1
$\begingroup$

Problem

I have been working through content in real analysis and came across across an inequality in a proof that is unclear to me.

We are told that a function $\psi$ is a step function on the interval $[a,b]$ where there exists partition $P=${$p_0,…p_k$} for which $\psi$ is constant on all intervals $(p_{i-1},p_i)$.

The proof begins with the inequality:

enter image description here

Note: we define $a_i=a+i\frac{(b-a)}n$

I understand that we can replace each value $\psi (a_i)$ with $\lVert \psi \lVert_\infty$ to bound the left hand side.

However, I’m not quite sure how they have derived the upper bound $2k\frac{(b-a)}n\lVert \psi\lVert_\infty$ from the information provided.

I would be grateful if anyone could shed some light on what I’m failing to understand here.

Edit

I have included more of the proof for clarity where the aim is to show that as n$\rightarrow$$\infty$ the left hand side converges to 0:

enter image description here

where $S[a,b]$ is the set of step functions on $[a,b]$.

$\endgroup$
8
  • $\begingroup$ What are the $a_i$? Are they randomly sampled from the $(p_{i-1},p_i)$ or are they chosen in a specific way? $\endgroup$
    – Glare
    Commented Apr 15, 2022 at 7:16
  • $\begingroup$ The $a_i$ aren’t defined to take any specific values, however, the eventual aim of the proof is to show that as n $\rightarrow \infty$ the left hand side of the inequality converges to 0. $\endgroup$
    – FD_bfa
    Commented Apr 15, 2022 at 7:23
  • $\begingroup$ There must be some constraint on the $a_i$'s or other variables, since otherwise, with the given information, the claim is not true in general. $\endgroup$ Commented Apr 15, 2022 at 8:02
  • $\begingroup$ I have edited my answer to include the full section of the proof, however, there doesn’t seem to be any more of a constraint than the ones I listed above. Unless I’m mistaken, or there is a mistake in the proof @AndreasLenz $\endgroup$
    – FD_bfa
    Commented Apr 15, 2022 at 8:09
  • 1
    $\begingroup$ The claim is wrong in general, here is a counterexample. Suppose $$ \psi(t) = \begin{cases}0 & a\leq t < \frac{a+b}{2} \\1 & \frac{a+b}{2} \leq t \leq b\end{cases}, $$ so $k=2$. Now, without any further restrictions on the $a_i$'s, we may set $a_i=a$ and $n=16$, for example. Therefore $\psi(a_i) = 0$ and the l.h.s evaluates to $\frac{b-a}{2}$, while the r.h.s becomes $\frac{b-a}{4}$ and so the claim is wrong. $\endgroup$ Commented Apr 15, 2022 at 8:30

1 Answer 1

1
$\begingroup$

The idea is the following. Consider another step function $g(t)$, whose discontinuities are at the points $a_i = a + i\frac{b-a}{n}$ and which has the step heights $g(a_i) = \psi(a_i)$ with a uniform step width $\frac{b-a}{n}$. This function has the integral $$ \int_a^b g(t) \mathrm{d}t = \sum_{i=1}^n \frac{b-a}{n} \psi(a_i). $$ The key observation of the authors is then, that $g(t)$ and $\psi(t)$ may only differ in intervals of size $\frac{b-a}{n}$ around the points where $\psi(t)$ is discontinuous. As $\psi(t)$ has at most $k$ discontinuities and the maximum difference between $g(t)$ and $\psi(t)$ is $2 ||\psi||_\infty$, the difference between the two integrals satisfies. $$ \left| \int_a^b \psi(t)\mathrm{d}t - \int_a^b g(t)\mathrm{d}t \right| \leq k\frac{b-a}{n} 2 ||\psi||_\infty. $$

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .