Quantum Chemistry with Qu&Co’s (now Pasqal) QUBEC on Amazon Braket

(Update: Pasqal and Qu&Co announced a merger on January 11, 2022.)

In this post, we discuss the progress and limitations of chemistry simulations on current quantum computers, and introduce Qu&Co‘s QUBEC, a quantum computational platform that is specifically designed for chemistry and materials science simulations. The post describes QUBEC’s architecture and how it integrates with Amazon Braket. Finally, we show how you can register for the QUBEC beta release program.

Introduction

Computational quantum chemistry has succeeded in supporting the development of novel drugs, new materials, and a better understanding of matter at the nanoscale. However, the complexity of configurations on real-world chemical systems makes it intractable for researchers to attain accurate results, so computational chemistry is used to extract in silico physical properties of molecules and solids.

The computational effort required to simulate systems accurately scales exponentially with the complexity of drug molecules and materials. Even using approximation methods, current supercomputers cannot achieve the level of accuracy that these simulations demand. Quantum computation (QC) has the potential to solve some of the most challenging computational problems faced in chemistry, allowing the scientific community to do chemical simulations that are intractable today. ^[1]

Current quantum hardware is still under active development, but improving at a very fast pace. ^[2] Even though today’s chemistry simulations based on QC are not at the accuracy levels of conventional supercomputers, it is essential that we start developing quantum-inspired computational tools. These will allow researchers to benefit from the power of QC once full-tolerant devices at scale are available.

Here at Qu&Co we are getting ready for that future. We are a European quantum software developer specialized in computational chemistry and multi-physics simulations. QUBEC, a fully managed cloud solution, enables quantum chemists and materials scientists to run chemical simulations using QC in just a few clicks. Built on Amazon Braket, QUBEC removes the heavy lifting from simulation execution, and combines it with an intuitive graphical interface for modeling molecular systems. It is the initial step towards integrating QC routines with the classical chemistry workflows already adopted by the industry.

The current state of chemistry simulations on quantum computers

Although quantum algorithms have been under active development for decades, only the recent development of variational approaches to quantum computing allowed for practical executions of small-scale experiments on gate-based quantum computing devices. The main challenge is quantum hardware noise. This largely arises from either qubit operations being different than expected (coherent errors) or the coupling with a noisy environment yielding dephasing and other errors (incoherent errors).

Like on conventional computers, “Error Correction” can prevent noise by encoding information redundantly in multiple bits or qubits, respectively. However, due to the fundamental laws of quantum mechanics, more qubit overhead is required to achieve error-protected computations. A recent proposal by the team at AWS Center for Quantum Computing, shows an architecture with potential to achieve quantum advantage at the order of tens to hundreds of thousands of physical components. In contrast, in the current ‘Noisy Intermediate Quantum’ or NISQ era, we only have access to tens of noisy physical qubits, and the challenge is to identify algorithms that can still make use of quantum operations to speed up computations.

A recent proposal, the Variational Quantum Eigensolver ^[3], showed a feasible route to use noisy hardware for near-term experiments. The proposal led to a surge in development of comparable variational algorithms in chemistry and beyond. In such algorithms, the quantum computer is used to prepare a complex wave function as a trial solution to a computational problem, such as finding the ground state of a chemical system. Observables on this wave function can be measured by repeatedly sampling (measuring) the qubits at the output of a parametrized quantum circuit. In the variational loop, the quantum circuit parameters are optimized with respect to an objective function until the desired wave function is prepared.  This combines the strength of the quantum processor in representing complex quantum states with the power of the classical processor to offset variational task. These algorithms are relatively robust against noise, because errors are partially mitigated by the variational aspect and sampled over many circuit evaluations. However, the computational scaling is expected to be worse than equivalent fault-tolerant quantum algorithms; the exact scaling of variational algorithms typically is hard to predict due to their heuristic nature.

To set expectations correctly, so far, no variational quantum algorithm has outperformed classical supercomputers in computational chemistry based on first principles (ab initio). However, recently quantum advantage has been demonstrated on a theoretical sampling task,^[4] and an increasing number of academic and industrial works are bringing down the resource requirements, devising better quantum circuit strategies and more efficient optimization protocols. The race is on, and many believe chemistry or materials science applications to be one of the candidates to show early examples of industry relevant quantum advantage on near-term hardware.

QUBEC: The quantum computational platform for chemistry and materials science

At Qu&Co, we developed QUBEC to offer our users a glimpse of what we believe the future of quantum chemistry simulations will look like. In that future, researchers and engineers will be able to apply quantum computational subroutines directly within their existing conventional workflows to benefit from the improved accuracy and scaling offered by future quantum processors. In an example workflow, one may compute a molecular geometry optimization first with mean-field self-consistent classical methods, and use that geometry as an input to an ab initio chemical simulation on a quantum backend. Then the results may be collected and analyzed using classical machine learning on high performance computing resources. In this way, workflows can make optimal use of the appropriate computing technologies where they make sense, all integrated in a seamless experience.

Furthermore, we know from experience that developing quantum computing solutions, and running quantum enhanced workloads is a highly specialized job.  Building such expertise in-house would not be cost-effective for most corporations, given the cost in terms of time and money for recruiting a dedicated team. Therefore, we believe quantum solutions should include a high-level of process automation. Our aim is to provide quantum computational chemistry to a corporate researcher in such a way that she can obtain good results from our solutions with little to no time overhead.

With these guiding principles in mind, Qu&Co built QUBEC as a fully managed platform running on AWS enabling users to execute chemistry simulations on current quantum computers, thanks to native integration with Amazon Braket. QUBEC automates the heavy-lifting necessary to run quantum computational tasks from automatic provisioning of the computing infrastructure to running pre- and post-processing classical calculations and performing error mitigation tasks. QUBEC needs a minimal amount of configuration and pre-existing quantum-computational know-how to achieve state-of-the-art results on current day processors. Also, QUBEC is interfaced with Maestro, a fully-featured solution for modelling chemical systems developed by Schrödinger for the conventional chemistry simulations market. The integration with such a graphical user interface (GUI) allows chemists and material scientists to start experimenting with quantum computational chemistry from within a familiar environment, benefiting the adoption rate of quantum (see Figure 1).

Figure 1 – Screenshot of QUBEC quantum computational platform for chemistry and materials-science, in this example accessed from within Schrödinger’s Maestro chemical modeling graphical user interface.

QUBEC currently offers two types of quantum computational backends. To calculate properties of smaller chemical systems, users can use the quantum processors offered by Amazon Braket and local simulators managed by QUBEC. QUBEC also includes an automated quantum resource estimator, Q-time, which returns the quantum resources (runtime and number of physical and logical qubits) needed to perform calculations on a future fault-tolerant quantum processor. This is a very important value proposition of QUBEC because it allows customers to better understand, from today, the requirements of the future quantum technologies needed to solve their most challenging industry-sized problems.

Given the fact that many potential end-users of QUBEC are relatively unfamiliar with quantum computations for chemistry, we conduct an extensive virtual onboarding session with all clients to help them get started. The purpose of the onboarding is to both illustrate the current product features and limitations and give a hands-on tutorial on QUBEC. After the virtual onboarding, customers can access the QUBEC graphical interface integrated with the latest version of Schrödinger’s Maestro platform. Users can also interact directly with QUBEC using the programmatic API written in Python, thus easily integrating into existing notebooks and workflows.

In Figure 2 we show an example usage of this API for extracting ground state energy of the LiH molecule. These are the steps required:

• Authenticate to the QUBEC platform using Qu&Co account’s username and password

# login to QUBEC platform
_ = enable_account("username", password="password")

• Define the chemical system giving atomic positions and basis set of the molecular wave function

# launch the job on QUBEC platform with UCCSD circuit
# using the Amazon Braket state vector simulator
experiment_input = {
    "problem": {
        "geometry": [
            ("Li", (-0.77000, -0.00520, 0.00000)),
            ("H",  (0.33000,   0.00220, 0.00000)),
        ],
        "basis_set": "3-21G"
    },
    "experiment": "vqa",
    "backend_type": "simulator",
    "provider": "braket",
    "n_shots": 25_000,
    "simulation_type": "tomography",
    "quantum_parameters": {
        "variational_proc": True,
        "initial_angles": "cc",
        "circuit_ansatz": "uccsd"
    }
}

• Instruct QUBEC to run the simulations on the Amazon Braket state vector simulator and wait for the simulation results

job = execute(**experiment_input)
job.wait_completion()
result = job.get_result()

• Print out the resulting ground state energy and its evolution at each iteration of the variational loop

# print the resulting ground state energy and compare with Hartree-Fock
print(f"Variational quantum algorithm ground state energy: {result.optimal_value} Ha\n----")
print(f"Hartree-Fock ground state energy: {result.hf_energy} Ha\n----")

# variationally optimized circuit parameters
print(f"Optimal circuit parameters: {result.optimal_parameters}")

# plot the evolution of the ground state energy at each variational iteration
result.plot()

Figure 2 – QUBEC output showcasing a typical variational quantum computational chemistry simulation.

QUBEC architecture

Under the hood, QUBEC uses a service-oriented architecture where each service deals with a single step in the multi-step workflow composing a quantum computing simulation such as input validation, pre-processing calculations with conventional quantum chemistry methods, and post-processing of the results obtained on Amazon Braket. QUBEC core computational services run onmazon Elastic Container Service (ECS). Clients can interact with the QUBEC via a purpose-built REST API. This REST API can be either accessed programmatically using a Python interface provided by Qu&Co or with the QUBEC GUI described above.

The QUBEC engine constructs the proprietary Qu&Co quantum circuits and executes the hybrid quantum-classical optimization loop customary for current variational quantum algorithms. Quantum computing tasks can be either executed on local simulators running within QUBEC or sent to Amazon Braket. To guarantee a smooth onboarding of new users, QUBEC offers access to Amazon Braket from an AWS account fully managed by Qu&Co. Alternatively, QUBEC can also leverage AWS Identity and Access Management (IAM) roles to launch Amazon Braket tasks on behalf of the user on their AWS account, allowing existing customers to seamlessly integrate QUBEC into their AWS environment. In the first case, the Amazon Braket data resides in a Qu&Co owned Amazon S3 bucket which can be accessed by the user via the QUBEC programmatic API. In the second case, the S3 bucket where the results from Amazon Braket tasks are stored can be accessed directly from the user’s AWS account. Figure 3 shows the high-level QUBEC architecture on AWS when Amazon Braket tasks are executed on behalf of the user on their account.

Figure 3 – QUBEC high-level architecture on AWS. This figure shows the QUBEC operating mode in which Amazon Braket tasks are executed on behalf of the user within their own AWS account.

QUBEC is tightly integrated with Amazon Braket; it leverages state vector (SV1), tensor network (TN1), and density matrix (DM1) simulators offered to efficiently execute large-scale variational quantum simulations. The high performance of these simulators allows QUBEC to carry out chemistry simulations on system sizes far beyond what is feasible with local simulators. Furthermore, QUBEC leverages the flexibility of Amazon Braket SDK to determine optimal qubit allocation to support the error mitigation routines implemented in QUBEC.

Conclusion

In this post, we briefly introduced concepts in quantum computing for chemistry, what the future of quantum computing promises for quantum chemistry simulations, and what are the limitations of quantum computational approaches today.

We also showed how the Qu&Co QUBEC platform, integrated with Amazon Braket, can offer a glimpse of what the future of quantum chemistry will look like. In that future, corporate researchers will be able to apply quantum computational subroutines directly in their existing conventional computational workflows, so that they can benefit from the improved accuracy and scaling offered by future quantum processors.

To request access to the beta release of QUBEC, register at www.quandco.com/product. For questions about our quantum-algorithm research activities, please contact us through www.quandco.com/contact; we are happy to talk.

The content and opinions in this post are those of the third-party author and AWS is not responsible for the content or accuracy of this post.

References

[1] Yudong Cao, Jonathan Romero, Jonathan P. Olson, Matthias Degroote, Peter D. Johnson, Mária Kieferová, Ian D. Kivlichan, Tim Menke, Borja Peropadre, Nicolas P. D. Sawaya, Sukin Sim, Libor Veis, and Alán Aspuru-Guzik, Quantum Chemistry in the Age of Quantum Computing, Chem. Rev. 2019, 119, 19, 10856–10915

[2] Christopher Chamberland, Kyungjoo Noh, Patricio Arrangoiz-Arriola, Earl T. Campbell, Connor T. Hann, Joseph Iverson, Harald Putterman, Thomas C. Bohdanowicz, Steven T. Flammia, Andrew Keller, Gil Refael, John Preskill, Liang Jiang, Amir H. Safavi-Naeini, Oskar Painter, Fernando G.S.L. Brandão,  Building a fault-tolerant quantum computer using concatenated cat codes, arxiv.org [2012.04108] (2020)

[3] Alberto Peruzzo, Jarrod McClean, Peter Shadbolt, Man-Hong Yung, Xiao-Qi Zhou, Peter J. Love, Alán Aspuru-Guzik & Jeremy L. O’Brien. A variational eigenvalue solver on a photonic quantum processor, Nat Commun 5, 4213 (2014).

[4] Arute, F. et al. Quantum supremacy using a programmable superconducting processor. Nature 574 (2019).