⚛️ quantum physics

$N$ -dimensional discrete Fourier transform via bosonic Hamiltonian

This paper proposes a novel bosonic Hamiltonian framework that realizes the $N$ -dimensional discrete Fourier transform in photonic integrated circuits using a single-stage multimode evolution, achieving a significantly improved scaling complexity of $\mathcal{O}(N\log\log{N})$ and enabling the construction of large-scale transforms with far fewer interferometers than traditional architectures.

Original authors: Edgar Barriga, Camila Muñoz, Alejandro Muñoz, Santiago Rojas-Rojas, Pablo Solano, Carla Hermann-Avigliano, Aldo Delgado

Published 2026-09-09

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Edgar Barriga, Camila Muñoz, Alejandro Muñoz, Santiago Rojas-Rojas, Pablo Solano, Carla Hermann-Avigliano, Aldo Delgado

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the world of modern computing, there is a mathematical tool so fundamental that it underpins everything from digital music compression to the security of online banking. This tool is the discrete Fourier transform, a method for taking a complex signal and breaking it down into its individual frequency components, much like a prism splitting white light into a rainbow of colors. While this process is routine for classical computers, a new frontier has opened in quantum physics, where scientists are trying to perform this same transformation on quantum states. Doing so efficiently is the key to unlocking powerful new algorithms, such as those capable of factoring large numbers to break encryption codes. However, building the physical machines to perform these quantum calculations has hit a wall. Current designs rely on stacking thousands of tiny optical switches in long, cascading chains, a method that becomes impossibly bulky and error-prone as the system grows larger. The challenge has been to find a way to perform this complex transformation in a single, compact step without the need for a massive assembly line of components.

A team of researchers in Chile has now proposed a solution that reimagines how light travels through these microscopic circuits. Instead of forcing photons through a long series of switches, they suggest guiding them through a single, carefully engineered interaction region where multiple paths cross and mix simultaneously. The researchers modeled these circuits as networks of waveguides, which are essentially tiny channels that carry light, similar to how fiber optic cables carry data but on a microscopic scale. In their new approach, the light waves in these channels interact with one another through a phenomenon called evanescent coupling, where the electromagnetic fields of neighboring waves overlap and exchange energy. By calculating the precise strength of these interactions and the speed at which light travels through each channel, the team discovered configurations where the light emerges perfectly transformed into the desired quantum pattern after just one pass.

The team explored two distinct ways to arrange these waveguides. The first approach treated the channels as the vertices of a regular polygon, where every channel interacts with every other channel. Using mathematical analysis, they found that this "all-to-all" connection works beautifully for systems with up to six channels. For a five-channel system arranged in a pentagon, for instance, the light naturally evolves into the correct pattern. However, this method hits a physical limit: as the number of channels increases, the distance between the outermost channels grows, and the strength of their interaction drops off exponentially. Beyond six channels, the interaction becomes too weak to be practical with current technology. To solve this, the researchers turned to a second approach based on graph theory, where they treated the waveguides as nodes in a network that does not require every channel to touch every other one.

In this graph-based model, the researchers realized that simply connecting the channels was not enough; they also needed to adjust the speed of light in specific channels by changing the material properties of the waveguides themselves. This added a new degree of freedom, allowing them to find solutions for much larger systems. Through extensive computer simulations, they identified specific geometric layouts that could perform the transformation for systems with up to eight channels, and they found partial solutions for systems with nine and ten channels. For larger systems, they proposed three guiding rules, or conjectures, to help narrow down the search for the correct configurations. These rules suggest that the number of connections between channels must fall within a specific range to work, a finding that helps filter out the billions of possible arrangements that would otherwise be impossible to test.

The significance of this work lies in its dramatic reduction of complexity. In traditional designs, the number of components required to perform this transformation grows with the square of the number of channels, meaning a system with a thousand channels would need a million components. The new method, by using these single-stage interactions as building blocks, reduces the growth rate to a much slower pace. The researchers calculated that using their new building blocks, they could assemble a system capable of handling over two thousand channels using only about two thousand components, a stark contrast to the millions required by older methods. This efficiency is crucial for applications like boson sampling, a test used to prove that quantum computers can outperform classical ones, which requires manipulating hundreds of photons simultaneously.

The team also addressed the practical reality of building these devices. They developed a method to test whether their theoretical solutions could actually be manufactured, considering the inevitable imperfections that occur when carving tiny structures into glass or crystal. They found that while many mathematical solutions exist, only a fraction of them are robust enough to survive the tiny errors in positioning that happen during fabrication. By applying a strict selection criterion, they identified specific layouts that are not only mathematically sound but also physically realizable with current technology. For the smaller systems, they provided exact blueprints, including the precise distances between channels and the required interaction lengths, showing that these devices could be built today using standard techniques like femtosecond laser writing.

Ultimately, this research offers a new path forward for quantum photonics. By moving away from the idea of chaining components together and instead designing a single, complex interaction that does the heavy lifting all at once, the researchers have shown that it is possible to scale up quantum circuits without hitting a wall of complexity. Their work provides the missing pieces for constructing larger, more efficient quantum devices, bringing the promise of powerful quantum algorithms closer to reality. The findings suggest that the future of quantum computing may not lie in building bigger machines with more parts, but in designing smarter, more integrated systems where light flows through a single, perfectly tuned landscape to perform its most complex tasks.

Technical Summary: N-dimensional Discrete Fourier Transform via Bosonic Hamiltonian

Problem Statement
The Discrete Fourier Transform (DFT) is a fundamental unitary operator in both classical algorithms and quantum information processing, notably serving as a key component in Shor's algorithm and Boson sampling. Implementing the $N$ -dimensional DFT (specifically the Quantum Fourier Transform Interferometer, or QFTI) in photonic integrated circuits (PICs) is currently constrained by existing architectures. Standard designs based on Reck and Clements architectures utilize cascades of Mach-Zehnder interferometers (MZIs), resulting in a spatial complexity scaling of $O(N^2)$ . While optimizations like the Barak et al. design (mapping the Cooley-Tukey algorithm) reduced this to $O(N \log N)$ , the resource requirements remain prohibitive for large-scale applications, such as Boson sampling with $N=2500$ modes, which would require approximately $3 \times 10^6$ interferometers. The challenge is to realize the $N$ -dimensional DFT in a single stage of multimode evolution to significantly reduce the number of required optical components.

Methodology
The authors propose realizing the $N$ -dimensional DFT through the evolution of coupled bosonic spatial modes in a waveguide array, governed by a bosonic Hamiltonian. The system is modeled as a graph where vertices represent waveguides and edges represent evanescent couplings. The evolution operator is given by $U = e^{-iCz}$ , where $C$ is the coupling matrix and $z$ is the propagation length. To achieve the DFT matrix $F_N$ , the authors supplement this evolution with phase shifters ( $\Phi_{in}$ and $\Phi_{out}$ ) before and after the interaction region.

The research employs two distinct modeling frameworks:

Polygonal Model (Complete Graphs): The authors analyze systems where every waveguide couples to every other waveguide (complete graphs). They derive analytical equations for the coupling coefficients to satisfy normalization and phase matching conditions. This approach yields analytical solutions for $N=4, 5, 6$ and numerical solutions up to $N=31$ . However, this model is limited by the exponential decay of evanescent coupling strength with distance, making it experimentally feasible only for $N \leq 6$ .
Graph-Based Model (Non-Complete Graphs): To overcome distance limitations, the authors explore non-complete connected graphs. Crucially, they introduce non-zero propagation constants ( $\beta_l \neq 0$ $β_{l} \neq = 0$ ) for specific waveguides within the Hamiltonian. This allows for the realization of the DFT in geometries that do not require all-to-all coupling.
- Numerical Strategy: The authors formulate a root-finding problem to satisfy the normalization condition $|U_{jk}|^2 = 1/N$ . They utilize the Levenberg-Marquardt algorithm to solve for coupling constants ( $C_{jk}$ ) and propagation constants ( $\beta_l$ ).
- Complex Hadamard Matrices (CHMs): The authors distinguish between solutions that merely satisfy normalization (CHMs) and those that match the specific phase structure of the DFT. They employ the concept of "defect" (an algebraic invariant) and "affine families" to verify if a solution corresponds to the DFT. For dimensions where the defect of the DFT is non-zero (e.g., $N=6$ ), they investigate non-affine parameters to ensure the correct phase structure.

Key Contributions and Results

Single-Stage Realization: The paper demonstrates that the $N$ -dimensional DFT can be achieved in a single stage of multimode evolution, requiring only phase shifters at the input and output, rather than deep cascades of MZIs.
Analytical and Numerical Solutions:
- Polygonal Model: Analytical solutions are provided for $N=4, 5, 6$ . Numerical solutions for the coupling matrix are found up to $N=31$ for the CHM condition, though experimental feasibility is limited to $N \leq 6$ due to coupling decay.
- Graph-Based Model: The authors report all valid geometrical configurations for $N \leq 8$ , partial exploration for $N=9$ , and selected cases for $N=10$ . They identify that for non-prime dimensions, different propagation constants are essential.
Discovery of Non-Affine Parameters: In the case of $N=6$ , the authors uncover the missing non-affine parameters required to construct the 6-dimensional DFT. They show that the 6-dimensional DFT belongs to a non-affine family of CHMs, resolving a gap in the understanding of the 6-dimensional Fourier matrix structure.
Scaling Law and Efficiency: By utilizing the new building blocks (the identified single-stage configurations) and combining them via Cooley-Tukey and Good-Thomas decompositions, the authors derive a new scaling law for the number of interferometers required. The complexity scales as $O(N \log \log N)$ $O (N lo g lo g N)$ .
- Specific Comparison: For a 2520-dimensional QFTI (relevant for Boson sampling), the proposed method requires only 1459 interferometers (using Good-Thomas decomposition) or 2625 interferometers (using Cooley-Tukey with $N=4,5$ blocks). This is a drastic reduction compared to the $\approx 3 \times 10^6$ interferometers required by Reck and Clements architectures.
Fabrication Tolerance: The authors establish a sensitivity criterion to filter numerical solutions based on physical admissibility. They demonstrate that multiple geometric configurations exist for the same $N$ , offering a "design resource" where specific layouts can be selected to maximize tolerance to fabrication errors (e.g., waveguide positioning).

Significance
The paper claims that these findings provide a pathway to assembling large-scale QFTIs with significantly reduced spatial complexity, making high-dimensional quantum information processing and Boson sampling more feasible with current photonic hardware. The work bridges the gap between theoretical Hamiltonian evolution and practical PIC design by:

Providing a systematic search for coupling geometries that realize the DFT.
Identifying the necessity of propagation constants in non-complete graphs.
Resolving the structural parameters of the 6-dimensional DFT.
Demonstrating a scaling law ( $O(N \log \log N)$ ) that allows for the assembly of massive interferometers (e.g., $N=2520$ ) with a number of components comparable to current experimental capabilities, unlike the exponential or quadratic scaling of previous methods.

The authors conclude that while the polygonal model is theoretically robust, the graph-based model with tailored propagation constants is the viable route for experimental realization in 3D waveguides for $N > 6$ . They note that the search for configurations for $N \geq 10$ is computationally intensive and guided by three conjectures regarding edge counts and regular graphs.

Technical Summary: N-dimensional Discrete Fourier Transform via Bosonic Hamiltonian

More like this