AQTIVATE Conference

28 Sept 2026, 09:00 → 2 Oct 2026, 19:00 Asia/Nicosia

Sun Hall Hotel

The AQTIVATE conference brings together experts in high performance computing, scalable algorithms and machine learning approaches, and quantum computing for physics, engineering and biology.

The conference is organized under the joint doctoral project "Advanced computing, QuanTum algorIthms and data-driVen Approaches for science, Technology and Engineering (AQTIVATE)". See the AQTIVATE project website for more details.

 Keynote speakers will cover the following topics:

• Modeling and algorithms towards exascale
• Machine learning approaches
• Quantum algorithms and tensor networks
• Applications from lattice QCD, computational fluid dynamics, and computational biology

 

Confirmed Speakers

Simone Bacchio (The Cyprus Institute, Cyprus)
Gunnar Bali (Universität Regensburg, Germany)
Tom Beck (National Center for Computational Sciences, USA)
Michele Buzzicotti (University of Rome Tor Vergata)
Chiara Calascibetta (Centre INRIA d’Université Côte d’Azur, France)
Paolo Carloni (Forschungszentrum Jülich and RWTH Aachen)
Jens Eisert (Freie Universität Berlin, Germany) 
Antonio Evangelista (University of Cyprus)
Jacob Finkenrath (University of Wuppertal, Germany)
Alessandro Gabbana (Los Alamos National Laboratory, USA and University of Ferrara, Italy)
Patrick Gallinari (Sorbonne University, France)
Ellinor Haglund (University of Hawaii, USA) 
Philipp Hauke (University of Trento, Italy) 
Michael Klein (Temple University, Philadelphia, USA) 
Thomas Lippert (Director of Jülich Supercomputing Centre, Germany) 
Sabrina Maniscalco (Algorithmiq Ltd, Finland) 
Enrique Rico Ortega (CERN-TH and Ikerbasque and UPV/EHU)
Michele Parrinello (IIT, Italy)
Luka Pavešić (University of Padova, Italy)
Jacob Schroder (University of New Mexico, USA)
Sauro Succi (IIT, Italy) Sergio Hoyas (Universitat Politècnica de València) 
Lipi Thukral (Institute of Genomics and Integrative Biology, India)

 

Funded by the European Union: The project AQTIVATE has received funding from the EU's research and innovation programme under the Marie Skłodowska-Curie Doctoral Networks action and GA No. 101072344

Sunhall_booking_form.pdf

Monday 28 September

09:00 → 09:30 Welcome

09:00 Welcome

Speaker: Constantia Alexandrou (University of Cyprus, and The Cyprus Institute)

09:30 → 11:00 Invited

09:30 Future HPC and Quantum Hardware

Speaker: Thomas Lippert

10:15 TBD

Speaker: Jiali Gao

11:00 → 11:30 Coffee break

11:30 → 13:00 Invited

11:30 Quantum Information

Speaker: Sabrina Maniscalco

12:15 Biology

Speaker: Ellinor Haglund

13:00 → 15:00 Lunch

15:00 → 16:30 ESR

15:00 Rethinking Algebraic Multigrid Methods: A Relaxation-Based Perspective

Algebraic multigrid methods are some of the most efficient methods used in the solution of large scale problems which often appear in engineering and physics theory and application. These methods rely entirely on the complementarity between a sequence of error relaxation processes each happening on a given scale within the hierarchy of scales present in the problem to be solved. The precise definition of true complementarity, although previously defined through numerical or heuristic arguments, is still illusive. In this talk, I will give a quick overview of a reformulation of the theory of algebraic multigrid through a novel coarse graining procedure applied to the relaxation process itself, one which allows us to better understand how multigrid operates and what true multiscale complementarity could look like.

Speaker: Rayan Moussa (University of Wuppertal)

15:30 Mass Splittings in Baryon Flavor-SU(2) Multiplets

Isospin breaking, arising from electromagnetic interactions and the difference between the up- and down-quark masses, leads to small but phenomenologically important mass differences between hadrons. As lattice QCD calculations reach percent-level precision, these effects need to be included systematically.

In this work, we study isospin-breaking mass splittings within spin-1/2 and spin-3/2 baryon flavor-SU(2) multiplets using lattice QCD. Electromagnetic and strong-isospin-breaking corrections are incorporated perturbatively around an isospin-symmetric lattice QCD calculation, using twisted-mass gauge ensembles tuned close to the physical isospin-symmetric point. Results at several lattice spacings allow us to investigate discretization effects and enable a controlled approach to the continuum limit. The use of several ensembles further allows us to investigate systematic uncertainties and compare our results with experimentally observed mass splittings.

Speaker: Christian Schneider (University of Cyprus)

16:00 The nucleons generalized form factors and Mellin moments up to fourth order

Nucleon Mellin moments of parton distribution functions and generalized parton distributions are computed up to the fourth order in lattice QCD. The computation is performed using one ensemble of twisted mass fermions at the physical pion mass point. We employ boosted frames to access the higher-order Mellin moments of generalized parton distributions. We also extract the generalized formfactors up to fourth order. These results establish benchmarks for future lattice studies and expand the understanding of the partonic structure of the proton.

Speaker: Christian Kummer

16:30 → 17:00 Coffee break

17:00 → 17:30 Invited

17:00 Computational Biophysics of Biomolecular Activation: From WAVE Complex Variants to GPCR Dynamics

Computer simulations can provide unique atomistic insights into the molecular mechanisms underlying protein activation. Here, I will focus on two neurobiologically relevant systems: the WAVE regulatory complex (WRC) and the adenosine A2A receptor (A2AR), a G protein-coupled receptor (GPCR). Molecular dynamics simulations of the WRC identified common mechanistic features of autism-associated hyperactive variants, enabling a billion-scale virtual screening campaign targeting one representative mutant. For A2AR, molecular dynamics simulations combined with DFT-based NMR calculations provided the first molecular-level interpretation of activation-dependent ¹⁹F NMR signals.

Speaker: Song Xie (Forschungszentrum Jülich)

17:30 → 18:30 ESR

17:30 QM/MM-derived force fields for metalloprotein simulations in drug discovery

Molecular dynamics simulations of protein–ligand systems are central to computer-aided drug discovery, but their reliability rests on the underlying force field. Metalloproteins are a particularly hard case: classical force fields describe metal coordination poorly, and reliable parameters are often unavailable. Quantum mechanics/molecular mechanics (QM/MM) molecular dynamics treats such sites accurately, yet reaches only picosecond time scales, far short of those relevant to ligand binding and drug design. Force matching (FM) addresses this gap, turning short QM/MM trajectories into system-specific classical force fields that retain reference accuracy at classical cost. In this talk I will present my work on force matching within the MiMiC multiscale framework. I will first describe MiMiCPy-FM, an automated implementation of QM/MM force matching that derives atomic charges and bonded parameters from MiMiC (CPMD/GROMACS) simulations and writes GROMACS topologies ready for classical MD. Its application to the Mg-based enzyme IDH1, a drug target in brain cancer, shows a seamless transition from picosecond QM/MM to microsecond classical dynamics that preserves the QM active-site structure. Fixed-charge force fields, however, cannot respond to a changing electrostatic environment, a limitation for metal sites and charged ligands. I will therefore present work extending force matching to polarizable force fields, coupling it to AMOEBA with Tinker-HP, validated on acetone in water.

Speaker: Mr Sachin Shivakumar (Forschungszentrum Jülich GmbH)

18:00 Generalized Parton Distribution Functions of the Nucleon from Twisted Mass Fermions with Physical Pion Mass

Studying hadrons can be challenging due to the non-perturbative nature of QCD. Parton Distribution Functions (PDFs), and their generalization Generalized Parton Distributions (GPDs), are objects that carry information about how quarks and gluons are distributed in hadrons such as protons and neutrons. The computation of GPDs from both theory and experiments is challenging, therefore extracting them from direct simulations of the theory on the Lattice is crucial. In this talk we present the PDFs and GPDs of the nucleon using an ensemble of $N_f$=2+1+1 twisted mass clover-improved fermions with masses tuned to their physical values and lattice spacing of $a=0.080$ fm. We analyze seven momentum boosts up to $P_z=1.69$ GeV and momentum transfer $-t$ up to $1.2 ~ \textrm{GeV}^2$. We explore both LaMET and Short Distance Factorization to obtain the light-cone PDFs and GPDs and their moments, respectively. We study the GPDs in the so-called "asymmetric frame", that allows for more efficient computation of multiple $-t$. Within the LaMET framework, we investigate an alternative method for the reconstruction of the x-dependent quasi-distributions using Bayes-Gauss-Fourier Transform and compare it to Backus-Gilbert.

Speaker: Gabriele Pierini (The Cyprus Institute, TU Berlin)

Tuesday 29 September

09:30 → 11:00 Invited

09:30 Machine Learning and Lattice QCD

I will review some of the recent machine learning applications in the field of lattice QCD. Based on that, I will discuss there possible application within application of lattice field theory simulations.

Speaker: Jacob Friedrich Finkenrath (Bergische Universitaet Wuppertal (DE))

10:15 Physics Aware Deep Learning

Speaker: Patrick Gallinari

11:00 → 11:30 Coffee break

11:30 → 12:00 Invited

11:30 Small-scale alignment and clustering of heavy inertial fibers in turbulence

The dynamics of non-spherical particles in turbulence are governed by the interplay between inertia and orientation-dependent hydrodynamic forces. We investigate rigid fibers dynamics by bridging two asymptotic descriptions: the small-size limit described by Jeffery’s equation, and the finite-length limit captured by slender-body theory. This framework enables systematic variation of aspect ratio, fiber length, and Stokes number. Using direct numerical simulations, we examine preferential alignment and small-scale clustering. At low Stokes numbers, fibers tend to align with the most unstable direction associated to the attractor towards which they converge and exhibit strong clustering consistent with convergence onto a fractal attractor. As the Stokes number increases, clustering is progressively reduced, with the attractor dimension approaching that of a uniform distribution, consistent with the expected behavior of inertial particles and the caustics mechanism.

This work was supported by the ANR through Project No. ANR-21-CE30-0040-01 and funded by the European Union’s Horizon Europe research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 101273291, project DragREACT. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Executive Agency (REA). Neither the European Union nor the European Research Executive Agency (REA) can be held responsible for them.

Speaker: Chiara Calascibetta (Université Côte d'Azur, CNRS, INPHYNI & Inria, Nice, France)

12:00 → 13:00 ESR

12:00 Probing techniques for disconnected quark loop estimation in Lattice QCD

Disconnected quark-loop diagrams contribute to many key hadronic observables, from flavor-singlet spectroscopy to nucleon structure and the muon anomalous magnetic moment, but require traces of the inverse Dirac operator, which can only be estimated stochastically. Probing methods reduce the variance of such estimates by exploiting the exponential decay of the quark propagator: noise vectors built from a distance-$d$ coloring of the lattice cancel all short-range off-diagonal contributions exactly. A widely used construction, hierarchical probing, achieves reusability of previous computations through nested colorings, but offers valid colorings only at power-of-two distances, with an exponentially increasing number of colors. We present an alternative coloring strategy that assigns colors through integer linear forms on the lattice coordinates, yielding valid distance-$d$ colorings at arbitrary distances, not only powers of two, and with fewer colors than hierarchical probing at equal distance. Numerical studies on Wilson-Dirac configurations, for disconnected loops Tr$[\Gamma(t)D^{-1}(t,t)]$ evaluated on single time slices, show that the new construction matches hierarchical probing at its complete levels while providing valid colorings at several intermediate budgets, allowing the accuracy to be improved continuously with the available computational cost.

Speaker: Mario Papace (Bergische Universität Wuppertal)

12:30 Benchmarking tensor network simulations of Shor's algorithm with coherent error propagation

Shor’s factorization algorithm which factors integers in time polynomial to their bit-length, is a major milestone in the race to demonstratable quantum advantage because of its exponential speedup over currently available classical algorithms. It also provides an ideal testbed for benchmarking classical simulations of quantum algorithms and is particularly interesting as a stress test for tensor-network simulations because of the entanglement generated within its quantum circuit representation. In this talk I will present our work on benchmarking the Shor’s algorithm on RSA-like semi-primes using the full gate-level Vedral, Barenco and Ekert (VBE) modular exponentiation circuit instead of a high-level arithmetic oracle on a tensor network powered quantum emulator and quantify its performance with three complementary figures of merit: a fidelity estimate derived from truncated singular values, a divergence metric comparing the distance between the true output distribution and the simulated one, and the normalized factoring success probability. In addition to illustrating how each of them behave with the bond dimension of the tensor network, I will also highlight how the simulation complexity relates to the bit-length and the multiplicative order for a particular semi-prime for ideal, noiseless circuits. This serves as the premise for the core of the talk: benchmarking under the effect of coherent errors, which is modelled using systematic unitary over-rotations applied stochastically, mimicking gate miscalibrations in quantum hardware. I will present the noise model in detail which rely on two main parameters: the over-rotation angle and the per-gate probability. Sweeping both, we map how the figures of merit behave, locate the thresholds where factoring fails, and find that coherent errors tend to drive the required bond dimension upward; thus establishing Shor's algorithm as a verifiable, tuneable probe of classical simulability at the edge of tractability.

Speaker: Ms Asmita Datta (University of Padova)

13:00 → 15:00 Lunch

15:00 → 15:30 Invited

15:00 Comp. biology

Speaker: Lipi Thukral

15:30 → 16:30 ESR

15:30 Bayesian Statistical Learning for VQE optimization

Variational quantum algorithms operate in a challenging statistical regime: objective functions and gradients must be inferred from a finite number of noisy quantum measurements. In this talk I present uncertainty-aware and bias-aware statistical learning methods for optimizing parametrized quantum circuits, with a focus on the Variational Quantum Eigensolver.

This framework strengthens the classical learning component of VQE through physics-informed Gaussian-process kernel learning, from which confidence-region methods, adaptive shot allocation, and the Bayesian parameter shift rule stem. A complementary bias analysis characterises when sequential minimal optimization energy estimates become unreliable and shows how this phenomenon can be exploited to speed up optimization without sacrificing statistical accuracy.

The methods are evaluated using exact classical simulations of spin-chain Hamiltonians, enabling quantitative comparisons between optimizers under a fixed cumulative shot budget. The results illustrate how explicit modeling of uncertainty and estimator bias can improve the efficiency and interpretability of variational quantum optimization.

Speaker: Samuele Pedrielli (Technische Universität Berlin, Università degli Studi di Padova)

16:00 Multi-scale Data Assimilation in Turbulence

"In chaotic systems, limited knowledge of initial conditions causes simulated dynamics to rapidly diverge from reality, severely undermining predictability and state estimation. However, integrating real-world observations, even if spatiotemporally sparse and noisy, effectively mitigates this divergence. The functional and optimal use of such imperfect data constitutes the core of Data Assimilation, which we leverage to improve state estimation in fully developed turbulence, a physical regime that is not only inherently chaotic, but characterized by complex interactions across multiple spatiotemporal scales."

Speaker: Francesco Fossella (Telecom-Paris & University of Rome "Tor Vergata")

16:30 → 17:00 Coffee break

17:00 → 18:30 Poster Session

17:00 Direct coarsest level solves in DDalphaAMG

Efficient solutions of large sparse linear systems arising in lattice quantum chromodynamics (LQCD) are essential for modern numerical simulations. In this work, we investigate the use of a direct solver, based on ScaLAPACK, as the coarsest-level solver within the DDalphaAMG software package. Instead of performing an iterative solve on the coarsest grid of the multigrid hierarchy, we employ a direct factorization at this level.

This modification renders the multigrid preconditioner effectively stationary, which is advantageous for Krylov methods and substantially simplifies the treatment of multiple right-hand sides. The direct coarse solve avoids repeated coarse-level iterations and provides a reusable factorization for subsequent solves.

Current performance results indicate that the proposed method achieves solve times competitive with state-of-the-art LQCD solvers. Ongoing work focuses on the optimization of coarse-grid parameters and data distributions in order to further reduce the cost of the direct coarse solve and improve overall scalability.

These results demonstrate that a direct coarsest-level solver is a promising strategy for stationary multigrid preconditioning in LQCD and can provide an efficient framework for applications involving many right-hand sides within the DDalphaAMG software package.

Speaker: Henning Leemhuis (Bergische Universität Wuppertal)

17:00 Flavor decomposition of the nucleon axial and tensor charges from lattice QCD

We determine the nucleon axial and tensor charges at the continuum limit by analyzing four $N_f=2+1+1$ twisted mass fermion ensembles, with all quark masses tuned to approximately their physical values. We include both connected and disconnected contributions to determine the isovector and isoscalar charges and their flavor decomposition. Systematic uncertainties from excited-state contamination and the continuum extrapolation are estimated using the Akaike Information Criterion. The axial charge provides crucial information on the intrinsic spin carried by quarks in the nucleon, while the tensor charge is an important input for searches for physics beyond the Standard Model.

Speaker: Christos Iona (University of Cyprus & The Cyprus Institute)

17:00 High-Fidelity CFD and Reduced-Order Modelling of Transient Capillary Filling in Microchannels

Surface tension driven flows in microchannels underpin numerous microfluidic technologies, including biomedical diagnostics, chemical processing and lab on a chip systems. High-fidelity computational fluid dynamics (CFD) simulations provide detailed predictions of the transient evolution of the liquid–gas interface and associated flow field, but their computational cost limits large-scale parametric studies and optimisation. This work presents a physics-based reduced-order model (ROM) for transient capillary filling, validated against CFD simulations.
High-fidelity CFD simulations are performed within the OpenFOAM and Basilisk frameworks using Eulerian–Eulerian and VoF formulations, respectively, to resolve transient capillary filling in rectangular microchannels. In parallel, a ROM is developed by coupling a nonlinear boundary-value problem based on general lubrication theory with Navier-slip regularisation to a Lucas–Washburn type ordinary differential equation governing liquid-column evolution. The resulting coupled BVP–ODE algorithm reconstructs the transient meniscus shape and curvature while simultaneously advancing the filling dynamics.
The ROM accurately reproduces the transient evolution of the meniscus shape, apparent contact angle and overall filling dynamics. Equivalent simulations are completed in approximately two minutes compared with several days for CFD, enabling rapid parametric studies while retaining good agreement with detailed interface dynamics.

Speaker: Georgios Pasias (University of Cyprus)

17:00 High-performance simulations of higher representations of Wilson fermions

We present HiRep v2, an open-source software suite for high-performance lattice field theory simulations with dynamical Wilson fermions in higher representations of $SU(N_g)$ gauge groups. This new version fully supports graphics processing unit (GPU) acceleration, optimizing both gauge configuration generation and measurements for NVIDIA and AMD GPUs. HiRep v2 integrates improved gauge and fermionic lattice actions, advanced inverters, and Monte Carlo algorithms, including the (Rational) Hybrid Monte Carlo ((R)HMC) with Hasenbusch acceleration. It exhibits excellent scalability across multiple GPUs and nodes with minimal efficiency loss, making it a robust tool for large-scale simulations in physics beyond the Standard Model.

Speaker: Sofie Martins (University of Graz)

17:00 Higher Mellin Moments of Pion and Kaon Parton Distributions from Gradient Flow

We present an ongoing lattice-QCD study of higher Mellin moments of pion and kaon parton distribution functions using the gradient-flow method. Building on previous work, we use twisted-mass ensembles with lighter pion masses and plan calculations at four lattice spacings. We investigate the flow-time dependence, continuum extrapolation, excited-state effects, and other systematic uncertainties with a more detailed analysis. Preliminary results for both the pion and kaon are expected to be presented at the conference.

Speaker: Bolun Hu

17:00 Imaginarity as a necessary resource for trainability in QAOA

The quantum approximate optimization algorithm (QAOA) prepares candidate solutions to combinatorial optimization problems by alternating between operations that encode the objective function and a mixing operation that shift probability between candidate solutions. Gradient-based training of such circuits requires estimating how the measured cost changes when a circuit angle is varied. We focus on the gradient with respect to the final mixing angle and ask which entries of the quantum state can contribute to it. For any diagonal cost Hamiltonian and any mixer that is real in the cost basis, this gradient is determined only by imaginary coherences between basis states connected by the mixer.
These mixer-edge coherences therefore give an upper bound on the gradient magnitude and are necessary for a nonzero gradient, although different edge contributions can cancel. For the common QAOA mixing operation that flips one bit at a time, the relevant coherences connect bitstrings that differ by one bit. We illustrate the bound on Max-Cut, a standard graph-partitioning problem, using exact simulations of small random regular graphs with common terminal noise channels. In this setting, the gradient is governed by how the noise changes the effective cost observable; coherence left in the final noisy state provides a separate diagnostic.

Speaker: Syed Muhammad Ali Hassan (The Cyprus Institute)

17:00 Machine Learning Identification of Hydrogeological Patterns in a Complex Dinaric Karst

Karst aquifers are complex systems characterised by nonlinear recharge responses and heterogeneous groundwater flow. This study investigates whether unsupervised machine learning can independently distinguish deep karst from shallow karst (fluviokarst) in the northern Croatian Dinaric region.
A multiparametric dataset comprising 5-year records of spring discharge, water temperature and precipitation, complemented by hydrochemical and stable isotope observations, is used to derive descriptors of groundwater response, storage, interaction between rock and water, as well as residence time. Principal component analysis and unsupervised clustering are applied without prior information on the hydrogeological classification of the monitored springs. The resulting data driven groups are compared with an independently established conceptual hydrogeological model.
The analysis examines whether distinct clusters correspond to springs draining the deep karst aquifers of the Kapela Mountain massif and the downstream shallow karst system. The contribution of hydrodynamic and hydrochemical variables is evaluated to determine which parameters provide the strongest discrimination between aquifer types.
By combining hydrogeological knowledge with machine learning, this study explores the potential of environmental time series and hydrochemical data for objective classification and data driven characterisation of complex karst groundwater systems.

Speaker: Ivana Boljat (Croatian Geological survey)

17:00 Multiplier-Based Colorings for Exact Truncated Hopping-Parameter Expansion Terms in Lattice-QCD Trace Estimation

We study trace estimation for disconnected quark loop contributions in lattice QCD, focusing on single-time-slice traces such as $ \operatorname{Tr}\left[\Gamma_5(t)D^{-1}(t,t)\right]$. We build on the standard use of hopping-parameter expansions (HPE) in trace estimation, where the truncated term has bounded graph range and can be evaluated exactly using distance-(d) probing.

We first investigate multiplier-based colorings as an alternative to hierarchical probing for the exact truncated HPE contribution. The advantage becomes most relevant at larger expansion orders, where exact evaluation requires higher coloring distances and therefore more probing vectors. Since multiplier-based colorings can achieve the same distance with fewer colors, higher HPE orders can be used at lower cost. This removes more short-distance contributions exactly before estimating the stochastic remainder, whose inverse application is solved with multigrid-preconditioned FGMRES.

We also study replacing the classical Neumann-series HPE polynomial by a GMRES polynomial. While the Neumann expansion is fixed by the chosen splitting, the GMRES polynomial is constructed from the action of the Dirac operator and can adapt more effectively to non-ideal spectral distributions.

Speaker: Jose Miguel Jimenez Merchan (Bergische Universität Wuppertal)

17:00 Nucleon helicity moment and Generalized form factors from lattice QCD at physical pion mass

We present the results of the nucleon helicity moments and generalized form factors in the isovector and isoscalar channels computed within lattice QCD. We analyzed four ensembles with $N_f=2+1+1$ twisted mass clover-improved fermions with the mass of the quarks tuned to their physical mass. The four ensembles have lattice spacings approximately $a=0.05 \,\text{fm}, 0.06 \,\text{fm}, 0.07 \,\text{fm}, 0.08 \,\text{fm}$ that allow us to obtain results at the continuum limit directly at physical pion mass. We computed axial-vector generalized form factors as a function of momentum transfer square $Q^2$. We study the $Q^2$ dependence of the GFFs using the dipole ansatz and model-independent $z-$expansion. We find the nucleon isovector helicity moment $\langle x\rangle_{\Delta u-\Delta d} = 0.1871(82)$ and isoscalar helicity moment $\langle x \rangle_{\Delta u+\Delta d} = 0.1186(60)$

Speaker: Priyajit Jana (University of Cyprus, The Cyprus Institute)

17:00 Probing the Nucleon Mass from Light to Charm

The origin and flavor structure of the nucleon mass are fundamental questions in strong interaction physics. The nucleon sigma terms quantify the response of the nucleon mass to variations of the quark masses and provide direct access to its scalar matrix elements. We present a lattice QCD determination of the light, strange, and charm sigma terms using twisted-mass fermions at the physical pion mass and multiple lattice spacings. Both connected and quark disconnected contributions are included, with particular attention to excited-state contamination and the continuum extrapolation. Studying the three flavors within a common framework allows us to probe qualitatively different regimes of QCD, ranging from chiral dynamics in the light quark sector and sea quark effects for strangeness to the onset of heavy quark dynamics for charm. We discuss the flavor dependence of the scalar response of the nucleon and the associated systematic effects, providing a unified view of nucleon sigma terms from light to charm.

Speaker: Srijit Paul (University of Cyprus)

17:00 Spin and momentum fraction carried by partons in the nucleon

We determine the momentum fraction and angular momentum carried by quarks and gluons in the proton in lattice QCD. We use four ensembles simulated with up, down, strange and charm quarks with their masses tuned to their physical values.
These ensembles have similar physical volume and different lattice spacings allowing us to take the continuum limit directly at the physical pion mass point.
We extract the quark and gluon momentum fractions and total angular momentum in the continuum limit as well as the intrinsic quark spin and orbital angular momentum contributions to the proton spin. We find the total momentum fraction $\langle x \rangle_N = 0.995(60)(29)$ and the total spin $J_N = 0.507(43)(65)$, showing that both the momentum and spin sum rules are satisfied. We compare our results to those extracted from phenomenological analyses.

Speaker: Yan Li

17:00 Study of the Λ-hyperon to proton transition in Lattice QCD

In this poster we present the study of Λ-hyperon to proton transition in Lattice QCD. We use the computed form factors from LQCD, and apply it to a number of problems. For example, we compute the decay rate of the electron semi-leptonic decay, and by combining it with the experimental data we obtain $V_{us}$. Furthermore, we use this decay channel to search for possible physics beyond the Standard Model, such as scalar and tensor interactions. Finally, we apply the same form factors to describe neutrino scattering and compare the predicted cross sections with measurements from several experiments.

Speaker: Andreas Konstantinou (University of Cyprus & The Cyprus Institute)

17:00 Surface wettability inverse design for programmable droplet transport and sorting

We present an inverse-design method that determines the chemical heterogeneity profile of a surface required to transport a droplet along a prescribed path. Droplet dynamics are simulated by employing a reduced-order model that reduces the lubrication equation to a system of ODEs for the Fourier harmonics of the contact line, and integrating this stiff system with an implicit adaptive Runge–Kutta scheme (Kvaerno 5/4) via diffrax. Designing the surface profile so that the simulated droplet reproduces a prescribed path is an inverse problem whose cost is dominated by the dimensionality of the design space. To make this problem tractable, we use adjoint gradients obtained using JAX's automatic-differentiation framework, which makes the computational cost largely independent of the number of design parameters. The chosen cost function penalizes the Fréchet distance between the simulated centroid trajectory and the prescribed path, as well as deformation of the contact line from a circle, the latter guarding against numerical instabilities. We validate the inverse-design method on reference paths of increasing complexity, from a straight line to a sinusoid, recovering centroid trajectories that accurately reproduce each target, with wall-clock times ranging from ~13 minutes to under 2 hours 40 minutes. We further demonstrate passive droplet sorting: droplets below or above a critical volume move in opposite directions under a single designed profile, all without the application of external fields or a droplet detection mechanism. The same differentiable-solver pattern applies broadly to PDE-constrained inverse design in computational fluid dynamics.

Speaker: Anastasios Zeniou (University of Cyprus)

17:00 SYSTEM-SIZE EFFECTS AND SCALING LAWS IN NANOSCALE DROPLET WETTING: A MOLECULAR DYNAMICS STUDY

Abstract

Wetting dynamics plays a central role in applications such as coating processes, microfluidics, and green energy technologies. Despite extensive theoretical work, several aspects of the influence of droplet size on wetting dynamics remain unclear, particularly how nanoscale dynamics evolve toward macroscopic behaviour as the system size increases.

In this contribution, we present a detailed molecular dynamics study of droplet wetting on silica surfaces, focusing on the role of surface wettability and finite-size effects on macroscopic scaling laws.

Initially spherical droplets with varying radii are placed in contact with silica substrates and simulated to capture their spreading behaviour. The substrate wettability is tuned by introducing hydroxyl ions at different concentrations, resulting in equilibrium contact angles spanning the hydrophilic regime from $0^\circ$ to approximately $90^\circ$. The surface ion density, $C \approx 0$--$9.4\,\mathrm{nm^{-2}}$, is varied at the silica--water interface. Complete wetting, corresponding to an equilibrium contact angle of approximately $0^\circ$, is observed at $C \approx 4.7\,\mathrm{nm^{-2}}$.

For each wettability condition, the temporal evolution of the droplet morphology is analysed in detail, with the final equilibrium configurations found to be consistent with spherical cap geometries. At early times, the simulations recover the well-established inertial scaling behaviour associated with neck formation, following a $t^{1/2}$ power law. At later times, the dynamics transition towards a viscous-dominated spreading regime characterized by Tanner's law, with a $t^{1/10}$ scaling, which emerges when the system size is sufficiently large.

Our results demonstrate how finite-size effects influence the emergence of macroscopic wetting dynamics from nanoscale molecular processes. The findings provide quantitative guidelines for designing molecular simulations across length scales and contribute toward a systematic multiscale framework for understanding wetting phenomena.

Speaker: Mr Naveen Kumar Subramanian (The Cyprus Institute)

17:00 The $\Delta$ resonance at the physical point

The $\Delta(1232)$ resonance is the lowest-lying baryon excitation and provides an important testing ground for our understanding of hadron structure. In this talk, we will present recent lattice-QCD results on the resonance at the physical pion mass.

Using gauge ensembles with $N_f=2+1+1$ dynamical quark flavors, we investigate both the finite-volume spectrum of the resonance and its electromagnetic structure through the ($N\rightarrow \Delta$) transition. Preliminary results for the magnetic dipole (M1), electric quadrupole (E2), and Coulomb quadrupole (C2) transition form factors will be presented, together with a discussion of excited-state effects and the extraction methodology based on optimized three-point correlation functions.

These studies represent an important step toward a first-principles determination of the properties and internal structure of the $\Delta$
resonance directly from QCD.

Speaker: Ferenc Pittler (The Cyprus Institute)

17:00 The Diagonal Kenney-Laub Rational Approximation to the Overlap Operator

We present a practical approach for implementing the overlap Dirac operator in Lattice QCD, combining the diagonal Kenney-Laub (KL) rational iterates for approximating the matrix sign function with the Brillouin operator as overlap kernel. The partial fraction decomposition of the KL iterates enables multi-shift conjugate gradient solvers, with all weights and shifts given by closed-form trigonometric expressions that depend only on the approximation order. No extreme eigenvalue estimates are required, neither for determining the decomposition constants nor for rescaling the overlap operator kernel.

Preliminary benchmarking against the Zolotarev optimal rational and Chebyshev polynomial approximations shows that the KL method delivers competitive efficiency with a qualitative advantage: convergence with approximation order is smooth and monotonic on all diagnostic quantities tested, including the Ginsparg-Wilson violation, PCAC mass, and critical bare mass, with indications of reduced statistical noise. The KL-Brillouin combination consistently reaches a target precision at lower computational cost than the Wilson-kernel counterpart, with the Brillouin kernel's improved spectral conditioning compensating for its higher per-application cost. With minimal parameter tuning, no spectral input, and predictable convergence behavior, the method offers a straightforward alternative for calculations in which exact or near-exact chiral symmetry is essential.

Speaker: Mr Stylianos Gregoriou (The Cyprus Institute)

Wednesday 30 September

09:30 → 11:00 Invited

09:30 Making quantum machine learning separations more physical

One of the core challenges of research in quantum computing is to understand whether quantum advantages can be found for problems of practical interest. In the field of quantum machine learning, we know for a few years now that proof-of-concept exponential advantages can be established in learning tasks derived from classically hard problems such as factoring or computing discrete logarithms. In this talk, I will expose recent efforts in making these learning separations more physically and practically relevant. Namely, I will present a first result establishing how non-local correlations can be at the origin of learning advantages in the experimentally-friendly realm of shallow-depth circuits. I will also explain how computationally-universal quantum many-body dynamics can be efficiently learned on a quantum computer, while remaining classically intractable under standard complexity assumptions.

Speaker: Sofiene Jerbi (Freie Universität Berlin - Helmholtz Zentrum Berlin)

10:15 A Principled Approach to Enhanced Sampling

Molecular Dynamics simulations are a valuable tool for investigating a wide variety of phenomena at the atomistic scale. However, it is this very high resolution that limits the time scales accessible to such simulations. Typically, this limitation is overcome by deploying methods that require knowledge of the system's metastable states and an approximation of the reaction coordinate, referred to as a collective variable.
Here, we propose a new enhanced sampling method which requires neither.
We build our approach on the observation that in rarely occurring events like phase transition or protein folding processes, there is a trade-off between enthalpy and entropy. It is rare fluctuations in these quantities which eventually result in a transition. Thus, we describe the state of the system via the distributions of single particle energies or of appropriately defined per-atom entropies. We then devise an algorithm that favours sampling the distribution tails, thus facilitating high-energy barrier crossing and the discovery of new states, with null or minimal prior knowledge of the system.
We demonstrate the generality of our approach by successfully simulating protein folding, ligand binding to receptors, and crystallising an undercooled metal melt starting just from the unfolded state, the apo state, and the liquid state respectively.

Speaker: MICHELE PARRINELLO (Fondazione istituto italiano di tecnologia)

11:00 → 11:30 Coffee break

11:30 → 13:00 Invited

11:30 Lattice Boltzmann Algorithms for complex fluids and complex flows

The Lattice Boltzmann Method (LBM) has evolved into one of the most versatile computational frameworks for simulating complex fluids and multiscale flow phenomena. Its kinetic formulation naturally accommodates multiphase and multicomponent flows, thermal transport, fluid–structure interactions, porous media flows, and turbulence, while offering excellent scalability on modern high-performance computing architectures.

This talk provides an overview of the current state of the art in advanced Lattice Boltzmann methods, with a particular focus on recent developments in collision operators and higher-order kinetic models.
Moreover, we will cover opportunities at the intersection between kinetic theory and machine learning. We will discuss the potential of learning novel collision operators using neural networks constrained by fundamental physical principles, including conservation laws and symmetry requirements. This physics-informed approach may open the way to a new generation of Lattice Boltzmann methods with enhanced stability and accuracy for complex fluid and multiscale flow simulations.

Speaker: Alessandro Gabbana (University of Ferrara / INFN Ferrara)

12:15 Physics HPC Lattice QCD

Speaker: Simone Bacchio

13:00 → 17:00 Free afternoon

Thursday 1 October

09:30 → 11:00 Invited

09:30 Hadron Structure

Speaker: Gunnar Bali

10:15 Quantum algorithms for classical fluids

In the recent years increasing attention has been directed towards the
the development of quantum algorithm for classical physics, most
notably fluid dynamics and other nonlinear transport phenomena.

Solving fluid dynamics on quantum computers is a steep challenge
on top of a challenge because, besides the notorious hurdles of
decoherence and loss of entanglement, the physics of fluids is
generally nonlinear and dissipative, hence it does not map
directly onto the unitary gates of quantum computers.
Hence, non trivial extra-steps need to be taken in order to formulate an
algorithm describing nonlinear and dissipative physics within a linear and
unitary mathematical harness.\
In this talk we shall mention the various techniques that have been proposed to circumvent the above problems, with special emphasis on the combination of Carleman linearization with block-encoding to cast fluid dynamics into a a infinite-dimensional linear and unitary framework amenable to quantum computing.

Speaker: Sauro Succi (Italian Institute of Technology)

11:00 → 11:30 Coffee break

11:30 → 12:30 Invited

11:30 Quantum Computing

Speaker: Ilaria Siloi

12:00 New directions in interpreting AI models

Mechanistic interpretability aspires to reverse engineer AI models by breaking down black-box weights and activations into human-understandable features and circuits. A leading approach for mechanistic interpretability is sparse dictionary learning, which trains an encoder to map activations into a sparse code and a decoder to reconstruct them. The corresponding architecture is called sparse-autoencoder (SAE), and it has been shown to uncover safety-relevant concepts such as deception, bias, and harmful content, enabling targeted interventions on model behavior. In this talk, we will go through applications of SAEs in language and vision models, as well as present new SAE variants based on tensor decompositions that enhance expressivity at a fraction of the parameter cost.

Speaker: Andreas Demou (The Cyprus Institute)

12:30 → 13:00 ESR

12:30 Wetting problems using advance mesoscale modeling

The study of wetting is a captivating problem laying at the intersection between physics, chemistry and engineering. The multi-scale nature of this phenomenon makes it challenging to model, calling for advanced numerical techniques. We present an immersed boundary-lattice Boltzmann (IB-LB) method [[1]] to tackle this task, improving on existing work [[2]] in scope and applicability, in order to reproduce droplets on a horizontal, homogeneous solid substrate ranging from hydrophobic to hydrophilic wetting regimes.

The droplet's non-ideal sharp interface, modelled via the immersed boundary (IB) method, is coupled to the inner and outer fluids resolved via the lattice Boltzmann (LB) method; the wetting interaction with the substrate is achieved through a force term designed with the key computational advantage of providing a regularization of the interface profile close to the contact line, avoiding abrupt curvature changes that would cause numerical instabilities.

Extensive model validations against analytical results for equilibrium droplet shape and scaling laws for droplet spreading dynamics are addressed. Furthermore, comparisons against other independent solvers are presented to investigate the hydrodynamic behaviour of the IB-LB method in relation to the implemented contact-line model [[3]]. Finally, the efficient generation of high-fidelity data with the IB-LB method is briefly explored in the context of machine learning for performance acceleration as an ongoing project.

---

This research is supported by the European Union's HORIZON MSCA Doctoral Networks programme, under Grant Agreement No. 101072344, project AQTIVATE (Advanced computing, QuanTum algorIthms and data-driVen Approaches for science, Technology and Engineering).

[[1]]: Bellantoni et al., (2025), Physical Review E 112, 025305, doi.org/10.1103/mp3p-8j22.
[[2]]: Pelusi et al., (2023), Physics of Fluids 35, 082126, doi.org/10.1063/5.0160096
[[3]]: Bellantoni et al., (2026), arXiv,2604.17463, doi.org/10.48550/arXiv.2604.17463

Speaker: Elisa Bellantoni (The Cyprus Institute, Tor Vergata University of Rome, Τélécom Paris)

13:00 → 14:30 Lunch break

14:30 → 15:00 Invited

14:30 Physics-constrained diffusion model for synthesis of 3D turbulent data

Synthesizing fully developed three-dimensional turbulent velocity fields remains a long-standing problem in fluid mechanics and an open challenge for generative modeling. This difficulty arises from the combination of extreme dimensionality, multiscale fluctuations, strong intermittency, and the need to satisfy exact physical constraints, including incompressibility and prescribed mass and momentum fluxes under given boundary conditions. We propose a physics-constrained diffusion model (PCDM) in which some a priori constraints can be incorporated directly into the generative dynamics. Using rotating turbulence as a paradigmatic open problem, with key applications to geophysical contexts, we show that the proposed framework enables stable and statistically faithful synthesis of inertial-range three-dimensional turbulent velocity fields, accurately reproducing anisotropic energy spectra, intermittent statistics, and physical constraints. By contrast, standard denoising diffusion probabilistic models without such constraints exhibit multiscale statistical deviations, violations of physical consistency, and substantially slower training convergence. These results point to broader implications for generative modeling of high-dimensional, physically constrained complex systems.

Speaker: Prof. Michele Buzzicotti (University of Rome Tor Vergata and INFN)

15:00 → 16:00 ESR

15:00 Solving QUBO problems with tensor networks

Quadratic unconstrained binary optimization (QUBO) problems arise naturally in spin glass models like the Sherrington-Kirkpatrick Hamiltonian, where strongly frustrated interactions create rugged energy landscapes. In this talk, I explore how the density matrix renormalization group (DMRG), the famous tensor network method used as a ground-state search algorithm, can be adapted as a heuristic QUBO solver. By introducing a weak transverse field, DMRG gains the ability to escape local minima. I will also discuss how these results compare to those of Gurobi optimizer, a state-of-the-art classical solver that struggles with strongly frustrated dense problems, and whether DMRG can offer an advantage.

Speaker: Rocco Barač (University of Padova & University of Cyprus)

15:30 Enabling Biomolecular Simulations with Neural Network Potentials in GROMACS

Hybrid machine learning/molecular mechanical (ML/MM) simulations are increasingly explored as an alternative to computationally expensive quantum mechanical (QM) methods, offering near first-principles accuracy at substantially reduced computational cost. Neural network potentials (NNPs) can reproduce QM-level energies and forces efficiently, yet their integration into established molecular dynamics (MD) engines has often been limited. To address this, we here present an NNP-interface implemented in the widely used MD code GROMACS. The interface enables NNPs trained in the PyTorch framework to contribute energies and forces during MD simulations, either for selected subsets or entire molecular systems. In particular, the design integrates NNP inference seamlessly into the extensive GROMACS molecular simulation ecosystem, providing users with the capability to straightforwardly combine NNPs with existing advanced sampling and free energy workflows.
In this talk, we will review the basic concepts of neural network potentials, and demonstrate the capabilities of our interface using several representative applications, including performance benchmarks, enhanced sampling of peptide torsional free energy landscapes, absolute solvation free energy calculations, and protein-ligand simulations.
Lastly, we will present ongoing work on a large membrane protein in the mammalian brain, illustrating how NNPs are expanding the possibilities of studying complex biomolecular processes.

Speaker: Lukas Müllender (KTH Royal Institute of Technology)

16:00 → 16:30 Coffee break

16:30 → 17:30 Invited

16:30 Dynamics of metastable states in quantum systems

Metastable states appear across many areas of physics, from condensed matter to cosmology. Their relaxation is described by the semi-classical 'critical bubble theory', developed more than 50 years ago. Despite its broad applicability, the quantum version of the theory has little experimental support, and understanding the relaxation of metastable states in quantum many-body systems remains a challenge.

I will present our recent progress in studying metastable-state decay in two-dimensional quantum spin systems. By combining tensor-network simulations with semiclassical field-theory calculations, we investigate false-vacuum decay through the nucleation and subsequent growth of bubbles of the competing phase. I will discuss how this approach connects microscopic many-body dynamics with the predictions of critical-bubble theory, and conclude with the prospects for using quantum simulators to explore these processes experimentally.

Speaker: Luka Pavesic (University of Padova)

17:00 Lattice QCD

Speaker: Antonio Evangelista

17:30 → 18:00 ESR

17:30 A posteriori LES closure of buoyancy driven turbulent flows

Buoyancy-driven turbulence arises across a wide range of natural and technological systems, from inertial confinement fusion to stellar explosions. Direct numerical simulation of all dynamically active scales is often prohibitively expensive, motivating large-eddy simulation and the development of accurate subgrid closures. We extend recent work on a solver-in-the-loop framework in which the governing equations are embedded directly into the training of neural-network closures for unresolved terms. The approach is investigated for homogeneous Rayleigh–Bénard convection and Rayleigh–Taylor turbulence in both two and three dimensions. We evaluate the learned models a posteriori through long-time integrations, focusing on the probability distributions of energy and scalar-variance fluxes, high-order statistics, and the limitations of conventional closure models in reproducing these quantities. Finally, we examine how the unrolled training horizon influences accuracy and stability, relating the required time in the loop to relevant time scales of the flow.

Speaker: André Freitas (University of Rome "Tor Vergata" & Institut Polytechnique de Paris)

18:00 → 18:40 Poster prize award

18:00 Poster prize award talk #1

18:20 Poster prize award talk #2

20:00 → 22:00 Conference dinner

Location TBD

Friday 2 October

09:30 → 11:30 Invited

09:30 Advances in Algebraic Multigrid Methods for PDEs

Algebraic multigrid (AMG) is a popular and effective solver for sparse linear systems arising from discretized partial differential equations (PDEs). The optimality and efficiency of AMG rests on the complementary relationship between relaxation (e.g., Gauss-Seidel) and interpolation, which when effective, results in optimal O($n$) scaling in the number of degrees-of-freedom $n$. Relaxation removes high-energy error, while interpolation maps low-energy error to a coarser space where it is reduced. While AMG is a relatively mature field for scalar PDEs that result in symmetric positive definite discretizations (SPD), such as the Poisson equation, there are a number of important matrix classes (e.g., nonsymmetric, indefinite, and discretizations of PDE systems) that remain problematic for AMG. In this talk, we review a number of developments that have extended the state-of-the-art for AMG beyond standard SPD problems. This includes (i) energy-minimization based interpolation and advanced algebraic coarsening strategies for convection-diffusion problems, highly anisotropic diffusion, discontinuous Galerkin discretizations, and indefinite problems, (ii) approximate ideal restriction (AIR) based methods for purely convective and highly nonsymmetric problems, (iii) adaptive AMG for detecting the near kernel of especially challenging matrices, and (iv) AMG approaches for PDE systems such as curl-curl and Stokes equations.

Speaker: Jacob Schroder (University of New Mexico)

10:15 QM/MM MD simulations of proton-coupled transporters

We will discuss some recent applications of our massively parallel QM/MM code MiMiC, developed within a consortium of european university. We will focus on transporters which exploit proton gradients across the membrane to transport organic molecules.
We will close with a perspective of using data science approaches to develop potentials and calculate accurate free energies using QM/MM MD data.

Speaker: Paolo Carloni (Forschungszentrum Jülich GmbH)

10:45 Hamiltonian methods from Quantum Information Technologies: from Z2 String Dynamics to Continuous U(1) Electrodynamics

Real-time, non-perturbative dynamics of gauge theories, from string formation and fragmentation to thermalization and jet production, lie largely beyond the reach of standard Euclidean Monte Carlo methods. Hamiltonian formulations of lattice gauge theories, combined with tools from quantum information science, offer a complementary route that provides direct access to time evolution, avoids the sign problem, and enforces gauge invariance at the level of the physical Hilbert space.

In this colloquium, I present three interconnected works that systematically develop this program, using the Z2 and U(1) gauge groups as a ladder of increasing complexity. In the first, we use matrix product state (MPS) methods to study the roughening transition of an electric flux string in a (2+1)-dimensional Z2 lattice gauge theory. Working in the Hamiltonian framework, we obtain the universal Lüscher correction to the confining potential, extract a central charge c = 1 consistent with an effective free-boson description of the rough string, confirm the restoration of rotational symmetry, and reveal qualitatively distinct real-time entanglement dynamics in the roughening and strongly-confined regimes, which are inaccessible to Euclidean approaches. In the second, we extend to the Z2-Higgs model with dynamical matter and implement it on a superconducting quantum processor with up to 144 qubits and a circuit depth of 192 two-qubit layers. Exploiting local gauge symmetry for error suppression and mitigation, we resolve a dynamical hierarchy between longitudinal string oscillations and transverse endpoint-bending modes, precursors to hadronization and meson rotational spectra, and observe multi-string fragmentation and recombination. In the third work, we propose an analog superconducting-circuit architecture that realizes compact U(1) lattice gauge theory using the intrinsic infinite-dimensional Hilbert space of Josephson-junction phase and charge variables. Gauss's law follows exactly from Kirchhoff's current conservation, with no truncation, penalty terms, or auxiliary stabilizers, and numerical diagonalization confirms the emergence of compact electrodynamics and coherent vortex excitations. Together, these results establish a complementary triad of classical tensor networks, digital NISQ hardware, and analog circuit design as a viable and scalable strategy for probing non-perturbative gauge dynamics in real time.

Speaker: Enrique Rico Ortega (CERN)

11:30 → 12:00 Coffee Break

12:00 → 13:00 ESR

12:00 HPC-oriented Molecular Dynamics simulations to address challenges in systems of pharmacological relevance

High-performance computing has turned atomistic Molecular Dynamics (MD) into a practical tool for resolving biomolecular processes with atomic resolution. For protein-ligand systems, these simulations expose the interactions governing binding and unbinding, a key issue in drug design. Drug efficacy, however, depends not only on binding affinity but also on residence time (RT), the kinetic parameter describing how long a ligand stays bound. Experiments can measure RT but cannot resolve its molecular basis, a gap atomistic MD can fill by mapping the free-energy landscape and dissociation pathway together.
That gap becomes acute when mutations in a target alter RT while leaving affinity essentially unchanged, implying that the transition-state (TS) ensemble along the unbinding pathway is selectively perturbed while the bound state is preserved. We examine this in the human adenosine A2A receptor bound to the antagonist ZM241385, a GPCR-ligand system relevant to Parkinson's disease and immuno-oncology. Using metadynamics-based enhanced sampling approaches, we reconstruct both the binding free-energy landscape and the dissociation kinetics of the wild-type receptor and mutants that reduce RT while preserving affinity. Our simulations reproduce the experimental RT differences and show that mutations act primarily by reshaping the TS region, providing a structural basis for rational, RT-optimized ligand design.

Speaker: Marta Devodier

12:30 Towards Understanding Performance of Task-Based Models on Modern HPC Systems

As the slowdown of Dennard scaling has shifted performance gains toward exploiting parallelism on increasingly heterogeneous many-core architectures, task-based programming models have emerged as a promising approach for managing complex workloads. Despite their growing adoption, however, understanding how runtime behavior translates into application performance remains a significant challenge.

To better understand the performance of task-based applications, we investigate the key performance drivers using both an irregular and a regular workload. Using LU decomposition as an irregular case study, we develop an application-specific analytical model that accurately predicts the optimal task granularity, eliminating the need for costly auto-tuning. We validate the analysis across OmpSs-2, GCC OpenMP, and HPX on both x86 and Arm platforms, demonstrating consistent performance trends across programming models and architectures. We then examine the remaining discrepancies between modeled and measured performance, attributing them to runtime scheduling decisions, including task placement and overheads. Finally, using a 2D stencil benchmark analyzed under the Roofline model, we isolate these runtime effects in a regular workload and show that task overprovisioning and NUMA-aware execution are critical for achieving high performance on modern multi-NUMA systems. Together, these two case studies illustrate how analytical modeling and runtime characterization provide complementary insights into the performance of task-based applications.

Speaker: Shiting Long

13:00 → 13:30 Closing