Loading [MathJax]/jax/output/HTML-CSS/fonts/TeX/fontdata.js

To measure or not to measure... a real problem!


This post will be more theoretical than usual. If you are afraid of mathematics, go away now before it is too late!

I will write about a set, the Vitali set, which arises in measure theory, to show that there is no coherent way of assigning a size to every subset of the real line. Because of that I will try to define size for subsets of the real line and then verify that such task is impossible.
I will also enforce a couple of (seemingly reasonable) restrictions on my definition.

I challenge you to read this post and to calmly follow my reasoning. Whenever something doesn't seem obvious, try to make it clear by yourself with a piece of paper and pen/pencil. If any doubts persist, drop me your question(s) in the comments and I will answer gladly!

Let us call m to the function that, given a subset of the real line, returns its size; that is, let us try to define m:P(R)→[0,∞].

Let us also suppose that the function m satisfies the following restrictions:
  • If I is an interval with endpoints a≤b, m(I)=b−a. That is, the size of an interval is what we intuitively call its length. For example, m([3,4])=4−3=1;
  • m(E+x)=m(E): the size of a set remains unchanged under translations, i.e. even if a set is "pushed over", its size remains the same;
  • If we have a finite number of sets, or a numerable quantity of them, and if they are all disjoint, then the size of their union must equal the sum of their individual sizes: m(∪∞n=1En)=∑∞n=1m(En) whenever all the En are disjoint.

Under these restrictions, can we create the function m?

Giuseppe Vitali, an italian mathematician, found out that the answer was "no". Vitali discovered a set V⊂R that allows one to conclude that we cannot create a function m satisfying the restrictions above. I will now present the construction of the Vitali set:

Let us take a real number x and define the set Cx={y∈R:y−x∈Q}, i.e. the set Cx contains all numbers that differ of x by a rational amount: Cx={x+q,∀q∈Q} (in fact, the set Cx represents the equivalence class of x under the equivalence relation x∼y⟺x−y∈Q).

Having defined the sets Cx for all real x, we pick out of every unique Cx, a value vx such that vx∈]0,1[.
I said "out of every unique Cx" because, for example, C0=C1; for this construction, we do not want repeated sets Cx. We will also say that the number vx is a representative of the set Cx and all its elements. As an example, 0.3 could be a representative of the set C0 and π−3.1 could be a representative of the set Cπ.
(By the axiom of choice) we have a set V - which we will call the Vitali set - where we have, from each unique set Cx, a single representative vx∈]0,1[.
We now ask what should be the value m(V). Either m(V)=0 or m(V)>0. We will build an auxiliary set which will allow one to conclude that both cases lead to incoherent results.

To build the auxiliary set, let us consider {r1,r2,⋯,rn,⋯} an enumeration of all the rationals in ]−1,1[ and let us define Vj=V+rj: Vj is the translation of the set V by rj. Then we define G=∪∞j=1Vj We notice that the sets Vj are all disjoint! Suppose y is an element lying simultaneously in Vj and in Vk (with j≠k). This means that
  1. xj=y−rj is in V because y∈Vj=V+rj;
  2. xk=y−rk is in V because y∈Vk=V+rk;
  3. From 1. and 2. we have that xj−xk=(y−rj)−(y−rk)=rk−rj is a rational number, i.e. xj−xk∈Q;
From 3. we get that xk∈Cxj but, by definition, V only has a representative from each Cx. Thus we obtain a contradiction and there can be no real number x lying in two different Vj: all the Vj are disjoint.

Because of the restrictions we imposed on m, we have that m(Vj)=m(V+rj)=m(V) because m is constant under translations; we also have that m(G)=m(∪∞j=1Vj)=∞∑j=1m(Vj)=∞∑j=1m(V) because the Vj are all disjoint and we are performing their union.

Now we can see that if it is the case that m(V)=0, then m(G)=0. But m(G)=0 is absurd because the interval ]0,1[ is inside G and m(]0,1[)=1 implying that m(G)≥1. To verify that ]0,1[⊂G, let x∈]0,1[. This element x belonged to some initial C∙ and therefore has a representative y in ]0,1[. If y is a representative of x, then x and y differ by a rational number. But x,y∈]0,1[ means that their distance is smaller than 1 and, in particular, x−y∈]−1,1[. Adding everything together, x−y∈Q and x−y∈]−1,1[ gives that x−y=rk for some k, as {r1,⋯,rn,⋯} was an enumeration of all the rationals in ]−1,1[.
If it was the case that m(V)>0, then m(G)=∞. But m(G)=∞ is another absurd given that G⊆]−1,2[ implying that m(G)≤m(]−1,2[)=3. To see that G⊆]−1,2[ we just have to remember that G is a numerable union of the sets Vj and every one of those is Vj=V+rj. We had that V⊂]0,1[ and rj∈]−1,1[ hence Vj⊆]−1,2[.
(the image above has some points from V in red, one of the rationals rj and the respective set Vj in blue; because every rj lies in ]−1,1[ it is obvious that Vj⊂]−1,2[)

Looking back at what we have, neither m(V)=0 nor m(V)>0 produce coherent results! We conclude that our function m cannot exist as I envisioned it in the beginning of the post... even though the restrictions I imposed seemed perfectly reasonable!

I understand if this post isn't the easiest to digest. If you have any doubts regarding the construction I described or anything related, drop a line on the comments section and I will be glad to answer!

  - RGS

To measure or not to measure... a real problem! To measure or not to measure... a real problem! Reviewed by Unknown on July 07, 2018 Rating: 5

No comments:

Powered by Blogger.