David Weir and Kari Rummukainen
David Weir
Tensor networks with the Hila framework – hila-TV (‘hila-tensoriverkko’)
The Hila framework [1] is a toolkit for performing very large-scale lattice field theory simulations.
Exploiting modern accelerated computing paradigms including parallelised GPU programming, it
supports lattices in arbitrary numbers of dimensions and performs well where the state at each site
in a configuration is represented by a large-dimensional matrix, such as in SU(N) lattice gauge
theories. Applications so far have been to Monte Carlo and real time simulations, but tensor
network states represent another fruitful area of study.
You will write a tensor network state layer for Hila that allows systems to be represented as matrix
product states and studied variationally using techniques such as the Density Matrix
Renormalisation Group (DMRG) [2]. Other GPU-accelerated frameworks for studying tensor
networks exist, such as iTensorGPU [3]; these will be used to validate the framework. We expect
that the multi-GPU parallelism made possible by Hila will provide a further speedup.
Given the focus of our group’s research we anticipate first investigating strongly coupled gauge
theories on the lattice [4,5]. As a starting point, you will study one-dimensional lattice QCD at finite
density [6], investigating the effect of the truncation of the gauge field state expansion on the
results, before going on to investigate the feasibility of simulating systems in higher spacetime
dimensions [7], or larger gauge groups. Your work will also enable large-scale numerical studies of
entanglement entropy in higher numbers of spatial dimensions.
The project will strengthen collaborative links between quantum information theorists and
theoretical particle physicists in Helsinki and beyond. It leverages Finland’s leadership in high
performance computing infrastructure at CSC, and represents a further step towards simulating
realistic lattice gauge theories on quantum computers.
Anticipated results:
References:
Jukka K Nurminen
Graph-Theoretic Compiler for Quantum Computers
With the recent advances in quantum computing comes a need for efficient compilers that
translate quantum software to the hardware. If quantum software is not efficiently compiled
to the target quantum hardware, quantum advantage may be unobtainable due to poor
communication within the quantum stack. Conversely, an efficient translation layer from
sotware to hardware enables quantum advantage.
Recently, quantum compilation algorithms have seen a shift from swapping qubits
(swapbased compilation), to regenerating a new quantum circuit (synthesis-based
compilation).
These algorithms show that quantum programs designed for near-term quantum computers
should be generated for a target machine directly, instead of generated for all-to-all
connectivity and subsequently routed using SWAP gates. The goal of this project is to build
up these compilation algorithms into a full-fledged open-source compiler.
Jukka K Nurminen
Feasibility of Contemporary Quantum Machine Learning Models
Quantum Machine learning (QML) is a novel outlook on how to use quantum computing in the
widely successful field of machine learning. Many of these contemporary QML models are mostly
tested on various existing classical datasets in “toy example” – scenarios, which causes many of
these results to be incomparable with each other. There also exist many theoretical results that seem
to indicate that there exist specific contexts where the usage of QML can make more sense. The aim
of the project is to investigate quantum machine learning challenges, especially in the large-scale,
realistic use case perspective, and develop solutions to key steps.