A remark on random and equidistributed

Nicolas BOULEAU

Lab oratoire de Mathematiques Appliquees UA CNRS

ENPC La Courtine NoisyleGrand Cedex

If is an equidistributed on the sdimensional cub e the average of

n

the values of a f on the sequence converges to the integral of f for f Riemann

integrable We study here the quite natural fact that a slight random p erturbation of the

sequence allows to integrate more irregular functions The notions used are dened in

and

Strongly distributed sequences

s

Let b e a probability measure on s IN equipp ed with its natural top ology

Except what envolves the convolution pro duct the sequel could b e extended to any lcd

space

s

Let X b e a sequence of random variables with values in dened on a probabi

n

lity space A IP Almost surely as means IPalmost surely If with probability

s

the sequence X is distributed on ie if almost surely

n

Z

N

X

s

f C IR f X f d

n

N

n

then X will b e said to b e almost surely distributed

n

Now the example of iid random variables X with common law shows that

n

some random sequences have the following stronger prop erty

Denition A random sequence X is said to be strongly distributed if for every

n

s

bounded measurable function f from into IR

Z

N

X

f X f d as

n

N

n

That a strongly distributed sequence X b e almost surely distributed comes

n

s

from the fact that if D is a countable dense subset of C IR such a sequence satises

Z

N

X

f X f d as f D

n

N

n

which implies easily

N

X

narrowly as

X

n

N

n

s

Prop osition Let X be a sequence of random variables with values in If for

n

s

every open A in it holds

N

X

X A as

A n

N

n

then X is strongly distributed

n

We shall use the following lemma

Lemma Property for open sets implies property for continuous f s

Pro of Let f b e continuous We may supp ose f Then

k k

G nG f

k k

n n

for an increasing sequence of op en sets G Approximating uniformly up to the function

k

f by

K

X

with A G nG f

k A k k k K

k

k

the fact that

Z

N

X

f X f d as

K n K

N

n

gives

Z Z

N N

X X

f d lim f X f X f d lim

n n

N N

n n

which shows the lemma

For the prop osition consider a b ounded measurable function f By the Lusin pro

p erty for every sequence decreasing to zero there exists a sequence of op en sets G

k k

and a sequence of continuous functions f such that

k

k G f f outside G k f k k f k

k k k k k k

Let us write

N N N

X X X

f X f X f f X

n k n k n

N N N

n n n

By the lemma the following inequalities hold as

N

X

R

f X f d k f k G lim

n k k

N

n

N

X

R

f X f d k f k G lim

n k k

N

n

Z

N

X

hense as lim f X f d

n

N

n

Remark The prop erty that b e satised for continuous functions f is weaker than

the prop erty for op en sets It is indeed equivalent to the almost sure distribution

of X And there exist almost surely distributed sequences which are not strongly

n

distributed consider a deterministic equidistributed sequence take X

n n n

and take f x

fxn xg

n

s

Prop osition Let be an equidistributed sequence on and V be a sequence

n n

s

of iid random variables with absolutely continuous common law on IR Then the

sequence X f V g is strongly distributed on the cube Here fxg means the

n n n s

fractional part of x component by component and is the in dimension

s

s

Pro of Let us dene absolutely continuous probability measures on the cub e by

n

Z

A f xgd x

n A n

s

IR

s

for every measurable subset A of

s

Lemma For every f Lebesgue measurable and bounded on the cube

N

X

f f lim

n s

N

N

n

Pro of The function

Z

g y f fy xgd x

s

IR

s s

is continuous and b ounded on IR as convolution pro duct of a function in L IR by a

s

function in L IR Hence

Z Z

N

X

g g y dy f y dy

n

s s

N

n

Now to achieve the pro of of the prop osition we use a classical argument

N

X

Let S f X f

N n n

N

n

2

a It holds S as

N

Indeed

2

N

X X

2

f f j IPN m jS

n n

N

N

n

N m

1

X

k f k

N

m

1

2

thus j as lim jS

N

N

b It holds S as

N

Indeed if M N M

jN M j M

2

j jS S k f k N k f k

N

M

M NM M

So by the lemma

Z

N

X

f X f d

n s

N

n

Extension

The preceding prop osition fails if is no more absolutely continuous as seen when

is a Dirac mass But if do es not charge p olar sets it is p ossible to integrate with X

n

quasicontinuous functions and even a little more For conveniency we work now with the

s s

torus T IRZ instead of the cub e

s

Prop osition Let be an equidistributed sequence on T and V be a sequence

n n

s

of iid random variables with values in T with common law which does not charge

sets of Newton capacity zero Let X V Then

n n n

Z

N

X

f X f d as

n s

s

N

T

n

s

for every f from T into IR with the fol lowing property

u v quasicontinuous and bounded such that

R

v ud u f v and

s

s

T

Pro of

a Let us supp ose rst that the probability b e a measure of nite energy integral

s

for the classical Dirichlet structure on T and that f b e a quasicontinuous version of an

s

element of H T Then approximating f in H by functions f in H and continuous

n

and denoting the translation by x give

x

1 1

C k f f k j f f j C k f f k

n x x n x x n

H H

s

and that implies that the convolution pro duct f is a continuous function on T

b Let us supp ose now f b e quasicontinuous and b ounded and the probability mea

sure do not charge p olar sets Then there exist probability measures of nite energy

P

integrals such that

p p p p p

p

1

b e the energynorms of the s Let C k U k

p p p

H

Let For each p let us choose a continuous function f and an op en set G such

p p

that

k f k k f k f f outside G C apG

p p p p

p

C

p

We have

X

f f f f

p p p p p

p

but

jf f j k f k e G k f k C apG C k f k

p p p p p p

p

where e G is the equilibrium p otential of G

p p

It follows that f is uniform limit of continuous functions and therefore is continuous

c Under the hypotheses of part b putting

n

n

it holds therefore by the equidistribution of

n

Z

N

X

f d f

s n

N

T

s

n

then the same argument that in the pro of of prop osition shows that

Z

N

X

f d as f X

s n

N

T

s

n

d At last if f satises prop erty we have

Z

N N

X X

ud lim uX lim f X

s n n

s

N N

T

n n

Z

N N

X X

f X v X v d lim lim

n n s

N N

T

s

n n

almost surely Hence

Z

N

X

f X f d as lim

n s

N

n

The functions satisfying can b e shown to b e the functions which are b ounded and

s

nely continuous at almost every p oint of T

s

REFERENCES

L Kuip ers and HN Niederreiter Uniform distribution of sequences Wiley

M Fukushima Dirichlet forms and Markov pro cesses NorthHolland