arXiv subjects

Zhi Zhu

Publications and source records attributed to Zhi Zhu.

3 recordsLinked to original sources

arXiv2026

MiroThinker: Pushing the Performance Boundaries of Open-Source Research Agents via Model, Context, and Interactive Scaling

We present MiroThinker v1.0, an open-source research agent designed to advance tool-augmented reasoning and information-seeking capabilities. Unlike previous agents that only scale up model size or context length, MiroThinker explores interaction scaling at the model level, systematically training the model to handle deeper and more frequent agent-environment interactions as a third dimension of performance improvement. Unlike LLM test-time scaling, which operates in isolation and risks degradation with longer reasoning chains, interactive scaling leverages environment feedback and external information acquisition to correct errors and refine trajectories. Through reinforcement learning, the model achieves efficient interaction scaling: with a 256K context window, it can perform up to 600 tool calls per task, enabling sustained multi-turn reasoning and complex real-world research workflows. Across four representative benchmarks-GAIA, HLE, BrowseComp, and BrowseComp-ZH-the 72B variant achieves up to 81.9%, 37.7%, 47.1%, and 55.6% accuracy respectively, surpassing previous open-source agents and approaching commercial counterparts such as GPT-5-high. Our analysis reveals that MiroThinker benefits from interactive scaling consistently: research performance improves predictably as the model engages in deeper and more frequent agent-environment interactions, demonstrating that interaction depth exhibits scaling behaviors analogous to model size and context length. These findings establish interaction scaling as a third critical dimension for building next-generation open research agents, complementing model capacity and context windows.

cs.CL↗

arXiv2024

Quasiphase transition of a single-file water chain influenced by atomic charges in a water model using orientational-biased replica exchange Monte Carlo simulations

The recently observed temperature-dependent quasiphase transition of the single-file water chain confined within a carbon nanotube in experiments has been validated by the simple lattice theory and molecular dynamics simulations. It has been pointed out that the atomic charges in water models are important, yet how the values will affect the structural details and thermodynamic properties of the quasiphase transition has not been fully revealed. In this work, we perform orientational-biased replica exchange Monte Carlo simulations in the canonical ensemble to explore the effect of atomic charges in the SPC/E water model on the quasiphase transition of a single-file water chain. Based on the atomic charge values reported in literature, three distinct quasiphases are reproduced, comprising a fully hydrogen-bonded water chain at lower temperatures, a more ordered dipolar orientation along the tube axis at intermediate temperatures, and a completely disordered structure at higher temperatures. Then by increasing the atomic charge values, we find that the fragmentation of the entire water chain into shorter water segments, the orientational ordering of water dipoles along the tube axis, and the transition towards complete disorder are all inhibited. Consequently, the transition temperatures between three quasiphases have been shifted to higher temperatures. The thermodynamic analysis demonstrates that the increased atomic charge values enhance the hydrogen bonding between neighbouring water molecules also the electrostatic attraction within the water chain, leading to a longer water dipole correlation length even at higher temperatures. These findings highlight the vital role of atomic charges in water models also the electrostatic interaction in regulating the orientational ordering of water molecules under nanoconfinement.

cond-mat.soft↗

arXiv2017

Anomalous behavior of membrane fluidity caused by copper-copper bond coupled phospholipids

Membrane fluidity, well-known to be essential for cell functions, is obviously affected by copper. However, the underlying mechanism is still far from being understood, especially on the atomic level. Here, we unexpectedly observed that a decrease in phospholipid (PL) bilayer fluidity caused by Cu2+ was much more significant than those induced by Zn2+ and Ca2+, while a comparable reduction occurred in the last two ions. This finding clearly disagrees with the placement in the periodic table of Cu just next to Zn and far from Ca. The physical nature was revealed to be a special attraction between Cu+ cations, which can induce a motif forming of two phospholipids coupled by Cu-Cu bond (PL-diCu-PL). Namely, upon Cu2+ ion binding to a negatively charged phosphate group of lipid, Cu2+ was reduced to Cu+. The special attraction of the cations then caused one Cu+ ion simultaneously binding to two lipids and another Cu+, resulting in the formation of PL-diCu-PL structure. In contrast, this attraction cannot occur in the cases of Zn and Ca ions due to their electron structure. Remarkably, besides lipids, the phosphate group widely exists in other biological molecules, including DNA, RNA, ADP and ATP, which would also induce the similar structure of Cu ions with the molecules. Our findings thus provide a new view for understanding the biological functions of copper and the mechanism underlying copper-related diseases.

physics.bio-ph↗