Algorithms like Quantum-Classical Auxiliary-Field Quantum Monte Carlo (QC-AFQMC) and Sample-Based Quantum Diagonalization (SQD) are driving a need for high-performance computing resources after quantum measurements are collected, a shift in how quantum and classical computers interact. These algorithms collect measurement samples from a quantum processing unit and then feed them into classical post-processing that can demand significant HPC resources. Benchen Huang, Sebastian Stern, and Tyler Takeshita explain that Amazon Braket now integrates with AWS Batch to orchestrate these scalable hybrid workflows, creating a compute pipeline for algorithms that decouple quantum and classical routines into sequential steps.
QC-AFQMC Algorithm Decomposes Quantum and Classical Stages
The Quantum-Classical Auxiliary-Field Quantum Monte Carlo (QC-AFQMC) algorithm relies on a distinct separation of computational tasks, initiating a sequence with quantum sampling before transitioning to extensive classical post-processing. This approach contrasts with other hybrid algorithms and necessitates high-performance computing (HPC) resources to manage the data generated after quantum measurements are completed. QC-AFQMC’s initial step involves using a quantum computer to prepare a trial state, designated as |Ψ T 〉, which then is the foundation for a Monte Carlo time evolution.
This evolution generates a population of “walkers,” each representing a sample in the simulation, and their subsequent evaluation, determining overlap and local energy relative to the trial state, is performed using classical shadows. “Evaluating a quantum-classical quantum Monte Carlo algorithm with Matchgate shadows,” details one area of ongoing exploration for the team, AWS says. The architecture supporting this workflow uses Amazon Braket, AWS Batch and AWS Lambda, orchestrated through AWS Step Functions to create a managed workflow.
This integration handles the asynchronous quantum step, the parallel classical processing of the walker data, and the final energy averaging, incorporating error handling and data transfer. The team demonstrated this approach with a prototypical QC-AFQMC implementation, showing how the system can manage the distinct phases of the algorithm. This decoupling allows for scalability, as the classical post-processing can be distributed across multiple compute nodes within AWS Batch.
The solution can be adapted to other algorithms, such as SQD, that share a similar workflow pattern. The availability of sample integration templates and examples in the Amazon Braket Examples repository further facilitates adoption and exploration of this hybrid approach for researchers.
Amazon Braket Hybrid Jobs Enable Quantum Sampling with Shadow Tomography
Amazon Braket streamlines hybrid quantum-classical workflows through a single API call, automatically allocating and releasing resources for both quantum processing units and classical computation. Collected shadow data from quantum sampling is then durably stored in Amazon S3 buckets, providing a centralized repository for analysis and future use. This integration allows users to package algorithm code and circuit designs into container images within Amazon ECR, deploy the workflow and retrieve results with a unified system.
AWS Step Functions orchestrates the entire process as a state machine, sequentially invoking Amazon Braket Hybrid Jobs for quantum sampling, AWS Batch for parallel classical post-processing, and AWS Lambda for energy averaging. This architecture ensures built-in sequencing, error handling, and smooth data transfer between each stage of the computation.
Monitoring a run is facilitated through the Step Functions console, which displays the status of each step with color-coded indicators, allowing users to quickly identify active stages and potential failures. For a recent demonstration, researchers estimated the ground-state energy of a hydrogen molecule using 4000 circuits, each executed with 100 shots on the Braket state-vector simulator SV1. These circuits were submitted as a single batch of parallel quantum tasks, highlighting the system’s ability to handle large-scale quantum computations.
AWS Batch Orchestrates Parallel Post-Processing of Monte Carlo Walkers
Algorithms using decoupled quantum and classical routines are increasingly reliant on substantial high-performance computing resources following quantum measurement, shifting the paradigm of quantum-classical interaction. This parallelization is inherent to the process, as each walker evolves independently, creating a natural fit for array-based workloads. The architecture detailed by the team maps each stage of the QC-AFQMC workflow to an appropriate AWS service, optimizing execution based on the task’s characteristics, according to the company.
Quantum sampling, involving the evaluation of a high volume of quantum circuits, is executed as an Amazon Braket Hybrid Job, which allocates both quantum processing units or circuit simulators and the classical resources needed to drive program execution. The classical post-processing phase, responsible for the parallel evolution of the Monte Carlo walkers, is run as an array job on AWS Batch, using its ability to distribute the workload efficiently.
Each Batch child job required 20 minutes to complete, while the initial quantum sampling step took 10 minutes. Once the Batch job finished, a Lambda function processed the data and returned the averaged ground-state energy in one minute, validating the pipeline’s end-to-end functionality by comparing results to classical AFQMC runs.
AWS Step Functions Automate Scalable QC-AFQMC Workflows
This decoupling represents a shift from custom scripting and manual polling for hybrid quantum-classical tasks, addressing challenges in operating and monitoring these complex processes. This automated sequencing, coupled with built-in error handling and data transfer, streamlines the execution of hybrid algorithms. The team’s implementation uses S3 for both input and output data, ensuring data durability throughout the workflow. This approach moves beyond the limitations of tying steps together with custom scripts, offering a scalable solution for hybrid quantum-classical computations.
The system’s design facilitates independent scaling of each computational stage, allowing resources to be allocated and released as needed. The team highlights the potential for broader application, noting that the architecture can accommodate variations in algorithm structure and resource requirements.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
