The Fourier transform of $1/p^3$
Take the following Fourier transform conventions $$ \tilde{f}(\mathbf{p}) =\int f(\mathbf{x}) \ e^{-i\mathbf{p}\cdot\mathbf{x}} \ d^3x $$ $$ f(\mathbf{x}) =\int \tilde{f}(\mathbf{p})\ e^{i\mathbf{p}\cdot\mathbf{x}} \ \frac{d^3p}{(2\pi)^3}. $$ In physics, it is useful to compute the transform of $1/p^3$ $$ I=\int \frac{1}{p^3}\ e^{i\mathbf{p}\cdot\mathbf{x}} \ \frac{d^3p}{(2\pi)^3}. $$ But I have no idea what it even mean since the integral is not well-defined.
1 answer
In short, in $\mathbb{R}^d$, if I define the Fourier transform as $\mathcal{F}(f)(x) = \int_{\mathbb{R}^d} f(y) \,e^{-2iπxy}\,\mathrm{d}y$ the result is $$ \boxed{\mathcal{F}\left(\frac{1}{\omega_d,|x|^d}\right) = \frac{\psi(d/2)-\gamma}{2} - \ln(|πx|)} $$ where $\omega_d = \frac{2\,\pi^{d/2}}{\Gamma(d/2)}$ is the size of the unit sphere (so $\omega_3 = 4\pi$), $\gamma$ is Euler–Mascheroni constant and $\psi$ is the digamma function. Since $\psi(3/2) = 2-\gamma-\ln(4)$, we deduce that in dimension $3$ $$ \mathcal{F}\left(\frac{1}{4\pi,|x|^3}\right) = 1-\gamma -\ln(|2πx|) $$ which with your Fourier transform convention gives $$ \boxed{\tilde{\mathcal{F}}\left(\frac{1}{|x|^3}\right) = \frac{1-\gamma -\ln(|x|)}{2\pi^2}.} $$
Now the details. So first, what is the meaning of $\frac{1}{|x|^d}$? One can define the following distribution $$ v_d := \mathrm{pf}\left(\frac{1}{|x|^d}\right) := \mathrm{div}\left(\frac{x\ln(|x|)}{|x|^d}\right) $$ where the derivative is taken in the sense of distributions. One easily verifies that the distribution $v_d$ coincides on $\Bbb R^d\setminus\{0\}$ with the function $$ v_d(x) = \frac{1}{|x|^d} \text{ for any } x≠ 0. $$ One also notices that it is a tempered distribution as the derivative of a function in $L^1 + L^\infty$.
Let $u_d = \frac{x\ln(|x|)}{|x|^d}$. Multiplying by a test function $\varphi\in C^\infty_c$, one gets for any $\lambda>0$ $$ \begin{align*} \langle v_d,\varphi\rangle &= -\int_{\mathbb{R}^d} u_d\cdot\nabla\varphi \newline &= -\int_{|x|\leq\lambda} u_d\cdot\nabla(\varphi(x)-\varphi(0)) - \int_{|x|>\lambda} u_d\cdot\nabla \varphi \end{align*} $$ which by integration by parts yields $$ \begin{align}\tag{1} \langle v_d,\varphi\rangle &= \int_{|x|\leq\lambda} \frac{\varphi(x)-\varphi(0)}{|x|^d},\mathrm{d}x \newline &\quad+ \int_{|x|>\lambda} \frac{\varphi(x)}{|x|^d} ,\mathrm{d}x + \omega_d \ln(\lambda) ,\varphi(0). \end{align} $$ One can take $\lambda = 1$ to get $$ \begin{align*} \langle v_d,\varphi\rangle &= \int_{|x|\leq 1} \frac{\varphi(x)-\varphi(0)}{|x|^d},\mathrm{d}x + \int_{|x|> 1} \frac{\varphi(x)}{|x|^d} ,\mathrm{d}x. \end{align*} $$ But with the formula (1) with $\lambda\neq 1$, one can also compute easily the distribution $v_d(\lambda x)$ defined by $$ \langle v_d(\lambda x),\varphi(x)\rangle = \frac{1}{|\lambda|^d}\langle v_d(x),\varphi(x/\lambda)\rangle $$ using Equation (1) with $\varphi(x)$ replaced by $\varphi(x/\lambda)$. Doing the change of variable $x/\lambda \to x$ yields $$ v_d(\lambda,\cdot) = \frac{1}{|\lambda|^d},v_d + \omega_d\frac{\ln(\lambda)}{|\lambda|^d},\delta_0. $$ Therefore, we can now use the scaling properties of the Fourier transform to get for any $r=1/\lambda>0$ $$ \begin{align*} (\mathcal{F}(v_d))(r\tilde{x}) &= r^{-d} (\mathcal{F}(v_d(y/r)))(\tilde{x}) \newline&= \mathcal{F}(v_d-\omega_d\ln(r)\delta_0)(\tilde{x}) \newline&= \mathcal{F}(v_d)(\tilde{x}) -\omega_d\ln(r). \end{align*} $$ Taking $\tilde{x} = \frac{x}{|x|}$ and $r=|x|$ gives $$ \boxed{\mathcal{F}(v_d)(x) = C_d -\omega_d\ln(|x|)} $$ where $C_d = \mathcal{F}(v_d)(\tilde{x})$ is a constant since the Fourier transform of a radial function is radial.
If you want to know the constant $C_d$, the usual trick is to multiply by a Gaussian and use the fact that we know the Fourier transform of a Gaussian. Here remark first that by the Fourier inversion theorem we have $$ \mathcal{F}(\ln(|x|)) = C_d, \delta_0 - \frac{v_d}{\omega_d} $$ Therefore $$ \begin{align*} C_d - \langle\mathcal{F}(\ln(|x|)), e^{-\pi|x|^2}\rangle &= \frac{1}{\omega_d}\langle v_d, e^{-\pi|x|^2}\rangle \newline &= \int_0^1 \frac{e^{-\pi r^2}-1}{r},\mathrm{d} r + \int_1^\infty \frac{e^{-\pi r^2}}{r},\mathrm{d} r \newline &= \int_0^\pi \frac{e^{-t}-1}{2,t},\mathrm{d} t + \int_\pi^\infty \frac{e^{-t}}{2,t},\mathrm{d} t \newline &= \frac{-\ln(\pi)}{2} + \int_0^\pi \frac{\ln(t),e^{-t}}{2},\mathrm{d} t + \int_\pi^\infty \frac{\ln(t),e^{-t}}{2},\mathrm{d} t \newline &= \frac{-\gamma - \ln(\pi)}{2} \end{align*} $$ where $\gamma = -\Gamma'(1) = -\psi(1)$ and I did a polar change of variable and the change $t = πr^2$ and then integrations by parts. But since $\mathcal{F}(e^{-\pi|x|^2}) = e^{-\pi|x|^2}$, we can also compute $$ \begin{align*} \langle\mathcal{F}(\ln(|x|)), e^{-\pi|x|^2}\rangle &= \int_{\mathbb{R}^d} \ln(|x|), e^{-\pi|x|^2},\mathrm{d} x \newline &= \omega_d \int_0^\infty \ln(r),e^{-\pi r^2} r^{d-1},\mathrm{d} r \newline &= \frac{1}{2,\Gamma(d/2)} \int_0^\infty (\ln(t)-\ln(\pi)),e^{-t}, t^{d/2-1},\mathrm{d} t \newline &= \frac{1}{2} ,\big(\psi(d/2) - \ln(\pi)\big). \end{align*} $$ with the same changes of variable. We deduce that $C_d = \frac{\psi(d/2)-\gamma}{2} - \ln(\pi)$. Tell me if you spot any errors!
0 comment threads
2 comment threads