The Borel-Kolmogorov Paradox and conditional expectations
Transcript of The Borel-Kolmogorov Paradox and conditional expectations
![Page 1: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/1.jpg)
The Borel-Kolmogorov Paradox
and
conditional expectations
Zalán GyenisRényi Institute of Mathematics, Budapest
Gábor Hofer-SzabóResearch Centre for the Humanities, Budapest
Miklós RédeiLondon School of Economics
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 1 / 25
![Page 2: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/2.jpg)
The Borel-Kolmogorov Paradox
What is the conditional probability that a randomly chosen point is on an
arc of a great circle of the sphere on the condition that it lies on that great
circle?
Tension:
1 Intuition: the conditional
probability is proportional to
the length of the arc.
2 Fact: since a great circle has
surface measure zero, Bayes'
formula cannot be used to
calculate the conditional
probability.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 2 / 25
![Page 3: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/3.jpg)
Main messages
1 Use conditional expectation to conditionalize.
2 This will allow for conditionalizing on probability zero events, as in
case of the Borel-Kolmogorov Paradox.
3 Using conditional expectation the Borel-Kolmogorov Paradox is not
paradoxical.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 3 / 25
![Page 4: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/4.jpg)
Historical remarks
The Borel-Kolmogorov Paradox was formulated by Borel in 1909
before Kolmogorov's (1933) measure-theoretic probability theory.
Kolmogorov's own resolution of the Paradox is based on the theory of
conditional expectation.
Since Kolmogorov's work, conditional expectation is the standard
device for conditionalization in probability theory.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 4 / 25
![Page 5: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/5.jpg)
Outline
1 Conditioning using conditional expectation
2 The Borel-Kolmogorov Paradox
3 Remarks
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 5 / 25
![Page 6: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/6.jpg)
Conditioning using conditional expectation
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 6 / 25
![Page 7: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/7.jpg)
Conditional expectation
Conditional expectation is a coarse-graining.
(X ,S, p): probability measure space
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 7 / 25
![Page 8: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/8.jpg)
Conditional expectation
Conditional expectation is a coarse-graining.
(X ,S, p): probability measure space
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 7 / 25
![Page 9: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/9.jpg)
Conditional expectation
L1(X ,S, p): set of p-integrable functions
p de�nes a functional φp(f ).=
∫Xf dp, f ∈ L1(X ,S, p)
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 8 / 25
![Page 10: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/10.jpg)
Conditional expectation
L1(X ,S, p): set of p-integrable functions
p de�nes a functional φp(f ).=
∫Xf dp, f ∈ L1(X ,S, p)
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 8 / 25
![Page 11: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/11.jpg)
Conditional expectation
(X ,A, pA): coarse-grained probability measure space
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 9 / 25
![Page 12: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/12.jpg)
Conditional expectation
E (·|A) : L1(X ,S, p)→ L1(X ,A, pA): conditional expectation
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 10 / 25
![Page 13: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/13.jpg)
Conditional expectation
De�nition
A map
E (·|A) : L1(X ,S, p)→ L1(X ,A, pA)
is called an A-conditional expectation if:
(i) E (f |A) is A-measurable for all f ∈ L1(X ,S, p);(ii) E (·|A) preserves the integration on elements of A:∫
Z
E (f |A)dpA =
∫Z
fdp ∀Z ∈ A.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 11 / 25
![Page 14: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/14.jpg)
Bayesian statistical inference
Problem of statistical inference (informal)
Given a probability measure p′A on A, what is its extension to S?
Reformulated in terms of functionals:
Problem of statistical inference
Given a continuous linear functional φ′A on L1(X ,A, pA), what is itsextension to L1(X ,S, p)?
There is no unique answer, but:
Bayesian statistical inference
Let the extension φ′ be:
φ′(f ).= φ′A(E (f |A)) ∀f ∈ L1(X ,S, p)
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 12 / 25
![Page 15: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/15.jpg)
Bayesian statistical inference
Problem of statistical inference (informal)
Given a probability measure p′A on A, what is its extension to S?
Reformulated in terms of functionals:
Problem of statistical inference
Given a continuous linear functional φ′A on L1(X ,A, pA), what is itsextension to L1(X ,S, p)?
There is no unique answer, but:
Bayesian statistical inference
Let the extension φ′ be:
φ′(f ).= φ′A(E (f |A)) ∀f ∈ L1(X ,S, p)
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 12 / 25
![Page 16: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/16.jpg)
Bayesian statistical inference
Problem of statistical inference (informal)
Given a probability measure p′A on A, what is its extension to S?
Reformulated in terms of functionals:
Problem of statistical inference
Given a continuous linear functional φ′A on L1(X ,A, pA), what is itsextension to L1(X ,S, p)?
There is no unique answer, but:
Bayesian statistical inference
Let the extension φ′ be:
φ′(f ).= φ′A(E (f |A)) ∀f ∈ L1(X ,S, p)
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 12 / 25
![Page 17: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/17.jpg)
Bayesian statistical inference
Problem of statistical inference (informal)
Given a probability measure p′A on A, what is its extension to S?
Reformulated in terms of functionals:
Problem of statistical inference
Given a continuous linear functional φ′A on L1(X ,A, pA), what is itsextension to L1(X ,S, p)?
There is no unique answer, but:
Bayesian statistical inference
Let the extension φ′ be:
φ′(f ).= φ′A(E (f |A)) ∀f ∈ L1(X ,S, p)
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 12 / 25
![Page 18: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/18.jpg)
Bayesian statistical inference
Problem of statistical inference (informal)
Given a probability measure p′A on A, what is its extension to S?
Reformulated in terms of functionals:
Problem of statistical inference
Given a continuous linear functional φ′A on L1(X ,A, pA), what is itsextension to L1(X ,S, p)?
There is no unique answer, but:
Bayesian statistical inference
Let the extension φ′ be:
φ′(f ).= φ′A(E (f |A)) ∀f ∈ L1(X ,S, p)
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 12 / 25
![Page 19: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/19.jpg)
Bayesian statistical inference
Problem of statistical inference (informal)
Given a probability measure p′A on A, what is its extension to S?
Reformulated in terms of functionals:
Problem of statistical inference
Given a continuous linear functional φ′A on L1(X ,A, pA), what is itsextension to L1(X ,S, p)?
There is no unique answer, but:
Bayesian statistical inference
Let the extension φ′ be:
φ′(f ).= φ′A(E (f |A)) ∀f ∈ L1(X ,S, p)
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 12 / 25
![Page 20: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/20.jpg)
Conditional probability
De�nition
The conditional probability p′(B) is simply the application of the
Bayesian statistical inference to the characteristic function χB :
p′(B).= φ′A(E (χB |A))
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 13 / 25
![Page 21: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/21.jpg)
Conditional probability
Remarks1 p′(B) depends on three factors: A, E(·|A) and φ′A
2 In the special case whenI A is generated by {A,A⊥},I p(A) 6= 0;I p′A(A) = 1,
the conditional probability can be given by the Bayes formula:
p′(B) =p(B ∩ A)
p(A)
3 Elements of A can also have zero prior probability p. Hence, it is
possible to obtain conditional probabilities with respect to probability
zero conditioning events if one uses conditional expectations.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 14 / 25
![Page 22: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/22.jpg)
Conditional probability
Remarks1 p′(B) depends on three factors: A, E(·|A) and φ′A2 In the special case when
I A is generated by {A,A⊥},I p(A) 6= 0;I p′A(A) = 1,
the conditional probability can be given by the Bayes formula:
p′(B) =p(B ∩ A)
p(A)
3 Elements of A can also have zero prior probability p. Hence, it is
possible to obtain conditional probabilities with respect to probability
zero conditioning events if one uses conditional expectations.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 14 / 25
![Page 23: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/23.jpg)
Conditional probability
Remarks1 p′(B) depends on three factors: A, E(·|A) and φ′A2 In the special case when
I A is generated by {A,A⊥},I p(A) 6= 0;I p′A(A) = 1,
the conditional probability can be given by the Bayes formula:
p′(B) =p(B ∩ A)
p(A)
3 Elements of A can also have zero prior probability p. Hence, it is
possible to obtain conditional probabilities with respect to probability
zero conditioning events if one uses conditional expectations.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 14 / 25
![Page 24: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/24.jpg)
The Borel-Kolmogorov Paradox
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 15 / 25
![Page 25: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/25.jpg)
The Borel-Kolmogorov Paradox
Let (S ,B(S), p) be the probability measure space on the sphere S with the
uniform probability p on S . What is the conditional probability of being on
an arc on a great circle C on condition of being on a great circle C of S?
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 16 / 25
![Page 26: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/26.jpg)
The Borel-Kolmogorov Paradox
O: generated by Borel measurable
sets of circles parallel to C
c
C
x
y
z
θc
Uniform conditional probability
M: generated by Borel measurable
sets of meridian circles containing C
North Pole
South Pole
φ
x
y
z
C
c
Non-uniform conditional probability
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 17 / 25
![Page 27: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/27.jpg)
The Borel-Kolmogorov Paradox
O: generated by Borel measurable
sets of circles parallel to C
c
C
x
y
z
θc
Uniform conditional probability
M: generated by Borel measurable
sets of meridian circles containing C
North Pole
South Pole
φ
x
y
z
C
c
Non-uniform conditional probability
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 17 / 25
![Page 28: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/28.jpg)
The Borel-Kolmogorov Paradox
O: generated by Borel measurable
sets of circles parallel to C
c
C
x
y
z
θc
Uniform conditional probability
M: generated by Borel measurable
sets of meridian circles containing C
North Pole
South Pole
φ
x
y
z
C
c
Non-uniform conditional probability
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 17 / 25
![Page 29: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/29.jpg)
The Borel-Kolmogorov Paradox
O: generated by Borel measurable
sets of circles parallel to C
c
C
x
y
z
θc
Uniform conditional probability
M: generated by Borel measurable
sets of meridian circles containing C
North Pole
South Pole
φ
x
y
z
C
c
Non-uniform conditional probability
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 17 / 25
![Page 30: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/30.jpg)
Remarks
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 18 / 25
![Page 31: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/31.jpg)
Remark 1
Standard view
�. . . we have di�erent conditional distributions depending on how
we describe the circle.� (Myrvold, 2014)
The Borel-Kolmogorov Paradox is paradoxical because the conditional
probabilities of the same event on the same conditioning events should
not depend on the di�erent parametrizations (violation of the Labelling
Irrelevance).
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 19 / 25
![Page 32: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/32.jpg)
Remark 1
Our view
Two Boolean algebras of random events are the same if there is an
isomorphism between them.
Proposition. There exists no isomorphism between O andM.
The Borel-Kolmogorov Paradox merely displays a sensitive dependence of
conditional probabilities of the same event on di�erent conditioning
Boolean subalgebras with respect to which conditional probabilities are
de�ned in terms of conditional expectations. These conditional probabilities
are answers to di�erent questions � not di�erent answers to the same
question.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 20 / 25
![Page 33: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/33.jpg)
Remark 1
Our view
Two Boolean algebras of random events are the same if there is an
isomorphism between them.
Proposition. There exists no isomorphism between O andM.
The Borel-Kolmogorov Paradox merely displays a sensitive dependence of
conditional probabilities of the same event on di�erent conditioning
Boolean subalgebras with respect to which conditional probabilities are
de�ned in terms of conditional expectations. These conditional probabilities
are answers to di�erent questions � not di�erent answers to the same
question.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 20 / 25
![Page 34: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/34.jpg)
Remark 1
Our view
Two Boolean algebras of random events are the same if there is an
isomorphism between them.
Proposition. There exists no isomorphism between O andM.
The Borel-Kolmogorov Paradox merely displays a sensitive dependence of
conditional probabilities of the same event on di�erent conditioning
Boolean subalgebras with respect to which conditional probabilities are
de�ned in terms of conditional expectations. These conditional probabilities
are answers to di�erent questions � not di�erent answers to the same
question.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 20 / 25
![Page 35: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/35.jpg)
Remark 2
Another standard view
Only the uniform conditional distribution is intuitively correct.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 21 / 25
![Page 36: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/36.jpg)
Remark 2
Our view
Both conditional distributions are intuitively correct.
One obtains a uniform distribution on the sphere by generating points:
uniformly on all circles in O.
c
C
x
y
z
θc
non-uniformly on all meridian circles
inM.North Pole
South Pole
φ
x
y
z
C
c
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 22 / 25
![Page 37: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/37.jpg)
Remark 2
Our view
Both conditional distributions are intuitively correct.
One obtains a uniform distribution on the sphere by generating points:
uniformly on all circles in O.
c
C
x
y
z
θc
non-uniformly on all meridian circles
inM.North Pole
South Pole
φ
x
y
z
C
c
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 22 / 25
![Page 38: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/38.jpg)
Remark 2
Our view
Both conditional distributions are intuitively correct.
One obtains a uniform distribution on the sphere by generating points:
uniformly on all circles in O.
c
C
x
y
z
θc
non-uniformly on all meridian circles
inM.North Pole
South Pole
φ
x
y
z
C
c
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 22 / 25
![Page 39: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/39.jpg)
Remark 2
Our view
Both conditional distributions are intuitively correct.
One obtains a uniform distribution on the sphere by generating points:
uniformly on all circles in O.
c
C
x
y
z
θc
non-uniformly on all meridian circles
inM.North Pole
South Pole
φ
x
y
z
C
c
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 22 / 25
![Page 40: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/40.jpg)
Remark 2
Our view
The intuition � that only the uniform conditional distribution is correct
� might come from the intutition that the uniform length measure is
singled out by pure probabilistic means.
To single out the uniform length measure one needs to use furthernon-probabilistic mathematical conditions, e.g.
I group-theoretic: it is the unique measure invariant with respect to the
subgroup of rotations on the sphere;I geometric: it is the restriction of the Lebesgue measure to the circle
as di�erentiable manifold.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 23 / 25
![Page 41: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/41.jpg)
Remark 2
Our view
The intuition � that only the uniform conditional distribution is correct
� might come from the intutition that the uniform length measure is
singled out by pure probabilistic means.
To single out the uniform length measure one needs to use furthernon-probabilistic mathematical conditions, e.g.
I group-theoretic: it is the unique measure invariant with respect to the
subgroup of rotations on the sphere;I geometric: it is the restriction of the Lebesgue measure to the circle
as di�erentiable manifold.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 23 / 25
![Page 42: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/42.jpg)
Remark 2
Our view
The intuition � that only the uniform conditional distribution is correct
� might come from the intutition that the uniform length measure is
singled out by pure probabilistic means.
To single out the uniform length measure one needs to use furthernon-probabilistic mathematical conditions, e.g.
I group-theoretic: it is the unique measure invariant with respect to the
subgroup of rotations on the sphere;I geometric: it is the restriction of the Lebesgue measure to the circle
as di�erentiable manifold.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 23 / 25
![Page 43: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/43.jpg)
Summary
1 The proper mathematical device to handle conditional probabilities is
the theory of conditional expectation.
2 This theory makes it possible to conditionalize on probability zero
events, as in the case of the Borel-Kolmogorov Paradox.3 Obtaining di�erent conditional probabilities is not paradoxical
becauseI they cannot be regarded as conditional probabilities of the same event
with respect to the same conditioning events;I both are intuitively correct if seen as describing generation of points
on sphere with uniform distribution.
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 24 / 25
![Page 44: The Borel-Kolmogorov Paradox and conditional expectations](https://reader030.fdocuments.in/reader030/viewer/2022012505/617fb24e202f2628dd5eac50/html5/thumbnails/44.jpg)
References
Z. Gyenis, G. Hofer-Szabó, and M. Rédei, �Conditioning using conditionalexpectation: the Borel-Kolmogorov Paradox,� (submitted).
A.N. Kolmogorov, Grundbegri�e der Wahrscheinlichkeitsrechnung, (Springer,Berlin, 1933); English translation: Foundations of the Theory of Probability,(Chelsea, New York, 1956).
W. Myrvold, �You can't always get what you want: Some considerations regardingconditional probabilities,� Erkenntnis, 80, 573-603 (2015).
M. Rescorla, �Some epistemological rami�cations of the Borel-KolmogorovParadox,� Synthese, 192, 735-767 (2015).
Gyenis, Hofer-Szabó and Rédei Borel-Kolmogorov 25 / 25