compute turbulence diagnostics

To address the challenge of intuitively correlating multiscale, multivariate data in multiphysics problems such as turbulence, this paper proposes a glyph-based visualization method integrating multiscale statistical information. The method innovatively combines curvelet transform—which enables anisotropic scale decomposition—with constrained level-set-driven Voronoi tessellation to locally aggregate statistical features across multiple physical fields. A composite glyph design encodes spatial position, scale hierarchy, and coupled physical quantities (e.g., velocity, temperature, reaction rate), embedded within an interactive visualization system. Experiments on turbulent combustion and incompressible channel flow datasets demonstrate that the approach effectively reveals spatially coherent patterns and dynamic interactions among physical fields across scales, significantly enhancing interpretability and exploratory efficiency of cross-scale physical mechanisms.

This work proposes a lightweight and scalable GPU-accelerated library designed to meet the demand for efficient streamline tracing in large-scale computational fluid dynamics simulations. Built upon an MPI-based data-parallel framework and featuring a modular API, the library supports diverse flow field representations and seamlessly integrates into existing MPI applications for both post hoc and in situ analysis. By innovatively combining GPU acceleration with distributed-memory architectures, the approach achieves high-performance streamline computation while also providing an interactive prototype tool for seed point placement to facilitate rapid visualization development. The code is released under a permissive open-source license, balancing performance, flexibility, and usability.

Neural operators often inadequately capture high-frequency turbulent dynamics, leading to distorted energy spectra and excessive smoothing. To address this, we propose a neural operator–diffusion model synergy: for the first time, conditional diffusion models leverage coarse-grained priors generated by Fourier- or graph-based neural operators to drive high-fidelity turbulent structure reconstruction. Our method integrates diffusion-based correction into autoregressive rollout and imposes POD spectral constraints, trained on Schlieren velocimetry experimental data. The key contribution is a novel coupling framework that bridges generative modeling with physics-informed operators, significantly enhancing spatiotemporal spectral fidelity and long-term forecast stability. Experiments demonstrate superior vortex-resolving accuracy on high-Reynolds-number jets and real experimental data, with predicted energy spectra showing markedly improved agreement with ground-truth distributions.

This study addresses the challenge of resolving highly unsteady thermal and viscous boundary layer dynamics in high-Rayleigh-number (Ra = 10¹²) thermal convection—a regime where conventional high-frequency data acquisition and post-hoc analysis are infeasible. To overcome these bottlenecks, we developed a scalable in situ analysis workflow on the JUWELS Booster supercomputer (840 nodes), coupling the GPU-accelerated spectral-element solver NekRS with the ASCENT in situ visualization framework across 3,360 GPUs—the first such large-scale integration. This enabled the largest-ever fully resolved three-dimensional turbulent direct numerical simulation at this Rayleigh number, capturing millisecond-scale boundary layer fluctuations in real time. The system achieves TB/s-level online feature extraction and visualization. Our approach revealed novel non-equilibrium boundary layer dynamics and establishes an extensible in situ analysis paradigm for studying complex heat and mass transfer under extreme conditions.

This work addresses the challenge of achieving both efficient compact representation and high physical fidelity in turbulent flow fields by proposing a continuous parameterization method based on learnable local Gaussian primitives. By superimposing anisotropic Gaussian kernels with adjustable positions, amplitudes, and scales, the approach enables a mesh-free, compact representation that supports accurate computation of derived quantities such as vorticity. The incorporation of a multi-resolution architecture and compactly supported Beta basis functions further enhances geometric expressiveness. In the Taylor–Green vortex benchmark, the method achieves compression ratios of 10³–10⁴ while preserving high-fidelity velocity fields, effectively recovering mid- to high-frequency structures and mitigating vorticity dissipation. These results indicate that geometric representational capacity—not merely parameter count—is the key limiting factor, laying the groundwork for structure-aware, physics-informed flow field compression.

This work addresses the scarcity of physically validated, high-quality three-dimensional turbulent flow data for training and evaluating neural operators. We develop a reproducible lattice Boltzmann method (LBM) data generation pipeline based on cumulant collision operators to simulate obstacle-induced turbulent channel flows across Reynolds numbers from 1,000 to 10,000, producing a high-fidelity dataset at a resolution of 1024×512×512. To our knowledge, this dataset is the first to include experimental validation and mesh convergence analysis, accompanied by a standardized benchmarking framework tailored for neural operators. Experimental comparisons demonstrate excellent agreement between LBM predictions and physical measurements in key quantities such as Strouhal number, drag coefficient, and turbulent fluctuations, enabling systematic evaluation of Fourier neural operators and U-Net variants in tasks including flow prediction, super-resolution, and error correction.

本文提出一种架构，通过分离意图解释、执行和解释，并基于领域本体约束分析链，解决了LLM辅助科学可视化中生成错误脚本的问题。

Existing metrics, such as relative L² error, often fail to comprehensively assess the numerical plausibility of learned PDE simulators. To address this limitation, this work proposes the first architecture-agnostic, post-hoc diagnostic framework that systematically audits a model’s behavior as an approximate evolution operator through structural indicators—including semigroup consistency, energy dynamics, and response to perturbations. Requiring only reference trajectories, predictions, equation metadata, and simulation configurations, the framework uniformly evaluates diverse architectures such as FNOs, DeepONets, U-Nets, and ResNets. Experiments across five canonical PDE benchmarks demonstrate that even when L² errors are low, structural metrics can exhibit significant degradation, thereby underscoring the necessity and efficacy of the proposed approach.

This study addresses the absence of open standards for CPU pipeline visualization tools and the difficulty in localizing performance bottlenecks. To this end, it proposes an open-source event stream format alongside Catscan, an interactive viewer. Methodologically, this work introduces a structured event stream based on transactional relationships, integrating typed event modeling, persistent highlighting techniques, and domain-specific search algorithms to enable microarchitectural trace analysis from symptoms down to individual instructions. Furthermore, it supports resource-oriented views synchronized with comparative trace alignment. By successfully reproducing industry-grade debugging workflows, this project provides the community with production-validated microarchitectural visualization infrastructure.

Existing physics-informed surrogate models struggle to accurately capture local multiscale flow structures. This work proposes a novel Transformer architecture that integrates a Fourier–wavelet hybrid spectral encoding with a physics-biased self-attention mechanism guided by partial differential equation (PDE) residual diagnostics. A self-supervised pretraining strategy is further introduced, combining masked physical prediction with equation consistency prediction. The proposed method substantially enhances the modeling of localized flow features and achieves state-of-the-art performance on canonical benchmarks for cylinder wake and fluid–structure interaction problems, yielding normalized mean squared errors of 0.05875 and 2.70×10⁻⁴, respectively, and a Pearson correlation coefficient of 0.97019. It faithfully reconstructs key flow structures, including near-body regions, the wake core, and far-wake zones.