This is one part of an old qualifier analysis problem that I am trying to complete.
Suppose $\phi$ is a continuous function on a finite interval $[0,b]$ and that $\phi(x)>0$ for all $x\in[0,b]$. Assume that $\phi(x)\sim A x^r$ for some constant $A>0$ and $r\geq0$. Define $$F(\varepsilon)=\int^b_0 \frac{dx}{\varepsilon + \phi(x)}$$
- If $0\leq r<1$, then $\lim_{\varepsilon\rightarrow0}=\int^b_0\frac{dx}{\phi(x)}<\infty$.
- If $r>1$, then $$F(\varepsilon)\sim \frac{\pi}{r A^{1/r}\sin(\pi/r)} \varepsilon^{-(1-\frac1r)}\quad\text{as}\quad\varepsilon\rightarrow0$$
- If $r=1$, then $$F(\varepsilon)\sim\frac{1}{A}\log(\varepsilon^{-1})\quad\text{as}\quad\varepsilon\rightarrow0$$
Part (1) is a simple application of monotone convergence.
When $r\geq1$, $\lim_{\varepsilon0}F(\varepsilon)=\infty$ by monotone convergence.
Part (2) I have done by first applying the of variable $u=\varepsilon^{-1}x^r$, followed by application of dominated convergence. All that gives $$\lim_{\varepsilon\rightarrow0}\varepsilon^{1-\frac1r}F(\varepsilon)=\int^\infty_0\frac{u^{\frac1r-1}}{1+Au}\,du$$ The later is a integral of the Mellin transform type and the expression in the problem can be obtained by complex contour integration.
Part (3) is where I am having difficulties. The asymptotic seems given in the problem seems correct. By assuming that $\phi(t)=At$, then a direct computation gives $$F(\varepsilon)=\frac{1}{A}\log\Big(\frac1A+\frac{b}{\varepsilon}\Big)\sim\frac{1}{A}\log(\varepsilon^{-1})$$ I thing dominated convergence still useful but a trick may be needed. For example, the function $x\mapsto x^{-1}\phi(x)$ can be extended to $[0,b]$ as a strictly positive continuous function on $[0,b]$ and so, there is $M>0$ such that $\phi(x)\geq Bx$ for $x\in[0,b]$. Hence $$\frac{1}{\varepsilon +\phi(x)}\leq\frac{1}{\varepsilon+Bx}$$
Any hints (not necessarily a complete solution) are appreciated. Thank you!
Hint:
Given any $0 < t < A$ there's a $\delta > 0$ such that $(A - t)x^r < \phi(x) < (A + t)x^r$ when $x < \delta$. Insert this into the $x < \delta$ portion of the integral and you'll get lower and upper bounds for this part of the integral which will be close to the asymptotics you need as $\epsilon \rightarrow 0$. The $x > \delta$ integral will be an error term as $\epsilon$ goes to zero. Letting $t \rightarrow 0$ gives the desired asymptotics.