2
$\begingroup$

I was reading R. Wong's book on Asymptotic Approximations of Integrals, and I'm having problems with the derivation of the asymptotic expansion of the exponential integral which he defined as follows: $$ Ei(z)=\int_{-\infty}^z \frac{e^t}{t} dt, \quad \mid arg(-z)\mid<\pi, $$ where the contour can be any path not crossing the branch cut on the positive real line.

After doing integration by parts, he ended up with: $$ Ei(z)=\frac{e^z}{z} \biggl [\sum_{k=0}^n \frac{k!}{z^k}+\epsilon _n(z) \biggr ], $$ where $$ \epsilon _n(z)=(n+1)!ze^{-z}\int_{-\infty}^z \frac{e^t}{t^{n+2}}dt, \ |arg(-z)|<\pi. $$

He then showed that $$|\epsilon _n(z)| \leq \frac{(n+1)!}{|z|^{n+1} (Sin \delta)^{n+2}},$$ where $\delta$ is some constant such that it ranges $(0,\pi)$, and such that $|arg(-z)| \leq \delta - \pi $; as I understood it, he introduced that delta so that the stipulation on the bounds of the argument is made explicit, and he also used it to get the error bound of epsilon.

From here on out, I have a couple of questions, first, does the last result above show that $\epsilon _n (z) = O(z^{-(n+1)})$? I think I should point out that the $Sin \delta $ above is not in absolute values, which I think is problematic, since provided the range of $\delta $, $Sin \delta $ could take on negative values and the inequality above will not be true anymore.

Second, after showing the inequality above, he concluded that as $z \rightarrow \infty $ in $|arg(-z)| \leq \delta - \pi $, then $$ Ei(z)\sim \frac{e^z}{z}\sum_{k=0}^{\infty} \frac{k!}{z^k}. $$

I'm not exactly sure why this is so, because, if assuming $\epsilon _n (z) = O(z^{-(n+1)})$ really is correct. Now, owing to the fact that there's still that $\frac{e^z}{z}$ factor outside the RHS of the second equation from above, then from what I understood, it should say that: $$ Ei(z) = \frac{e^z}{z} \sum_{k=0}^n \frac{k!}{z^k} + O \left (\frac{e^z}{z^{(n+2)}} \right), $$

which doesn't really fit the Poincare asymptotic expansion. Or is because I shouldn't see it in light of the Poincare asymptotic expansion but in terms of some other asymptotic scale?

Excuse my lengthy post, I'm quite new to the subject and I'm still having trouble figuring things out.

$\endgroup$
4
$\begingroup$

After doing integration by parts, he ended up with $$ \operatorname{Ei}(z)=\frac{e^z}{z} \biggl [\sum_{k=0}^n \frac{k!}{z^k}+\epsilon _n(z) \biggr ]. $$ [...] He then showed that $$|\epsilon _n(z)| \leq \frac{(n+1)!}{|z|^{n+1} (\sin \delta)^{n+2}},$$ where $\delta$ is some constant such that it ranges $(0,\pi)$, and such that $|\arg(-z)| \leq \delta - \pi$.

[...] I think I should point out that the $\sin \delta$ above is not in absolute values, which I think is problematic, since provided the range of $\delta $, $\sin \delta $ could take on negative values and the inequality above will not be true anymore.

If $\delta \in (0,\pi)$ then $\sin \delta > 0$, so there's no need to worry about the absolute value.

Does the last result above show that $\epsilon _n (z) = O(z^{-(n+1)})$?

The number $\delta$ is constant with respect to $z$ here, so there really is a constant

$$ C = \frac{(n+1)!}{(\sin \delta)^{n+2}} $$

such that

$$ |e_n(z)| \leq C |z|^{-(n+1)}. $$

This is exactly what it means to write $e_n(z) = O(z^{-(n+1)})$.

Now, owing to the fact that there's still that $\frac{e^z}{z}$ factor outside the RHS of the second equation from above, then from what I understood, it should say that: $$ Ei(z) = \frac{e^z}{z} \sum_{k=0}^n \frac{k!}{z^k} + O \left (\frac{e^z}{z^{(n+2)}} \right), $$ which doesn't really fit the Poincare asymptotic expansion. Or is because I shouldn't see it in light of the Poincare asymptotic expansion but in terms of some other asymptotic scale?

There are two ways to think about it. One way is to agree that it shows that

$$ \sum_{k=0}^n \frac{k!}{z^k}+\epsilon _n(z) = \sum_{k=0}^n \frac{k!}{z^k}+O(z^{-(n+1)}) $$

is a Poincaré expansion, so what the formula

$$ \operatorname{Ei}(x) \sim \frac{e^x}{x} \sum_{k=0}^{\infty} \frac{k!}{z^k} $$

means is precisely that

$$ \operatorname{Ei}(x) = \frac{e^x}{x} \left[\sum_{k=0}^{n} \frac{k!}{z^k}+O(z^{-(n+1)})\right] $$

for every $n \in \mathbb N$.

The other way is to expand things as you've done and write

$$ \operatorname{Ei}(z) = \sum_{k=0}^n \frac{k!e^z}{z^{k+1}} + O\left( \frac{e^z}{z^{(n+2)}} \right). \tag{$*$} $$

This is now a generalized asymptotic expansion $\sum_k g_k(z)$, where the summands

$$ g_k(z) = \frac{k!e^z}{z^{k+1}} $$

form an asymptotic sequence. Indeed we have

$$ O\left( \frac{e^z}{z^{(n+2)}} \right) \subset o(g_n(z)) $$

as $z \to \infty$, so a consequence of $(*)$ is that

$$ \operatorname{Ei}(z) = \sum_{k=0}^n g_k(z) + o(g_n(z)) $$

for every fixed $n \in \mathbb N$ as $z \to \infty$. This is precisely the statement that

$$ \operatorname{Ei}(z) \sim \sum_{k=0}^\infty g_k(z) $$

as $x \to \infty$, that is

$$ \operatorname{Ei}(z) \sim \sum_{k=0}^\infty \frac{k!e^z}{z^{k+1}} \tag{$**$} $$

as $x \to \infty$. Now, the interpretation would be equivalent either way (i.e. the Big O statements one could obtain from the asymptotic series) so one could formally factor out $e^z/z$ from this "sum" and thus write the result as

$$ \operatorname{Ei}(z) \sim \frac{e^z}{z}\sum_{k=0}^\infty \frac{k!}{z^k} \tag{$***$} $$

as $z \to \infty$.

You should convince yourself that $(**)$ and $(***)$ really are equivalent statements.

$\endgroup$
  • $\begingroup$ Right, I was thinking about $Cos \delta $ in the range $(0, \pi)$. I guess my real problem is that I didn't recognize that $g_k (z)$ is the asymptotic scale here. I'm actually having problems recognizing the validity of big-O's and little-o's. Anyway, thanks for the answer, it's clear for me now. $\endgroup$ – Kurome Jul 1 '14 at 21:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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