Search

With the emergence of Foundational Machine Learning Interatomic Potential (FMLIP) models trained on extensive datasets, transferring data between different ML architectures has become increasingly important. In this work, we examine the extent to which training data optimised for one machine-learning forcefield algorithm may be re-used to train different models, aiming to accelerate FMLIP fine-tuning and to reduce the need for costly iterative training. As a test case, we train models of an organic liquid mixture that is commonly used as a solvent in rechargeable battery electrolytes, making it an important target for reactive MLIP development. We assess model performance by analysing the properties of molecular dynamics trajectories, showing that this is a more stringent test than comparing prediction errors for fixed datasets. We consider several types of training data, and several popular MLIPs - notably the recent MACE architecture, a message-passing neural network designed for high efficiency and smoothness. We demonstrate that simple training sets constructed without any ab initio dynamics are sufficient to produce stable models of molecular liquids. For simple neural-network architectures, further iterative training is required to capture thermodynamic and kinetic properties correctly, but MACE performs well with extremely limited datsets. We find that configurations designed by human intuition to correct systematic model deficiencies transfer effectively between algorithms, but active-learned data that are generated by one MLIP do not typically benefit a different algorithm. Finally, we show that any training data which improve model performance also improve its ability to generalise to similar unseen molecules. This suggests that trajectory failure modes are connected with chemical structure rather than being entirely system-specific., Comment: 23 pages, 7 figures

We show that unconventional relaxation dynamics of special initial states in one-dimensional arrays of Rydberg atoms produce non-generic decay of the initial-state survival probability at early times. Using the PXP hamiltonian as a minimal model of the Rydberg blockade, we prove that the early-time survival probability for states exhibiting quantum many-body scarring decays at a characteristic rate, whose finite-size scaling is determined solely by scars. We numerically investigate the effects of both revival-enhancing and ergodicity-restoring perturbations and find results consistent with the limiting cases of integrable and ergodic dynamics, respectively. Since the survival probability is easily accessible experimentally at early times, our findings enable us to probe the presence of scars at time scales much shorter than that of thermalization.

This paper focuses on the presentation and evaluation of the high-level quantum programming language Eclipse Qrisp. The presented framework, used for developing and compiling quantum algorithms, is measured in terms of efficiency for its implementation of the Quantum Approximation Optimization Algorithm (QAOA) Module. We measure this efficiency and compare it against two alternative QAOA algorithm implementations using IBM's Qiskit toolkit. The evaluation process has been carried out over a benchmark composed of 15 instances of the well-known Maximum Cut Problem. Through this preliminary experimentation, Eclipse Qrisp demonstrated promising results, outperforming both versions of its counterparts in terms of results quality and circuit complexity., Comment: 7 pages, 2 figures and 2 tables. Paper accepted for being presented at the 3rd International Conference on Emergent Quantum Technologies (ICEQT'24)

We introduce a tensor network algorithm for the solution of $p$-spin models. We show that bond compression through rank-revealing decompositions performed during the tensor network contraction resolves logical redundancies in the system exactly and is thus lossless, yet leads to qualitative changes in runtime scaling in different regimes of the model. First, we find that bond compression emulates the so-called leaf-removal algorithm, solving the problem efficiently in the "easy" phase. Past a dynamical phase transition, we observe superpolynomial runtimes, reflecting the appearance of a core component. We then develop a graphical method to study the scaling of contraction for a minimal ensemble of core-only instances. We find subexponential scaling, improving on the exponential scaling that occurs without compression. Our results suggest that our tensor network algorithm subsumes the classical leaf removal algorithm and simplifies redundancies in the $p$-spin model through lossless compression, all without explicit knowledge of the problem's structure., Comment: 12 pages, 12 figures; Updated results; Corrected typos; Updated to the published version

We introduce fermionic machine learning (FermiML), a machine learning framework based on fermionic quantum computation. FermiML models are expressed in terms of parameterized matchgate circuits, a restricted class of quantum circuits that map exactly to systems of free Majorana fermions. The FermiML framework allows for building fermionic counterparts of any quantum machine learning (QML) model based on parameterized quantum circuits, including models that produce highly entangled quantum states. Importantly, matchgate circuits are efficiently simulable classically, thus rendering FermiML a flexible framework for utility benchmarks of QML methods on large real-world datasets. We initiate the exploration of FermiML by benchmarking it against unrestricted PQCs in the context of classification with random quantum kernels. Through experiments on standard datasets (Digits and Wisconsin Breast Cancer), we demonstrate that FermiML kernels are on-par with unrestricted PQC kernels in classification tasks using support-vector machines. Furthermore, we find that FermiML kernels outperform their unrestricted candidates on multi-class classification, including on datasets with several tens of relevant features. We thus show how FermiML enables us to explore regimes previously inaccessible to QML methods.

We introduce a quantum-inspired approximation algorithm for MaxCut based on low-depth Clifford circuits. We start by showing that the solution unitaries found by the adaptive quantum approximation optimization algorithm (ADAPT-QAOA) for the MaxCut problem on weighted fully connected graphs are (almost) Clifford circuits. Motivated by this observation, we devise an approximation algorithm for MaxCut, \emph{ADAPT-Clifford}, that searches through the Clifford manifold by combining a minimal set of generating elements of the Clifford group. Our algorithm finds an approximate solution of MaxCut on an $N$-vertex graph by building a depth $O(N)$ Clifford circuit. The algorithm has runtime complexity $O(N^2)$ and $O(N^3)$ for sparse and dense graphs, respectively, and space complexity $O(N^2)$, with improved solution quality achieved at the expense of more demanding runtimes. We implement ADAPT-Clifford and characterize its performance on graphs with positive and signed weights. The case of signed weights is illustrated with the paradigmatic Sherrington-Kirkpatrick model, for which our algorithm finds solutions with ground-state mean energy density corresponding to $\sim94\%$ of the Parisi value in the thermodynamic limit. The case of positive weights is investigated by comparing the cut found by ADAPT-Clifford with the cut found with the Goemans-Williamson (GW) algorithm. For both sparse and dense instances we provide copious evidence that, up to hundreds of nodes, ADAPT-Clifford finds cuts of lower energy than GW., Comment: Revised version, several typos corrected, extended appendices

The energy spectrum of topological semimetals contains protected degeneracies in reciprocal space that correspond to Weyl, Dirac, or multifold fermionic states. To exploit the unconventional properties of these states, one has to access the electronic structure of the three-dimensional bulk. In this work, we present the first joint theory-experiment study of the electronic structure of a candidate topological semimetal with resonant inelastic X-ray scattering (RIXS). We resolve the bulk electronic states of FeSi using momentum-dependent RIXS at the Fe $L_3$ edge. We observe a broad excitation continuum devoid of sharp features, consistent with particle-hole scattering in an underlying electronic band structure. Using density functional theory (DFT), we calculate the electronic structure of FeSi and derive a band theory formulation of RIXS in the fast collision approximation to model the scattering process with zero adjustable parameters. While band theory predicts an excitation continuum with broad spectral features similar to the observed ones, discrepancies between theory and experiment suggest the presence of low-energy processes that DFT alone does not account for. This first study of RIXS in a topological semimetal shows that RIXS is a useful tool for revealing unanticipated behavior of bulk electronic states in this class of materials., Comment: v2: published version

Coexistence of localized and extended excitations is central to the macroscopic properties of correlated materials. For 5d transition metal compounds, electron correlations alone generally do not lead to a metal-insulator (Mott) transition, with insulating behavior usually resulting from their coupling with magnetic ordering and/or structural distortions. 1$T$-TaS$_2$ is a prototypical example of such correlated insulating behavior, with a high-symmetry metallic phase transforming into a distorted, charge density wave (CDW) insulating state at low temperatures. The relevance of the localized electron physics at play in 3d compounds to these 5d transition metal compounds remains an open question. We resolved this standing controversy in 1$T$-TaS$_2$ combining resonant inelastic X-ray spectroscopy and first-principles calculations. We observed five electronic excitations arising from the interband transitions of the Ta 5d orbitals and the S 3p ligand state, with none of the excitations on the order of the Mott gap. These excitations cannot be explained within the framework of standard multiplet calculations that assume a localized wavefunction, but instead, are captured by a band theory framework accounting for the low symmetry of the crystal field in the CDW state. Our findings suggest that the electronic property of 1$T$-TaS$_2$ is dominated by both plasmonic quasiparticles and inter-band transitions associated with a Drude-type response, with no resonance associated with a putative Mott transition. Our discovery provides new insights into the electron localization and the onset of insulating behavior in 5d transition metal materials.

Many elusive quantum phenomena emerge from the interaction of a quantum system with its classical environment. Quantum simulators enable us to program this interaction by using measurement operations. Measurements generally remove part of the quantum entanglement built between the qubits inside a simulator. While in simple cases entanglement may disappear at a constant rate as we measure qubits one by one, the evolution of entanglement under measurements for a given class of quantum states is generally unknown. Here, we show that consecutive measurements of the qubits in a quantum simulator can lead to criticality, separating two phases of entanglement. We prepare an entangled superposition of ground states to a classical spin model and show that progressively measuring the qubits drives the simulator into a spin glass phase of entanglement. By entangling and measuring up to 48 qubits in this fashion, we determine the vitrification point and its critical exponent, which obey spin glass theory exactly. Our findings show that measurements alone can trigger entanglement criticality and suggest that coupling to a classical environment can drive critical phenomena in more general quantum states., Comment: Updated to reflect published version

We introduce a methodology for generating random multi-qubit stabilizer codes based on solving a constraint satisfaction problem (CSP) on random bipartite graphs. This framework allows us to enforce stabilizer commutation, $X/Z$ balancing, finite rate, sparsity, and maximum-degree constraints simultaneously in a CSP that we can then solve numerically. Using a state-of-the-art CSP solver, we obtain convincing evidence for the existence of a satisfiability threshold. Furthermore, the extent of the satisfiable phase increases with the number of qubits. In that phase, finding sparse codes becomes an easy problem. Moreover, we observe that the sparse codes found in the satisfiable phase practically achieve the channel capacity for erasure noise. Our results show that intermediate-size finite-rate sparse quantum codes are easy to find, while also demonstrating a flexible methodology for generating good codes with custom properties. We therefore establish a complete and customizable pipeline for random quantum code discovery.

We define an ensemble of random Clifford quantum circuits whose output state undergoes an entanglement phase transition between two volume-law phases as a function of measurement rate. Our setup maps exactly the output state to the ground space of a spin glass model. We identify the entanglement phases using an order parameter that is accessible on a quantum chip. We locate the transition point and evaluate a critical exponent, revealing spin glass criticality. Our work establishes an exact statistical mechanics theory of an entanglement phase transition., Comment: Updated to reflect published version

Solid-state drives (SSDs) are extensively used to deploy persistent data stores, as they provide low latency random access, high write throughput, high data density, and low cost. Tree-based data structures are widely used to build persistent data stores, and indeed they lie at the backbone of many of the data management systems used in production and research today. In this paper, we show that benchmarking a persistent tree-based data structure on an SSD is a complex process, which may easily incur subtle pitfalls that can lead to an inaccurate performance assessment. At a high-level, these pitfalls stem from the interaction of complex software running on complex hardware. On one hand, tree structures implement internal operations that have nontrivial effects on performance. On the other hand, SSDs employ firmware logic to deal with the idiosyncrasies of the underlying flash memory, which are well known to lead to complex performance dynamics. We identify seven benchmarking pitfalls using RocksDB and WiredTiger, two widespread implementations of an LSM-Tree and a B+Tree, respectively. We show that such pitfalls can lead to incorrect measurements of key performance indicators, hinder the reproducibility and the representativeness of the results, and lead to suboptimal deployments in production environments. We also provide guidelines on how to avoid these pitfalls to obtain more reliable performance measurements, and to perform more thorough and fair comparison among different design points.

Computational memory (CM) is a promising approach for accelerating inference on neural networks (NN) by using enhanced memories that, in addition to storing data, allow computations on them. One of the main challenges of this approach is defining a hardware/software interface that allows a compiler to map NN models for efficient execution on the underlying CM accelerator. This is a non-trivial task because efficiency dictates that the CM accelerator is explicitly programmed as a dataflow engine where the execution of the different NN layers form a pipeline. In this paper, we present our work towards a software stack for executing ML models on such a multi-core CM accelerator. We describe an architecture for the hardware and software, and focus on the problem of implementing the appropriate control logic so that data dependencies are respected. We propose a solution to the latter that is based on polyhedral compilation., Comment: Accepted at SPMA '20

With rapid improvements in NVM storage devices, the performance bottleneck is gradually shifting to the network, thus giving rise to the notion of "data movement wall". To reduce the amount of data movement over the network, researchers have proposed near-data computing by shipping operations and compute-extensions closer to storage devices. However, running arbitrary, user-provided extensions in a shared, disaggregated storage environment presents multiple challenges regarding safety, isolation, and performance. Instead of approaching this problem from scratch, in this work we make a case for leveraging the Linux kernel eBPF framework to program disaggregated NVM storage devices. eBPF offers a safe, verifiable, and high-performance way of executing untrusted, user-defined code in a shared runtime. In this paper, we describe our experiences building a first prototype that supports remote operations on storage using eBPF, discuss the limitations of our approach, and directions for addressing them.

Tensor networks represent the state-of-the-art in computational methods across many disciplines, including the classical simulation of quantum many-body systems and quantum circuits. Several applications of current interest give rise to tensor networks with irregular geometries. Finding the best possible contraction path for such networks is a central problem, with an exponential effect on computation time and memory footprint. In this work, we implement new randomized protocols that find very high quality contraction paths for arbitrary and large tensor networks. We test our methods on a variety of benchmarks, including the random quantum circuit instances recently implemented on Google quantum chips. We find that the paths obtained can be very close to optimal, and often many orders or magnitude better than the most established approaches. As different underlying geometries suit different methods, we also introduce a hyper-optimization approach, where both the method applied and its algorithmic parameters are tuned during the path finding. The increase in quality of contraction schemes found has significant practical implications for the simulation of quantum many-body systems and particularly for the benchmarking of new quantum chips. Concretely, we estimate a speed-up of over 10,000$\times$ compared to the original expectation for the classical simulation of the Sycamore `supremacy' circuits., Comment: 22 pages, 10 figures

We study the level-spacing statistics in the entanglement spectrum of output states of random universal quantum circuits where qubits are subject to a finite probability of projection to the computational basis at each time step. We encounter two phase transitions with increasing projection rate: The first is the volume-to-area law transition observed in quantum circuits with projective measurements; The second separates the pure Poisson level statistics phase at large projective measurement rates from a regime of residual level repulsion in the entanglement spectrum within the area-law phase, characterized by non-universal level spacing statistics that interpolates between the Wigner-Dyson and Poisson distributions. By applying a tensor network contraction algorithm introduced in Ref. [1] to the circuit spacetime, we identify this second projective-measurement-driven transition as a percolation transition of entangled bonds. The same behavior is observed in both circuits of random two-qubit unitaries and circuits of universal gate sets, including the set implemented by Google in its Sycamore circuits., Comment: 7 pages, 7 figures