arXiv subjects

Di Qi

Publications and source records attributed to Di Qi.

At least 19 recordsLinked to original sources

arXiv2026

Mode Coverage in Normalizing Flow Boltzmann Generators via Log-Ratio Variation

Normalizing flow Boltzmann generators retain a tractable pushforward density, but training with forward KL depends on target samples that may be biased or omit modes. As a result, a flow can miss target mass while its observed importance weights give a high effective sample size. We introduce the log-ratio variation $\X_\omega$, the mean absolute pairwise difference of the target-to-pushforward log-density ratio under a weighting measure $\omega$, and use it to define KLXX, a new loss function. Two log-ratio variations are added to the forward KL (denoted by the two X's): one weighted by the target to improve accuracy, the other by a mixture of quench and temper samples with pushforward samples to search candidate modes. We derive the Fisher--Rao gradient flow of KLXX, where both variations contribute nonpositive dissipation, and a fixed-surrogate error bound for KLXX. We use KLXX in an adaptive-staging Boltzmann generator, with importance reweighting at every stage. We bound the sampling error of its inference scheme when the stage weights are essentially bounded, and prove it asymptotically unbiased in the sample size. In the numerical tests, KLXX improves mode coverage over forward KL. It also improves the generator's per-stage diagnostics against the loss that built the schedule. The observables the generator recovers are close to independent references. The log-ratio variations thus supply information that the forward KL loss usually omits.

stat.ML↗

arXiv2026

The finite expression method for turbulent dynamics with high-order moment recovery

Turbulent dynamical systems are characterized by nonlinear interactions and stochastic effects that generate coupled statistical quantities, such as non-zero higher-order moments, which are difficult to capture from data with accuracy. We propose a two-stage data-driven modeling framework that combines symbolic regression with generative models to jointly identify the governing dynamics and predict their key statistical quantities. In Stage I of the framework, the Finite Expression Method (FEX) is adopted to discover closed-form expressions of the deterministic dynamics, recovering nonlinear interaction terms and external forcing without predefined libraries. In Stage II, generative models are introduced to learn the residual stochastic components as a refined correction to the model error from the Stage I approximation, enabling accurate characterization of higher-order statistics. Theoretical analysis establishes the consistency of the symbolic estimator and quantifies the estimation error in terms of data size and numerical discretization. The model performance is verified through detailed numerical experiments on the stochastic triad models across multiple regimes, demonstrating that the framework successfully recovers interaction terms and forcing expressions, and accurately predicts statistical moments up to order five. These results highlight the potential of integrating interpretable symbolic discovery with data-driven stochastic modeling for complex turbulent systems.

cs.LG↗

arXiv2026

Jeffreys Flow: Robust Boltzmann Generators for Rare Event Sampling via Parallel Tempering Distillation

Sampling physical systems with rough energy landscapes is hindered by rare events and metastable trapping. While Boltzmann generators already offer a solution, their reliance on the reverse Kullback--Leibler divergence frequently induces catastrophic mode collapse, missing specific modes in multi-modal distributions. Here, we introduce the Jeffreys Flow, a robust generative framework that mitigates this failure by distilling empirical sampling data from Parallel Tempering trajectories using the symmetric Jeffreys divergence. This formulation effectively balances local target-seeking precision with global modes coverage. We show that minimizing Jeffreys divergence suppresses mode collapse and structurally corrects inherent inaccuracies via distillation of the empirical reference data. We demonstrate the framework's scalability and accuracy on highly non-convex multidimensional benchmarks, including the systematic correction of stochastic gradient biases in Replica Exchange Stochastic Gradient Langevin Dynamics and the massive acceleration of exact importance sampling in Path Integral Monte Carlo for quantum thermal states.

cs.LG↗

arXiv2026

Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters

We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency. We focus on what matters most when building agents: sharp reasoning and fast, reliable execution. Step 3.5 Flash pairs a 196B-parameter foundation with 11B active parameters for efficient inference. It is optimized with interleaved 3:1 sliding-window/full attention and Multi-Token Prediction (MTP-3) to reduce the latency and cost of multi-round agentic interactions. To reach frontier-level intelligence, we design a scalable reinforcement learning framework that combines verifiable signals with preference feedback, while remaining stable under large-scale off-policy training, enabling consistent self-improvement across mathematics, code, and tool use. Step 3.5 Flash demonstrates strong performance across agent, coding, and math tasks, achieving 85.4% on IMO-AnswerBench, 86.4% on LiveCodeBench-v6 (2024.08-2025.05), 88.2% on tau2-Bench, 69.0% on BrowseComp (with context management), and 51.0% on Terminal-Bench 2.0, comparable to frontier models such as GPT-5.2 xHigh and Gemini 3.0 Pro. By redefining the efficiency frontier, Step 3.5 Flash provides a high-density foundation for deploying sophisticated agents in real-world industrial environments.

cs.CL↗

arXiv2026

DockSmith: Scaling Reliable Coding Environments via an Agentic Docker Builder

Reliable Docker-based environment construction is a dominant bottleneck for scaling execution-grounded training and evaluation of software engineering agents. We introduce DockSmith, a specialized agentic Docker builder designed to address this challenge. DockSmith treats environment construction not only as a preprocessing step, but as a core agentic capability that exercises long-horizon tool use, dependency reasoning, and failure recovery, yielding supervision that transfers beyond Docker building itself. DockSmith is trained on large-scale, execution-grounded Docker-building trajectories produced by a SWE-Factory-style pipeline augmented with a loop-detection controller and a cross-task success memory. Training a 30B-A3B model on these trajectories achieves open-source state-of-the-art performance on Multi-Docker-Eval, with 39.72% Fail-to-Pass and 58.28% Commit Rate. Moreover, DockSmith improves out-of-distribution performance on SWE-bench Verified, SWE-bench Multilingual, and Terminal-Bench 2.0, demonstrating broader agentic benefits of environment construction.

cs.AI↗

arXiv2026

STEP3-VL-10B Technical Report

We present STEP3-VL-10B, a lightweight open-source foundation model designed to redefine the trade-off between compact efficiency and frontier-level multimodal intelligence. STEP3-VL-10B is realized through two strategic shifts: first, a unified, fully unfrozen pre-training strategy on 1.2T multimodal tokens that integrates a language-aligned Perception Encoder with a Qwen3-8B decoder to establish intrinsic vision-language synergy; and second, a scaled post-training pipeline featuring over 1k iterations of reinforcement learning. Crucially, we implement Parallel Coordinated Reasoning (PaCoRe) to scale test-time compute, allocating resources to scalable perceptual reasoning that explores and synthesizes diverse visual hypotheses. Consequently, despite its compact 10B footprint, STEP3-VL-10B rivals or surpasses models 10$\times$-20$\times$ larger (e.g., GLM-4.6V-106B, Qwen3-VL-235B) and top-tier proprietary flagships like Gemini 2.5 Pro and Seed-1.5-VL. Delivering best-in-class performance, it records 92.2% on MMBench and 80.11% on MMMU, while excelling in complex reasoning with 94.43% on AIME2025 and 75.95% on MathVision. We release the full model suite to provide the community with a powerful, efficient, and reproducible baseline.

cs.CV↗

arXiv2025

An Asymptotic Approach for Modeling Multiscale Complex Fluids at the Fast Relaxation Limit

We present a new asymptotic strategy for general micro-macro models which analyze complex viscoelastic fluids governed by coupled multiscale dynamics. In such models, the elastic stress appearing in the macroscopic continuum equation is derived from the microscopic kinetic theory, which makes direct numerical simulations computationally expensive. To address this challenge, we introduce a formal asymptotic scheme that expands the density function around an equilibrium distribution, thereby reducing the high computational cost associated with the fully coupled microscopic processes while still maintaining the dynamic microscopic feedback in explicit expressions. The proposed asymptotic expansion is based on a detailed physical scaling law which characterizes the multiscale balance at the fast relaxation limit of the microscopic state. An asymptotic closure model for the macroscopic fluid equation is then derived according to the explicit asymptotic density expansion. Furthermore, the resulting closure model preserves the energy-dissipation law inherited from the original fully coupled multiscale system. Numerical experiments are performed to validate the asymptotic density formula and the corresponding flow velocity equations in several micro-macro models. This new asymptotic strategy offers a promising approach for efficient computations of a wide range of multiscale complex fluids.

math-ph↗

arXiv2025

Resonance-Driven Intermittency and Extreme Events in Turbulent Scalar Transport with a Mean Gradient

We study the statistical properties of passive tracer transport in turbulent flows with a mean gradient, emphasizing tracer intermittency and extreme events. An analytically tractable model is developed, coupling zonal and shear velocity components with both linear and nonlinear stochastic dynamics. Formulating the model in Fourier space, a simple explicit solution for the tracer invariant statistics is derived. Through this model we identify the resonance condition responsible for non-Gaussian behavior and bursts in the tracer. Resonant conditions, that lead to a peak in the tracer variance, occur when the zonal flow and the shear flow phase speeds are equivalent. Numerical experiments across a range of regimes, including different energy spectra and zonal flow models, are performed to validate these findings and demonstrate how the velocity field and stochasticity determines tracer extremes. These results provide additional insight into the mechanisms underlying turbulent tracer transport, with implications for uncertainty quantification and data assimilation in geophysical and environmental applications.

cs.CE↗

arXiv2025

Step1X-3D: Towards High-Fidelity and Controllable Generation of Textured 3D Assets

While generative artificial intelligence has advanced significantly across text, image, audio, and video domains, 3D generation remains comparatively underdeveloped due to fundamental challenges such as data scarcity, algorithmic limitations, and ecosystem fragmentation. To this end, we present Step1X-3D, an open framework addressing these challenges through: (1) a rigorous data curation pipeline processing >5M assets to create a 2M high-quality dataset with standardized geometric and textural properties; (2) a two-stage 3D-native architecture combining a hybrid VAE-DiT geometry generator with an diffusion-based texture synthesis module; and (3) the full open-source release of models, training code, and adaptation modules. For geometry generation, the hybrid VAE-DiT component produces TSDF representations by employing perceiver-based latent encoding with sharp edge sampling for detail preservation. The diffusion-based texture synthesis module then ensures cross-view consistency through geometric conditioning and latent-space synchronization. Benchmark results demonstrate state-of-the-art performance that exceeds existing open-source methods, while also achieving competitive quality with proprietary solutions. Notably, the framework uniquely bridges the 2D and 3D generation paradigms by supporting direct transfer of 2D control techniques~(e.g., LoRA) to 3D synthesis. By simultaneously advancing data quality, algorithmic fidelity, and reproducibility, Step1X-3D aims to establish new standards for open research in controllable 3D asset generation.

cs.CV↗

arXiv2025

Data Assimilation Models for Computing Probability Distributions of Complex Multiscale Systems

We introduce a data assimilation strategy aimed at accurately capturing key non-Gaussian structures in probability distributions using a small ensemble size. A major challenge in statistical forecasting of nonlinearly coupled multiscale systems is mitigating the large errors that arise when computing high-order statistical moments. To address this issue, a high-order stochastic-statistical modeling framework is proposed that integrates statistical data assimilation into finite ensemble predictions. The method effectively reduces the approximation errors in finite ensemble estimates of non-Gaussian distributions by employing a filtering update step that incorporates observation data in leading moments to refine the high-order statistical feedback. Explicit filter operators are derived from intrinsic nonlinear coupling structures, allowing straightforward numerical implementations. We demonstrate the performance of the proposed method through extensive numerical experiments on a prototype triad system. The triad system offers an instructive and computationally manageable platform mimicking essential aspects of nonlinear turbulent dynamics. The numerical results show that the statistical data assimilation algorithm consistently captures the mean and covariance, as well as various non-Gaussian probability distributions exhibited in different statistical regimes of the triad system. The modeling framework can serve as a useful tool for efficient sampling and reliable forecasting of complex probability distributions commonly encountered in a wide variety of applications involving multiscale coupling and nonlinear dynamics.

math.NA↗

arXiv2025

Predicting 3D representations for Dynamic Scenes

We present a novel framework for dynamic radiance field prediction given monocular video streams. Unlike previous methods that primarily focus on predicting future frames, our method goes a step further by generating explicit 3D representations of the dynamic scene. The framework builds on two core designs. First, we adopt an ego-centric unbounded triplane to explicitly represent the dynamic physical world. Second, we develop a 4D-aware transformer to aggregate features from monocular videos to update the triplane. Coupling these two designs enables us to train the proposed model with large-scale monocular videos in a self-supervised manner. Our model achieves top results in dynamic radiance field prediction on NVIDIA dynamic scenes, demonstrating its strong performance on 4D physical world modeling. Besides, our model shows a superior generalizability to unseen scenarios. Notably, we find that our approach emerges capabilities for geometry and semantic learning.

cs.CV↗

arXiv2024

A Closed-Form Nonlinear Data Assimilation Algorithm for Multi-Layer Flow Fields

State estimation in multi-layer turbulent flow fields with only a single layer of partial observation remains a challenging yet practically important task. Applications include inferring the state of the deep ocean by exploiting surface observations. Directly implementing an ensemble Kalman filter based on the full forecast model is usually expensive. One widely used method in practice projects the information of the observed layer to other layers via linear regression. However, when nonlinearity in the highly turbulent flow field becomes dominant, the regression solution will suffer from large uncertainty errors. In this paper, we develop a multi-step nonlinear data assimilation method. A sequence of nonlinear assimilation steps is applied from layer to layer recurrently. Fundamentally different from the traditional linear regression approaches, a conditional Gaussian nonlinear system is adopted as the approximate forecast model to characterize the nonlinear dependence between adjacent layers. The estimated posterior is a Gaussian mixture, which can be highly non-Gaussian. Therefore, the multi-step nonlinear data assimilation method can capture strongly turbulent features, especially intermittency and extreme events, and better quantify the inherent uncertainty. Another notable advantage of the multi-step data assimilation method is that the posterior distribution can be solved using closed-form formulae under the conditional Gaussian framework. Applications to the two-layer quasi-geostrophic system with Lagrangian data assimilation show that the multi-step method outperforms the one-step method with linear stochastic flow models, especially as the tracer number and ensemble size increase.

physics.flu-dyn↗

arXiv2024

Oscillatory solutions at the continuum limit of Lorenz 96 systems

In this paper, we study the generation and propagation of oscillatory solutions observed in the widely used Lorenz 96 (L96) systems. First, period-two oscillations between adjacent grid points are found in the leading-order expansions of the discrete L96 system. The evolution of the envelope of period-two oscillations is described by a set of modulation equations with strictly hyperbolic structure. The modulation equations are found to be also subject to an additional reaction term dependent on the grid size, and the period-two oscillations will break down into fully chaotic dynamics when the oscillation amplitude grows large. Then, similar oscillation solutions are analyzed in the two-layer L96 model including multiscale coupling. Modulation equations for period-three oscillations are derived based on a weakly nonlinear analysis in the transition between oscillatory and nonoscillatory regions. Detailed numerical experiments are shown to confirm the analytical results.

nlin.CD↗

arXiv2024

Coupled Stochastic-Statistical Equations for Filtering Multiscale Turbulent Systems

We present a new strategy for filtering high-dimensional multiscale systems characterized by high-order non-Gaussian statistics using observations from leading-order moments. A closed stochastic-statistical modeling framework suitable for systematic theoretical analysis and efficient numerical simulations is designed. Optimal filtering solutions are derived based on the explicit coupling structures of stochastic and statistical equations subject to linear operators, which satisfy an infinite-dimensional Kalman-Bucy filter with conditional Gaussian dynamics. To facilitate practical implementation, we develop a finite-dimensional stochastic filter model that approximates the optimal filter solution. We prove that this approximating filter effectively captures key non-Gaussian features, demonstrating consistent statistics with the optimal filter first in its analysis step update, then at the long-time limit guaranteeing stable convergence to the optimal filter. Finally, we build a practical ensemble filter algorithm based on the approximating filtering model, which enables accurate recovery of the true model statistics. The proposed modeling and filtering strategies are applicable to a wide range challenging problems in science and engineering, particularly for statistical prediction and uncertainty quantification of multiscale turbulent states.

math-ph↗

arXiv2024

The Maintenance of Coherent Vortex Topology by Lagrangian Chaos in Drift-Rossby Wave Turbulence

This work introduces the "potential vorticity bucket brigade," a mechanism for explaining the resilience of vortex structures in magnetically confined fusion plasmas and geophysical flows. Drawing parallels with zonal jet formation, we show how inhomogeneous patterns of mixing can reinforce, rather than destroy non-zonal flow structure. We accomplish this through an exact stochastic Lagrangian representation of vorticity transport, together with a near-integrability property, which relates coherent flow topology to fluid relabeling symmetries. We demonstrate these ideas in the context of gradient-driven magnetized plasma turbulence, though the tools we develop here are model-agnostic and applicable beyond the system studied here.

physics.plasm-ph↗

arXiv2024

Mean Field Games for Controlling Coherent Structures in Nonlinear Fluid Systems

This paper discusses the control of coherent structures in turbulent flows, which has broad applications among complex systems in science and technology. Mean field games have been proved a powerful tool and are proposed here to control the stochastic Lagrangian tracers as players tracking the flow field. We derive optimal control solutions for general nonlinear fluid systems using mean field game models, and develop computational algorithms to efficiently solve the resulting coupled forward and backward mean field system. A precise link is established for the control of Lagrangian tracers and the scalar vorticity field based on the functional Hamilton-Jacobi equations derived from the mean field models. New iterative numerical strategy is then constructed to compute the optimal solution with fast convergence. We verify the skill of the mean field control models and illustrate their practical efficiency on a prototype model modified from the viscous Burger's equation under various cost functions in both deterministic and stochastic formulations. The good model performance implies potential effectiveness of the strategy for more general high-dimensional turbulent systems.

math.OC↗

arXiv2024

Slot-guided Volumetric Object Radiance Fields

We present a novel framework for 3D object-centric representation learning. Our approach effectively decomposes complex scenes into individual objects from a single image in an unsupervised fashion. This method, called slot-guided Volumetric Object Radiance Fields (sVORF), composes volumetric object radiance fields with object slots as a guidance to implement unsupervised 3D scene decomposition. Specifically, sVORF obtains object slots from a single image via a transformer module, maps these slots to volumetric object radiance fields with a hypernetwork and composes object radiance fields with the guidance of object slots at a 3D location. Moreover, sVORF significantly reduces memory requirement due to small-sized pixel rendering during training. We demonstrate the effectiveness of our approach by showing top results in scene decomposition and generation tasks of complex synthetic datasets (e.g., Room-Diverse). Furthermore, we also confirm the potential of sVORF to segment objects in real-world scenes (e.g., the LLFF dataset). We hope our approach can provide preliminary understanding of the physical world and help ease future research in 3D object-centric representation learning.

cs.CV↗

arXiv2023

Nearly integrable flows and chaotic tangles in the Dimits shift regime of plasma edge turbulence

Transitionally turbulent flows frequently exhibit spatiotemporal intermittency, reflecting a complex interplay between driving forces, dissipation, and transport present in these systems. When this intermittency manifests as observable structures and patterns in the flow, the characterization of turbulence in these systems becomes challenging due to the nontrivial correlations introduced into the statistics of the turbulence by these structures. In this work, we use tools from dynamical systems theory to study intermittency in the Dimits shift regime of the flux-balanced Hasegawa-Wakatani (BHW) equations, which models a transitional regime of resistive drift-wave turbulence relevant to magnetically confined fusion plasmas. First, we show in direct numerical simulations that turbulence in this regime is dominated by strong zonal flows and coherent drift-wave vortex structures which maintain a strong linear character despite their large amplitude. Using the framework of generalized Liouville integrability, we develop a theory of integrable Lagrangian flows in generic fluid and plasma systems and discuss how the observed zonal flows plus drift waves in the BHW system exhibit a form of ``near-integrability'' originating from a fluid element relabeling symmetry. We further demonstrate that the BHW flows transition from integrability to chaos via the formation of chaotic tangles in the aperiodic Lagrangian flow, and establish a direct link between the `lobes' associated with these tangles and intermittency in the observed turbulent dissipation. This illustrates how utilizing tools from deterministic dynamical systems theory to study convective nonlinearities can explain aspects of intermittent spatiotemporal structure exhibited by the statistics of turbulent fields.

physics.plasm-ph↗