Researchers at the German Aerospace Center and University of Cologne have devised new methods to encode data into quantum states, overcoming a key obstacle to realizing practical quantum computation. The work introduces deterministic and scalable amplitude-encoding techniques that simplify fully controlled multiplexer circuits to one- and two-qubit gates, a reduction typically considered complex.

These approaches use tensor networks to encode data exceeding the capacity of classical memory, and benchmarking shows up to orders of magnitude improvement in circuit depth compared to existing methods. By significantly reducing gate depth for encoding, this work enables the experimental execution of applications on current and near-term hardware.

Scalable Unitary State-Encoding for Large Data Sets

Disjoint Sum of Products synthesis simplifies each mask independently, enabling construction of encoding circuits for complex functions and data sets with controlled fidelity error bounds. When classical memory capacity is exceeded, the team uses tensor networks to represent data, with the Tensor Train serving as the simplest example, allowing evaluation of encoding equations even for very large datasets. This approach circumvents the limitations of directly storing exponentially growing angle sets required for quantum state preparation, a common bottleneck in quantum computing protocols.

The ability to encode arbitrary complex data or queryable functions beyond classical memory limits represents an advancement in handling massive datasets for quantum algorithms. Encoding the state over 20 qubits in a $2^{10}$ × $2^{10}$ grid, measured amplitudes were binned into 64× 64 bins for visibility, and the target state was plotted over the same bins for comparison, details of which are available in Section IV. 9. The developed methods offer tunable trade-offs between state-fidelity and circuit depth, providing a practical and scalable solution for deterministic amplitude-encoding.

Quantum Data Encoding Bottlenecks and Potential Speedups

The approach allows for scaling to larger qubit numbers when the underlying data exhibits exploitable structure, a key factor in overcoming limitations of classical data loading for quantum algorithms. Analysis revealed the methods can prepare quantum states of moderate size. The team benchmarked performance against standard state preparation methods implemented in the package, finding varying degrees of circuit-depth reduction across all functions analyzed. The Hybrid method consistently outperformed at larger qubit counts, suggesting a potential for significant gains in computational efficiency as quantum systems grow.

This improvement is particularly relevant given that algorithmic speedups do not automatically translate into overall computational advantage without efficient data preparation and readout. The work addresses a fundamental bottleneck in quantum computing: the efficient encoding of classical data into quantum states, a challenge that has received substantial scientific attention and economic investment.

While quantum algorithms promise exponential speedups, the cost of preparing and reading out data can negate those benefits, necessitating encoding methods with low gate-depth, high fidelity and feasible qubit requirements. The team’s approach offers deterministic or high-probability success, which supports reliable quantum computation, and is agnostic to the input data structure.

Harrow-Hassidim-Lloyd Algorithm Limitations for Large N

The efficiency of encoding data into quantum states directly impacts the viability of algorithms like the Harrow-Hassidim-Lloyd algorithm, as polynomial scaling in the number of elements can negate potential speedups. Recognizing this limitation, the work introduces a method for determining which components of the Walsh-Hadamard spectrum can be safely truncated without significantly impacting state fidelity, effectively sparsifying the encoding process.

This flexibility is important for adapting to the constraints of specific quantum hardware and applications. The team’s methods are designed to be deterministic, ensuring reliable state preparation and scalable, allowing for the encoding of increasingly complex datasets.

Fully Controlled Multiplexers and Generic State Preparation

Fully controlled multiplexer circuits can now be simplified to use only one- and two-qubit gates through newly developed algorithms. This reduction in gate complexity stems from formally mapping the elimination of redundant gate controls to a Boolean logic minimization problem, a process that streamlines circuit construction. The approach allows for the creation of more efficient circuits, potentially accelerating computations that rely on complex state preparation.

These algorithms also use a single ancilla qubit in conjunction with the Walsh-Hadamard spectrum of gate angles to map multi-controlled rotation gates into sequences of simpler operations. By analyzing the angles through this spectrum, researchers can effectively translate complex rotations into a series of one- and two-qubit operations, further reducing the overall circuit depth. This technique is particularly beneficial for generic state preparation, an important step in many quantum protocols, and avoids the need for amplitude amplification or mid-circuit measurement.

The researchers’ method systematically truncates branches within a binary-tree graph representing the hierarchical dependence of angles in multiplexer-based state preparation. This truncation focuses on eliminating branches with small probability amplitude weight, effectively prioritizing the most influential angles in the final state. This selective pruning aims to preserve fidelity while minimizing circuit complexity.

The process identifies qubit layouts where this elimination occurs early in the tree structure, leading to an exponential reduction in the number of angles required for encoding. A generalized control string, denoted as c, is used to represent partially controlled rotations, identifying sets of binary strings that can be covered by the control qubits.

Optimizing Circuit Depth with Tensor-Network Representations

The Boolean Tree-Binary Encoder (BT-BE) algorithm achieves a circuit depth of 1128 while maintaining an infidelity of 6. 5 x 10^-4 when encoding data up to 36 qubits, a level of performance previously difficult to attain without significantly increasing ancillary qubit requirements. This optimization stems from a systematic approach to simplifying multiplexer circuits, reducing their complexity by examining the underlying structure of gate angles within each layer.

The team’s work demonstrates a method for approximating continuous multiplexer angles with a finite basis, generating fixed-angle Boolean masks that streamline the encoding process. Analysis of the BT-BE algorithm reveals a direct relationship between binary precision and both circuit depth and encoding fidelity for a fixed number of qubits.

This sparser spectrum is achieved through the application of Tensor Contraction Isometry (TCI) to the encoded state, resulting in a maximum bond dimension of 24 with a cutoff of 10^-10 for the Singular Value Decomposition (SVD) spectrum. 5 x 10^-4. The researchers state.

Depth Reduction Compared to Qiskit’s State Preparation

Benchmarking reveals that the new algorithms consistently outperform standard state preparation methods in terms of circuit depth, particularly as the number of qubits increases. For a function family evaluated on a uniform grid, the total one- and two-qubit gate depth of the Binary-Tree based methods, BT-BE, BT-WHT and BT-Hybrid, demonstrates a clear advantage over the standard approach, with scaling fits indicating sustained efficiency gains. The ratio of CX-depth, a measure of circuit complexity, for each Binary-Tree method is significantly lower than that of the standard state preparation, indicating a substantial simplification of the circuits.

A comparison using a function discretized on a much smaller grid revealed a striking difference, and despite being 2¹² times smaller in size, the standard method required a circuit over 10. This disparity highlights the efficiency gains achieved by the new methods in handling large datasets.

The analysis of circuit depth versus the number of qubits further confirms the scalability of the approach, with the Binary-Tree methods consistently exhibiting lower depths than the standard state preparation. These results demonstrate a practical and scalable set of methods for deterministic amplitude-encoding, enabling the experimental execution of applications previously limited by circuit complexity.

Encoding Fidelity Benchmarks on IQM Hardware

The work’s ability to prepare states on IQM hardware with high accuracy further validates the approach, shortening the time horizon for robust quantum utility. Unlike prior approximation methods limited to real, positive functions and burdened by barren plateaus, this approach offers a practical implementation for hardware, addressing both angle calculation and redundant gate elimination. The team benchmarked fidelity by calculating the overlap between the prepared state and the target state, reporting infidelity as I = 1-F, where F represents the fidelity.

Demonstrations on IQM hardware with a square lattice connectivity yielded excellent fidelity, even with minimal noise mitigation techniques applied. The researchers report that the resulting circuits are not only smaller in depth but also scale more favorably than the theoretical limit of Θ(2^n/n) expected for exact encoding, achieving this with only a modest increase in infidelity.

This combination of reduced complexity and maintained accuracy is important for executing applications on current and near-term quantum hardware, preventing calculations from being overwhelmed by noise before completion. The team normalized functions over flattened indices using uniform grids containing $2^{n_x}$ and $2^{n_y}$ points, where $n_{x}$ = ⌈n/2⌉ and $n_{y}$ = ⌊n/2⌋, allowing for encoding of two-dimensional functions f(x,y). This work addresses the challenges of constructing encoding circuits for large datasets under well-controlled fidelity error bounds.

👉 More information
🗞 Scalable Unitary State-Encoding for Generic Large Data
✍️ Vittorio Pagni, Peter K. Schuhmacher, Michael Felderer, Michael Epping and Kevin Lively (German Aerospace Center (DLR), Institute of Software Technology, Sankt Augustin, Germany)
🧠 ArXiv: https://arxiv.org/pdf/2611.09760

More like this