Conveners
Algorithms, machines, and code development: Session 4-1 A
- Issaku Kanamori (RIKEN)
Algorithms, machines, and code development: Session 4-2 D
- Tilo Wettig (University of Regensburg)
Algorithms, machines, and code development: Session 5-2 C
- Seyong Kim (Sejong University)
A variation of the Domain Wall operator with an additional parameter α will be introduced. The conditioning of the new Domain Wall operator depends on α, whereas the corresponding 4D propagator does not. The new and the conventional Domain Wall operator agree for α = 1. By tuning α, speed ups of the linear system solvers of around 20% could be achieved.
We compare two of different methods for coarsening domain wall fermions and discuss progress towards a multigrid method for DWF with the ability to set up and solve during gauge configuration generation.
We present a multiple right-hand side (rhs) implementation of the Adaptive Aggregation-based Domain Decomposition Multigrid method (DD$\alpha$AMG) using twisted mass fermions.
Our implementation extends the strong scaling region of DD$\alpha$AMG and simplifies vectorization, which would otherwise require using vector extensions. This multiple rhs implementation is thus better suited to take...
We report our recent extension of Bridge++, a general-purpose code set for numerical simulations of lattice gauge theories, to the latest vector processor, NEC SX-Aurora TSUBASA. The Bridge++ project aims at developing a readable, extensible, and portable workbench with sufficiently high performance. Based on the code set we investigate fast algorithms for parallel numerical calculations, and...
QPACE 4 is the latest member of the QCD PArallel Compute Engine (QPACE) series, which was deployed in Regensburg in June 2020. It features 64 Fujitsu A64FX model FX700 CPUs interconnected by InfiniBand EDR. The A64FX is the first CPU supporting the Arm Scalable Vector Extension (SVE). In this contribution we discuss the implementation of SVE in the Grid Lattice QCD framework and show Grid...
The supercomputer Fugaku is a new supercomputer in Kobe, Japan, co-developed by RIKEN with Fujitsu, and the top system of the latest June 2020 TOP500 supercomputers. I will introduce the supercomputer Fugaku and a Lattcie QCD simulation library, QCD Wide SIMD library (QWS) for Fugaku. I will also present some tuning methods for Fugaku, QWS performance on Fugaku, and tentative benchmark results...
As parallel systems become massive, the neighboring communication in lattice QCD becomes more and more important.
In this talk, I will focus on the implementation of neighboring communication in QCD Wide SIMD library (QWS) for the supercomputer Fugaku.
We adopt the double buffering algorithm and implement it on top of a wrapper library to call the uTofu API, which is a low level interface...
Lattice QCD is one of the major scientific work-loads on supercomputer installations. Most of the computer time is spent in an iterative solver of a large, sparse set of linear equations. One of the simplest examples of such an iterative solver is the conjugate gradient algorithm. In this talk, we present an optimized implementation of this algorithm in the context of Lattice QCD for Xilinx...
The most computational cost in typical lattice QCD simulation is doing the invertion to obtain the propagator. While it is a huge waste to free the propagators in RAM after the contraction. This work explores a field selection algorithm for the correlation functions. The field selection algorithm constructs the correlation function by selecting point on the lattice. It is found that almost the...
Tempered Lefschetz thimble method (TLTM) [Fukuma-Umeda(1703.00861)] is an algorithm towards solving the numerical sign problem. There, the integration region is deformed into the complex space following the antiholomorphic gradient flow equation, and the system is parallel-tempered using the flow time as a tempering parameter so as to solve both sign and ergodicity problems simultaneously. In...
Monte Carlo simulations of finite density QCD is plagued by the sign problem. The tempered Lefschetz thimble method (TLTM) [Fukuma-Umeda(1703.00861)] is a promising algorithm towards solving the sign problem, where the integration region is deformed into the complex space and the system is parallel-tempered with the flow time so as to solve both sign and ergodicity problems simultaneously. In...
Statistical modeling plays a key role in lattice field theory calculations. Examples including extracting masses from correlation functions or taking the chiral-continuum limit of a matrix element. We discuss the method of model averaging, a way to account for uncertainty due to model variations, from the perspective of Bayesian statistics. Statistical formulas are derived for model-averaged...
We present a synergistic approach between machine learning and histogram reweighting to discover and study phase transitions in physical systems. We treat the output of a neural network, designed for phase classification, as an observable in a statistical system enabling its extrapolation over continuous ranges in parameter space using histogram reweighting. The approach, which leads to...
The quantum link Hamiltonian was introduced two decades ago as an alternative to Wilson’s Euclidean lattice QCD with gauge fields represented by bi-linearfermion/anti-fermion operators, and later generalized as D-theory. Recasting as a Hamiltonian in Minkowski space for real time evolution, D-theory leads naturally to quantum algorithms. We investigate the simplest toy model of U(1) compact...
Due to the existence of sign problem in the Lattice QCD simulation with finite chemical potential, the traditional Monte-Carlo simulations on classical supercomputers are confronted with significant difficulties on achieving high precision. On the other hand, with the fast development of quantum computers, it might be possible to provide the ultimate solution to sign problem in the future....
