arXiv subjects

Sebastian Keller

Publications and source records attributed to Sebastian Keller.

16 recordsLinked to original sources

arXiv2026

GPU-Native Compressed Neighbor Lists with a Space-Filling-Curve Data Layout

We have developed a compressed neighbor list for short-range particle-particle interaction based on a space- filling curve (SFC) memory layout and particle clusters. The neighbor list can be constructed efficiently on GPUs, supporting NVIDIA and AMD hardware, and has a memory footprint of only 4 bytes per particle to store approximately 200 neighbors. Compared to the highly-optimized domain-specific neighbor list implementation of GROMACS, a molecular dynamics code, it has a comparable cluster overhead and delivers similar performance in a neighborhood pass. Thanks to the SFC-based data layout and the support for varying interaction radii per particle, our neighbor list performs well for systems with high density contrasts, such as those encountered in many astrophysical and cosmological applications. Due to the close relation between SFCs and octrees, our neighbor list seamlessly integrates with octree-based domain decomposition and multipole-based methods for long-range gravitational or electrostatic interactions. To demonstrate the coupling between long- and short-range forces, we simulate an Evrard collapse, a standard test case for the coupling between hydrodynamical and gravitational forces, on up to 1024 GPUs, and compare our results to the analytical solution.

cs.CE↗

arXiv2025

Tidal disruption events with SPH-EXA: resolving the return of the stream

In a tidal disruption event (TDE), a star is disrupted by the tidal field of a massive black hole, creating a debris stream that returns to the black hole, forms an accretion flow, and powers a luminous flare. Over the last few decades, several numerical studies have concluded that shock-induced dissipation occurs as the stream returns to pericentre (i.e., pre-self-intersection), resulting in efficient circularisation of the debris. However, the efficacy of these shocks is the subject of intense debate. We present high-resolution simulations (up to 10^10 particles) of the disruption of a solar-like star by a 10^6M_sun black hole with the new, GPU-based, smoothed-particle hydrodynamics code SPH-EXA, including the relativistic apsidal precession of the stellar debris orbits; our simulations run from initial disruption to the moment of stream self-intersection. With 10^8 particles - corresponding to the highest-resolution SPH simulations of TDEs in the pre-existing literature - we find significant, in-plane spreading of the debris as the stream returns through pericenter, in line with previous works that suggested this is a significant source of dissipation and luminous emission. However, with increasing resolution this effect is dramatically diminished, and with 10^10 particles there is effectively no change between the incoming and the outgoing stream widths. Our results demonstrate that the paradigm of significant dissipation of kinetic energy during pericentre passage is incorrect, and instead it is likely that debris circularisation is mediated by the originally proposed, stream-stream collision scenario.

astro-ph.HE↗

arXiv2025

From particles to precision. Simulating subsonic turbulence with Smoothed Particle Hydrodynamics

The numerical simulation of subsonic turbulence with smoothed particle hydrodynamics (SPH) has traditionally been hampered by zeroth-order (E0) errors, inaccurate gradient evaluations, and excessive numerical dissipation. We aim to investigate whether a modern SPH formulation can overcome these challenges by comparing its results to those obtained using moving-mesh and meshless finite-volume methods such as AREPO and GIZMO. For this purpose, we used SPH-EXA, a highly scalable, natively GPU-accelerated, state-of-the-art SPH code. Our results show that SPH-EXA accurately reproduces the Kolmogorov inertial range scaling in the subsonic regime with increasing resolution, closely matching the results of the reference methods. We also identify accurate grad-h terms as critical: a noisy standard implementation can imprint spurious granulation in the density field once dissipation is sufficiently reduced. These results demonstrate that, with appropriate methodological advances, SPH can achieve a level of fidelity in modeling subsonic turbulence comparable to the most advanced Eulerian and moving-mesh approaches.

astro-ph.IM↗

arXiv2023

Cornerstone: Octree Construction Algorithms for Scalable Particle Simulations

This paper presents an octree construction method, called Cornerstone, that facilitates global domain decomposition and interactions between particles in mesh-free numerical simulations. Our method is based on algorithms developed for 3D computer graphics, which we extend to distributed high performance computing (HPC) systems. Cornerstone yields global and locally essential octrees and is able to operate on all levels of tree hierarchies in parallel. The resulting octrees are suitable for supporting the computation of various kinds of short and long range interactions in N-body methods, such as Barnes-Hut and the Fast Multipole Method (FMM). While we provide a CPU implementation, Cornerstone may run entirely on GPUs. This results in significantly faster tree construction compared to execution on CPUs and serves as a powerful building block for the design of simulation codes that move beyond an offloading approach, where only numerically intensive tasks are dispatched to GPUs. With data residing exclusively in GPU memory, Cornerstone eliminates data movements between CPUs and GPUs. As an example, we employ Cornerstone to generate locally essential octrees for a Barnes-Hut treecode running on almost the full LUMI-G system with up to 8 trillion particles.

astro-ph.IM↗

arXiv2022

Application Experiences on a GPU-Accelerated Arm-based HPC Testbed

This paper assesses and reports the experience of ten teams working to port,validate, and benchmark several High Performance Computing applications on a novel GPU-accelerated Arm testbed system. The testbed consists of eight NVIDIA Arm HPC Developer Kit systems built by GIGABYTE, each one equipped with a server-class Arm CPU from Ampere Computing and A100 data center GPU from NVIDIA Corp. The systems are connected together using Infiniband high-bandwidth low-latency interconnect. The selected applications and mini-apps are written using several programming languages and use multiple accelerator-based programming models for GPUs such as CUDA, OpenACC, and OpenMP offloading. Working on application porting requires a robust and easy-to-access programming environment, including a variety of compilers and optimized scientific libraries. The goal of this work is to evaluate platform readiness and assess the effort required from developers to deploy well-established scientific workloads on current and future generation Arm-based GPU-accelerated HPC systems. The reported case studies demonstrate that the current level of maturity and diversity of software and tools is already adequate for large-scale production deployments.

cs.DC↗

arXiv2017

An efficient relativistic density-matrix renormalization group implementation in a matrix-product formulation

We present an implementation of the relativistic quantum-chemical density matrix renormalization group (DMRG) approach based on a matrix-product formalism. Our approach allows us to optimize matrix product state (MPS) wave functions including a variational description of scalar-relativistic effects and spin-orbit coupling from which we can calculate, for example, first-order electric and magnetic properties in a relativistic framework. While complementing our pilot implementation (S. Knecht et al., J. Chem. Phys., 140, 041101 (2014)) this work exploits all features provided by its underlying non-relativistic DMRG implementation based on an matrix product state and operator formalism. We illustrate the capabilities of our relativistic DMRG approach by studying the ground-state magnetization as well as current density of a paramagnetic $f^9$ dysprosium complex as a function of the active orbital space employed in the MPS wave function optimization.

physics.chem-ph↗

arXiv2017

Interfacial exchange interactions and magnetism of Ni2MnAl/Fe bilayers

Based on a multi-scale calculations, combining ab-initio methods with spin dynamics simulations, we perform a detailed study of the magnetic behavior of Ni2MnAl/Fe bilayers. Our simulations show that such a bilayer exhibits a small exchange bias effect when the Ni2MnAl Heusler alloy is in a disordered B2 phase. Additionally, we present an effective way to control the magnetic structure of the Ni2MnAl antiferromagnet, in the pseudo-ordered B2-I as well as the disordered B2 phases, via a spin-flop coupling to the Fe layer.

cond-mat.mes-hall↗

arXiv2017

Second-Order Self-Consistent-Field Density-Matrix Renormalization Group

We present a matrix-product state (MPS)-based quadratically convergent density-matrix renormalization group self-consistent-field (DMRG-SCF) approach. Following a proposal by Werner and Knowles (JCP 82, 5053, (1985)), our DMRG-SCF algorithm is based on a direct minimization of an energy expression which is correct to second-order with respect to changes in the molecular orbital basis. We exploit a simultaneous optimization of the MPS wave function and molecular orbitals in order to achieve quadratic convergence. In contrast to previously reported (augmented Hessian) Newton-Raphson and super-configuration-interaction algorithms for DMRG-SCF, energy convergence beyond a quadratic scaling is possible in our ansatz. Discarding the set of redundant active-active orbital rotations, the DMRG-SCF energy converges typically within two to four cycles of the self-consistent procedure

physics.chem-ph↗

arXiv2016

A nonorthogonal state-interaction approach for matrix product state wave functions

We present a state-interaction approach for matrix product state (MPS) wave functions in a nonorthogonal molecular orbital basis. Our approach allows us to calculate for example transition and spin-orbit coupling matrix elements between arbitrary electronic states provided that they share the same one-electron basis functions and active orbital space, respectively. The key element is the transformation of the MPS wave functions of different states from a nonorthogonal to a biorthonormal molecular orbital basis representation exploiting a sequence of non-unitary transformations following a proposal by Malmqvist (Int. J. Quantum Chem. 30, 479 (1986)). This is well-known for traditional wave-function parametrizations but has not yet been exploited for MPS wave functions.

physics.chem-ph↗

arXiv2016

New Approaches for ab initio Calculations of Molecules with Strong Electron Correlation

Reliable quantum chemical methods for the description of molecules with dense-lying frontier orbitals are needed in the context of many chemical compounds and reactions. Here, we review developments that led to our newcomputational toolbo x which implements the quantum chemical density matrix renormalization group in a second-generation algorithm. We present an overview of the different components of this toolbox.

physics.chem-ph↗

arXiv2016

Spin-adapted Matrix Product States and Operators

Matrix product states (MPSs) and matrix product operators (MPOs) allow an alternative formulation of the density matrix renormalization group algorithm introduced by White. Here, we describe how non-abelian spin symmetry can be exploited in MPSs and MPOs by virtue of the Wigner--Eckart theorem at the example of the spin-adapted quantum chemical Hamiltonian operator.

physics.chem-ph↗

arXiv2015

Pair Correlations in Doped Hubbard Ladders

Hubbard ladders are an important stepping stone to the physics of the two-dimensional Hubbard model. While many of their properties are accessible to numerical and analytical techniques, the question of whether weakly hole-doped Hubbard ladders are dominated by superconducting or charge-density-wave correlations has so far eluded a definitive answer. In particular, previous numerical simulations of Hubbard ladders have seen a much faster decay of superconducting correlations than expected based on analytical arguments. We revisit this question using a state-of-the-art implementation of the density matrix renormalization group algorithm that allows us to simulate larger system sizes with higher accuracy than before. Performing careful extrapolations of the results, we obtain improved estimates for the Luttinger liquid parameter and the correlation functions at long distances. Our results confirm that, as suggested by analytical considerations, superconducting correlations become dominant in the limit of very small doping.

cond-mat.str-el↗

arXiv2015

An Efficient Matrix Product Operator Representation of the Quantum-Chemical Hamiltonian

We describe how to efficiently construct the quantum chemical Hamiltonian operator in matrix product form. We present its implementation as a density matrix renormalization group (DMRG) algorithm for quantum chemical applications in a purely matrix product based framework. Existing implementations of DMRG for quantum chemistry are based on the traditional formulation of the method, which was developed from a viewpoint of Hilbert space decimation and attained a higher performance compared to straightforward implementations of matrix product based DMRG. The latter variationally optimizes a class of ansatz states known as matrix product states (MPS), where operators are correspondingly represented as matrix product operators (MPO). The MPO construction scheme presented here eliminates the previous performance disadvantages while retaining the additional flexibility provided by a matrix product approach; for example, the specification of expectation values becomes an input parameter. In this way, MPOs for different symmetries - abelian and non-abelian - and different relativistic and non-relativistic models may be solved by an otherwise unmodified program.

physics.comp-ph↗

arXiv2015

Bayesian Markov Blanket Estimation

This paper considers a Bayesian view for estimating a sub-network in a Markov random field. The sub-network corresponds to the Markov blanket of a set of query variables, where the set of potential neighbours here is big. We factorize the posterior such that the Markov blanket is conditionally independent of the network of the potential neighbours. By exploiting this blockwise decoupling, we derive analytic expressions for posterior conditionals. Subsequently, we develop an inference scheme which makes use of the factorization. As a result, estimation of a sub-network is possible without inferring an entire network. Since the resulting Gibbs sampler scales linearly with the number of variables, it can handle relatively large neighbourhoods. The proposed scheme results in faster convergence and superior mixing of the Markov chain than existing Bayesian network estimation techniques.

stat.ML↗

arXiv2014

Matrix Product State applications for the ALPS project

The density-matrix renormalization group method has become a standard computational approach to the low-energy physics as well as dynamics of low-dimensional quantum systems. In this paper, we present a new set of applications, available as part of the ALPS package, that provide an efficient and flexible implementation of these methods based on a matrix-product state (MPS) representation. Our applications implement, within the same framework, algorithms to variationally find the ground state and low-lying excited states as well as simulate the time evolution of arbitrary one-dimensional and two-dimensional models. Implementing the conservation of quantum numbers for generic Abelian symmetries, we achieve performance competitive with the best codes in the community. Example results are provided for (i) a model of itinerant fermions in one dimension and (ii) a model of quantum magnetism.

cond-mat.str-el↗

arXiv2014

Self-Consistent Embedding of Density-Matrix Renormalization Group Wavefunctions in a Density Functional Environment

We present the first implementation of a density matrix renormalization group algorithm embedded in an environment described by density functional theory. The frozen density embedding scheme is used with a freeze-and-thaw strategy for a self-consistent polarization of the orbital-optimized wavefunction and the environmental densities with respect to each other.

physics.chem-ph↗