2
Also, Z 2
1
f (x y)Kn (y) dy 0
2
so that Z 2 Z
1 1
n (f, x) = f (x y)Kn (y) dy f (x y)Kn (y) dy.
2 0 2
43
Thus, we get from Lemma 2 (c) that for all n N ,
Z 2 Z 2
1 1
|n (f , x) f (x)| = f (x y)Kn (y) dy Kn (y) dy f (x)
2 0 2 0
Z 2
1
= (f (x y) f (x)) Kn (y) dy
2 0
1
Z Z 2
1
(f (x y) f (x)) Kn (y) dy + (f (x y) f (x)) Kn (y) dy
2 2
Z Z 2
1 1
|f (x y) f (x)| |Kn (y)| dy + (|f (x y)| + |f (x)|) |Kn (y)| dy
2 2
Z Z 2
1 1
< |Kn (y)| dy + (2M ) dy
2 2 2 4M
Z 2
1
< + dy = .
2 2 2 0
Corollary 1. Let f (x) satisfy (i) and (ii), and let > 0. Then there exists a trignometric polynomial
n
X
P (x) = aj eijx
j=n
Proof. Since n (f, x) is a trignometric polynomial, the Corollary follows from Theorem 26 by
taking P (x) = n (f, x) with n sufficiently large.
We now prove Weyls Theorem (Theorem 24). It suffices to show that if R is irrational
and 0 a b 1, then
|{r n : 2{r} [2a, 2b]}|
lim = b a. (13)
n n
Lemma 3. Let R with irrational. Suppose f : R C satisfies (i) and (ii) of the previous
section. Then n Z 2
1X 1
lim f (2r) = f (x) dx.
n n 2 0
r=1
Proof. Let
n Z 2
1X 1
Gn (f ) = f (2r) f (x) dx.
n r=1 2 0
44
We want to show that limn Gn (f ) = 0. First, we consider f (x) = eisx where s is an integer. If
s = 0, then f (x) 1 and
n Z 2
1X 1
Gn (f ) = Gn (1) = 1 dx = 0.
n r=1 2 0
Thus, limn Gn (f ) = 0 in the case f (x) = eisx with s = 0. Now, suppose s 6= 0. Then
1 X n Z 2
2irs 1 isx
|Gn (f )| = e e dx
n 2 0
r=1
n
1 X
2irs
= e
n r=1
1 2is e2ins 1
= e e2is 1
n
2
.
n|e2is 1|
m n Z 2 !
X 1 X 2irs 1
= as lim e eisx dx
s=m
n n r=1
2 0
m
X
as lim Gn eisx = 0.
=
n
s=m
45
such that if n N , then |Gn (P )| < /3. Hence, for all n N ,
and
n Z 2
1X 1 1
f (2r) f (x) dx = (2(b a) 2) = (b a) 2.
n r=1 2 0 2
46
Hence, for all n N ,
|{r n : 2{r} [2a, 2b]}|
(b a) < 2,
n
The following are two problems related to this subject. The details of their solutions are omitted
here.
(1) Let d {0, 1, 2, . . . , 9}. What is the proportion of times that 2n begins with the digit d as n
runs through the positive integers? More specifically, compute
Homework:
2. Let {1 , 2 , . . . } be a sequence of real numbers. Prove that the sequence is uniformly dis-
tributed modulo one if and only if for every a and b with 0 a b 1,
3. Let
P
j=0 aj be a divergent series with aj > 0 for each j 0. Prove that the Cesaro sum of
the series is infinite.
(a) 1 1 + 1 1 + 1 1 + .
1 1 1
(b) 1 + + .
2 4 8
5. Prove the following:
47
Theorem 27. Let {k } be a sequence of real numbers satisfying
n
1 X 2imk
lim e =0
n n
k=1
(Hint: Look at the proof of Theorem 24 with k replacing k and decide what changes need
to be made. You do not need to rewrite the proofs if you point out clearly where the changes
need to be made and what the changes are.)
6. Prove that the sequence { n} n=1 is uniformly distributed modulo 1.
48