Measure theory conjecture


By Brad Rodgers on Sunday, September 29, 2002 - 03:35 am:
Suppose we have a set S with measure S ' , and S ' > 0. If


1
S '
òS xm dx=1/(m+1)

,

for all positive m, does that imply that S ' =1? For all natural numbers m?
Brad


By Michael Doré on Sunday, September 29, 2002 - 02:28 pm:

Note that if c > 1 then the set SÇ[c,¥) is null. For if there is a subset of S of measure r > 0 each element of which is ³ c then for even m, your integral would be ³ r cm which tends to ¥ as m®¥, a contradiction. This means that SÇ[1,¥) is null, by the countable union property. Similarly SÇ(-¥,-1] is null. So S is completely contained in [-1,1] apart from a null set.

Now let e > 0 and let f be a continuous function which is 1 on [-1,-e] and 0 on [0,1] (and linear in between). For any d > 0, by Weierstrass' theorem we can find a polynomial p such that |p(x)-f(x)| < d/2 for all x in [-1,1]. Therefore |q(x)-f(x)| < d for all x in [-1,1] where q(x)=p(x)-p(0). So q is a polynomial with no constant coefficient.

Now we have:

1/S ' òS xm dx=ò01 xm dx

for all positive integers m. Therefore:

|1/S ' òS q(x) dx|=|ò01 q(x) dx| < d

It quickly follows that [-1,-e]ÇS is null so [-1,0]ÇS is null. So S is completely contained in [0,1] apart from a null set.

Using similar approximation techniques you can show that for any [a,b] contained in [0,1] then measure SÇ[a,b]=S ' (b-a). My guess is that it should follow quickly that S is the whole of [0,1] apart from a null set but I haven't worked out the details.


By Michael Doré on Sunday, September 29, 2002 - 05:00 pm:

Actually we can simplify this somewhat. As above S[-1,1] is null. Now we're given:

1/S ' òS xm dx=ò01 xm dx

for all positive integers m. However this relationship is also trivially true for m=0 (a fact I managed to miss earlier!) Therefore for any polynomial P we have:

1/S ' òS P(x) dx=ò01 P(x)dx

So by Weierstrass this is true for any continuous function P (uniformly approximate the continuous function by polynomials on [-1,1] and it doesn't matter what happens outside [-1,1]).

Certainly S is null on [-1,0] (let P(x) = 0 for positive x and P(x) = -x/e on [-e,0] and P(x) = 1 for x < -e).

So S is contained in [0,1] except for a null set and:

1/S ' òS P(x) dx = ò01 P(x) dx

for all continuous P. Then for any subinterval [a,b] of [0,1] set P(x) = 1 on [a,b] and make it fall off to 0 very rapidly outside [a,b]. Taking the limit we have 1/S ' òSÇ[a b] 1 dx = òab 1 dx so the measure of S Ç[a,b] is S ' (b-a) as I stated earlier.

Is this enough to conclude that S is the same as [0,1] apart from a null difference? I think something like this came up earlier in a thread in which it was asked whether a set existed such that its Lesbegue measure on any interval [a,b] was (b-a)/2. As I remember Dan Goodman said there was a theorem which said no, the Lesbegue density of any measurable set is 0 or 1 almost everywhere. In this case this implies that S ' = 1 and the measure of S Ç[a,b] is S ' (b-a) = b-a for every subinterval [a,b] of [0,1]. In particular the measure of S on [0,1] is 1 and so S is almost the same as [0,1] (i.e. almost all elements of S are in [0,1] and almost all elements on [0,1] are in S).


By Michael Doré on Sunday, September 29, 2002 - 11:02 pm:

Of course we can do this last bit without using the density theorem. S has measure S ' so it has outer measure S ' . Therefore for any e > 0 we can cover S with countably many disjoint open intervals I1, I2, ... such that
|S ' - å
|Ii || < e

. Therefore:


|S ' - 1/S ' å
measure S ÇIi | < e

So by the countable union property:

|S ' - 1/S ' measure (S ÇI1 ) È(S ÇI2) È...| < e But the set (S ÇI1) È(S ÇI2 ) È... is just S (all its elements are in S and all elements of S are in one of the Iis). So:

|S ' - 1| < e This holds for every e > 0 so S ' = 1 and S is [0,1] up to a null difference.


By Ian Short on Monday, September 30, 2002 - 12:55 pm:

That seems like a really good solution!