Reconstructing QCD spectral functions with Gaussian processes

Horak, Jan; Pawlowski, Jan M.; Rodríguez‐Quintero, J.; Turnwald, Jonas; Urban, Julian M.; Wink, Nicolas; Zafeiropoulos, Savvas

doi:10.1103/physrevd.105.036014

Cited by 54 publications

(45 citation statements)

References 86 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…By using physically motivated toy spectral functions, we generated training data sets and trained a feedforward neural network on these functions. In contrast to alternative approaches [24,35], our network assures that the output spectral function will have the correct UV asymptotics, as dictated by the renormalization group. We examined the performance of different neural network architectures and illustrated their reconstruction accuracy on our data sets with unseen testing data.…”

Section: Discussionmentioning

confidence: 99%

“…Our findings suggest that a neural network, trained on toy propagators, can recognize if and where pairs of complex poles are present which can be of great importance to expanding our knowledge of confined gluon 2-point functions. Its computation speed is also much faster than traditional methods, as highlighted in [35]. Future work could examine the use of different neural networks (e.g.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Neural network approach to reconstructing spectral functions and complex poles of confined particles

Lechien

Dudal

2022

SciPost Phys.

View full text Add to dashboard Cite

Reconstructing spectral functions from propagator data is difficult as solving the analytic continuation problem or applying an inverse integral transformation are ill-conditioned problems. Recent work has proposed using neural networks to solve this problem and has shown promising results, either matching or improving upon the performance of other methods. We generalize this approach by not only reconstructing spectral functions, but also (possible) pairs of complex poles or an infrared (IR) cutoff. We train our network on physically motivated toy functions, examine the reconstruction accuracy and check its robustness to noise. Encouraging results are found on both toy functions and genuine lattice QCD data for the gluon propagator, suggesting that this approach may lead to significant improvements over current state-of-the-art methods.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Neural network approach to reconstructing spectral functions and complex poles of confined particles

Lechien

Dudal

2022

SciPost Phys.

View full text Add to dashboard Cite

show abstract

“…It has been argued that in Landau gauge this also applies to the gluon propagator [70][71][72]. While high precision spectral reconstructions are not in contradiction to this assumption and do work for the gluon propagator [31,[73][74][75], extensions with complex conjugate poles are also commonly entertained in reconstructions, see e.g. [68,[76][77][78][79][80][81][82][83][84].…”

Section: Gluon Propagatormentioning

confidence: 99%

Renormalised spectral flows

Braun¹,

Chen²,

Fu³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

We derive renormalised finite functional flow equations for quantum field theories in real and imaginary time that incorporate scale transformations of the renormalisation conditions, hence implementing a flowing renormalisation. The flows are manifestly finite in general non-perturbative truncation schemes also for regularisation schemes that do not implement an infrared suppression of the loops in the flow. Specifically, this formulation includes finite functional flows for the effective action with a spectral Callan-Symanzik cutoff, and therefore gives access to Lorentz invariant spectral flows. The functional setup is fully non-perturbative and allows for the spectral treatment of general theories. In particular, this includes theories that do not admit a perturbative renormalisation such as asymptotically safe theories. Finally, the application of the Lorentz invariant spectral functional renormalisation group is briefly discussed for theories ranging from real scalar and Yukawa theories to gauge theories and quantum gravity.

show abstract

“…Our hunch is that the generation of decorrelated configurations in LQCD using methods such as normalizing flows [46] will be one of the first successes beyond the examples of predicting expensive to calculate correlation functions in terms of cheaper ones as discussed in [40]. Another exciting application of ML in the analysis of data is to estimate the real-time spectral function from lattice data for the Euclidean 2-point function [41,49,50,42], a notoriously difficult problem since the latter is the Laplace transform of the former.…”

Section: Contrasting Two Black Boxes: Lattice Qcd and MLmentioning

confidence: 99%

AI and Theoretical Particle Physics

Gupta¹,

Bhattacharya²,

Yoon³

2022

Preprint

View full text Add to dashboard Cite

Theoretical particle physicists continue to push the envelope in both high performance computing and in managing and analyzing large data sets. For example, the goals of sub-percent accuracy in predictions of quantum chromodynamics (QCD) using large scale simulations of lattice QCD and in finding signals of rare events and new physics in exabytes of data produced by experiments at the high luminosity large hadron collider (LHC) require new tools beyond just developments in hardware. Machine learning and artificial intelligence offer the promise of dramatically reducing the computational cost and time. This chapter reviews selected areas where AI/ML tools could have a major impact, provides an overview of the challenges, and discusses how new ideas such as normalizing flows can speed up the generation of gauge configurations needed in lattice QCD calculations; the growth of ML in surrogate models and pattern matching to reduce the cost of event generators and in the analysis of experimental data; and in the search for viable vacua in the landscape of string theories. While such approaches transform aspects of particle theory into computational problems, and thus black boxes, we argue that physics-aware development of these tools combined with algorithms that ensure that the results are bias free will continue to require a deep understanding of the physics. We see this broader transformation as akin to formulating and extracting observables from simulations of lattice QCD, a numerical integration of the path integral formulation of QCD that nevertheless requires a deep understanding of the underlying quantum field theory, the standard model of particle physics and effective field theory methods.

show abstract

Reconstructing QCD spectral functions with Gaussian processes

Cited by 54 publications

References 86 publications

Neural network approach to reconstructing spectral functions and complex poles of confined particles

Neural network approach to reconstructing spectral functions and complex poles of confined particles

Renormalised spectral flows

AI and Theoretical Particle Physics

Contact Info

Product

Resources

About