arXiv subjects

Li Fu

Publications and source records attributed to Li Fu.

At least 19 recordsLinked to original sources

arXiv2026

PAPERBENCHX: Can AI Reproduce Science Across Disciplines?

Before AI can be trusted to discover new science, it should be able to reproduce what science already knows. Reproduction provides a rigorous checkpoint for an AI scientist: the target claim is known, yet recovering it still requires translating an incomplete scientific description into an executable experiment and producing evidence that can be independently verified. Existing research-reproduction benchmarks largely focus on AI and machine learning, where experiments share relatively homogeneous software stacks. Extending this capability across scientific disciplines introduces a different challenge: agents must operate domain-specific scientific software, make physical and numerical choices, diagnose long-running computations, and distinguish valid evidence from merely plausible outputs. We introduce PAPERBENCHX, a benchmark of 93 expert-constructed paper-reproduction tasks spanning 12 research areas across electromagnetics, photonics, chemistry, materials science, biology, and robotics. Each task provides the source paper, a public task specification, and a domain-native scientific environment involving specialized scientific software. Agents must reconstruct a key experiment, execute and diagnose it, and deliver a runnable workflow that supports the target scientific claim. To evaluate scientific validity rather than surface-level agreement, we replay every submission from scratch and score only regenerated evidence against expert-validated criteria. Across 10 frontier agent configurations, the best configuration fully reproduces only 14% tasks under the 90% threshold. Meanwhile, mean rubric scores range from 27.53% to 59.08%, indicating that substantial partial progress rarely translates into complete scientific reproduction. These results reveal a large gap between paper understanding and reliably producing reproducible scientific evidence.

cs.AI↗

arXiv2026

fix uvt and fix pimd/uvt: A Unified LAMMPS Framework for Constant-Potential Constant-Temperature Molecular Dynamics

Accurate simulations of electrochemical interfaces require the simultaneous treatment of constant-potential conditions, nuclear quantum effects, and sufficient configurational sampling. Integrating these capabilities within a general and efficient molecular dynamics (MD) framework remains challenging. In this work, we implement fix uvt and fix pimd/uvt in LAMMPS for constant-potential classical MD and path integral molecular dynamics (PIMD), respectively. We organize the class hierarchy to reuse LAMMPS's existing Nosé--Hoover chain thermostat routines and share nuclear propagation routines across PIMD integrators. A common interface connects these integrators to models that provide the electron-number derivative of the potential energy. We present three examples with accompanying input commands to guide users through constant-potential classical MD, thermostatted PIMD, and constant-potential PIMD, covering analytical models, liquid water, and electrochemical interfaces described by machine learning potentials. This work provides practical tools and guidance for large-scale constant-potential simulations incorporating nuclear quantum effects.

physics.chem-ph↗

arXiv2026

Navigating Sparse Singlet Fission Chemical Space: An Intelligent Generative-Predictive Paradigm

Singlet fission (SF) offers a promising route to surpass the Shockley-Queisser limit by converting a photoexcited singlet exciton into two triplet excitons, thereby enhancing photovoltaic energy conversion efficiency. However, realizing efficient SF process requires stringent energetic requirements among low-lying excited states that render SF molecules intrinsically rare within the vast chemical space. This extreme sparsity presents a grand challenge for molecular discovery. Due to low hit rates and trial-and-error computational waste on nonviable structures, conventional high-throughput virtual screening faces significant constraints, even when accelerated by machine learning models. Here, we establish a synergistic generative-predictive framework for the targeted inverse design of SF molecules by integrating a structure generator, a properties predictor and a multi-criteria validation workflow. By continuously coupling generative exploration with SF predictive models, the framework progressively enriches SF species and achieves a success rate of approximately 90% in generating molecules that satisfy the target SF energetic criteria. High-throughput evaluation of about 100 million generated structures with time-dependent density functional theory (TDDFT) validation of just a random 1% subset confirmed a 90.8% success rate for SF candidates. All together, we constructed an SF database of 283,559 candidates with favorable energetics of excited states and synthetic accessibility. From it, we identified a key fragment strongly associated with the requirements for SF, namely, CN([O])N(C)[O]. These findings establish an efficient route for overcoming the sparsity difficulty in SF molecular discovery and provide interpretable design principles for the development of novel excited-state functional materials.

cond-mat.mtrl-sci↗

arXiv2026

Dual-Form ASR: Semantics-Aware Inverse Text Normalization for Chinese Speech Recognition

Modern automatic speech recognition (ASR) scenarios require both spoken-form transcripts for faithful transcription and readable written-form transcripts with inverse text normalization (ITN). However, these forms are typically produced by cascaded modules, where a spoken-form ASR output is rewritten by a separate ITN component, making written-form ASR-ITN vulnerable to recognition errors and decoupling normalization from acoustic-contextual modeling, especially for semantically dependent numeric expressions. In this paper, we propose Dual-Form ASR (DF-ASR), a framework that extends spoken-form ASR capability to semantics-aware written-form ITN through paired spoken-form and written-form supervision while retaining prompt-level selection between transcript forms. The dual-form supervision is constructed via a large language model (LLM)-driven generate-and-judge workflow, and training is further enhanced by ITN-MWER, a sequence-level objective that assigns higher cost to errors on normalization-sensitive spans. We also introduce a decision-aware REQUIRE-ITN/\FORBID-ITN protocol to separately measure required normalization and forbidden-span preservation. On manually annotated Chinese subsets from SpeechIO, DF-ASR consistently outperforms open-source ASR-ITN systems, remains competitive with strong closed-source references, and preserves reliable prompt-level control between spoken-form and written-form outputs.

cs.CL↗

arXiv2026

Automated Discovery of Metainterfaces with Tailored Friction Laws

Providing dry solid contacts with on-demand macroscale frictional behaviour remains a formidable challenge in tribology, haptics or robotics. Metainterfaces created from surfaces with engineered asperity-based topographies can achieve such friction control. However, only few friction behaviours were demonstrated because suitable topographies were identified based on human intuition. Here, we introduce a numerical-optimisation-based inverse design framework to automatically discover new metainterfaces satisfying specified relationships between friction and normal forces (friction law). To illustrate the framework's versatility, we first expand the range of achievable friction coefficients at a constant material pair; we next unlock power-law friction laws with arbitrary exponents between 2/3 and 1.35; we then achieve bilinear laws with a smaller slope in the second segment than in the first. We validate relevant cases experimentally. By enabling systematic exploration of large parameter spaces, not limited to topography but potentially incorporating the individual asperities' bulk material or surface physicochemistry, our automated framework offers design solutions for any physically possible friction law. It also provides new insights into the elusive relationship between local interfacial properties and macroscopic friction.

physics.class-ph↗

arXiv2026

fix pimd/langevin: An Efficient Implementation of Path Integral Molecular Dynamics in LAMMPS

Path integral molecular dynamics (PIMD), which maps a quantum particle onto a fictitious classical system of ring polymers and propagates the "beads" of this extended classical system using molecular dynamics, is widely used to capture nuclear quantum effects (NQEs) in molecular simulations. Accurate PIMD calculations typically require a large number of beads and are therefore computationally demanding. While software packages such as i-PI offer comprehensive PIMD functionality, the high efficiency of simulations driven by machine learning interatomic potentials, such as Deep Potential (DP), calls for more efficient PIMD implementations that fully exploit modern massively parallel supercomputers. Here we present fix pimd/langevin, an efficient PIMD implementation in LAMMPS that supports commonly used features and leverages the Message Passing Interface architecture of LAMMPS to achieve high computational efficiency. We demonstrate the usage and validate the correctness of our code using liquid water as a representative example, and provide a comprehensive overview of the supported features. Then we discuss several important technical aspects of the implementation. Using DP simulations of water as a benchmark, we show that our implementation achieves several-fold acceleration compared to i-PI. Finally, we report strong and weak scaling results that demonstrate the favorable parallel performance of our code.

physics.chem-ph↗

arXiv2026

Efficient Screening of Organic Singlet Fission Molecules Using Graph Neural Networks

Singlet fission (SF) provides a promising strategy for surpassing the Shockley-Queisser limit in photovoltaics. However, the identification of efficient SF materials is hindered by the limited availability of suitable molecular candidates and the high computational costs associated with conventional quantum-chemical methods for excited states. In this study, we introduce a high-throughput screening framework that integrates a graph neural network (GNN) with multi-level validation to accelerate the discovery of SF-active molecules. Trained on a previously reported FORMED database, the GNN achieves state-of-the-art accuracy in predicting SF-relevant excited-state properties, demonstrating a mean absolute error of about 0.1 eV for S1, T1, and T2 excitation energies. This capability facilitates the efficient screening of over 20 million molecular structures from both OE62 and QO2Mol databases. Our framework significantly reduces the computational demand associated with Time-Dependent Density Functional Theory validation by four orders of magnitude and identifies 180 potential SF molecules along with more than 1000 conformers. Subsequent assessments regarding synthetic accessibility, GW approximation and Bethe-Salpeter equation calculations further highlight a subset of experimentally feasible candidates among these SF candidates. The approach presented herein exemplifies an effective strategy for accelerating the discovery of functional molecules with optoelectronic applications.

cond-mat.mtrl-sci↗

arXiv2025

PAC: Pronunciation-Aware Contextualized Large Language Model-based Automatic Speech Recognition

This paper presents a Pronunciation-Aware Contextualized (PAC) framework to address two key challenges in Large Language Model (LLM)-based Automatic Speech Recognition (ASR) systems: effective pronunciation modeling and robust homophone discrimination. Both are essential for raw or long-tail word recognition. The proposed approach adopts a two-stage learning paradigm. First, we introduce a pronunciation-guided context learning method. It employs an interleaved grapheme-phoneme context modeling strategy that incorporates grapheme-only distractors, encouraging the model to leverage phonemic cues for accurate recognition. Then, we propose a pronunciation-discriminative reinforcement learning method with perturbed label sampling to further enhance the modelś ability to distinguish contextualized homophones. Experimental results on the public English Librispeech and Mandarin AISHELL-1 datasets indicate that PAC: (1) reduces relative Word Error Rate (WER) by 30.2% and 53.8% compared to pre-trained LLM-based ASR models, and (2) achieves 31.8% and 60.5% relative reductions in biased WER for long-tail words compared to strong baselines, respectively.

cs.CL↗

arXiv2025

Electrochemistry-Enhanced Dynamic Paths Sampling Unveiling Nuclear Quantum Effects in Electrocatalysis

Proton-coupled electron transfers (PCET) are elementary steps in electrocatalysis. However, accurate calculations of PCET rates remain challenging, especially considering nuclear quantum effects (NQEs) under a constant potential condition. Statistical sampling of reaction paths is an ideal approach for rate calculations, however, is always limited by the rare-event issue. Here we develop an electrochemistry-driven quantum dynamics approach enabling realistic enhanced paths sampling under constant potentials without a priori defined reaction coordinates. We apply the method in modeling the Volmer step of the hydrogen evolution reaction, and demonstrate that the NQEs exhibit more than one order of magnitude impact on the computed rate constant, indicating an essential role of NQEs in electrochemistry.

physics.chem-ph↗

arXiv2025

Defining implication relation for classical logic

In classical logic, "P implies Q" is equivalent to "not-P or Q". It is well known that the equivalence is problematic. Actually, from "P implies Q", "not-P or Q" can be inferred ("Implication-to-Disjunction" is valid), whereas from "not-P or Q", "P implies Q" cannot be inferred in general ("Disjunction-to-Implication" is not generally valid), so the equivalence between them is invalid in general. This work aims to remove the incorrect Disjunction-to-Implication from classical logic (CL). The logical system (the logic IRL) this paper proposes has the expected properties: (a) CL is obtained by adding Disjunction-to-Implication to IRL, and (b) Disjunction-to-Implication is not derivable in IRL; while (c) fundamental laws in classical logic, including law of excluded middle (LEM) and principle of double negation, law of non-contradiction (LNC) and ex contradictione quodlibet (ECQ), conjunction elimination and disjunction introduction, and hypothetical syllogism and disjunctive syllogism, are all retained in IRL.

math.LO↗

arXiv2025

Efficiently charting the space of mixed vacancy-ordered perovskites by machine-learning encoded atomic-site information

Vacancy-ordered double perovskites (VODPs) are promising alternatives to three-dimensional lead halide perovskites for optoelectronic and photovoltaic applications. Mixing these materials creates a vast compositional space, allowing for highly tunable electronic and optical properties. However, the extensive chemical landscape poses significant challenges in efficiently screening candidates with target properties. In this study, we illustrate the diversity of electronic and optical characteristics as well as the nonlinear mixing effects on electronic structures within mixed VODPs. For mixed systems with limited local environment options, the information regarding atomic-site occupation in-principle determines both structural configurations and all essential properties. Building upon this concept, we have developed a model that integrates a data-augmentation scheme with a transformer-inspired graph neural network (GNN), which encodes atomic-site information from mixed systems. This approach enables us to accurately predict band gaps and formation energies for test samples, achieving Root Mean Square Errors (RMSE) of 21 meV and 3.9 meV/atom, respectively. Trained with datasets that include (up to) ternary mixed systems and supercells with less than 72 atoms, our model can be generalized to medium- and high-entropy mixed VODPs (with 4 to 6 principal mixing elements) and large supercells containing more than 200 atoms. Furthermore, our model successfully reproduces experimentally observed bandgap bowing in Sn-based mixed VODPs and reveals an unconventional mixing effect that can result in smaller band gaps compared to those found in pristine systems.

cond-mat.mtrl-sci↗

arXiv2024

UME: Upcycling Mixture-of-Experts for Scalable and Efficient Automatic Speech Recognition

Recent advancements in scaling up models have significantly improved performance in Automatic Speech Recognition (ASR) tasks. However, training large ASR models from scratch remains costly. To address this issue, we introduce UME, a novel method that efficiently Upcycles pretrained dense ASR checkpoints into larger Mixture-of-Experts (MoE) architectures. Initially, feed-forward networks are converted into MoE layers. By reusing the pretrained weights, we establish a robust foundation for the expanded model, significantly reducing optimization time. Then, layer freezing and expert balancing strategies are employed to continue training the model, further enhancing performance. Experiments on a mixture of 170k-hour Mandarin and English datasets show that UME: 1) surpasses the pretrained baseline by a margin of 11.9% relative error rate reduction while maintaining comparable latency; 2) reduces training time by up to 86.7% and achieves superior accuracy compared to training models of the same size from scratch.

eess.AS↗

arXiv2024

Neural2Speech: A Transfer Learning Framework for Neural-Driven Speech Reconstruction

Reconstructing natural speech from neural activity is vital for enabling direct communication via brain-computer interfaces. Previous efforts have explored the conversion of neural recordings into speech using complex deep neural network (DNN) models trained on extensive neural recording data, which is resource-intensive under regular clinical constraints. However, achieving satisfactory performance in reconstructing speech from limited-scale neural recordings has been challenging, mainly due to the complexity of speech representations and the neural data constraints. To overcome these challenges, we propose a novel transfer learning framework for neural-driven speech reconstruction, called Neural2Speech, which consists of two distinct training phases. First, a speech autoencoder is pre-trained on readily available speech corpora to decode speech waveforms from the encoded speech representations. Second, a lightweight adaptor is trained on the small-scale neural recordings to align the neural activity and the speech representation for decoding. Remarkably, our proposed Neural2Speech demonstrates the feasibility of neural-driven speech reconstruction even with only 20 minutes of intracranial data, which significantly outperforms existing baseline methods in terms of speech fidelity and intelligibility.

cs.SD↗

arXiv2024

Do self-supervised speech and language models extract similar representations as human brain?

Speech and language models trained through self-supervised learning (SSL) demonstrate strong alignment with brain activity during speech and language perception. However, given their distinct training modalities, it remains unclear whether they correlate with the same neural aspects. We directly address this question by evaluating the brain prediction performance of two representative SSL models, Wav2Vec2.0 and GPT-2, designed for speech and language tasks. Our findings reveal that both models accurately predict speech responses in the auditory cortex, with a significant correlation between their brain predictions. Notably, shared speech contextual information between Wav2Vec2.0 and GPT-2 accounts for the majority of explained variance in brain activity, surpassing static semantic and lower-level acoustic-phonetic information. These results underscore the convergence of speech contextual representations in SSL models and their alignment with the neural network underlying speech perception, offering valuable insights into both SSL models and the neural basis of speech and language processing.

q-bio.NC↗

arXiv2023

The first Ka-band (26.1-35 GHz) blind line survey towards Orion KL

We conducted a Ka-band (26.1--35 GHz) line survey towards Orion KL using the TianMa 65-m Radio Telescope (TMRT). It is the first blind line survey in the Ka band, and achieves a sensitivity of mK level (1--3 mK at a spectral resolution of $\sim$1 km s$^{-1}$). In total, 592 Gaussian features are extracted. Among them, 257 radio recombination lines (RRLs) are identified. The maximum $Δn$ of RRLs of H, He and C are 20, 15, and 5, respectively. Through stacking, we have detected the $β$ lines of ion RRLs (RRLs of C$^+$ with possible contribution of other ions like O$^+$) for the first time, and tentative signal of the $γ$ lines of ion RRLs can also be seen on the stacked spectrum. Besides, 318 other line features were assigned to 37 molecular species, and ten of these species were not detected in the Q-band survey of TMRT. The vibrationally excited states of nine species were also detected. Emission of most species can be modeled under LTE. A number of transitions of E-CH3OH ($J_2-J_1$) display maser effects, which are confirmed by our modeling, and besides the bumping peak at $J\sim 6$ there is another peak at $J\sim 13$. Methylcyanoacetylene (CH$_3$C$_3$N) is detected in Orion KL for the first time. This work emphasizes that the Ka band, which was long-ignored for spectral line surveys, is very useful for surveying RRLs and molecular lines simultaneously.

astro-ph.GA↗

arXiv2023

Ionic blockade in a charged single-file water channel

The classical continuum theories fail to describe the ionic transport in Angstrom channels, where conduction deviates from Ohm's law, as attributed to dehydration/self-energy barrier and dissociation of Bjerrum ion-pairs in previous work. Here we found that the cations are strongly bound to the surface charge that blockade the ionic transport in a single-file water channel, causing nonlinear current-voltage responses. The presence of free ions significantly increased the probability of bound ions being released, resulting in an ionic current. We found that ionic conduction gradually becomes Ohmic as surface charge density increases, but the conduction amplitude decreased due to increased friction from bound ions. We rationalized the ionic transport by 1D Kramers' escape theory framework, which well described nonlinear ionic current, and the impact of surface charge density on turning to Ohmic system. Our results possibly provide an alternative view of ionic blockade in Angstrom channels.

cond-mat.soft↗

arXiv2023

OTF: Optimal Transport based Fusion of Supervised and Self-Supervised Learning Models for Automatic Speech Recognition

Self-Supervised Learning (SSL) Automatic Speech Recognition (ASR) models have shown great promise over Supervised Learning (SL) ones in low-resource settings. However, the advantages of SSL are gradually weakened when the amount of labeled data increases in many industrial applications. To further improve the ASR performance when abundant labels are available, we first explore the potential of combining SL and SSL ASR models via analyzing their complementarity in recognition accuracy and optimization property. Then, we propose a novel Optimal Transport based Fusion (OTF) method for SL and SSL models without incurring extra computation cost in inference. Specifically, optimal transport is adopted to softly align the layer-wise weights to unify the two different networks into a single one. Experimental results on the public 1k-hour English LibriSpeech dataset and our in-house 2.6k-hour Chinese dataset show that OTF largely outperforms the individual models with lower error rates.

eess.AS↗

arXiv2023

UFO2: A unified pre-training framework for online and offline speech recognition

In this paper, we propose a Unified pre-training Framework for Online and Offline (UFO2) Automatic Speech Recognition (ASR), which 1) simplifies the two separate training workflows for online and offline modes into one process, and 2) improves the Word Error Rate (WER) performance with limited utterance annotating. Specifically, we extend the conventional offline-mode Self-Supervised Learning (SSL)-based ASR approach to a unified manner, where the model training is conditioned on both the full-context and dynamic-chunked inputs. To enhance the pre-trained representation model, stop-gradient operation is applied to decouple the online-mode objectives to the quantizer. Moreover, in both the pre-training and the downstream fine-tuning stages, joint losses are proposed to train the unified model with full-weight sharing for the two modes. Experimental results on the LibriSpeech dataset show that UFO2 outperforms the SSL-based baseline method by 29.7% and 18.2% relative WER reduction in offline and online modes, respectively.

eess.AS↗