Servicenavigation

Detailseite

Zurück

Projekt Druckansicht

TRR 55:  Hadronenphysik mit Gitter-QCD

Fachliche Zuordnung Physik
Informatik, System- und Elektrotechnik
Mathematik
Förderung Förderung von 2008 bis 2020
Projektkennung Deutsche Forschungsgemeinschaft (DFG) - Projektnummer 35592816
 
Erstellungsjahr 2021

Zusammenfassung der Projektergebnisse

The overarching goal of SFB/TRR-55 was to combine three different research fields, namely Lattice QCD, advanced mathematical algorithms and novel developments in High Performance Computing in such a way • that an improved level of precision is realized in Lattice QCD and that a much extended spectrum of questions in hadron physics, both at zero and finite temperature, can be answered. • that new algorithms are not only developed but also widely used in practical applications, such that their performance can directly steer further innovation. • that new HPC hard- and software is developed, in close cooperation with major international computer companies, which provides an unprecedented cost-performance ratio for Lattice QCD calculations. To reach this goal one of the main design criteria was to achieve much improved energy efficiency. All of these goals have been achieved, see subsection 1.3. Over the 12 years SFB/TRR-55 existed all three, highly dynamical fields have spawned most interesting new developments which have been taken up by SFB/TRR-55 and which have both extended and modified its scope of research activities compared to the first proposal. Some of these are: • At the time SFB/TRR-55 was created many fundamental issues of QCD thermodynamics were still unclear, e.g. the QCD phase diagram and the QCD equation of state. • At the time of the first proposal in 2007 the community had very high hopes concerning the use of chiral fermions (overlap or domain wall) which, therefore, played an important role in that proposal. It was hoped that chiral fermions would lead to such large systematic improvements (in particular improving control of the continuum limit) that this would more than compensate for their larger computational cost. The results of us and many other groups moderated these expectations. Today, simulations with chiral and e.g. Wilson fermions are performed with comparable computer resources and one has learned which problems can be investigated best with which fermion type. Consequently, overlap fermions played a reduced role in the continuation proposals. • As was finally announced January 2020 the heavy-ion and polarized proton-proton program at the RHIC accelerator of Brookhaven National Laboratory will come to an end in few years, getting replaced by electron-proton/nucleus physics performed at the Electron-Ion-Collider (EIC) which will recycle much of the RHIC infrastructure. This project is realized jointly with JLab, the Thomas Jefferson National Accelerator Facility. In addition the electron-proton/nucleus accelerator JLab@12 GeV has started operating in 2017. At CERN a roughly ten year long upgrade phase of LHC to the High Luminosity LHC (HL-LHC) has started. Somewhat disappointingly the present LHC has not yet identified the signals for Beyond the Standard Model (BSM) physics which one had hoped for. The much increased luminosity will significantly extend its discovery potential along what is called the ’precision frontier’ just as the EIC will do for hadron structure physics. Both developments offer new opportunities and pose increased challenges for highprecision Lattice QCD because in both cases the main source of systematic theoretical uncertainties comes from non-perturbative QCD. Another important development is the impressive increase in experimental precision reached in searches for axions and other astrophysically relevant particles. Our collaboration has positionned itself to optimally profit from all these developments, see subsection 1.5. • To fulfill the increased demands on high precision QCD also requires significant progress along the lowenergy frontier much of which has to come from the lattice. This development is illustrated best by the rapidly increasing importance of the FLAG reports, see e.g. [X1] of which Dr. Sara Collins and Dr. Stephan Dürr were co-authors. page 5 of 135 Part A Research section Research achievements and outcomes • For all phenomenological applications, methods of nonperturbative renormalization were developed further in a systematic drive to reduce the associated systematic and statistical errors to the precision goals aimed at. • The major new German research facility FAIR approaches completion. Therefore, it was a particularly timely and relevant issue to develop techniques to describe the QCD plasma at nonvanishing temperatures and chemical potentials, which will be the main focus point of the CBM (Compressed Baryon Matter) experiment at FAIR. We have made a significant progress towards this important direction. • Adaptive algebraic multigrid methods have evolved into an important solver technology for linear systems where the required operator hierarchy does not arise naturally from the physical model. Adaptive algebraic multigrid methods can be used in a variety of contexts, and their use when solving the Wilson-Dirac equation has algorithmically reduced computational efforts by at least one order of magnitude. This crucially contributed to our today’s capability to simulate at physical quark masses. The adaptive algebraic multigrid approach has a variety of different facets which require a deeper mathematical understanding as well as the careful selection of parameters. Research is still ongoing, but for lattice simulations we have already demonstrated significant progress in aggregation-based coarsening and in the bootstrap principle for the setup. • The problem of an increasingly slow memory speed vs. a high processor speed became even more pronounced on current architectures during the lifetime of the SFB. As an algorithmic reaction to this problem, block methods—like, e.g., treating several linear systems with the same matrix simultaneously—have become increasingly important. The numerical linear algebra community has contributed various novel techniques on how to combine different Krylov subspaces in the linear system context and analyzed their influence on the speed of convergence, and so did we. • SFB/TRR-55 co-designed five large HPC systems (QPACE 1 through QPACE 4 and iDataCool) together with major computer companies (IBM, Intel, Eurotech, Fujitsu, Arm) and supercomputing centers (JSC, LRZ, RIKEN R-CCS). These efforts, which generated substantial international visibility, were accompanied by the development of algorithms and high-performance software. As a result, we obtained close-to-optimal sustained performance for key Lattice QCD (and other) applications at highly competitive cost-performance and power-performance ratios. A major theme was the early adoption of new computing architectures such as IBM’s Cell processor, Intel’s Xeon Phi processors (Knights Corner and Knights Landing) and Fujitsu’s A64FX. Through our close collaboration with the processor manufacturers we had access to chip simulators and early engineering samples so that our codes could be developed and optimized before the new processors came to market. Another major theme was energy efficiency, which has become a major design goal in HPC. Of our five systems, three were large enough to make the TOP 500 list of the largest HPC systems worldwide. On the Green 500 list, which orders these machines by energy efficiency, our systems were No. 1 (QPACE 1 in 2009 and 2010), No. 15 (QPACE 2 in 2015) and No. 5 (QPACE 3 in 2016). The iDataCool system we developed together with IBM went one step further and demonstrated the possibility of hot-water cooling with energy reuse. It is certainly fair to say that SFB/TRR-55 and thus DFG have contributed significantly to crucial trends in HPC developments on par with major players in the field.

Projektbezogene Publikationen (Auswahl)

 
 

Zusatzinformationen

© 2024 DFG Kontakt / Impressum / Datenschutz

Textvergrößerung und Kontrastanpassung