Axioms of measure

axiomsdefinitionmeasure-theory

Let $X$ be a set. $\mathscr{F}$ be a $\sigma$-algebra of subsets of $X$. Let $\mu$ be an extended real valued function defined on $\mathscr{F}$ satisfying

$(i) E \in \mathscr{F} \implies \mu(E) \geq 0$

$(ii) \mu(\phi) = 0$

$(iii) E_1,E_2,\ldots \in \mathscr{F}, E_i \cap E_j=\phi \implies \mu(\cup_{i=1}^{\infty}E_i)=\sum_{i=1}^{\infty}\mu(E_i)$

Then $\mu$ is a measure. My question is that : Do we really need $(ii)$? Does not it come directly from $(iii)$? $\mu(\phi)=\mu(\phi \cup\phi)=\mu(\phi)+\mu(\phi) \implies \mu(\phi)=0.$

Wikipedia says that we require at least one $E \in \mathscr{F}$ to have finite measure to write $\mu(E)=\mu(E \cup\phi)=\mu(E)+\mu(\phi) \implies \mu(\phi)=0.$ I do not understand this requirement when we can do it without it!

P.S. I have a feeling that the problem arises because we are allowing $\mu$ to take $+\infty$. However, my point still stands for finite measures $(\mu(X)<\infty)$!

Best Answer

The problem is that $0$ is not the only extended real number $a$ such that $a+a=a$, since $\infty+\infty=\infty$. So you could also have $\mu(\emptyset)=\infty$ instead of $\mu(\emptyset)=0$. (In that case, you would have $\mu(A)=\infty$ for all $A$ since you can write $A=A\cup\emptyset$, so as long as any set has finite measure this can't happen.)