Leveraging Quantum
The third pillar of the CLARA Testbed Infrastructure consists of quantum acceleration for the CLARA Testbed. This will include leveraging the LUMI-Q quantum computer VLQ and the LRZ quantum computer Euro-Q-Exa for research in molecular modelling and machine learning acceleration.
This task is focused on the design, implementation, and evaluation of an experimental, hybrid quantum-classical solution aiming to exploit Quantum Processing Units as accelerators of computation, esp. for computationally complex tasks of protein-ligand docking, molecular dynamics, and related ML problems.
We aim to design the whole workflow to be ready for the advanced NISQ era, providing a quantum advantage shortly.
The goals
of this task are:
toolkits
Installation of the necessary SW toolkits
Installation of the necessary SW toolkits, notably Qiskit as mid-level API enabling direct work with quantum circuits and Pennylane for higher-level implementation, mostly of ML parts.
Implementation of VQE-based approach
Implementation of VQE-based approach for obtaining molecular properties like potential energy surfaces, electronic couplings, etc.
models
Implementation of necessary ML models
Implementation of necessary ML models utilizing QPUs both for their training and subsequent evaluation. The models will be predominantly used for the representation of the above mentioned molecular properties, i.e. the PESs and their gradients to obtain a continuous, smooth representation of the discrete data, aiming to work around the runtime overhead of the on-the-fly computational approach.
Application of the implemented software tools:
- in a state vector simulation regime, i.e. without any sampling noise or decoherence present. The purpose of this step is to demonstrate that the whole method is working without any inherent QC-related problems;
- in a simulation with a sampling noise without any decoherence channel. With the addition of the sampling noise, the simulation will change its character significantly, which may require a convenient optimisation strategy;
- in a simulation with a sampling noise and a set of decoherence channels (depolarising and dephasing ones) to gain further information about the behaviour of our models in the simulated environment;
- in a simulation with full “fake backend”, getting very close to a run on a real hybrid quantum-classical infrastructure;
- on the real hybrid quantum-classical computational infrastructure.