arXiv subjects

Jinping Li

Publications and source records attributed to Jinping Li.

8 recordsLinked to original sources

arXiv2026

InfMem: Learning System-2 Memory Control for Long-Context Agent

Reasoning over ultra-long documents requires synthesizing sparse evidence scattered across distant segments under strict memory constraints. While streaming agents enable scalable processing, their passive memory update strategy often fails to preserve low-salience bridging evidence required for multi-hop reasoning. We propose InfMem, a control-centric agent that instantiates System-2-style control via a PreThink-Retrieve-Write protocol. InfMem actively monitors evidence sufficiency, performs targeted in-document retrieval, and applies evidence-aware joint compression to update a bounded memory. To ensure reliable control, we introduce a practical SFT-to-RL training recipe that aligns retrieval, writing, and stopping decisions with end-task correctness. On ultra-long QA benchmarks ranging from 32k to 1M tokens, InfMem consistently outperforms MemAgent across backbones. Specifically, InfMem improves average absolute accuracy by 10.17, 11.84, and 8.23 points on Qwen3-1.7B, Qwen3-4B, and Qwen2.5-7B, respectively, while reducing inference time by 3.9x on average and by up to 5.1x through adaptive early stopping. Code is available at https://github.com/UCMP13753/InfMem.

cs.CL↗

arXiv2023

Deep learning numerical methods for high-dimensional fully nonlinear PIDEs and coupled FBSDEs with jumps

We propose a deep learning algorithm for solving high-dimensional parabolic integro-differential equations (PIDEs) and high-dimensional forward-backward stochastic differential equations with jumps (FBSDEJs), where the jump-diffusion process are derived by a Brownian motion and an independent compensated Poisson random measure. In this novel algorithm, a pair of deep neural networks for the approximations of the gradient and the integral kernel is introduced in a crucial way based on deep FBSDE method. To derive the error estimates for this deep learning algorithm, the convergence of Markovian iteration, the error bound of Euler time discretization, and the simulation error of deep learning algorithm are investigated. Two numerical examples are provided to show the efficiency of this proposed algorithm.

math.NA↗

arXiv2022

Adaptive Model Refinement Approach for Bayesian Uncertainty Quantification in Turbulence Model

The Bayesian uncertainty quantification technique has become well established in turbulence modeling over the past few years. However, it is computationally expensive to construct a globally accurate surrogate model for Bayesian inference in a high-dimensional design space, which limits uncertainty quantification for complex flow configurations. Borrowing ideas from stratified sampling and inherited sampling, an adaptive model refinement approach is proposed in this work, which concentrates on asymptotically improving the local accuracy of the surrogate model in the high-posterior-density region by adaptively appending model evaluation points. To achieve this goal, a modification of inherited Latin hypercube sampling is proposed and then integrated into the Bayesian framework. The effectiveness and efficiency of the proposed approach are demonstrated through a two-dimensional heat source inversion problem and its extension to a high-dimensional design space. Compared with the prior-based method, the adaptive model refinement approach has the ability to obtain more reliable inference results using fewer evaluation points. Finally, the approach is applied to parametric uncertainty quantification of the Menter shear-stress transport turbulence model for an axisymmetric transonic bump flow and provides convincing numerical results.

physics.data-an↗

arXiv2018

The First-principles Study on the Mechanics, Optical and Phonon Properties of Carbon Chains

The two kinds of carbyne, i.e., cumulene and polyyne, are investigated by the first principles, where the mechanical properties, electronic structure, optical and phonon properties of the two carbynes are calculated. The results on the crystal binding energy and the formation energy show that polyyne is more stable and harder than cumulene, and both are difficult to be synthesized from diamond or graphite. The tensile stiffness, bond stiffness and Young's modulus of cumulene are 94.669 eV/Å, 90.334GPa and 60.62GPa, respectively; while the corresponding values of polyyne are 94.939eV/Å, 101.42GPa and 60.06GPa, respectively. The supercell calculation shows that carbyne is the most stable at N=5, where N is the supercell number. It indicates that the carbon chain with 10 atoms is the most stable. The calculation on the electronic band structure shows that cumulene is conductor, while polyyne is semiconductor with a band gap as 0.37eV. The dielectric function of carbynes varies in different direction, consistent with the one dimensional nature of the carbon chains. In the phonon dispersion of cumulene there are imaginary frequencies with the lowest value down to -3.817THz, which indicates that cumulene could be unstable at room temperature and normal pressure.

cond-mat.mtrl-sci↗

arXiv2016

A small shoulder of optical absorption in polycrystalline HfO2 by LDA+U approach

The dielectric function of the wide-gap optical material HfO2 is investigated by the local density approximation (LDA) +U approach. We focus on the origin of the shoulder-like structure near the edge of the band gap in the imaginary part of the dielectric function, which has been observed on the thin films of monoclinic HfO2. A comparison study on the three polymorphs of hafnia shows that regardless of the underlying crystal structure, the existence of the shoulder is directly controlled by the value of the shortest length of Hf-O bonds. The proposition is further supported by the numerical simulations of isostatic pressing. A possible implication in high-pressure measurements is suggested accordingly.

cond-mat.mtrl-sci↗

arXiv2016

First-principles study of electronic structure, optical and phonon properties of α-ZrW2O8

ZrW2O8 exhibits isotropic negative thermal expansions over its entire temperature range of stability, yet so far its physical properties and mechanism have not been fully addressed. In this article, the electronic structure, elastic, thermal, optical and phonon properties of α-ZrW2O8 are systematically investigated from first principles. The agreements between the generalized gradient approximation (GGA) calculation and experiments are found to be quite satisfactory. The calculation results can be useful in relevant material designs, e.g., when ZrW2O8 is employed to adjust the thermal expansion coefficient of ceramic matrix composites.

cond-mat.mtrl-sci↗

arXiv2013

Optical properties of monoclinic HfO2 studied by first-principles local density approximation +U approach

The band structures and optical properties of monoclinic HfO2 are investigated by the local density approximation (LDA)+U approach. With the on-site Coulomb interaction being introduced to 5d orbitals of Hf atom and 2p orbitals of O atom, the experimental band gap is reproduced. The imaginary part of the complex dielectric function shows a small shoulder at the edge of the band gap, coinciding with the experiments. This intrinsic property of crystallized monoclinic HfO2, which is absent in both the tetragonal phase and cubic phase, can be understood as a consequence of the reconstruction of the electronic states near the band edge following the adjustment of the crystal structure. The existence of a similar shoulder-like-structure in the monoclinic phase of ZrO2 is predicted.

cond-mat.mtrl-sci↗

arXiv2013

First-principles generalized gradient approximation (GGA) +Ud+Up studies of electronic structures and optical properties in cubic HfO2

The electronic structures and optical properties of cubic HfO2 are calculated by means of generalized gradient approximation (GGA) +U approach. Without on-site Coulomb interactions, the band gap of cubic HfO2 is 2.92eV, much lower than the experimental value (5.7eV). Introducing the Coulomb interactions of 5d orbitals on Hf atom (Ud) and of 2p orbitals on O atom (Up), we can reproduce the experimental value of the band gap. The calculated dielectric function of cubic HfO2 by the GGA+ Ud + Up approach predicts the presence of a shoulder structure below the main peak of the absorption spectrum. These indicate that the GGA +Ud +UP approach is a convenient and powerful method to calculate and predict the electronic structures and the optical properties of wide-gap optical materials.

cond-mat.mtrl-sci↗