Suppose $$\hat{H}_1 = U\left(\hat{n}_{d,\uparrow} - \frac{1}{2}\right)\left(\hat{n}_{d,\downarrow} - \frac{1}{2}\right)\tag{1}$$ To decouple the many-body operator, many articles suggest using the discrete Hubbard-Stratonovich transformation $$e^{-\Delta \tau \hat{H}_1}=\frac{1}{2} \sum_{\sigma= \pm 1} e^{\lambda \sigma\left(\hat{n}_{d, \uparrow}-\hat{n}_{d, \downarrow}\right)}\tag{2}$$ where $\cosh\lambda=e^{\Delta\tau U/2}$.
All the articles make references to J. E. Hirsch's paper on the transformation, but the formula presented in the paper is $$\exp(-\Delta\tau U n_\uparrow n_\downarrow) = \frac{1}{2} \sum_{\sigma=\pm1} \exp\left[2a\sigma(n_\uparrow - n_\downarrow) - \frac{1}{2}U\Delta\tau(n_\uparrow + n_\downarrow)\right]\tag{3}$$ where $\tanh^2 a = \tanh(\Delta\tau U/4)$. The paper said it's easily to prove this, but I don't see where it comes from.
How does one derive equation (3), and use it to derive equation (2)?