Research Area Faculty

• Prof. Dr. Constantia Alexandrou
• Assoc. Prof. Dr. Giannis Koutsou

Research Area Overview

Research Highlights

Research Highlight 1

Title: Innovative Algorithms and High-Performance Computing for Lattice QCD

Related people: C. Alexandrou, G. Koutsou, S. Bacchio, F. Pittler, S. Gregoriou, J. Finkenrath, K. Hadjiannakou, S. Yamamoto

Graphical Abstract

Description: The development of innovative algorithms and the exploitation of cutting-edge computing technologies are central to our research. The selected figure illustrates the advancements over the years in large-scale lattice quantum chromodynamics (QCD) simulations, conducted on Europe’s most powerful supercomputers. Notably, the introduction of multigrid methods in Ref. [1,2] achieved a 50× speed-up, while the use of 4th-order integrators in Ref. [3] enhanced scaling, resulting in a 2× speed-up on large lattices. Furthermore, enabling the software to run on GPUs, as shown in Ref. [4], provided an additional 5× to 7× speed-up. These innovations have allowed us to utilize LUMI, Europe’s largest supercomputer at CSC Finland, for our simulations.

 

Overview

The simulation of Quantum Chromodynamics (QCD) via its Euclidean-time, discrete formulation on a lattice, is one of the most compute-intensive applications in scientific computing, consuming substantial fractions of computer time at leadership HPC facilities internationally. A major challenge being addressed at CaSToRC is in advancing lattice QCD calculations to use improved, scalable algorithms and new computing technologies, such that progress in the field exceeds what would be dictated by computational technological improvements alone. These activities are conducted in collaboration with other groups within the Extended Twisted Mass Collaboration (ETMC).

 

Scientific Achievement

Novel approaches that have been developed and implemented in our calculations include algebraic multigrid solvers that accelerate calculations using physical quark masses, improved integration schemes employed in the molecular dynamics update of our Monte Carlo, efficient implementations in parallel GPU codes, capable of scaling to hundreds of GPU nodes, and investigations of proof-of-concept techniques such as for the inclusion of generative models within the Monte Carlo update process. The research is made possible through collaborations with researchers in applied mathematics and computer science and software developers, e.g. at NVIDIA.

 

Significance and Impact

Developing innovative algorithms and software optimized for cutting-edge technology ensures our competitiveness on an international scale, particularly in accessing the latest and most powerful supercomputers worldwide. Our group consistently secures resources on European systems funded by EuroHPC, including extreme-scale access to Leonardo, LUMI, and Marenostrum 5. Additionally, we actively participate in early access projects, such as those for Alps at CSCS and Jupiter at JSC, the latter being Europe’s first exascale supercomputer.

 

Research Details

Efficient simulation methods have enabled us to produce state-of-the-art ensembles that rank among the most accurate descriptions of the QCD vacuum worldwide. Achieving this level of precision is crucial for lattice QCD simulations, as results must be corrected for three key effects before making physical predictions. First, ensembles must be simulated at physical parameters; the use of multigrid methods has allowed us to efficiently reach the physical light quark mass [1,2]. Second, simulations must account for finite-size effects by using very large volumes, made more feasible through improved scaling with 4th-order integrators [3]. Finally, simulations must approach the continuum limit by decreasing the lattice spacing while maintaining a constant physical volume, a challenge addressed by leveraging GPU computing to produce some of the finest ensembles globally [4].

 

References

1. C. Alexandrou, S. Bacchio, J. Finkenrath, A. Frommer, K. Kahl, and M. Rottmann, ‘Adaptive aggregation-based domain decomposition multigrid for twisted mass fermions’, Phys. Rev. D, vol. 94, no. 11, p. 114509, Dec. 2016, doi: 10.1103/PhysRevD.94.114509.
2. S. Bacchio, C. Alexandrou, and J. Finkerath, ‘Multigrid accelerated simulations for Twisted Mass fermions’, EPJ Web Conf., vol. 175, p. 02002, 2018, doi: 10.1051/epjconf/201817502002.
3. C. Alexandrou et al., ‘Simulating twisted mass fermions at physical light, strange, and charm quark masses’, Phys. Rev. D, vol. 98, no. 5, p. 054518, Sep. 2018, doi: 10.1103/PhysRevD.98.054518.
4. B. Kostrzewa et al., ‘Twisted mass ensemble generation on GPU machines’, in Proceedings of The 39th International Symposium on Lattice Field Theory — PoS(LATTICE2022), Bonn, Germany: Sissa Medialab, Jan. 2023, p. 340. doi: 10.22323/1.430.0340.

 

Research Highlight 2

Title: Understanding the Fundamental Constituents of Matter

Related people: C. Alexandrou, S. Bacchio, G. Koutsou, F. Pittler, G. Pierini, B. Prasad, L. A. Rodriguez Chacon, J. Finkenrath, K. Hadjiyiannakou

Graphical Abstract

Description: Understanding the fundamental constituents of matter is a major task in modern theoretical physics. An example is the proton spin, expected to be 1/2, but distributed among its fundamental constituents—quarks and gluons. Experiments in the late 1980s measured the contribution of quark spins to the proton’s total spin and found it surprisingly small, accounting for only about 30%, far less than expected. This discrepancy, known as the proton spin puzzle. Our Lattice QCD simulations have been instrumental in addressing this problem. By accurately calculating the contributions of quark and gluon spins we have provided a theoretical resolution to the puzzle. The above figure, from Ref. [1] depicts how the spin of the proton distributed among its fundamental constituents, demonstrating that the total sum is indeed 1/2.

 

Overview

A fundamental challenge in modern physics is precisely calculating the internal structure of protons and neutrons from first principles, which involves overcoming the immense computational complexity of simulating quantum interactions between quarks and gluons at nuclear energy scales. Our research focuses on advancing lattice QCD calculations by developing techniques to reduce computational noise and improve precision, enabling more accurate theoretical predictions for quantities measured in current and future experiments.

 

Scientific Achievement

A key innovation in these calculations lies in the algorithmic and technical advancements that enable direct evaluation of nucleon structure properties using simulations at physical quark mass values. This approach eliminates the uncontrolled systematic uncertainties introduced when extrapolating results from heavier-than-physical quark masses. The research is made possible through a close collaboration with the University of Cyprus and is conducted within the Extended Twisted Mass Collaboration (ETMC). This partnership unites experts in quantum field theory, scalable algorithms for solving large systems of equations, high-performance computing for large-scale simulations, and advanced statistical analysis.

 

Significance and Impact

This research directly informs ongoing and upcoming experiments that test the Standard Model (SM) of particle physics while searching for subtle discrepancies that could indicate new physics beyond the SM. Key examples include neutrino experiments like NOνA, MINERνA, and DUNE; electron scattering experiments at facilities such as Jefferson Lab, MAMI in Mainz, and CERN; direct searches for dark matter candidates, such as weakly interacting massive particles (WIMPs); and the forthcoming experimental program at the Electron-Ion Collider (EIC) at Brookhaven National Laboratory.

 

Research Details

Methodology: Large-scale Markov-chain Monte Carlo simulations of QCD are employed to generate representative configurations of the QCD vacuum at multiple lattice spacings, enabling extrapolation to the continuum limit. These simulations incorporate physical masses for the up, down, strange, and charm quarks. Their feasibility relies on advanced linear solvers, such as those utilizing Algebraic Multi-grid methods, and highly optimized codes capable of parallel processing across hundreds of GPUs.

Results: Key results include nucleon charges—axial, tensor, and scalar—and their associated form factors, which describe how protons and neutrons couple to currents with specific quantum numbers. Other significant outcomes are the nucleon σ-terms, which quantify the quark contributions to nucleon mass, and the spin decomposition of the proton, confirming that QCD predicts the nucleon spin sums to 1/2 through the spin and angular momentum of its quark and gluon constituents. These results serve as critical tests of our understanding of quantum chromodynamics at low energy scales.

 

References

1. C. Alexandrou et al., ‘Complete flavor decomposition of the spin and momentum fraction of the proton using lattice QCD simulations at physical pion mass’, Phys. Rev. D, vol. 101, no. 9, p. 094513, May 2020, doi: 10.1103/PhysRevD.101.094513.
2. C. Alexandrou et al., ‘Nucleon axial, tensor, and scalar charges and σ -terms in lattice QCD’, Phys. Rev. D, vol. 102, no. 5, p. 054517, Sep. 2020, doi: 10.1103/PhysRevD.102.054517.
3. C. Alexandrou et al., ‘Nucleon axial and pseudoscalar form factors using twisted-mass fermion ensembles at the physical point’, Phys. Rev. D, vol. 109, no. 3, p. 034503, Feb. 2024, doi: 10.1103/PhysRevD.109.034503.
4. C. Alexandrou et al., ‘Moments of the nucleon transverse quark spin densities using lattice QCD’, Phys. Rev. D, vol. 107, no. 5, p. 054504, Mar. 2023, doi: 10.1103/PhysRevD.107.054504.

Selected Publications

• C. Alexandrou et al., ‘Inclusive Hadronic Decay Rate of the τ Lepton from Lattice QCD: The ūs Flavor Channel and the Cabibbo Angle’, Phys. Rev. Lett. 132 261901, 2024, doi: 10.1103/PhysRevLett.132.261901
• S. Bacchio, ‘Novel approach for computing gradients of physical observables’, Phys. Rev. D, vol. 108, no. 9, p. L091508, Nov. 2023, doi: 10.1103/PhysRevD.108.L091508.
• S. Bacchio, P. Kessel, S. Schaefer, and L. Vaitl, ‘Learning trivializing gradient flows for lattice gauge theories’, Phys. Rev. D, vol. 107, no. 5, p. L051504, Mar. 2023, doi: 10.1103/PhysRevD.107.L051504.
• C. Alexandrou et al., ‘Nucleon axial and pseudoscalar form factors using twisted-mass fermion ensembles at the physical point’, Phys. Rev. D, vol. 109, no. 3, p. 034503, Feb. 2024, doi: 10.1103/PhysRevD.109.034503.
• C. Alexandrou et al., ‘Elastic nucleon-pion scattering amplitudes in the Δ channel at physical pion mass from lattice QCD’, Phys. Rev. D, vol. 109, no. 3, p. 034509, Feb. 2024, doi: 10.1103/PhysRevD.109.034509.