Conveners
Algorithms and Machines
- Michael Endres (MIT)
Algorithms and Machines
- Mike Peardon
Algorithms and Machines
- Guido Cossu (Higgs Centre for Theoretical Physics, School of Physics and Astronomy, University of Edinburgh)
Algorithms and Machines
- Tilo Wettig
Ms
Barbara De Palma
(Istituto Nazionale di Fisica Nucleare (INFN) and Università di Pavia)
7/26/16, 4:30 PM
Algorithms and Machines
Talk
We report lattice simulations of $\phi^4_2$ and $O(N)\,\phi^4$ models, performed by means a Monte Carlo method based on the all-order strong coupling expansion (worm algorithm). The investigation of the non–perturbative features of the $\phi^4$ continuum limit in two dimensions lead us to the result $g/\mu^2 = 11.15 \pm 0.06_{stat} \pm 0.03_{syst}$ for the critical coupling. Furthermore we...
Ms
Emilie Huffman
(Duke University)
7/26/16, 4:50 PM
Algorithms and Machines
Talk
We solve a variety of sign problems for models in lattice field theory using the Hamiltonian formulation, including Yukawa models and simple lattice gauge theories. The solutions emerge naturally in continuous time and use the dual representation for the bosonic fields. These solutions allow us to construct quantum Monte Carlo methods for these problems. The methods could provide an...
Mr
Alessandro Nada
(Università di Torino & INFN, Torino)
7/26/16, 5:10 PM
Algorithms and Machines
Talk
Jarzynski's equality is a well-known result in statistical mechanics, relating free-energy differences between equilibrium ensembles with fluctuations in the work performed during non-equilibrium transformations from one ensemble to the other.
In this talk, an extension of this relation to lattice gauge theory will be presented, along with numerical results for the $Z_2$ gauge model in three...
Dr
Roberto Pellegrini
(The University of Edinburgh)
7/26/16, 5:30 PM
Algorithms and Machines
Talk
The LLR algorithm is a recent proposal for computing the density of states in lattice gauge theory. This algorithm has been tested in several bosonic models at zero and finite chemical potential with impressive results. Its original formulation is based on the simulation of the theory on restricted action intervals using local Monte Carlo updates.I will discuss a new version of the method...
Prof.
Biagio Lucini
(Swansea University)
7/26/16, 5:50 PM
Algorithms and Machines
Talk
It has been shown that the recently proposed LLR method is very efficient at overcoming strong metastabilities that arise near first-order phase transition points. Here we present a systematic study of the performance of the algorithm near (pseudo)critical points on q-state Potts models for q as large as 20, in two and three dimensions. In particular, we shall focus our study on the ergodicity...
Mr
Francesco Sanfilippo
(University of Southampton)
7/26/16, 6:10 PM
Algorithms and Machines
Talk
Metadynamics is a class of powerful algorithms in which the time evolution of a system is modified introducing a history-dependent potential associated with the past values of observables of choice. This has the effect of driving the system away from previously occupied states, ultimately speeding up the evolution of the system.
These methods are widely used in biochemistry and computational...
Stefan Schaefer
(NIC, DESY)
7/27/16, 9:00 AM
Algorithms and Machines
Talk
The numerical computation of many hadronic correlation functions is exceedingly difficult due to the exponentially decreasing signal-to-noise ratio with the distance between source and sink. Multilevel integration methods, using independent updates of separate regions in space-time, are known to be able to solve such problems but have so far been available only for pure gauge theory.
We...
Marco Cè
(Scuola Normale Superiore, Pisa, Italy & INFN, Sezione di Pisa, Italy)
7/27/16, 9:20 AM
Algorithms and Machines
Talk
The numerical computation of many hadronic correlation functions is exceedingly
difficult due to the exponentially decreasing signal-to-noise ratio with the
distance between source and sink. Multilevel integration methods, using independent
updates of separate regions in space-time, are known to be able to solve such
problems but have so far been available only for pure gauge theory.
We...
Dr
Matthias Rottmann
(University of Wuppertal)
7/27/16, 9:40 AM
Algorithms and Machines
Talk
In this talk we report on the publication of the DDalphaAMG solver library. We describe its features, show examples of its application and give an overview of future developments.
Mr
Simone Bacchio
(University of Cyprus - University of Wuppertal)
7/27/16, 10:00 AM
Algorithms and Machines
Talk
We extend the Adaptive Aggregation Based Domain Decomposition Multigrid (DD-$\alpha$AMG) algorithm to $N_f=2$ twisted mass fermions. We show numerical results for an $N_f=2$ ensemble of twisted fermions with a clover term simulated at the physical value of the pion mass. We fine-tuned the parameters to achieve a speedup comparable to the one obtained for clover fermions. We also present a...
Mr
David Murphy
(Columbia University)
7/27/16, 10:20 AM
Algorithms and Machines
Talk
As algorithmic developments have driven down the cost of simulating degenerate light quark pairs the relative cost of simulating single quark flavors with the Rational Hybrid Monte Carlo (RHMC) algorithm has become more expensive. TWQCD has proposed an exact one-flavor algorithm (EOFA) that allows for HMC simulations of a single quark flavor without taking a square root of the fermion...
Mr
Andrea Bussone
(Univeristy of Southern Denmark & CP3 Origins)
7/27/16, 10:40 AM
Algorithms and Machines
Talk
We present a cheap strategy to optimize Hybrid Monte Carlo parameters
in simulations of QCD and QCD-like theories. We specialize to the case of
mass-preconditioning, with multiple-time-step Omelyan integrators.
Starting from properties of the shadow Hamiltonian we show how the
optimal setup for the integrator can be chosen once the forces and
their variances are measured, assuming that...
Dr
Meifeng Lin
(Brookhaven National Laboratory)
7/27/16, 11:30 AM
Algorithms and Machines
Talk
We present updated strategies and results of combining hand-tuning with the R-Stream source-to-source auto-parallelizing compiler to transform the serial implementation of the domain wall fermion Dslash kernel in CPS into an efficient parallel code targeting the Intel Xeon CPUs. The R-Stream compiler performs preliminary optimizations of the input Dslash code, including a novel iteration space...
Mr
Arjun Gambhir
(College of William and Mary/JLab)
7/27/16, 11:50 AM
Algorithms and Machines
Talk
Computing disconnected diagrams on the lattice involves taking the trace of the inverse of the Dirac operator. This is a computationally challenging problem, however recent algorithmic improvements such as low mode averaging and hierarchical probing have increased the efficiency of this trace estimation. We detail an algorithm that builds upon hierarchical probing by deflating the near null...
Mr
Matthias Puhr
(University of Regensburg)
7/27/16, 12:10 PM
Algorithms and Machines
Talk
We present a method for the numerical calculation of derivatives of
functions of general complex matrices which also works for implicit
matrix function approximations such as Krylov-Ritz type algorithms. An
important use case for the method is the overlap Dirac operator at
finite quark chemical potential. The evaluation of the overlap Dirac
operator at finite chemical potential calls for...
Dr
James Osborn
(ANL)
7/28/16, 2:00 PM
Algorithms and Machines
Talk
We present a new lattice field theory software framework designed with
ease of use, flexibility, efficiency and portability in mind. The
framework is written using the Nim programming language which offers
many of the features one would find in a high-level scripting
language, while in fact being a strongly-typed language with full
control over low-level optimizations. This allows us to...
Dr
Mathias Wagner
(NVIDIA)
7/28/16, 2:20 PM
Algorithms and Machines
Talk
Analysis tasks in Lattice QCD often requires solving linear equations for multiple right hand sides for a constant gauge field. Recently deflation methods have become more widely used and proven to be very efficient. They however do require the calculation and storing of eigenvectors which is either not always feasible or too expensive. Here we present results for an implementation of a block...
Dr
Kate Clark
(NVIDIA)
7/28/16, 2:40 PM
Algorithms and Machines
Talk
The past decade has witnessed a dramatic acceleration of lattice quantum chromodynamics calculations in nuclear and particle physics. This has been due to both significant progress in accelerating the iterative linear solvers using multi-grid algorithms, and due to the throughput improvements brought by GPUs. Deploying hierarchical algorithms optimally on GPUs is non-trivial owing to the lack...
Dr
Evan Weinberg
(Boston University)
7/28/16, 3:00 PM
Algorithms and Machines
Talk
As the push towards the exascale enables increasingly accurate lattice calculations, the inversion of the Dirac matrix becomes a superlinearly growing expense. Adaptive algebraic multigrid (AAMG) methods for all fermion discretizations are essential to address this phenomena of critical slowing down. As a preconditioner, AAMG expedites Dirac matrix inversions with manageable start up costs....
Dr
Ken-Ichi Ishikawa
(Hiroshima University)
7/28/16, 3:20 PM
Algorithms and Machines
Talk
The most computationally demanding part of Lattice QCD simulations is
solving quark propagators.
Quark propagators are typically obtained with a linear equation solver utilizing HPC machines.
The success of Lattice QCD simulations owes much to the development of numerical algorithms and
optimization for the quark solver, and evolution of HPC machines.
The CCS QCD Benchmark is a benchmark...
Dr
Ruizi Li
(Indiana University)
7/28/16, 3:40 PM
Algorithms and Machines
Talk
We review our work done to optimize the staggered conjugate gradient(CG) algorithm in the MILC code for use with the Intel Knights Landing (KNL) architecture. KNL is the second generation Intel Xeon Phi processor. It is capable of massive thread parallelism, data parallelism and high on-board memory bandwidth, and is being adopted in supercomputing centers for scientific research. The CG...