MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

These questions comes from theorem 19.C, page 81-82, in Halmos' "Measure Theory", as the image below shows.

alt text

Question 1): The 4th line of the proof says "we restrict our attention to finite valued functions" and the proof is carried out for finite f and g. Why can we "restrict our attention to finite valued functions"? How to extend the conclusion from finite case to extended real valued case?

Question 2): measurable functions, by definition in page 76-77, are defined on the whole X. But it is possible that f + g has no definition at some point x of X, e.g. $f(x)=+\infty$ but $g(x)=-\infty$. The product fg has the same situation. Of course we can assume in advance that f + g must be defined on the whole X in order for the theorem to hold, but violation of such an assumption occurs immediately: in the equation in the last line of the proof, $fg=[(f+g)^2-(f-g)^2]/4$, even if we assume f + g is defined on the whole X, we cannot guarantee that f - g and $(f+g)^2-(f-g)^2$ is meaningful for all x in X; in the paragraph that follows the proof, $f^+=f\cup 0=(f+|f|)/2$, if $f(x)=-\infty$ for some x, we cannot apply the theorem to obtain the measurability of (f+|f|)/2 and in turn of $f^+$ because f+|f| is not definable on whole X. So we have to allow that f + g (and fg) has domain smaller than X, but this violates the definition of measurable function in page 76-77. What, then, does the conclusion of the theorem "so also are f + g and fg" mean exactly on earth? and how to apply it to $fg=[(f+g)^2-(f-g)^2]/4$ and $f^+=(f+|f|)/2$?


share|cite|improve this question
Regarding Question 1: Halmos is not saying that the proof of the general case will follow from the finite valued cases; rather, he states clearly that the case in which $f$ or $g$ take infinite values requires an examination "of a small number of cases"; e.g., for $(f+g)(x)=\infty$ to hold, either $f(x)=\infty$, or $g(x)=\infty$; so the inverse image of $\infty$ will just be a union of two measurable sets, etc. After you "throw away" the points where either $f$ or $g$ is $\pm\infty$, it all takes place where there are no problems. Then you check what happens there directly. – Arturo Magidin May 11 '10 at 2:31
OK, trying again with working markup... Regarding question 2: I think it's being assumed that $f+g$ and $fg$ are actually meaningful. But really this just comes back to question one - if you have $f(x)=\pm\infty$ or $g(x)=\pm\infty$, this is handled separately. The rest of the time, everything is finite and there are no domain problems. – Harry Altman May 11 '10 at 7:13
Get Taylor's GENERALIZED THEORY OF FUNCTIONS AND INTEGRATION or Rao's book and a lot of your confusion will disappear. – The Mathemagician May 11 '10 at 18:49
up vote 1 down vote accepted

I see no explicit requirement on pages 76-77 that the function be defined on all of $X$, though I agree that it does seem implicit. It seems to me that the obvious interpretation here is that the statement says that the natural domain of $f+g$ is a measurable subset, and the function is measurable on the domain. Likewise with $fg$, which may not be defined at points where one of the two is $\pm\infty$ and the other is $0$. Also, I did not say anything about points where $f$ or $g$ are not defined. So, what do I mean?

For $f+g$: Let $M=f^{-1}(\infty)$, $N=f^{-1}(-\infty)$, $K=g^{-1}(\infty)$, and $L=g^{-1}(-\infty)$. Each is measurable, since $f$ and $g$ are measurable. Now let $Y = X - ((N\cap K)\cup (M\cap L))$. This is the natural domain of $f+g$, and is measurable. Consider first the restrictions to $Y-(M\cup N\cup K\cup L)$, applying the proof given. Then you work out $(f+g)^{-1}(\infty)$ and $(f+g)^{-1}(-\infty)$ directly in terms of $K$, $L$, $M$, and $N$ (and other sets) to get that $f+g$ is measurable on its natural domain. A similar argument, being careful with the case $f(x)=0$ and $g(x)=0$ this time, will yield the case of $fg$.

As to Question 2, since Halmos is assuming throughout that $f(x)$ and $g(x)$ are finite valued functions, the situation does not arise: the equation involves only finite quantities and all operations are defined. This handles the situation in $X-(K\cup L\cup M\cup N)$. Then you need to consider the situation when you are working in $(M\cup N)\cap g^{-1}(0)$, $(M\cup N)\setminus g^{-1}(0)$, and the remaining two cases.

share|cite|improve this answer
we have to assume that the whole space X is a measurable set in order for the natural domain of f+g, $X-[(N\cap K)\cup(M\cap L)]$, to be measurable. I think this hypothesis is indispensable for the conclusion to hold. Because this book is based on sigma-ring instead of sigma-algebra although in many modern textbooks this is a default setting, I think this hypothesis should be mentioned. But, unfortunately, it is omitted in the theorem. – zzzhhh May 12 '10 at 11:42

too long to be a comment, so I have to post it as an answer.

@Arturo Magidin: Do you mean: by "throw away" points where f and g are not defined or take infinite value, we do not need to abide by the requirement in the definition of measurable function in page 76-77 of this book that the domain of f be the whole X? We only need to grasp the essential idea that $N(f)\cap f^{-1}(M)$ is measurable for all extended Borel set M, that is, in particular, $f^{-1}(\{+\infty\}), f^{-1}(\{-\infty\})$ and $N(f)\cap f^{-1}(M)=f^{-1}(M-\{0\})$ is measurable for all real Borel set M, without caring if f + g and fg are definable on whole X. I think this is what I should learn from this theorem. So, denoting $F=[(f+g)^2-(f-g)^2]/4$ from the last line of the proof or (f+|f|)/2 for $f^+$, after checking inverse images of $\{+\infty\}$ and $\{-\infty\}$, since for any real subset E we have $(fg)^{-1}(E)$ or $(f^+)^{-1}(E)=F^{-1}(E)$ which lie in the subset of X where both f (x) and g (x) are defined and finite (note this subset is contained in the actual domain of F which may be smaller than X), we have $(fg)^{-1}(M-\{0\})$ and $(f^+)^{-1}(M-\{0\})=F^{-1}(M-\{0\})$ are measurable. For product fg, this is what we want, although fg is not necessarily definable for the whole X. For $f^+$, this completes the proof that $f^+$ is a measurable function. Then my second question is therefore answered. Is the above words what your reply intended and correct, Prof Magidin?

PS: Note that $(f+g)^{-1}(\{+\infty\})=[f^{-1}(\{+\infty\})\cap g^{-1}((-\infty,+\infty])]\cup[g^{-1}(\{+\infty\})\cap f^{-1}((-\infty,+\infty])]$ which is measurable by the 3rd paragraph in page 77. We also need to prove that $N(f+g)=[N(f)\cup N(g)]-C$ where $C=\{x|f(x)=-g(x)\}$ is measurable to apply Theorem 19.A. Because $C\subseteq N(f)\cup N(g)$, so $C=[N(f)\cup N(g)]\cap C$ which is measurable by Theorem 19.A, so N (f + g) is measurable. Note that C may contain (all) points that f + g is not meaningful.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.