When can I set $d=4$ in dimensional regularization? I am using dimensional regularization to extract the divergence of some complicated integral. I work in $d=2\omega$ dimensions, with $\omega\approx 2$. After I extract the divergence, I have an expression of the form
$$f(\omega)\Gamma(\omega-2)\int_{-\infty}^{\infty}d\tau_3 d\tau_4 \frac{1}{(x_{13}^2)^{\omega-1}}\frac{1}{(x_{24}^2)^{\omega-1}}\frac{1}{(x_{34}^2)^{\omega-2}}\tag{1}$$
with $x_{ij}:=x_i-x_j$. Now I know how to compute
$$\int_{-\infty}^{\infty}d\tau_3 \frac{1}{(x_{13}^2)^{\omega-1}}\tag{2}$$
but the last factor spoils it. However, the integral seems finite, so if I send $\omega\to 2$ now, the last factor is simply $1$ and the integral is easy to compute.
Am I allowed to send $\omega\to 2$ for just one part of the integral, if the latter is finite? More generally, can I send $\omega\to 2$ for parts of a computation if they are finite in this number of dimensions?
Note that although this is a mathematical question, I felt that this was belonging to the physics page since (1) dimensional regularization is a tool that is used a lot in QFT, (2) the computation is directly related to a physics research, and other people probably thought about this question before in the physics community.
Clarification about the notation:
I forgot a few details about the remaining integrals: $x_{3\mu}$ and $x_{4\mu}$ are, respectively, defined as $(0,0,0,\tau_3)$ and $(0,0,0,\tau_4)$, while $x_1=(1,0,0,0)$ and $x_2=(x_2^1,x_2^2,0,0)$. Note that I work in Euclidean space. Thus, the integrals can be written as:
$$\int_{-\infty}^{\infty} d\tau_3 d\tau_4 \frac{1}{(x_1^2+\tau_3^2)^{\omega-1}}\frac{1}{(x_2^2+\tau_4^2)^{\omega-1}}\frac{1}{(x_{34}^2)^{\omega-2}}\tag{3}$$
If I set $\omega=2$, the integrals decouple and are elementary integrals. This maybe shows why my question arised in the first place.
 A: The important quantities in dimensional regularization are precisely the poles you will obtain in the limit $\omega \rightarrow 2$ and their associated residues. In other words, your bare correlation functions will involve integrals with some divergences,
$$
I = \sum_{n = 1}^m\frac{a_n}{(\omega - 2)^n} + \mathrm{finite},
$$
and the renormalization of your theory consists of getting rid of those poles, and in order to do so, you'll need to specify the constants $a_n$. 
Now the issue is that, within intermediate calculations, you may have multiple poles contributing. For example, consider the function
$$
\frac{f(\omega)}{(\omega - 2)^2},
$$
where $f(2)$ is finite. The problem here is that when you set $\omega$ equal to $2$ within this function, you are actually missing out on a first-order pole in $\omega$. Instead, you should write
$$
\frac{f(\omega)}{(\omega - 2)^2} = \frac{f(2)}{(\omega - 2)^2} + \frac{f'(2)}{(\omega - 2)} + \mathrm{finite}
$$
where $f'(\omega) = df(\omega)/d\omega$.
This is a potential issue in your case, since your integrals are multiplying $\Gamma(\omega - 2)$ which already has a pole, but I don't quite understand your notation (how are the $\tau_i$'s and $x_{ij}$'s related?). But if the integrals are all finite for $\omega = 2$ then you are safe with the replacement.
