Denis A. Drozhzhin

Quantum computation with $d$-level quantum systems, also known as qudits, benefits from the possibility to use a richer computational space compared to qubits. However, for an arbitrary qudit-based hardware platform, the issue is that a generic qudit operation has to be decomposed into the sequence of native operations $-$ pulses that are adjusted to the transitions between two levels in a qudit. Typically, not all levels in a qudit are simply connected to each other due to specific selection rules. Moreover, the number of pulses plays a significant role, since each pulse takes a certain execution time and may introduce error. In this paper, we propose a resource-efficient algorithm to decompose single-qudit operations into the sequence of pulses that are allowed by qudit selection rules. Using the developed algorithm, the number of pulses is at most $d(d{-}1)/2$ for an arbitrary single-qudit operation. For specific operations, the algorithm could produce even fewer pulses. We provide a comparison of qudit decompositions for several types of trapped ions, specifically $^{171}\text{Yb}^+$, $^{137}\text{Ba}^+$ and $^{40}\text{Ca}^+$ with different selection rules, and also decomposition for superconducting qudits. Although our approach deals with single-qudit operations, the proposed approach is important for realizing two-qudit operations since they can be implemented as a standard two-qubit gate that is surrounded by efficiently implemented single-qudit gates.

Here we present the results of benchmarking a quantum processor based on trapped $^{171}$Yb$^{+}$ ions by performing basic quantum machine learning algorithms. Using a quantum-enhanced support vector machine algorithm with up to five qubits we perform a supervised binary classification on two types of datasets: small binary digit images and weighted graphs with a ring topology. For the first dataset, images are intentionally selected so that they could be classified with 100% accuracy. This allows us to specifically examine different types of quantum encodings of the digit dataset and study the impact of experimental noise. In the second dataset, graphs are divided into two categories based on the spectral structure of their Ising Hamiltonian models, which is related to the NP-hard problem. For this problem we consider an embedding of an exponentially large Hamiltonian spectrum into an entangled state of a linear number of qubits. For both problems, we study various levels of circuit optimization and found that, for all experiments conducted, we achieve classifiers with 100% accuracy on both training and testing datasets. This demonstrates that the quantum processor has the ability to correctly solve the basic classification task under consideration. As we expect, with the increase in the capabilities of quantum processors, they can be utilized for solving machine learning tasks.