Large-scale numerical optimization of bosonic quantum information processing
Title: Large-scale numerical optimization of bosonic quantum information processing
SNIC Project: SNIC 2022/22-998
Project Type: SNIC Small Compute
Principal Investigator: Patric Holmvall <patric.holmvall@physics.uu.se>
Affiliation: Uppsala universitet
Duration: 2022-10-18 – 2023-05-01
Classification: 10302
Keywords:

Abstract

The interconversion of quantum states is at the heart of quantum information processing. Restrictions given by the quantum systems employed lead to a hierarchy in the operations one can apply, because some are easier or less costly to apply than others. The consequence is that some operations and states are resources that are costly to obtain. For continuous variable or bosonic systems the resource of importance is most often non-Gaussianity, where Gaussian operations and states are considered easy to obtain or perform and thus free. The interconversions in this framework then are between resource state using free operations. Thus, finding the optimal conversion or a bound on the convertibility between resources is of utmost importance for any application involving quantum information. However, current studies in this framework are restricted to 1 to 1 deterministic conversions. We want to extend this by studying "N to M" conversions (N>M) and lift the requirement to be deterministic. A lot of interesting application such as distillation protocols and cascading protocols fall into this category. In detail we want to use a particle swarm optimizer (PSO), which is a corner-stone algorithm in machine-learning and AI research, to find the parameters that bring our input resource as close as possible to the desired states. The problem is highly parallelizable, something that we have exploited to the fullest in our code FidelityOptim. FidelityOptim is written in CUDA and C++ and runs on GPUs, with a hierarchy of parallelization. At the bottom level, GPU threads are used to parallelize the expensive calculation of the fidelity. The next level uses CUDA streams to parallelize over the particles in the PSO, the machine-learning part of the problem. On the top level, our code can also use multiple GPUs to form different swarms, another level to the machine learning. The code has been run successfully on SNIC clusters before (C3SE 2022/1-6 and its predecessors), to produce a publication in PRX Quantum (https://journals.aps.org/prxquantum/abstract/10.1103/PRXQuantum.2.010327) and more recently in Physical Review A (https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.062446). We are extending the FidelityOptim code with new functionality to conduct the research described above.