The usual way to investigate the order of vanishing of a modular form at a cusp other than $\infty$ is to find an element of $\mathrm{SL}_2(\mathbb{Z})$ that maps $\infty$ to your cusp and "recenter" your form at $\infty$ using this element. If your element is $\gamma$, then look at $j(\gamma,z)^{-1}f(\gamma z)$ (or something like this...) where $j$ is the cocyle by which the form $f$ transforms. With any luck you can work out a formula for this from which the behavior at $\infty$ (and hence the behavior of $f$ at your chosen cusp) is evident.

For $\theta$ the easiest way to do this is to look at a table of "theta functions with characteristics" (like the one in Mumford's *Tata Lectures on Theta*, Volume I if I recall correctly). The result is that the $q$-expansions of $\theta$ at $\infty$, $0$, and $1/2$ are, respectively, $\sum q^{n^2}$, $\sum (q^{1/4})^{n^2}$, and $q^{1/4}\sum q^{n^2+n}$, where the sums are over $\mathbb{Z}$ and the latter two are really only defined up to some constant (perhaps just a fourth root of unity).

Now each cusp on a modular curve comes with a "width" $h$ - its ramification index over $X(1)$, and usually the $q$-expansion of a modular form at a cusp is an expansion in $q^{1/h}$. In this case the cusp $1/2$ has width $1$, and clearly something else is going on here. It appears that $\theta$ vanishes to order $1/4$ at this cusp!

While this turns out to be an oddly useful perspective (at least it has to me), it's nonsense on its face. To remedy this you need to be careful about setting up $\theta$ as a section of some complex-analytic line bundle on $X_0(4)$. As usual, this thing is just cooked up using the cocyle by which $\theta$ transforms and when you work out the local picture around the cusp $1/2$ you'll see that the section $\theta$ vanishes to order $1$ there as a section of this bundle.

To read more about the goofy phenomenon at $1/2$ search around for "irregular cusp." The underlying reason for the problem is that, from a moduli point of view, the object classified by this cusp has non-trivial automorphisms (so complex-analytically $X_0(4)$ is better thought of as an orbifold). On the other hand, one can see this disparity on the power of $q$ needed to expand a modular form very explicitly in the integral weight case by messing with local calculations for the sheaf $\omega^{\otimes k}$ around the cusp. This was done very nicely in some early draft of Brian Conrad's book on the Ramanujan-Petersson conjectures. I don't know what the status of that book is these days though...

5more comments