Compactifying linear optical unitaries using multiport beamsplitters

We show that any $N$-dimensional unitary matrix can be realized using a finite sequence of concatenated identical fixed multiport beamsplitters (MBSs) and phase shifters (PSs). Our construction is based on a Lie group theorem applied to existing decompositions. Using the Bell-Walmsley-Clements framework, we prove that any $N$-dimensional unitary requires $N+2$ phase masks, $N-1$ fixed MBSs, and $N-1$ BSs. Our scheme requires only $\mathcal{O}(N)$ fixed, identical components (MBSs and BSs) compared to the $\mathcal{O}(N^2)$ fixed BSs required by conventional schemes (e.g., Clements), all while keeping the same number of PSs. Experimentally, these MBS can be realized as a monolithic element via femtosecond laser writing, offering superior performance through reduced insertion losses. As an application, we present a reconfigurable linear optical circuit that implements a three-dimensional unitary emerging in the unambiguous discrimination of two nonorthogonal qubit states.

💡 Research Summary

The paper “Compactifying linear optical unitaries using multiport beamsplitters” introduces a novel architecture for implementing arbitrary N‑dimensional unitary transformations in linear‑optical quantum information processing. Traditional universal interferometers, such as the Reck and Clements schemes, decompose a target unitary into a cascade of two‑port beamsplitters (BS) and phase shifters (PS). While these methods are mathematically complete, they require O(N²) fixed BSs (N(N‑1) in total) and N² tunable phases, leading to deep interferometers with substantial insertion loss and limited scalability.

Motivated by recent advances in fabricating multiport beamsplitters (MBS) – devices that couple N input modes to N output modes in a single monolithic element – the authors propose to replace the large number of two‑port BSs with a small set of identical, fixed MBSs. An MBS can be built from N‑1 conventional BSs and internal fixed phases, but when fabricated as a single element (e.g., by femtosecond laser waveguide writing) it behaves as a loss‑optimized black box with a fixed unitary matrix.

The theoretical core relies on a Lie‑group theorem: any element of a connected Lie group can be expressed as a finite product of elements drawn from a generating subset. By applying this theorem to the unitary group U(N) and using the Bell‑Walmsley‑Clements (BWC) framework, the authors show that a universal set of operations can be generated with only N‑1 identical MBSs, N‑1 conventional BSs, and N+2 phase masks. The phase masks are placed before, after, and between the MBS layers, providing the necessary tunable degrees of freedom. Consequently, the total number of fixed components scales linearly with N, a dramatic reduction from the quadratic scaling of conventional designs.

The paper provides explicit constructions for low dimensions. For N=2, a single BS together with two phase shifters constitutes a basic block; cascading two such blocks reproduces any SU(2) operation. For N=3, a tritter (the three‑port analogue of a BS) serves as the elementary block. Using four identical tritters the authors can synthesize any U(3). They demonstrate this by implementing the three‑dimensional unitary that appears in the optimal unambiguous discrimination (USD) of two non‑orthogonal qubit states. Numerical analysis confirms that three tritters are insufficient, establishing the optimality of the four‑tritter construction.

For higher dimensions (N≥4), the BWC architecture generalizes straightforwardly: each layer consists of a single MBS acting on all N modes, interleaved with phase masks. The authors prove that this layout requires exactly N+2 phase masks, N‑1 fixed MBSs, and N‑1 ordinary BSs. They also discuss practical fabrication routes, emphasizing that monolithic MBSs can be produced with femtosecond laser direct writing, which dramatically reduces insertion loss compared with a cascade of separate BSs.

The work highlights several implications. First, the linear scaling of fixed components enables more compact, low‑loss interferometers, which is crucial for large‑scale photonic quantum processors, boson‑sampling devices, and multi‑photon entanglement generation. Second, the approach is compatible with existing reconfigurable platforms because the number of tunable phase shifters remains unchanged (N+2), preserving programmability while simplifying hardware. Third, the authors suggest that similar reductions could be achieved in architectures that employ discrete Fourier transform blocks or other higher‑dimensional static units.

In summary, the authors present a rigorous proof and concrete designs showing that any N‑dimensional unitary can be realized with O(N) identical, fixed multiport beamsplitters and a modest number of phase shifters. This compactification of linear optical unitaries offers a clear path toward scalable, low‑loss photonic quantum technologies.