arXiv subjects

Adam G. M. Lewis

Publications and source records attributed to Adam G. M. Lewis.

12 recordsLinked to original sources

arXiv2022

Large scale quantum chemistry with Tensor Processing Units

We demonstrate the use of Google's cloud-based Tensor Processing Units (TPUs) to accelerate and scale up conventional (cubic-scaling) density functional theory (DFT) calculations. Utilizing 512 TPU cores, we accomplish the largest such DFT computation to date, with 247848 orbitals, corresponding to a cluster of 10327 water molecules with 103270 electrons, all treated explicitly. Our work thus paves the way towards accessible and systematic use of conventional DFT, free of any system-specific constraints, at unprecedented scales.

physics.comp-ph↗

arXiv2022

Density Matrix Renormalization Group with Tensor Processing Units

Google's Tensor Processing Units (TPUs) are integrated circuits specifically built to accelerate and scale up machine learning workloads. They can perform fast distributed matrix multiplications and therefore be repurposed for other computationally intensive tasks. In this work we demonstrate the use of TPUs for accelerating and scaling up the density matrix renormalization group (DMRG), a powerful numerical approach to compute the ground state of a local quantum many-body Hamiltonian. The cost of DMRG scales with system size $N$ as $O(ND^3)$, where the so-called bond dimension $D$ regulates how expressive the underlying matrix product state (MPS) variational ansatz is. We consider lattice models in two spatial dimensions, with square lattices of size $10\times 10$ (free fermions) and $20\times 20$ (transverse field Ising model), for which the required MPS bond dimension is known to scale at least as $\exp(\sqrt{N})$. Using half of a TPU v3 pod (namely $1,\!024$ TPU v3 cores) we reached an unprecedentedly large bond dimension $D = 2^{16} = 65,\!536$, for which optimizing a single MPS tensor took about 2 minutes.

cond-mat.str-el↗

arXiv2022

Dynamics of Transmon Ionization

Qubit measurement and control in circuit QED rely on microwave drives, with higher drive amplitudes ideally leading to faster processes. However, degradation in qubit coherence time and readout fidelity has been observed even under moderate drive amplitudes corresponding to few photons populating the measurement resonator. Here, we numerically explore the dynamics of a driven transmon-resonator system under strong and nearly resonant measurement drives, and find clear signatures of transmon ionization where the qubit escapes out of its cosine potential. Using a semiclassical model, we interpret this ionization as resulting from resonances occurring at specific resonator photon populations. We find that the photon populations at which these spurious transitions occur are strongly parameter dependent and that they can occur at low resonator photon population, something which may explain the experimentally observed degradation in measurement fidelity.

quant-ph↗

arXiv2022

Simulation of quantum many-body dynamics with Tensor Processing Units: Floquet prethermalization

Tensor Processing Units (TPUs) are specialized hardware accelerators developed by Google to support large-scale machine-learning tasks, but they can also be leveraged to accelerate and scale other linear-algebra-intensive computations. In this paper we demonstrate the usage of TPUs for massively parallel, classical simulations of quantum many-body dynamics on long timescales. We apply our methods to study the phenomenon of Floquet prethermalization, i.e., exponentially slow heating in quantum spin chains subject to high-frequency periodic driving. We simulate the dynamics of L=34 qubits for over $10^5$ Floquet periods, corresponding to circuits with millions of two-qubit gates. The circuits simulated have no additional symmetries and represent a pure-state evolution in the full $2^L$-dimensional Hilbert space. This is achieved by distributing the computation over 128 TPU cores. On that size TPU cluster, we find speedups in wall-clock runtime of 230x and 15x when compared to reference CPU and single-GPU simulations, respectively, for shorter 30-qubit simulations that can be handled by all three platforms. We study the computational cost of the simulations, as a function of both the number of qubits and the number of TPU cores used, up to our maximum capacity of L=40 qubits, which requires a ``full pod" of 2048 TPU cores with tens of terabytes of memory in total. For these simulations, an 8-TPU-core machine is comparable to a single A100 GPU, and thus the full TPU pod is comparable to a machine with hundreds of GPUs. However, the TPU pod is more energy and cost efficient, and readily accessible (via Google Cloud), unlike such large many-GPU configurations. We also study the accumulation of numerical error as a function of circuit depth in very deep circuits. Our work demonstrates that TPUs can offer significant advantages for state-of-the-art simulations of quantum many-body dynamics.

quant-ph↗

arXiv2021

Large Scale Distributed Linear Algebra With Tensor Processing Units

We have repurposed Google Tensor Processing Units (TPUs), application-specific chips developed for machine learning, into large-scale dense linear algebra supercomputers. The TPUs' fast inter-core interconnects (ICI)s, physically two-dimensional network topology, and high-bandwidth memory (HBM) permit distributed matrix multiplication algorithms to rapidly become computationally bound. In this regime, the matrix-multiply units (MXU)s dominate the runtime, yielding impressive scaling, performance, and raw size: operating in float32 precision, a full 2048-core pod of third generation TPUs can multiply two matrices with linear size $N= 220= 1 048 576$ in about 2 minutes. Via curated algorithms emphasizing large, single-core matrix multiplications, other tasks in dense linear algebra can similarly scale. As examples, we present (i) QR decomposition; (ii) resolution of linear systems; and (iii) the computation of matrix functions by polynomial iteration, demonstrated by the matrix polar factorization.

physics.comp-ph↗

arXiv2020

Classical Simulations of Quantum Field Theory in Curved Spacetime I: Fermionic Hawking-Hartle Vacua from a Staggered Lattice Scheme

We numerically compute renormalized expectation values of quadratic operators in a quantum field theory (QFT) of free Dirac fermions in curved two-dimensional (Lorentzian) spacetime. First, we use a staggered-fermion discretization to generate a sequence of lattice theories yielding the desired QFT in the continuum limit. Numerically-computed lattice correlators are then used to approximate, through extrapolation, those in the continuum. Finally, we use so-called point-splitting regularization and Hadamard renormalization to remove divergences, and thus obtain finite, renormalized expectation values of quadratic operators in the continuum. As illustrative applications, we show how to recover the Unruh effect in flat spacetime and how to compute renormalized expectation values in the Hawking-Hartle vacuum of a Schwarzschild black hole and in the Bunch-Davies vacuum of an expanding universe described by de Sitter spacetime. Although here we address a non-interacting QFT using free fermion techniques, the framework described in this paper lays the groundwork for a series of subsequent studies involving simulation of interacting QFTs in curved spacetime by tensor network techniques.

gr-qc↗

arXiv2020

Hadamard Renormalization of a 2-Dimensional Dirac Field

The Hadamard renormalization procedure is applied to a free, massive Dirac field $ψ$ on a 2 dimensional Lorentzian spacetime. This yields the state-independent divergent terms in the Hadamard bispinor $G^{(1)}(x, x') = \frac{1}{2} \left\langle \left[ \barψ(x'), ψ(x) \right] \right\rangle$ as $x$ and $x'$ are brought together along the unique geodesic connecting them. Subtracting these divergent terms within the limit assigns $G^{(1)}(x, x')$, and thus any operator expressed in terms of it, a finite value at the coincident point $x' = x$. In this limit, one obtains a quadratic operator instead of a bispinor. The procedure is thus used to assign finite values to various quadratic operators, including the stress-energy tensor. Results are presented covariantly, in a conformally-flat coordinate chart at purely spatial separations, and in the Minkowski metric. These terms can be directly subtracted from combinations of $G^{(1)}(x, x')$ - themselves obtained, for example, from a numerical simulation - to obtain finite expectation values defined in the continuum.

gr-qc↗

arXiv2019

Quantum Computation with Machine-Learning-Controlled Quantum Stuff

We describe how one may go about performing quantum computation with arbitrary "quantum stuff", as long as it has some basic physical properties. Imagine a long strip of stuff, equipped with regularly spaced wires to provide input settings and to read off outcomes. After showing how the corresponding map from settings to outcomes can be construed as a quantum circuit, we provide a machine learning algorithm to tomographically "learn" which settings implement the members of a universal gate set. At optimum, arbitrary quantum gates, and thus arbitrary quantum programs, can be implemented using the stuff.

quant-ph↗

arXiv2018

Automatic generation of CUDA code performing tensor manipulations using C++ expression templates

We present a C++ library, TLoops, which uses a hierarchy of expression templates to represent operations upon tensorial quantities in single lines of C++ code that resemble analytic equations. These expressions may be run as-is, but may also be used to emit equivalent low-level C or CUDA code, which either performs the operations more quickly on the CPU, or allows them to be rapidly ported to run on NVIDIA GPUs. We detail the expression template and C++-class hierarchy that represents the expressions and which makes automatic code-generation possible. We then present benchmarks of the expression-template code, the automatically generated C code, and the automatically generated CUDA code running on several generations of NVIDIA GPU.

cs.MS↗

arXiv2018

GPU-Accelerated Simulations of Isolated Black Holes

We present a port of the numerical relativity code SpEC which is capable of running on NVIDIA GPUs. Since this code must be maintained in parallel with SpEC itself, a primary design consideration is to perform as few explicit code changes as possible. We therefore rely on a hierarchy of automated porting strategies. At the highest level we use TLoops, a C++ library of our design, to automatically emit CUDA code equivalent to tensorial expressions written into C++ source using a syntax similar to analytic calculation. Next, we trace out and cache explicit matrix representations of the numerous linear transformations in the SpEC code, which allows these to be performed on the GPU using pre-existing matrix-multiplication libraries. We port the few remaining important modules by hand. In this paper we detail the specifics of our port, and present benchmarks of it simulating isolated black hole spacetimes on several generations of NVIDIA GPU.

gr-qc↗

arXiv2017

Fundamental frequencies and resonances from eccentric and precessing binary black hole inspirals

Binary black holes which are both eccentric and undergo precession remain unexplored in numerical simulations. We present simulations of such systems which cover about 50 orbits at comparatively high mass ratios 5 and 7. The configurations correspond to the generic motion of a nonspinning body in a Kerr spacetime, and are chosen to study the transition from finite mass-ratio inspirals to point particle motion in Kerr. We develop techniques to extract analogs of the three fundamental frequencies of Kerr geodesics, compare our frequencies to those of Kerr, and show that the differences are consistent with self-force corrections entering at first order in mass ratio. This analysis also locates orbital resonances where the ratios of our frequencies take rational values. At the considered mass ratios, the binaries pass through resonances in one to two resonant cycles, and we find no discernible effects on the orbital evolution. We also compute the decay of eccentricity during the inspiral and find good agreement with the leading order post-Newtonian prediction.

gr-qc↗

arXiv2016

Redshift factor and the first law of binary black hole mechanics in numerical simulations

The redshift factor $z$ is an invariant quantity of fundamental interest in post-Newtonian and self-force descriptions of compact binaries. It connects different approximation schemes, and plays a central role in the first law of binary black hole mechanics, which links local quantities to asymptotic measures of energy and angular momentum in these systems. Through this law, the redshift factor is conjectured to have a close relation to the surface gravity of the event horizons of black holes in circular orbits. We propose and implement a novel method for extracting the redshift factor on apparent horizons in numerical simulations of quasicircular binary inspirals. Our results confirm the conjectured relationship between $z$ and the surface gravity of the holes and that the first law holds to a remarkable degree for binary inspirals. The redshift factor enables tests of analytic predictions for $z$ in spacetimes where the binary is only approximately circular, giving a new connection between analytic approximations and numerical simulations.

gr-qc↗