Dear Uncle Colin

Help! An emergency has arisen and my L’Hôpital’s Rule is out of order – but I still need to find limx0+ln(x)cot(x)! This is not a drill!

-- Help! Inexplicable Limitations Bring Everyone Ridiculous Trouble

Don’t panic, HILBERT! Don’t panic.

Now, for the benefit of anyone who’s not constrained by HILBERT’s inability to use L’Hôpital’s rule 1, that is the obvious way to solve it. When

L=limxx0uv is undefined (because u=0 and v=0 or u= and v=), then L=limxx0dudxdvdx.

In this case, limx0+ln(x)cot(x)=limx0+1xcosec2(x)=limx0+sin2(x)x.

Since for all x>0, sin(x)<x, 0<sin2(x)<x2. That means 0<sin2(x)x<x, so as x0+, L0. Sandwich rule for the win!

However.

However, as HILBERT says, L’Hôpital’s rule isn’t allowed in this urgent problem. So how do we approach it?

The first thing is to rewrite the limit as:

limx0+ln(x)tan(x). It already looks nicer.

We can apply a similar sort of reasoning as before using the sandwich rule, but it needs a bit of setting up: it turns out that 1x<ln(x)0 for 0<x1.

The upper end of the inequality is obvious; the lower end not so much, so let’s show it: it’s equivalent to showing 1<xln(x) in this range. Differentiating y=x12ln(x) gives dydx=12x12ln(x)+x121x. This vanishes when 1x(ln(x)+2)=0, or x=e2, at which point y>1. (It works out to be -0.736 or so). That’s the only turning point, and it’s a global minimum on the domain where it’s defined, so the inequality holds on 0<x1.

Lucky!

So, we know that 1x<ln(x)0 for 0<x1, which means tan(x)x<ln(x)tan(x)0.

Now all we need to do is show that tan(x)x goes to 0 as x0. It turns out that tan(x)<2x for 0<x<1, so tan(x)x>2xx=2x, which goes to 0 as x0.

So that’s it done! We successfully sandwiched ln(x)tan(x) between 0 and 2x, both of which go to 0 as x0.

Hope that saves the day!

-- Uncle Colin

Footnotes:

1. discovered by Bernoulli, of course