The Advanced Simulations of Quantum Computations is a 2-day workshop on October 17-18, 2021 from 10:45 am – 4:45 pm MDT.
Abstract: As quantum computing hardware is steadily evolving towards the quantum advantage regime, classical simulation of quantum circuits is becoming more and more challenging, yet crucial for the verification and validation of the new hardware and algorithms. In recent years, we observed fast progress in new advanced techniques enabling more efficient simulations of rather large quantum circuits. Importantly, these techniques and algorithms were also able to take better advantage of modern classical high performance computing platforms based on the heterogeneous accelerated node architectures.
This workshop will bring together participants from the national labs, academia, industry and open-source software community to share recent results in algorithms and software for large-scale quantum circuit simulations across a broad range of methods, covering state-vector, tensor network, graphical model, stabilizer, and pulse-level simulations. We seek to provide an open platform for state-of-the-art development efforts, exchanging ideas and best practices, and fostering research collaboration in an attempt to stimulate the formation of an inclusive research community focused around the this topic.
Target Audience: The workshop attendees are expected to be quantum simulator developers and/or users from the US and other countries. The level of experience of simulators will range from novice to experienced developers.
To cope with the complexity of simulating quantum computations, several approaches are currently investigated---ranging from straight-forward realizations of the corresponding matrix-vector multiplications through arrays to dedicated HPC-solutions exploiting massive hardware power. In this talk, an alternative is presented utilizes decision diagrams. These provide a compact representation of quantum states and operations for many (practically-relevant) instances and allow for an efficient simulation. The talk will review the underlying ideas of decision diagrams and how to put them into use for corresponding simulation schemes. This will be followed by a discussion on the performance of the resulting solutions (featuring both, the strength but also the weaknesses of decision diagram-based simulation) as well as a brief overview of corresponding open-source implementations.
Faithful simulations of quantum processors are essential for computer-aided quantum computer design. In this talk, we will demonstrate that the simulation of superconducting qubits can be made differentiable with respect to both the design and control parameters. In addition, we can evaluate the gradient of typical design objectives in a single reverse computation. This leads to a speedup proportional to the number of parameters in the objective function against the finite difference method. We can thus utilize the gradients to optimize design and control parameters jointly and efficiently, extending the scope of quantum optimal control. For example, we can use this approach to design chips for more robust gate schemes or specific purposes like a near-term quantum application or a quantum error correction scheme.
In this talk, I will introduce the concept of quantum circuit simulators, why they are needed, the most recent developments, and our work on parallel quantum simulators designed to run on large supercomputers with the eventual goal to run at scale on exa-scale supercomputer Aurora and Polaris. Our most recent simulator QTEnsor is based on the tensor network representation of quantum circuits. I will compare the performance of QTEnsor against other simulators and discuss future developments.
12:15 pm - 12:30 am
Classical Simulations of Quantum Computing Devices are crucial tools for developing and understanding performance aspects of algorithms and quantum devices themselves. A very large number of such software tools have appeared in recent years involving a variety of techniques (e.g. tensor networks, solving the Master equation) to solve particular problems (e.g. optimal control, open quantum systems, gate based algorithms) involving several computing languages (e.g. C++, Python, Julia). Given the rapid growth of these simulation ecosystems, it is prudent to consider evaluating these systems with an eye towards capability and usability and how well they match the needs of the community. Are there missing capabilities? Are there usability concerns that, if resolved, could expand the reach of the software? In this talk, we will propose how to do such an evaluation and steps the community could take with results.
Developing state-of-the-art classical simulators of quantum circuits is of utmost importance to test and evaluate early quantum technology and understand the true potential of full-blown error-corrected quantum computers. To support a unified and optimized use of multiple techniques across platforms, we developed HybridQ, a highly extensible platform designed to provide a common framework to integrate multiple state-of-the-art techniques to run on a variety of hardware. The powerful tools developed in HybridQ allow users to manipulate, develop, and extend noiseless and noisy circuits for different hardware architectures. HybridQ supports large-scale high-performance computing (HPC) simulations, automatically balancing workload among different processor nodes and enabling the use of multiple backends to maximize parallel efficiency. Everything is then glued together by a simple and expressive language that allows seamless switching from one technique to another as well as from one hardware to the next, without the need to write lengthy translations, thus greatly simplifying the development of new hybrid algorithms and techniques.
This talk will be divided into two parts. First, I will walk through several examples that illustrate the power and ease of use of Cirq as a python framework for programming quantum computers. While Cirq can run computations on real quantum computers, it acts as an interface that is agnostic to its backend. I will show how it can be used to simulate medium scale quantum circuits, thanks to the high-performance simulator qsim. In the second part of the talk, I will give an update on the newest methods we have developed at Google for the simulation of relatively large scale quantum computations over instances that were until now considered infeasible to simulate.
Despite fascinating developments in NISQ based quantum computing recently, simulations of quantum programs in classical HPC systems are still essential in validating quantum algorithms, understanding the noise effect, and designing robust quantum algorithms. To allow efficient large-scale noise-enabled simulation on state-of-the-art heterogeneous supercomputers, we developed NWQSim, a quantum circuit simulation environment that provides support for frontends such as Q#, Qiskit, Cirq, OpenQASM, etc., and backends such as X86/Power CPUs, NVIDIA/AMD GPUs, and Xeon-Phis, through state-vector and density matrix. NWQSim can scale out to more than a thousand GPUs/CPUs on ORNL Summit and has been tested on ORNL Spock, ANL Theta and NERSC Cori, achieving 10x over existing approaches. In this talk, I will describe the various techniques that enable high-performance, scalability and portability. Our work has been accepted at SC-20 and SC-21. NWQSim is supported by the U.S. DOE Quantum Science Center (QSC).
In this talk, we present examples of the challenges of realistic device simulations, going beyond standard Markovian noise models. We start with the triple-dot decoherence-free-subspace qubit, where one qubit is comprised of three quantum dots. In this system, only native ‘gate’, exchange between neighboring quantum dots, is used to compile all qubit operations. Noise in these systems is typically highly non-Markovian, comprising both leakage and 1/f qualities. Even with these challenges, accurate simulations can be performed and compare favorably with experiment. Time permitting, we will also discuss pulse design in neutral atom qubits using realistic device simulations.
To be added later
Recent years have seen the emergence of a large variety of Noisy, Intermediate Scale Quantum (NISQ) processors with a few tens of noisy qubits. In this presentation, I will show that the ability to numerically simulate the noisy behavior of such processors – whether digital or analog -- with high-performance classical simulations is crucial in order to design, optimize and test quantum algorithms for these chips, and hence identify potential applications.
We outline key optimizations of the open source vm6502q/qrack (“Qrack”) quantum computer simulator framework, so that these techniques and design concepts may be adapted to other software in the domain. We include a brief overview of validation and performance metrics for the framework. We define a general rule for open-ended optimizing gate replacement in any “ket” based simulation: the norm of the projection of an optimized test case state upon a control case state must be 1, therefore preserving all Hermitian operator expectation values. Oriented by a diagrammatic map of the composable “layers” of Qrack, we discuss CPU/GPU simulation switching thresholds, “paging” simulations as across segmented GPU maximum allocation boundaries for higher qubit widths, extension of Aaronson’s stabilizer tableau algorithm by buffered “gate fusion” of universal gates with recourse to ket simulation, and the “Qrack::QUnit” layer’s novel ket and stabilizer based Schmidt decomposition and controlled gate buffer commutation techniques—all operating together transparently in the default optimal “layer stack” of Qrack.
Simulators are an essential component of quantum computing frameworks in which they play the role of backends complementary to quantum hardware. Here we discuss two aspects of the interaction between simulators and the rest of the quantum computing system. First, we demonstrate the advantage of co-developing simulators and compilers by proposing a specialized compiler pass to reduce the simulation time for arbitrary circuits. Second, we increase the realism of the simulations by describing gates not as unitary matrices, but as general quantum channels. In this way, device characterization like process tomography can be readily used in simulation. In both cases, we present concrete implementations within the Intel Quantum Simulator, an open-source, high-performance, state-vector-based simulator.
Quantum circuit simulation on classical computers, especially large-scale HPC systems, is useful for characterizing, benchmarking, and validating near-term quantum hardware. In this talk, I will present the tensor network quantum virtual machine (TNQVM), which is a configurable tensor network-based simulator supporting both exact and approximate quantum circuit simulation. I will describe the various tensor network representations enabling the flexibility and scalability of the simulator. I will also report performance benchmarks of TNQVM on GPU-accelerated platforms, including the Summit supercomputer.
Stabilizer states are a rich class of quantum states that can be efficiently represented and manipulated on classical computers. This feature makes stabilizer states a useful basis for simulating quantum computations that do not deviate too far from a sequence of Clifford operations, for instance noisy quantum error correction circuits, noise-free circuits with few T gates, and some low-depth variational circuits. In this talk I will review several approaches to stabilizer-based simulation and some of the tradeoffs involved. I will present a hybrid method, combining several previously known techniques, for efficiently simulating near-Clifford circuits with near-Clifford noise. Finally, I will present results showing the application of this method to variational quantum computations.
Quantum Circuit Simulators (QCS) play a key role in the development of algorithms, potential applications, and research into more capable quantum processors. It is therefore important for QCS to offer the ability to execute larger quantum volumes in a reasonable amount of time on classical computers. Today, the most commonly used simulation methods are state vector and tensor network methods, which are both computationally intensive and benefit greatly from GPU acceleration. This is why we are introducing the NVIDIA cuQuantum SDK that will provide APIs to accelerate QCS that rely on both methods. In this talk, we will look at the early capabilities of cuQuantum libraries and demonstrate the performance impact on some benchmark simulations.
We look at some finer details of automatically extracting performance from large scale quantum circuit tensor network contractions, especially on the GPU. We also consider the possibilities of augmenting the exact contraction framework with compression to yield general approximate contraction strategies for many problems including quantum circuits.
Quantum computing is currently generating great excitement and receiving large investments by major tech companies. One recent experiment by a group at Google specifically performed a task that would definitely be impossible for a regular, classical computer. But when taking into account the analog nature of Google's device, which incurs a small error at every step, is the real difficulty of the task unchanged?
Quantum computers promise to open new opportunities in the simulation of quantum systems and machine learning. Superconducting qubit quantum computers are currently one o the leading platforms to perform, but their performance is significantly limited by the quality of the two qubit quantum gates. In this talk we use learning control to create high fidelity cross resonance gates using the software package Boulder Opal. We present the entire learning control development cycle. We start by using Boulder Opal to create a realistic pulse level simulation of a quantum compute with cross resonance gates. The underlying simulations are implemented in Tensorflow and are accelerated with GPUs. A series of reinforcement learning algorithms are then tested and compared using this simulation tool. The best reinforcement learning algorithm is then picked and implemented on IBM-Q. IBM-Q provides access to superconducting quantum computers that can be programmed at the individual pulse level using the OpenPulse standard. We demonstrate that reinforcement learning is able to produce cross resonance gates with a 40% less errors than the IBM-Q defaults.