I was reading this research paper Quantum interference enables constant time quantum information processing and was confused by one particular expression involving the Hamiltonian of a beam splitter.
We consider two interfering modes $a$ and $b$ on a beam splitter device. The hamiltonian of the beam splitter is then
$$\hat{H}_{B S}=\frac{i \hbar}{2}\left(\hat{a}^{\dagger} \hat{b} e^{-i \varphi}-\hat{a} \hat{b}^{\dagger} e^{i \varphi}\right)$$
where $\hat{a}$, and $\hat b$ are respective annihilation operators of states $a$ and $b$.
Then the unitary transformation generated by the beam splitter is given by $$U_{bs}=e^{-i \theta H_{B S} / \hbar}$$ where $\sin\left(\frac{\theta}{2}\right)=\sqrt{r}$ (the reflection coefficient).
My question given this how is it possible to derive the expression given on page 32 for the annihilation operatorrs for output states $a_{r}$ and $a_{t}$
$$\begin{array}{l}{a_{r}=U_{B S}^{\dagger} a U_{B S}=a \cos \frac{\theta}{2}+b e^{-i \varphi} \sin \frac{\theta}{2}} \\ {a_{t}=U_{B S}^{\dagger} b U_{B S}=-a e^{i \varphi} \sin \frac{\theta}{2}+b \cos \frac{\theta}{2}}\end{array}$$
How can we derive this result using only the information given above? I tried evaluating the matrix exponential in Mathematica for some low dimensional cases and got several contradictions (mostly sign errors) which just made me more confused.