Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Here is Bartle's proof of Hake's Theorem found in "A Modern Theory of Integration". I think there is a mistake in the highlighted line:

The Theorem: $f:[a,b]\to \mathbb{R}$ is gauge integrable if and only if it is gauge integrable in $[a,x]\ \forall x\in [a,b)$ and the limit $\lim_{x\to b}\int_a^xf$ is finite. In that case \begin{equation}\int_a^bf=\lim_{x\to b^-}\int_a^xf\end{equation}

The problem is in the $(\impliedby)$ proof:

Let $(c_n)$ be a strictly increasing sequence with $c_0=a$, $c_n\to b$. As $\lim_{n\to +\infty}\int_{c_0}^{c_n}f=\lim_{x\to b^-}\int_a^xf=L\in \mathbb{R}$ and $c_n\to b$, $\exists r\in \mathbb{N}$ so that \begin{equation}\forall x\in (c_r,b)\ \left|\int_a^xf-L\right|<\epsilon\text{ and }\left|f(b)\right|(b-c_r)<\epsilon\end{equation} By the integrability in $[c_{n-1},c_n]$, there exists a gauge $\delta_n$ on $[c_{n-1},c_n]$ so that if $\dot{\mathcal{P}}_n<<\delta_n$, \begin{equation}\left|S(f,\dot{\mathcal{P}}_n)-\int_{c_{n-1}}^{c_n}f\right|<\frac{\epsilon}{2^n}\end{equation} Without loss of generality, \begin{align*}\delta_1(c_0)&\le \frac12(c_1-c_0)\tag{i}\\ \delta_{n+1}(c_n)&\le \min\left\{\delta_n(c_n),\frac12(c_n-c_{n-1}),\frac12(c_{n+1}-c_n)\right\}\tag{ii}\\ \delta_n(x)&\le \min\left\{\frac12(x-c_{n-1}),\frac12(c_n-x)\right\}\text{ for }x\in (c_{n-1},c_n)\tag{iii} \end{align*} We define a gauge in $[a,b]$ by \begin{equation}\delta(x)=\begin{cases}\delta_n(x)&\text{ if, }x\in [c_n,c_{n+1})\\ b-c_N&\text{ if, }x=b\end{cases}\end{equation} Let $\mathcal{P}=\left\{a=x_0<...<x_m=b\right\}$ partition $[a,b]$ with tags $t_i$ so that $\dot{\mathcal{P}}<<\delta$. As $b\not\in\cup_{n=0}^{\infty}[c_n,c_{n+1})=[a,b)$, this forces $b$ to be a tag of $[x_{m-1},b]$. But then, as $t_m=b$, \begin{equation}c_r=b-\delta(b)<x_{m-1}\end{equation} Let $s\in \mathbb{N}$ be the smallest integer with $x_{m-1}\le c_s$ so that $r\le s$. If $k=1,...,s-1$ then condition (iii) implies that the point $c_k$ must be a tag of the sub-interval in $\mathcal{P}$ that contains it. ...

My argument: Suppose $c_k\in [x_p,x_{p+1}]$. But, $[x_p,x_{p+1}]$ may also contain other points of the sequence (finitely many). Let $c_{q},c_{q+1},...,c_k,c_{k+1},...,c_l$ be all these points. Suppose $t_p$ is the tag of $[x_p,x_{p+1}]$. If $t_p\neq c_q,c_{q+1},...,c_l$ and $t_p>c_k$ then $t_p\in (c_i,c_{i+1})$ with $c_i\ge c_k$. Condition (iii) and the fact that $\dot{\mathcal{P}}<<\delta$ imply that $$c_k> t_p-\delta(t_p)=t_p-\delta_i(t_p)\ge t_p-\frac{t_p-c_i}2=\frac{t_p+c_i}2\ge \frac{c_k+c_k}2=c_k$$ which is a contradiction. Similarly if $t_p<c_k$. What I have shown is that the tag must be a point of the sequence. But why must it be $c_k$ like Bartle says?. This fact is crucial to the proof. Bartle then continues:

Using the right-left procedure we may assume $c_0,c_1,...,c_{s-1}\in \mathcal{P}$.

That can't be done (I think). Indeed consider the simple case where the only points of the sequence in $[x_p,x_{p+1}]$ are $c_k$ and $c_{k+1}$. The tag $t_p$ must be one of $c_k,c_{k+1}$. Suppose it were $c_k$. Then $c_k\in \mathcal{P}$ and $c_{k+1}\in (c_k,x_{p+1})$. How is it then possible to prove $c_{k+1}$ is the tag of $[c_k,x_{p+1}]$?. $c_k$ seems like an equally strong candidate for this.

EDIT: I also think there is a gap in "As $b\not\in\cup_{n=0}^{\infty}[c_n,c_{n+1})=[a,b)$, this forces $b$ to be a tag of $[x_{m-1},b]$." Why can't a point $c_n$ be the tag?

share|improve this question
add comment

2 Answers 2

I don't have the book, so I can only go on what you've quoted. But based on that, the indexing seems off in your definition of $\delta$: the gauge $\delta_n$ is defined on $[c_{n-1},c_n]$, so we should have $\delta(x) = \delta_{n+1}(x)$ if $x\in [c_n,c_{n+1})$, not $\delta_n(x)$. Maybe this is a Bartle's mistake, or maybe you made a typo in copying it.

Anyway, with this correction, I think that condition (ii) solves the problems. Condition (ii) guarantees that $\delta(c_n)$ is smaller than the distance from $c_n$ to any other point of the sequence. Thus, a sub-interval of which $c_n$ is the tag cannot contain any other $c_k$. Similarly, it cannot contain $b$, since $b$ is greater than all the $c_k$'s.

share|improve this answer
add comment

Ok, I'm not sure but... The fact that if there is almost a point $c_{k}$ of the sequence in $[x_{k-1},x_{k}]$ then one of them must be the tag seems to be true but in addition we have the following:

By the definition of $\delta$: $\delta_{1}(c_{0})\leq \frac{1}{2}(c_{1} - c_{0})$ $\;$. Suppose that there are two points $c_{0}$ and $c_{1}$ of the sequence. One of them must be the tag , let $c_{0}$ be the tag. For the $\delta$-fine condition: $$ [x_{k-1},x_{k}] \subseteq [t_{k}-\delta(c_{0}) , t_{k}+\delta(c_{0})] $$ Therefore $$ | x_{k}- x_{k-1} | \leq 2 \delta(c_{0}) \leq |c_{1} - c_{0}| $$ which is a contradiction except when $c_{0}=x_{k-1}$ and $c_{1}=x_{k}$ in which case we use the left-right procedure to obtain both subintervals that have $c_{0}$ and $c_{1}$ as tags.

Conclusion: There cannot be more than one element c_{k} of the sequence in each subinterval $[x_{k-1},x_{k}]$ containing $c_{k}$ and this $c_{k}$ must be the tag.

share|improve this answer
add comment

Your Answer

 
discard

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.