Confusion & potential error in Probability Theory (For Scientists and Engineers)

[mdekstrand]

First, thank you for providing these studies. I have been looking for probability theory intro material to provide to my data science students & advisees, and I think your studies will be quite useful for their intermediate-level study of probability.

There are two things that tripped me up, though.

The first is the pushforward map definition. Since $f$ is not restricted to be one-to-one at this point, isn't $f^{-1}$ a set? Use later in 4.1 would be consistent with this understanding. If $f^{-1}$ is a set, $f^{-1} \in A$ didn't make sense to me; $f^{-1} \cap A \ne \emptyset$ seems to be the meaning here.

The second is the definition of absolutely continuous in 4.3. The statement:

A measure $\nu$ is absolutely continuous with respect to another measure $\mu$ when $\nu$ allocates zero volume only to those sets which $\mu$ also allocates zero volume

My translation of this, consistent with the 'only if' in the following formula, would be that $\nu$ absolutely continuous w.r.t. $\mu$ means $\nu(A) = 0 \implies \mu(A) = 0$ ($\nu(A) = 0$ only if $\mu(A) = 0$). However, this is the converse of the definition I find in Athreya and Lahiri (p. 53) and the Encyclopedia of Mathematics: $\nu$ is absolutely continuous w.r.t. $\mu$ if $\mu(A) = 0 \implies \nu(A) = 0$. If I'm understanding correctly, it should say:

A measure $\nu$ is absolutely continuous with respect to another measure $\mu$ when $\nu$ allocates zero volume to all sets to which $\mu$ also allocates zero volume

It's quite possible I'm missing something, as I am pretty new to measure theory, but I don't see how these definitions don't contradict.

[betanalpha]

First, thank you for providing these studies. I have been looking for probability theory intro material to provide to my data science students & advisees, and I think your studies will be quite useful for their intermediate-level study of probability.

Thanks!

The first is the pushforward map definition. Since $f$ is not restricted to be one-to-one at this point, isn't $f^{-1}$ a set? Use later in 4.1 would be consistent with this understanding. If $f^{-1}$ is a set, $f^{-1} \in A$ didn't make sense to me; $f^{-1} \cap A \ne \emptyset$ seems to be the meaning here.

f^{-1} is definitely meant to be treated as a set everywhere, except in a few places where I explicitly assume a 1-1 function and consider only point pullbacks, f^{-1}(y). For example in the intro to Section 4.1, right before Section 4.1.1, I talk about how a pullback can be computed on discrete spaces by pulling back each element in the set one by one.

Looking at the current version I see f^{-1} = A, where A \in \mathcal{X}, the sigma-algebra of the input space, everywhere. Is there a particular place in the text with confusion notation, or should I try to make the role of A more clear?

The second is the definition of absolutely continuous in 4.3. The statement:

A measure $\nu$ is absolutely continuous with respect to another measure $\mu$ when $\nu$ allocates zero volume only to those sets which $\mu$ also allocates zero volume

My translation of this, consistent with the 'only if' in the following formula, would be that $\nu$ absolutely continuous w.r.t. $\mu$ means $\nu(A) = 0 \implies \mu(A) = 0$ ($\nu(A) = 0$ only if $\mu(A) = 0$). However, this is the converse of the definition I find in Athreya and Lahiri (p. 53) and the Encyclopedia of Mathematics https://www.encyclopediaofmath.org/index.php/Absolute_continuity: $\nu$ is absolutely continuous w.r.t. $\mu$ if $\mu(A) = 0 \implies \nu(A) = 0$. If I'm understanding correctly, it should say:

A measure $\nu$ is absolutely continuous with respect to another measure $\mu$ when $\nu$ allocates zero volume to all sets to which $\mu$ also allocates zero volume

It's quite possible I'm missing something, as I am pretty new to measure theory, but I don't see how these definitions don't contradict.

Thanks! The “only” should indeed by “every”.

I’ll include a fix when I update the case study next.

[mdekstrand]

Looking at the current version I see f^{-1} = A, where A \in \mathcal{X}, the sigma-algebra of the input space, everywhere. Is there a particular place in the text with confusion notation, or should I try to make the role of A more clear?

The definition of f_*(A) in 1.4 was the place where I found the confusion.

Thank you very much!

[betanalpha]

Okay, thanks. I’ll add some clarifying text in the next edit.

On Feb 4, 2020, at 1:03 PM, Michael Ekstrand [email protected] wrote:

Looking at the current version I see f^{-1} = A, where A \in \mathcal{X}, the sigma-algebra of the input space, everywhere. Is there a particular place in the text with confusion notation, or should I try to make the role of A more clear?

The definition of f_*(A) in 1.4 was the place where I found the confusion.

Thank you very much!

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/betanalpha/knitr_case_studies/issues/31?email_source=notifications&email_token=AALU3FRX2CWYVCSIRYSS3XDRBGUYRA5CNFSM4KJJTRC2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEKYTLGY#issuecomment-582038939, or unsubscribe https://github.com/notifications/unsubscribe-auth/AALU3FW7PNPQATHB44KBKV3RBGUYRANCNFSM4KJJTRCQ.