70

I am having trouble computing integration w.r.t. counting measure. Let $(\mathbb{N},\scr{P}(\mathbb{N}),\mu)$ be a measure space where $\mu$ is counting measure. Let $f:\mathbb{N}\rightarrow{\mathbb{R}}$ be a non-negative bounded measurable function. Then, what is $\int_{\mathbb{N}}fd\mu$? What's gonna happen if we remove the boundedness in $f$, i.e. just let $f$ be an arbitrary non-negative measurable function?

What happens if we relace $\mathbb{N}$ by a general set $X$?

  • 17
    If $\mu$ is the counting measure on $\mathbb{N}$, then it would seem that $$\int_{\mathbb{N}} f,\text{d}\mu = \sum_{n=1}^\infty f(n)$$ – Nicholas Stull Apr 22 '14 at 04:29

2 Answers2

95

Counting measure is just summation!

To see this, you can approach from a few different angles; how about we consider the Monotone Convergence Theorem. To that end, for $n\in\mathbb{N}$, define $f_n:\mathbb{N}\to\mathbb{R}$ by $$ f_n(k)=\begin{cases}f(k) & \text{if }1\leq k\leq n\\ 0 & \text{else}\end{cases}. $$ Then clearly, as $n\to\infty$, $f_n\to f$ pointwise; it is also monotone increasing, because $f(k)\geq0$ for all $k\in\mathbb{N}$. So, by the MCT, $$ \int_{\mathbb{N}} f_n\,d\mu\to\int_{\mathbb{N}} f\,d\mu\text{ as }n\to\infty. $$ Now, consider these $f_n$. Note that we can write $$ \mathbb{N}=\{1\}\cup\{2\}\cup\cdots\cup\{n\}\cup\{n+1,n+2,\ldots\}, $$ and that these sets are all measurable. So, $$ \begin{align*} \int_{\mathbb{N}} f_n\,d\mu&=\int_{\{1\}}f_n\,d\mu+\cdots+\int_{\{n\}}f_n\,d\mu+\int_{\{n+1,n+2,\ldots\}}f_n\,d\mu\\ &=\int_{\{1\}}f_n(1)\,d\mu+\cdots+\int_{\{n\}}f_n(n)\,d\mu+\int_{\{n+1,n+2,\ldots\}}0\,d\mu, \end{align*} $$ where we have used that $f_n$ is constant on each of these sets, by definition. So, we see that $$ \int_{\mathbb{N}} f_n\,d\mu=1\cdot f_n(1)+1\cdot f_n(2)+\cdots+1\cdot f_n(n)+0=f(1)+\cdots+f(n). $$ So, we have that $$ \int_{\mathbb{N}} f\,d\mu=\lim_{n\to\infty}\int_{\mathbb{N}} f_n\,d\mu=\lim_{n\to\infty}(f(1)+\cdots+f(n))=\sum_{k=1}^{\infty}f(k). $$ The boundedness is of no consequence here -- since our terms are non-negative, the series either converges or diverges to $\infty$; in either case, the integral is exactly the sum.

Nick Peterson
  • 32,430
  • 2
    @ Nicholas: Your arguments really help me to grasp the concept of counting measure. Thanks a lot! What's gonna happen if we replace $\mathbb{N}$ by a general set $X$? –  Apr 22 '14 at 04:49
  • 4
    @Algebra The same. $\int f,\mathrm d\mu=\sum_{x\in X}f(x)$ – Hagen von Eitzen Apr 22 '14 at 05:05
  • 4
    @Algebra I'm glad that you found it helpful! Also: Hagen von Eitzen is correct regarding general counting measures. – Nick Peterson Apr 22 '14 at 23:54
  • 1
    Can we do the same for $f:\mathbb{R}\rightarrow\mathbb{R}$? But mass is defined on points $supp(f)\subset\mathbb{N}\subset\mathbb{R}$. – ZHU Feb 17 '18 at 23:10
  • Yes, you can. The same argument applies. – Nick Peterson Feb 19 '18 at 02:15
  • 2
    @NickPeterson Can we do it like this instead: $\int_{\mathbb{N}} f \mathrm{d} \mu = \int_{\cup_{n \in \mathbb{N}} {n} } f \mathrm{d} \mu = \sum_{n \in \mathbb{N}} \int_{ {n} } f \mathrm{d} \mu = \sum_{n \in \mathbb{N}} \int_{ {n} } f(n) \mathrm{d} \mu = \sum_{n \in \mathbb{N}} f(n) \mu({n}) = \sum_{n \in \mathbb{N}} f(n)$ ? – goblinb May 20 '20 at 03:00
  • 1
    Really nice explanation! +1 – RFZ Jul 17 '20 at 23:13
  • Hello! @NickPeterson I'm very new to measure theory, and I have trouble seeing why $$ \int_{{1}} f_n(1) d\mu = 1\cdot f_n(1)$$ Could you throw some light on what basic principle is being used here? I know what a counting measure is defined to be, $\mu(E) = |E|$ where $E$ is whatever set in the underlying $\sigma$-algebra. Is it generally the case that: $$\int_E d\mu = \mu(E)$$ Thanks for your help! – stoic-santiago Mar 05 '21 at 17:14
  • 2
    @strawberry-sunshine Yes, $\int_E 1,d\mu=\mu(E)$ for any measurable set. And more generally, for any fixed $\alpha$, $\int_E\alpha,d\mu=\alpha\int_E1,d\mu=\alpha\mu(E)$. – Nick Peterson Mar 05 '21 at 21:20
  • 1
    I'm wondering the same as goblinb above: Why can't we just directly consider $\int_{\cup_{n\geq 1}{n}}f(n)=\sum_n f(n)$ and why wouldn't that be true even if f not necessarily non neg? – Christopher.L Nov 12 '21 at 20:57
  • @NickPeterson I am not sure, but it seems that the measure $\mu$ has a role, because maybe the sum is $\int_{\mathbb{N}} f \text{d}\mu = \sum_{k=1}^\infty f(k)\mu({k})$, clearly it can not be true that $\mu({k}) = 1$ for any $k$ (we must have $\mu({k}) < 1$ for some $k$). – Davius Jan 03 '22 at 21:55
  • @Davius I don't follow your question. By definition, the counting measure on a given measurable space is the measure that assigns to each measurable set a measure equal to its cardinality ($\infty$ if it is not a finite set). – Nick Peterson Jan 04 '22 at 02:12
  • @NickPeterson Ok, now I understand. I misunderstood that $\mu$ was an arbitrary measure on the natural numbers. Thank you very much. – Davius Jan 05 '22 at 19:13
  • @NickPeterson hey! I see that you answered a lot here and I thought I could ask you the following. I have the counting measure $\mu$ on $\mathcal P(X)$ and I have to determine $\mathcal M(X,\mu)$, i.e. to determine the set of all measurable functions. How could I do this? And is it different than having a counting measure $\mu$ on the set $X$ instead of on the power set of $X$ like in my case? Thanks! – syphracos Jan 20 '22 at 19:20
1

$\color{red}{\mathbf{\text{general case}:}}$

if $X$ be arbitrary set and if $(X,P(X),\mu)$ be counting measure space on $P(X)$ such that

$\underset{\{x\}\xrightarrow{\hspace{1.2cm}}1}{\mu:P(x)\rightarrow [0,\infty]}$

and if

$f:X\rightarrow [0,\infty]$

be an arbitrary function on $X$, then integral of $f$ is equal to

$\int_Xfd\mu\color{blue}=\underset{\underset{\text{$\phi$ is simple}}{\phi\leq f}}{\sup}\left(\int_{X}\phi\; d\mu\right)=\underset{\underset{\underset{\text{F is finite}}{F\subseteq X}}{\phi\leq f}}{\sup}\left(\underset{x\in F}{\sum} \phi(x)\right) =\underset{\underset{\text{F is finite}}{F\subseteq X}}{\sup}\left(\underset{x\in F}{\sum} f(x)\right) \color{red}=\underset{x\in X}{\sum}f(x)$

in fact equality in blue, is definition of integral and equality in red is definition of sum of arbitrary positive function on arbitrary set $X$.

now, let $f:X\rightarrow \mathbb{C}$. if $$\int_X |f|d\mu<\infty$$ then integral of $f$ is defined by:

\begin{align*} &\int_X fd\mu\\ &=\Bigg[\int_X f^+_{Re}d\mu-\int_X f^-_{Re}d\mu\Bigg]+i\Bigg[\int_X f^+_{Im}d\mu-\int_X f^-_{Im}d\mu \Bigg]\\ &=\Bigg[\underset{x\in X}{\sum}f^+_{Re}(x)+ \underset{x\in X}{\sum}f^-_{Re}(x)\Bigg]+i\Bigg[ \underset{x\in X}{\sum}f^+_{Im}(x) - \underset{x\in X}{\sum}f^-_{Im}(x) \Bigg] \end{align*}

$\color{red}{\textbf{the case $X=\mathbb{N}$}}$

now, if $X=\mathbb{N}$, then for function $f\geq 0$, we have

$\int_{\mathbb{N}}fd\mu=\underset{n\in \mathbb{N}}{\sum}f(n)$

and integral for complex function, is defined as above.