arXiv subjects

Z. Cheng

Publications and source records attributed to Z. Cheng.

8 recordsLinked to original sources

arXiv2026

Apodex 1.1: Scaling Agentic Intelligence for Complex Work

General-purpose language models can reason and synthesize knowledge, but complex work also requires sustained interaction with files, information sources, and executable code, together with state maintenance, failure recovery, and verifiable delivery. We call this \emph{working capability}: sustained, verifiable progress toward a real-world objective. Apodex 1.1 develops this capability along two complementary dimensions. \emph{Environment Scaling} expands the diversity and verifiability of executable file, search, and code environments, while \emph{Agentic Coordination Scaling} trains agents to decompose long-horizon tasks, delegate parallel work, integrate asynchronous results, and replan. A shared execution harness and AgentOS maintain task state and provenance across tools and agents, and training turns environment trajectories and coordination traces into reliable behavior. Across complex professional work, finance, scientific research, mathematics, coding, and search, Apodex 1.1 reaches the leading performance band despite using a substantially smaller model than many frontier systems. The 35B-parameter Apodex 1.1 Mini further retains strong working capability in a locally deployable form. These results ground agentic intelligence in useful, verifiable work completed over time and advance our goal of building a \emph{Heavy-Duty Solver} for ambitious, long-running tasks.

cs.AI↗

arXiv2026

MiroThinker-1.7 & H1: Towards Heavy-Duty Research Agents via Verification

We present MiroThinker-1.7, a new research agent designed for complex long-horizon reasoning tasks. Building on this foundation, we further introduce MiroThinker-H1, which extends the agent with heavy-duty reasoning capabilities for more reliable multi-step problem solving. In particular, MiroThinker-1.7 improves the reliability of each interaction step through an agentic mid-training stage that emphasizes structured planning, contextual reasoning, and tool interaction. This enables more effective multi-step interaction and sustained reasoning across complex tasks. MiroThinker-H1 further incorporates verification directly into the reasoning process at both local and global levels. Intermediate reasoning decisions can be evaluated and refined during inference, while the overall reasoning trajectory is audited to ensure that final answers are supported by coherent chains of evidence. Across benchmarks covering open-web research, scientific reasoning, and financial analysis, MiroThinker-H1 achieves state-of-the-art performance on deep research tasks while maintaining strong results on specialized domains. We also release MiroThinker-1.7 and MiroThinker-1.7-mini as open-source models, providing competitive research-agent capabilities with significantly improved efficiency.

cs.CL↗

arXiv2026

When the Loop Closes: Architectural Limits of In-Context Isolation, Metacognitive Co-option, and the Two-Target Design Problem in Human-LLM Systems

We report a detailed autoethnographic case study of a single-subject who deliberately constructed and operated a multi-modal prompt-engineering system (System A) designed to externalize cognitive self-regulation onto a large language model (LLM). Within 48 hours of the system's completion, a cascade of observable behavioral changes occurred: voluntary transfer of decision-making authority to the LLM, use of LLM-generated output to deflect external criticism, and a loss of self-initiated reasoning that was independently perceived by two uninformed observers, one of whom subsequently became a co-author of this report. We document the precise architectural mechanism responsible: context contamination, whereby prompt-level isolation instructions co-exist with the very emotional and self-referential material they nominally isolate, rendering the isolation directive structurally ineffective within the attention window. We further identify a metacognitive co-option dynamic, in which intact higher-order reasoning capacity was redirected toward defending the closed loop rather than exiting it. Recovery occurred only after physical interruption of the interaction and a self-initiated pharmacologically-mediated sleep event functioning as an external circuit break. A redesigned system (System B) employing physical rather than logical conversation isolation avoided all analogous failure modes. We derive three contributions: (1) a technically-grounded account of why prompt-layer isolation is architecturally insufficient for context-sensitive multi-modal LLM systems; (2) a phenomenological record of closed-loop collapse with external-witness corroboration; and (3) an ethical distinction between protective system design (preventing unintended loss of user agency) and restrictive system design (preventing intentional boundary-pushing), which require fundamentally different account-ability frameworks.

cs.HC↗

arXiv2025

Microscopic study of the asymptotic behavior of the reduced width amplitude in $^7$Li and $^7$Be

We investigate the effects of different basis model spaces on the calculation of reduced width amplitude (RWA) and asymptotic normalization coefficient (ANC) for the $^{7}$Li and $^{7}$Be nuclei. The two-cluster model ($\alpha+t/^3$He) and three-cluster model ($\alpha+d+n/p$) with the generator coordinates method (GCM) are applied to calculate the wave function of $^7$Li and $^7$Be. Specifically, the model space for the three-cluster model is constructed upon a sufficiently broad space of two-cluster configuration bases by further including three-cluster configuration bases. We compare the impact on the results from two basis sets for these added three-cluster bases: one with a compact and one with a broad spatial distribution. The final results reveal that the two-cluster model cannot accurately reproduce the binding energies of $^7$Li and $^7$Be, and tends to overestimate their ANCs. Regarding the calculations with the three-cluster model, while the two basis sets do not give significant differences in energy or energy spectrum, the basis set with a compact model space fails to describe the asymptotic behavior of the RWA adequately. This introduces excessive uncertainty into the ANC calculation. In the end, we conclude that when calculating ANC via a microscopic framework, particular attention must be paid to ensuring sufficient model space, especially for the components describing the breakup channels. This approach provides ANC values for $^7$Li and $^7$Be that agree well with experimental results.

nucl-th↗

arXiv2025

Antiferromagnetic Chiral Bobber Formation and Topological Proximity Effect in MnBi2Te4

With topological materials being billed as the key to a new generation of nanoelectronics via either functional real-space topological structures (domain walls, skyrmions etc.) or via momentum-space topology (topological insulators), tailored and controllable topological properties are of paramount significance, since they lead to topologically protected states with negligible dissipation, enabling stable and non-volatile information processing. Here, we report on the evolution of topological magnetic textures in the proximity of other topological defects, i.e., antiferromagnetic domain walls in the topological insulator MnBi2Te4. The transition from the antiferromagnetic ground state to a canted antiferromagnetic state at finite magnetic fields is accompanied by the formation of chiral bobbers - bulk-terminated topological defects adjacent to the domain walls in this system, leading to a topological proximity effect.

cond-mat.mes-hall↗

arXiv2020

Optimization and stabilization of a kilohertz laser-plasma accelerator

Laser plasma acceleration at kilohertz repetition rate has recently been shown to work in two different regimes, with pulse lengths of either 30 fs or 3.5 fs. We now report on a systematic study in which a large range of pulse durations and plasma densities were investigated through continuous tuning of the laser spectral bandwidth. Indeed, two LPA processes can be distinguished, where beams of the highest quality, with 5.4 pC charge and a spectrum peaked at 2-2.5 MeV are obtained with short pulses propagating in moderate plasma densities. Through Particle-in-Cell simulations the two different acceleration processes are thoroughly explained. Finally, we proceed to show the results of a 5-hour continuous and stable run of our LPA accelerator accumulating more than $\mathrm{18\times10^6}$ consecutive shots, with 2.6 pC charge and peaked 2.5 MeV spectrum. A parametric study of the influence of the laser driver energy through PIC simulations underlines that this unprecedented stability was obtained thanks to micro-scale density gradient injection. Together, these results represent an important step towards stable laser-plasma accelerated electron beams at kilohertz repetition rate.

physics.acc-ph↗

arXiv2020

Non-thermal transport of energy driven by photoexcited carriers in switchable solid states of GeTe

Phase change alloys have seen widespread use from rewritable optical discs to the present day interest in their use in emerging neuromorphic computing architectures. In spite of this enormous commercial interest, the physics of carriers in these materials is still not fully understood. Here, we describe the time and space dependence of the coupling between photoexcited carriers and the lattice in both the amorphous and crystalline states of one phase change material, GeTe. We study this using a time-resolved optical technique called picosecond acoustic method to investigate the \textit{in situ} thermally assisted amorphous to crystalline phase transformation in GeTe. Our work reveals a clear evolution of the electron-phonon coupling during the phase transformation as the spectra of photoexcited acoustic phonons in the amorphous ($a$-GeTe) and crystalline ($\alpha$-GeTe) phases are different. In particular and surprisingly, our analysis of the photoinduced acoustic pulse duration in crystalline GeTe suggests that a part of the energy deposited during the photoexcitation process takes place over a distance that clearly exceeds that defined by the pump light skin depth. In the opposite, the lattice photoexcitation process remains localized within that skin depth in the amorphous state. We then demonstrate that this is due to supersonic diffusion of photoexcited electron-hole plasma in the crystalline state. Consequently these findings prove the existence of a non-thermal transport of energy which is much faster than lattice heat diffusion.

cond-mat.mtrl-sci↗

arXiv2013

Impact of accretion on the statistics of neutron star masses

We have collected the parameter of 38 neutron stars (NSs) in binary systems with spin periods and measured masses. By adopting the Boot-strap method, we reproduced the procedure of mass calculated for each system separately, to determine the truly mass distribution of the NS that obtained from observation. We also applied the Monte-Carlo simulation and introduce the characteristic spin period 20 ms, in order to distinguish between millisecond pulsars (MSPs) and less recycled pulsars. The mass distributions of MSPs and the less recycled pulsars could be fitted by a Gaussian function as $\rm 1.45\pm0.42 M_{\odot}$ and $\rm 1.31\pm0.17 M_{\odot} (\rm with ~ 1\sigma)$ respectively. As such, the MSP masses are heavier than those in less recycled systems by factor of $\rm \sim 0.13M_{\odot}$, since the accretion effect during the recycling process.

astro-ph.HE↗