Conveners
Algorithms and artificial intelligence
- En-Hung Chao (JGU Mainz)
Algorithms and artificial intelligence
- James Osborn
Algorithms and artificial intelligence
- Shailesh Chandrasekharan (Duke University)
Algorithms and artificial intelligence: I
- Masafumi Fukuma (Kyoto University)
Algorithms and artificial intelligence: II
- Letizia Parato (University of Colorado - Boulder)
Algorithms and artificial intelligence
- Matteo Saccardi (Colorado State University)
Algorithms and artificial intelligence
- Thomas Spriggs (Delft University of Technology)
Algorithms and artificial intelligence
- Daniel Hackett
Algorithms and artificial intelligence
- Francesco Di Renzo (University of Parma and INFN)
Algorithms and artificial intelligence
- Fernando Romero Lรณpez (Uni Bern)
Algorithms and artificial intelligence
- Xiao-Yong Jin
-
Shunsuke Yasunaga (Institute of Science Tokyo / RIKEN)27/07/2026, 14:001Algorithms and artificial intelligenceContributed talk
Domain-wall fermions provide a lattice formulation that preserves chiral symmetry to a high degree by introducing an additional fifth dimension. In practical simulations, however, the extent of this direction must remain finite, which leads to residual chiral symmetry breaking characterized by the residual mass. Increasing the fifth-dimensional size can reduce this effect, but it also...
Go to contribution page -
Thamirys de Oliveira (National Yang Ming Chiao Tung University)27/07/2026, 14:20Algorithms and artificial intelligenceContributed talk
In this talk we introduce a machine learning approach to tune an effective relativistic heavy quark action for applications in lattice QCD.
Go to contribution page
The effective action is the so-called "RHQ" action, which has three open parameters that require non-perturbative tuning to match physics observables associated with Lorentz symmetry, spin-averaged masses, and hyperfine splittings in charmonium and... -
Simon Pfahler (University of Regensburg)27/07/2026, 14:402Algorithms and artificial intelligenceContributed talk
Gauge-field generation in lattice QCD is dominated by repeated solves of the Dirac equation. While multigrid preconditioners reduce solve costs significantly, their expensive setup phase limits overall efficiency. We introduce a gauge-equivariant neural network that accelerates this setup by using approximate low modes from previous configurations, exploiting the autocorrelation along the...
Go to contribution page -
Mr Jonah Eick (Columbia University)27/07/2026, 15:002Algorithms and artificial intelligenceContributed talk
We report on the current status of the renormalization-group preconditioned conjugate gradient (RGPCG) for domain wall fermions (DWF), an algorithm that leverages the correspondence between low modes of the Dirac operator on ensembles related by RG-blocking to produce a preconditioner with small setup overhead. We have studied this approach for Mรถbius DWFs on a fine $a^{-1} = 2$ GeV lattice,...
Go to contribution page -
Dr Peter Boyle27/07/2026, 15:205Algorithms and artificial intelligenceContributed talk
We study and visualize the process of topology change in field transformation hybrid Montecarlo, correlating gauge and fermion forces with changes in the chiral mode structure of the Domain Wall fermion operator. We see that in each trajectory there are multiple changes of topology by both flowed gauge measures and by the Domain wall fermion index. These appear to correspond to small scale...
Go to contribution page -
Ankur Singha (Technical University Berlin)27/07/2026, 16:102Algorithms and artificial intelligenceContributed talk
We propose a multilevel generative sampling framework for lattice gauge theories designed to overcome topological freezing by explicitly sampling all relevant topological sectors. The target distribution is decomposed into coarse and fine scales using an RG-inspired blocking strategy. At each level, the generative sampler produces plaquette degrees of freedom and reconstructs consistent...
Go to contribution page -
Roberto Dionisio (University of Pisa, INFN sez. di Pisa)27/07/2026, 16:307Algorithms and artificial intelligenceContributed talk
Efficient sampling across topological sectors is one of the central algorithmic challenges in lattice field theory. Approaches based on interpolating distributions, such as parallel tempering with defects and learned-flow methods, improve ergodicity by connecting an easily sampled reference system to the target theory, but their efficiency depends strongly on the choice of interpolation path....
Go to contribution page -
Gianluca Fuwa (Bergische Universitรคt Wuppertal)27/07/2026, 16:503Algorithms and artificial intelligenceContributed talk
As the continuum limit is approached, conventional update algorithms in lattice QCD and other topologically non-trivial theories suffer from a particularly severe form of critical slowing down, caused by high action barriers separating distinct topological sectors. Parallel tempered Metadynamics (PT-MetaD) has been shown to overcome this problem by pairing, in the simplest case, the physical...
Go to contribution page -
Victor Granados-Pinto (Universitรคt Bern)27/07/2026, 17:104Algorithms and artificial intelligenceContributed talk
In Lattice QCD, standard Markov Chain Monte Carlo (MCMC) algorithms exhibit topological freezing: As the continuum limit is approached, the autocorrelation times of topological observables increase exponentially. As a result, ergodicity is effectively lost and statistical error estimates become unreliable. A method to mitigate this is Parallel Tempering on Boundary Conditions (PTBC). The...
Go to contribution page -
James Osborn28/07/2026, 14:004Algorithms and artificial intelligenceContributed talk
We present a generalization of the dynamics used in HMC that allows for an arbitrary range of auxiliary field content along with freedom in choosing the dynamics (including non-symplectic dynamics), constrained mainly by the Metropolis-Hastings acceptance requirement. We will discuss how the generalized dynamics relates to other HMC variations that have appeared in the literature. We will...
Go to contribution page -
Robert Mawhinney (Columbia University)28/07/2026, 14:201Algorithms and artificial intelligenceContributed talk
In this talk we study the impact of altering the standard Hybrid Monte Carlo in numerous small subvolumes of a larger lattice. A primary alteration considered is to fix each subvolume to a maximal tree gauge, such as axial gauge, and then choose conjugate momenta that are not site-local in the subvolume, but rather have spatial coherence acress the subvolume. We focus on pure SU(3) gauge...
Go to contribution page -
Masafumi Fukuma (Kyoto University)28/07/2026, 14:402Algorithms and artificial intelligenceContributed talk
The Worldvolume Hybrid Monte Carlo (WV-HMC) method is designed to address the numerical sign problem while avoiding the ergodicity issues inherent in the Lefschetz thimble method. Using a general prescription for applying WV-HMC to group manifolds, I demonstrate its successful application to lattice gauge theories with complex actions.
Go to contribution page -
Xiao-Yong Jin28/07/2026, 15:003Algorithms and artificial intelligenceContributed talk
We explore how a change of variables impacts HMC sampling in lattice gauge theories. Using a 2D U(1) pure gauge system as a testbed, we introduce a local change of variables while keeping the computational cost of the Jacobian low. We analyze how different formulations affect the effective action and the effective force, comparing them to existing field transformation techniques in the...
Go to contribution page -
Yacob OZDALKIRAN (IJCLab, Orsay)28/07/2026, 15:202Algorithms and artificial intelligenceContributed talk
Based on https://arxiv.org/abs/2606.21217, we present two adaptations of the Event-Chain Monte Carlo algorithm for pure gauge lattice QCD. The algorithms rely on continuous deterministic updates interrupted by stochastic events, creating an irreversible, rejection-free Markov process that satisfies global balance. We detail the construction of these algorithms and present numerical results on...
Go to contribution page -
Prof. Shailesh Chandrasekharan (Duke University)28/07/2026, 16:102Algorithms and artificial intelligence
We present a continuous-time path-integral Monte Carlo method for computing the low-lying spectrum of generic quantum lattice Hamiltonians, motivated in part by applications to qubit regularizations of quantum field theories. The method is based on projecting the thermal density matrix, $e^{-\beta H}$, onto a subspace spanned by a chosen set of linearly independent states. It is free of...
Go to contribution page -
Dominic Schuh (University of Bonn)28/07/2026, 16:102Algorithms and artificial intelligence
The Hubbard model at finite chemical potential is a cornerstone for understanding doped correlated systems, but simulations are severely limited by the sign problem. In the auxiliary-field formulation, the spin basis mitigates the sign problem, yet severe ergodicity issues have limited its use. We extend recent advances with normalizing flows at half-filling to finite chemical potential by...
Go to contribution page -
Felicitas Freche28/07/2026, 16:303Algorithms and artificial intelligenceContributed talk
In this work, we extend the normalizing-flow-based generalized density-of-states (NF-gDoS) method to the doped Hubbard model. The Hubbard model is known to exhibit a sign problem in the presence of a chemical potential. The NF-gDoS framework is attractive because it reformulates the original complex-weight sampling problem by separating the complex phase from the Boltzmann distribution and...
Go to contribution page -
Dr En-Hung Chao (Massachusetts Institute of Technology)28/07/2026, 16:302Algorithms and artificial intelligenceContributed talk
We propose a hybrid reweighting method to calculate expectation values of observables in the presence of external fields coupled to dynamical fermions. The reweighting factor is calculated from the ratio of fermion determinants with and without the external field, evaluated on a background of Wilson-flowed dynamical gauge fields. Because the Wilson flow greatly reduces ultraviolet...
Go to contribution page -
Thomas Hauschild (Forschungszentrum Juelich)28/07/2026, 16:50Algorithms and artificial intelligenceContributed talk
Auxiliary-field quantum Monte Carlo methods provide a powerful route to unbiased simulations of strongly correlated quantum systems. While determinant quantum Monte Carlo has been highly successful, its standard formulation is naturally adapted to particle-number-conserving Hamiltonians. Pfaffian quantum Monte Carlo extends this framework to systems, that do not conserve particle...
Go to contribution page -
Janik Kreit (University of Bonn)28/07/2026, 16:504Algorithms and artificial intelligence
Generative modeling of strongly correlated fermionic systems has emerged as a promising complement to traditional quantum Monte Carlo methods. Building on recent demonstrations that normalizing flows can learn the Boltzmann distribution of the Hubbard model, we systematically investigate the path toward larger lattice sizes and lower temperatures. We identify the key challenges that arise in...
Go to contribution page -
Finn Temmen (Forschungszentrum Jรผlich, IAS-4)28/07/2026, 17:101Algorithms and artificial intelligenceContributed talk
Stochastic methods are indispensable tools for studying strongly correlated fermionic systems due to their far more favorable volume scaling than direct approaches such as exact diagonalization (ED) or tensor network methods. However, this improved scaling comes at the cost of new challenges, most notably the sign problem and long autocorrelation times, which severely restrict the accessible...
Go to contribution page -
Petar Sinilkov (IAS-4, Forschungszentrum Jรผlich)28/07/2026, 17:101Algorithms and artificial intelligenceContributed talk
Our central advance is an algorithm that simultaneously stabilizes the determinant and force calculations that holds to machine precision for very large $\beta$ (in units of inverse hopping) while scaling equally well with standard determinant QMC simulations of these systems. Furthermore, our formalism retains the full time-displaced (2-point) Greens function, allowing us to construct any...
Go to contribution page -
Alessandro De Santis (Helmholtz-Institut Mainz, Johannes Gutenberg-Universitรคt Mainz)29/07/2026, 09:006Algorithms and artificial intelligenceContributed talk
Spectral reconstruction is one of the most challenging and important problems in lattice QCD, as spectral functions are directly related to a wide range of phenomenologically relevant observables. In this talk, I present a novel strategy based on reformulating the reconstruction problem within the framework of Operator Learning using DeepONet neural networks. The network is trained in a...
Go to contribution page -
Demetrianos Gavriel29/07/2026, 09:201Algorithms and artificial intelligenceContributed talk
Reconstructing spectral information from Euclidean lattice correlators is a central inverse problem in lattice QCD. The correlator is related to the underlying spectrum through a Laplace-type transform, but inverting this relation from finite and noisy data is severely ill-conditioned. In this contribution, we discuss two complementary strategies for approaching this problem. The first...
Go to contribution page -
Francesco Di Renzo29/07/2026, 09:402Algorithms and artificial intelligenceContributed talk
Euclidean lattice correlators can be written as Laplace transforms of given spectral functions. Solving for the latter would thus simply amount to computing anti-transforms, which is unfortunately the prototype of an ill-posed problem. If one numerically approximates the integral transform via (e.g.) Gauss-Laguerre quadratures, the inverse problems is turned into a linear system, which is, not...
Go to contribution page -
Jose Alejandro Vidal Maxia (University of Parma & INFN)29/07/2026, 10:002Algorithms and artificial intelligenceContributed talk
We will discuss the analytical characterisation of the Laplace transform inversion problem turned into a finite-dimensional linear system. This is done via a Gauss-Laguerre quadrature rule and regularised by means of a Tikhonov prescription. This, in the light of solutions obtained by, e.g., M. Hansen, et al. & M. Saccardi et al. via the implementation of other methods/regulation prescriptions...
Go to contribution page -
Benjamin Luke (Baylor University)29/07/2026, 10:202Algorithms and artificial intelligenceContributed talk
Estimating the trace of the inverse of a large quark matrix is a longstanding problem in lattice QCD. Hutchinson's method is the standard for this calculation, which uses random noise vectors to achieve a desired tolerance. Doing so requires solving a system of linear equations $Ax = b$, where $b$ is the random noise vector. In this work, we introduce novel solving algorithms for use in...
Go to contribution page -
Patrick Oare (Brookhaven National Laboratory)29/07/2026, 11:102Algorithms and artificial intelligenceContributed talk
The Dirac operator is a fundamental object in lattice gauge theory. Its spectral properties carry direct physical significance: its low modes limit convergence in Krylov-based linear solvers, and real (zero) eigenvalues of the Wilson (overlap) Dirac operator correspond to topologically non-trivial field configurations. While the spectrum of Hermitian Dirac operators is real and easy to compute...
Go to contribution page -
Chulwoo Jung29/07/2026, 11:302Algorithms and artificial intelligenceContributed talk
Despite the intrinsic ambiguity in defining the topological index $Q$ on a discrete lattice, fermionic definitions based on either the Hermitian or non-Hermitian WilsonโDirac operator yield an integer-valued topological index for any lattice QCD configuration. Here, we focus on configurations in which the topological index changes during the Hybrid Monte Carlo evolution, and investigate the...
Go to contribution page -
choi minjae29/07/2026, 11:502Algorithms and artificial intelligenceContributed talk
The low-lying modes of the Wilson Dirac operator are key ingredients in deflation and multigrid algorithms for lattice QCD, where they define the near-null spaces responsible for critical slowing down. In this talk, we present a method for learning such low modes with a gauge-equivariant neural network trained on an ensemble of gauge configurations, using a RayleighโRitz loss that minimizes...
Go to contribution page -
Ho Hsiao (Center for Computational Sciences, University of Tsukuba)29/07/2026, 12:101Algorithms and artificial intelligenceContributed talk
Machine learning offers an alternative to conventional iterative algorithms for lattice gauge fixing, with the potential to reduce computational cost for large lattice volumes. Building upon our previous work, we perform a systematic scan of convolutional neural network architectures for lattice gauge fixing, in which the gauge transformation matrices are constructed from Wilson lines with...
Go to contribution page -
Renzo Kapust (Universitรคt Heidelberg)30/07/2026, 14:001Algorithms and artificial intelligenceContributed talk
In this talk we present a novel generative architecture for systems with complex probability distributions. In general, these sampling tasks come with two challenges: resolving sign problems and efficient sampling. The architecture is based on physics-informed kernels (PIKs) introduced in arXiv:2510.26678, and aims at resolving both challenges. Key to the complex PIK-architecture is its...
Go to contribution page -
Fernando Romero Lรณpez (Uni Bern)30/07/2026, 14:206Algorithms and artificial intelligenceContributed talk
The generating functional in quantum field theory provides the natural framework for constructing correlation functions as derivatives with respect to source operators. In this talk, I will present a methodology that leverages machine-learned normalizing flows to reduce the variance of arbitrary N-point correlation functions of bosonic operators in lattice gauge theory calculations by encoding...
Go to contribution page -
Elia Cellini (University of Edinburgh)30/07/2026, 14:404Algorithms and artificial intelligenceContributed talk
In recent years, the combination of neural network architectures and out-of-equilibrium methods has emerged as a powerful framework to build a new generation of numerical algorithms in various scientific fields. In lattice field theory, a prominent example is Stochastic Normalizing Flows, which have been used to tackle topological freezing in $\textrm{SU}(3) $ gauge theory. In this talk, we...
Go to contribution page -
Octavio Vega (University of Illinois Urbana-Champaign)30/07/2026, 15:005Algorithms and artificial intelligenceContributed talk
Sampling with flows and diffusion models has emerged as a promising alternative to MCMC in lattice field theory. A central obstacle to their practical adoption is the degradation of sample quality as the lattice volume grows, and identifying a scalable prescription for applying deep generative models to the lattice setting is still an open problem. We adapt and extend a framework based on...
Go to contribution page -
Letizia Parato (University of Colorado - Boulder)30/07/2026, 15:205Algorithms and artificial intelligenceContributed talk
Critical slowing down remains a major challenge for lattice simulations near continuous phase transitions, where the correlation length diverges. One possible strategy to address this issue is to generate large-volume configurations from smaller lattices, where local update algorithms remain efficient. In this setting, the renormalization group provides a natural framework: the coarse...
Go to contribution page -
Simone Romiti30/07/2026, 16:101Algorithms and artificial intelligence
The Hamiltonian formulation of lattice gauge theories solves several of the problems in Euclidean Monte Carlo simulations. However, the main obstacle is the exponential growth of the Hilbert space with the lattice volume, and one needs to find an effective representation that fits the computer memory. Most of the present calculations rely on explicit truncation methods, usually suited only for...
Go to contribution page -
Suryansh Rajawat (University of Maryland)30/07/2026, 16:302Algorithms and artificial intelligenceContributed talk
We present a variational approach to quantum field theory based on wavefunctions parameterized by neural networks, as a stepping stone towards real-time and finite-density regimes, inaccessible to path-integral Monte Carlo. Working in the Hamiltonian formulation on a spatial lattice, we optimize a neural-network ansatz with variational Monte Carlo to obtain the ground-state and excited-state...
Go to contribution page -
Joshua Lin (Argonne National Laboratory)30/07/2026, 16:504Algorithms and artificial intelligenceContributed talk
In the Euclidean path-integral formalism, the theta-term introduces a sign problem which makes direct investigation difficult. On the other hand, the theta-term can be included in a Hamiltonian formulation straightforwardly - as long as one can deal with the infinite dimensional Hilbert space. In this contribution we present an investigation into the 3+1d SU(3) Yang Mills action with nonzero...
Go to contribution page -
Rohan Misra (Boston University)30/07/2026, 17:102Algorithms and artificial intelligenceContributed talk
Placing lattice field theories on curved spacetimes requires more than a choice of triangulation: the UV couplings must encode the real-space geometry seen by long-distance observables. Motivated by Brower and Owen's solution for the 2D Ising model on a latticized two-sphere [
Go to contribution pagearXiv:2407.00459], the affine-plane Ising construction [arXiv:2209.15546] and the affine conjecture... -
Erik Lundstrum (Columbia University)31/07/2026, 14:00Algorithms and artificial intelligenceContributed talk
We recently completed a one-loop calculation of the RI/SMOM to $\overline{\text{MS}}$ renormalization scheme conversion factors for the $\Delta S = 1$ four-quark operators with four active quark flavors entirely by prompting an agentic AI system to generate Mathematica code. These are required perturbative inputs for lattice QCD simulations of weak processes such as the RBC/UKQCD calculation...
Go to contribution page -
Tatsuya Wada (YITP/Kyoto University)31/07/2026, 14:20Algorithms and artificial intelligenceContributed talk
The hopping-parameter expansion (HPE) of the logarithm of the Wilson-fermion determinant expresses the coefficient of $\kappa^n$ as a sum over closed loops of length $n$. It is widely used in studies of heavy-quark QCD and in stochastic estimators of the fermion determinant. Although the expansion through sixth order, corresponding to LO and NLO, is well established, higher-order terms have...
Go to contribution page -
Pietro Butti (QTC, University of Southern Denmark)31/07/2026, 14:404Algorithms and artificial intelligenceContributed talk
The signal-to-noise problem limits the reach of many lattice calculations. We present a variational framework that recasts it as a transport problem: the loss of signal reflects a mismatch between the distribution one samples and the one needed to measure an observable, and can be reduced by transporting configurations to close that gap.
Go to contribution page
The optimal transport is typically determined either... -
Urs Wenger (University of Bern)31/07/2026, 15:001Algorithms and artificial intelligenceContributed talk
Classically perfect fixed-point (FP) actions based on the renormalization group allow to reliably extract continuum physics from Monte Carlo simulations at coarse lattice spacings, thereby avoiding topological freezing. While these FP actions are very complicated, machine-learned gauge-equivariant neural networks enable accurate parametrizations and efficient simulations. In this talk I...
Go to contribution page -
Daniel Nogradi31/07/2026, 15:20Algorithms and artificial intelligenceContributed talk
The 4-dimensional hypercubic lattice has a symmetry group with 384 elements. The 16-cell honeycomb lattice has 3 times as many elements, hence a larger symmetry. Motivated by this fact we discretize lattice QCD on this lattice and observe better scaling for gauge observables and better chiral properties for the Wilson-type Dirac operator. Even though a lattice site has 24 neighbors and not 8...
Go to contribution page -
Katsumasa Nakayama (RIKEN)31/07/2026, 16:101Algorithms and artificial intelligenceContributed talk
We apply the tensor renormalization group method to SU(N) non-Abelian gauge theories in three- and four-dimensional spacetime. Using a sampling approach for the gauge group elements, we obtain expectation values that are consistent with perturbation theory in both the strong- and weak-coupling regions. We propose several distributions for the random sampling and compare their performance at...
Go to contribution page -
Takaaki Kuwahara (The University of Tokyo)31/07/2026, 16:30Algorithms and artificial intelligenceContributed talk
The tensor renormalization group (TRG) is a numerical method that is, in principle, free from the sign problem and is regarded as a promising candidate for first-principles calculations of finite-density QCD. While many successful studies have been carried out for scalar and fermionic field theories, its application to gauge theories remains under development. In this talk, we propose a method...
Go to contribution page -
Benjamin Izett (ETH Zรผrich)31/07/2026, 16:50Algorithms and artificial intelligenceContributed talk
We review the string formulation of gauge fields on a finite lattice with periodic boundary conditions, as proposed by Batrouni and Halpern in 1984. Simulation results using the Hybrid Monte Carlo algorithm are reported. The properties of the molecular dynamics force under gauge transformations are compared with the Wilson formulation and we discuss the application to Fourier acceleration.
Go to contribution page -
Shuhei Yamamoto (BNL)31/07/2026, 17:104Algorithms and artificial intelligenceContributed talk
The Field-Transformation Hybrid Monte-Carlo (FTHMC) algorithm potentially mitigates critical slowing down by combining the HMC with an invertible field transformation, originally proposed by Lรผscher and motivated as trivializing the theory. In our previous study, using a single Jacobian-computable smearing step resembling stout smearing in 2+1 domain-wall fermion simulations, we found a...
Go to contribution page