We introduce ChemPro, a progressive benchmark with 4100 natural language question-answer pairs in Chemistry, across 4 coherent sections of difficulty designed to assess the proficiency of Large Language Models (LLMs) in a broad spectrum of general chemistry topics. We include Multiple Choice Questions and Numerical Questions spread across fine-grained information recall, long-horizon reasoning, multi-concept questions, problem-solving with nuanced articulation, and straightforward questions in a balanced ratio, effectively covering Bio-Chemistry, Inorganic-Chemistry, Organic-Chemistry and Physical-Chemistry. ChemPro is carefully designed analogous to a student's academic evaluation for basic to high-school chemistry. A gradual increase in the question difficulty rigorously tests the ability of LLMs to progress from solving basic problems to solving more sophisticated challenges. We evaluate 45+7 state-of-the-art LLMs, spanning both open-source and proprietary variants, and our analysis reveals that while LLMs perform well on basic chemistry questions, their accuracy declines with different types and levels of complexity. These findings highlight the critical limitations of LLMs in
To enhance large language models (LLMs) for chemistry problem solving, several LLM-based agents augmented with tools have been proposed, such as ChemCrow and Coscientist. However, their evaluations are narrow in scope, leaving a large gap in understanding the benefits of tools across diverse chemistry tasks. To bridge this gap, we develop ChemToolAgent, an enhanced chemistry agent over ChemCrow, and conduct a comprehensive evaluation of its performance on both specialized chemistry tasks and general chemistry questions. Surprisingly, ChemToolAgent does not consistently outperform its base LLMs without tools. Our error analysis with a chemistry expert suggests that: For specialized chemistry tasks, such as synthesis prediction, we should augment agents with specialized tools; however, for general chemistry questions like those in exams, agents' ability to reason correctly with chemistry knowledge matters more, and tool augmentation does not always help.
The spatial distribution of the chemical reservoirs in protoplanetary disks is key to elucidate the composition of planets, especially habitable ones. However, the partitioning of the main elements among the refractory and volatile phases is still elusive. Key parameters such as the carbon-to-oxygen C/O elemental ratio and the ionization fraction remain poorly constrained, with the latter potentially orders of magnitude lower than in the interstellar medium. Moreover, the thermal structure of the gas is also poorly known, despite its deep influence on gas-phase chemistry. In this context, ortho-to-para ratios could provide selective and sensitive probes. Recent ALMA observations have measured the spatially resolved column density of ortho-and para-H2CO in the transition disk orbiting TW Hya and derived the radial profile of the ortho-to-para ratio. Yet, current disk models do not include the nuclear-spin-resolved chemistry required to interpret these observations. The present work aims to fill this gap, by combining a parametric disk physical model of TW Hya with the UGAN network, updated to include a comprehensive description of the nuclear-spin-resolved chemistry of formaldehyde.
Biological theory usually treats time as an external chronological variable against which growth, aging, and ecological change are parametrized. Yet living systems also generate an internal measure of duration through physiological cycling and irreversible entropy production, and the regularities of allometric lifespan scaling, biological clocks, life-history evolution, ecological synchronization, and disease are ordinarily studied in isolation rather than within a single thermodynamic internal-time framework. The Principle of Biological Time Equivalence (PBTE) proposes such a framework.
Multimodal scientific reasoning remains a significant challenge for large language models (LLMs), particularly in chemistry, where problem-solving relies on symbolic diagrams, molecular structures, and structured visual data. Here, we systematically evaluate 40 proprietary and open-source multimodal LLMs, including GPT-5, o3, Gemini-2.5-Pro, and Qwen2.5-VL, on a curated benchmark of Olympiad-style chemistry questions drawn from over two decades of U.S. National Chemistry Olympiad (USNCO) exams. These questions require integrated visual and textual reasoning across diverse modalities. We find that many models struggle with modality fusion, where in some cases, removing the image even improves accuracy, indicating misalignment in vision-language integration. Chain-of-Thought prompting consistently enhances both accuracy and visual grounding, as demonstrated through ablation studies and occlusion-based interpretability. Our results reveal critical limitations in the scientific reasoning abilities of current MLLMs, providing actionable strategies for developing more robust and interpretable multimodal systems in chemistry. This work provides a timely benchmark for measuring progress in
Observable performance is commonly used to characterize biological systems, yet aggregated outputs may remain insufficient for uniquely resolving observational conditions, and richer multivariate representations may retain substantial ambiguity. This article proposes a representational bootstrap framework for adaptive biological systems. Bootstrap is used in a methodological and epistemological sense, not as statistical resampling. New analytical levels emerge when the active representation becomes insufficient for the question under investigation. The framework comprises five successive levels: observable performance, conceptual dynamic organization, exploratory multivariate representation, observed longitudinal centroid displacement, and internal approximation of observed displacement. Three previously reported gait-occlusion studies are used as a methodological case sequence rather than as new experimental evidence. The revised first study showed persistent static representational non-identifiability: neither the scalar score nor the exploratory embedding uniquely resolved the occlusal probes. The second study shifted the question toward M1-M2 centroid displacement in a common P
This paper surveys foundation models for AI-enabled biological design, focusing on recent developments in applying large-scale, self-supervised models to tasks such as protein engineering, small molecule design, and genomic sequence design. Though this domain is evolving rapidly, this survey presents and discusses a taxonomy of current models and methods. The focus is on challenges and solutions in adapting these models for biological applications, including biological sequence modeling architectures, controllability in generation, and multi-modal integration. The survey concludes with a discussion of open problems and future directions, offering concrete next-steps to improve the quality of biological sequence generation.
Accelerated materials discovery is critical for addressing global challenges. However, developing new laboratory workflows relies heavily on real-world experimental trials, and this can hinder scalability because of the need for numerous physical make-and-test iterations. Here we present MATTERIX, a multiscale, graphics processing unit-accelerated robotic simulation framework designed to create high-fidelity digital twins of chemistry laboratories, thus accelerating workflow development. This multiscale digital twin simulates robotic physical manipulation, powder and liquid dynamics, device functionalities, heat transfer and basic chemical reaction kinetics. This is enabled by integrating realistic physics simulation and photorealistic rendering with a modular graphics processing unit-accelerated semantics engine, which models logical states and continuous behaviors to simulate chemistry workflows across different levels of abstraction. MATTERIX streamlines the creation of digital twin environments through open-source asset libraries and interfaces, while enabling flexible workflow design via hierarchical plan definition and a modular skill library that incorporates learning-based
Primates exhibit a robust deviation from canonical allometric scaling: at fixed body mass, their lifespans exceed those of non-primate mammals by factors of two to three. A rhesus macaque (8 kg) lives 25-40 years, whereas a cat of similar mass rarely exceeds 18 years. This statistically significant clade-level excess cannot be explained by standard metabolic or ecological models. We provide a thermodynamic explanation within the Principle of Biological Time Equivalence (PBTE), where lifespan is determined by a finite cycle budget governed by entropy production. We show that primates reduce entropy production per physiological cycle through increased neural energy allocation. The neural power fraction acts as a control parameter, extending the effective lifetime cycle count. Three mechanisms, predictive regulation, enhanced repair, and behavioral buffering, jointly suppress dissipation. This yields a quantitative neuro-metabolic multiplier that explains primate longevity and provides testable predictions linking brain energetics, entropy production, and lifespan.
Efficient chemical kinetic model inference and application in combustion are challenging due to large ODE systems and widely separated time scales. Machine learning techniques have been proposed to streamline these models, though strong nonlinearity and numerical stiffness combined with noisy data sources make their application challenging. Here, we introduce ChemKANs, a novel neural network framework with applications both in model inference and simulation acceleration for combustion chemistry. ChemKAN's novel structure augments the generic Kolmogorov Arnold Network Ordinary Differential Equations (KAN-ODEs) with knowledge of the information flow through the relevant kinetic and thermodynamic laws. This chemistry-specific structure combined with the expressivity and rapid neural scaling of the underlying KAN-ODE algorithm instills in ChemKANs a strong inductive bias, streamlined training, and higher accuracy predictions compared to standard benchmarks, while facilitating parameter sparsity through shared information across all inputs and outputs. In a model inference investigation, we benchmark the robustness of ChemKANs to sparse data containing up to 15% added noise, and superfl
Many biological processes can be thought of as the result of an underlying dynamics in which the system repeatedly undergoes distinct and abortive trajectories with the dynamical process only ending when some specific process, purpose, structure or function is achieved. A classic example is the way in which microtubules attach to kinetochores as a prerequisite for chromosome segregation and cell division. In this example, the dynamics is characterized by apparently futile time histories in which microtubules repeatedly grow and shrink without chromosomal attachment. We hypothesize that for biological processes for which it is not the initial conditions that matter, but rather the final state, this kind of exploratory dynamics is biology's unique and necessary solution to achieving these functions with high fidelity. This kind of cause and effect relationship can be contrasted to examples from physics and chemistry where the initial conditions determine the outcome. In this paper, we examine the similarities of many biological processes that depend upon random trajectories starting from the initial state and the selection of subsets of these trajectories to achieve some desired func
Cyanopolyynes, a family of nitrogen containing carbon chains, are common in the interstellar medium and possibly form the backbone of species relevant to prebiotic chemistry. Following their gas phase formation, they are expected to freeze out on ice grains in cold interstellar regions. In this work we present the hydrogenation reaction network of isolated HC_{3}N, the smallest cyanopolyyne, that consists over-a-barrier radical-neutral reactions and barrierless radical-radical reactions. We employ density functional theory, coupled cluster and multiconfigurational methods to obtain activation and reaction energies for the hydrogenation network of HC_{3}N. This work explores the reaction network of the isolated molecule and constitutes a preview on the reactions occurring on the ice grain surface. We find that the reactions where the hydrogen atom adds to the carbon chain at carbon atom opposite of the cyano-group give the lowest and most narrow barriers. Subsequent hydrogenation leads to the astrochemically relevant vinyl cyanide and ethyl cyanide. Alternatively, the cyano-group can hydrogenate via radical-radical reactions, leading to the fully saturated propylamine. These results
Are biological self-organising systems more ``intelligent'' than artificial intelligence (AI)? If so, why? I address this question using a mathematical framework that defines intelligence in terms of adaptability. Systems are modelled as stacks of abstraction layers (\emph{Stack Theory}) and compared by how effectively they delegate agentic control down their stacks. I illustrate this using computational, biological, military, governmental, and economic systems. Contemporary AI typically relies on static, human-engineered stacks whose lower layers are fixed during deployment. Put provocatively, such systems resemble inflexible bureaucracies that adapt only top-down. Biological systems are more intelligent because they delegate adaptation. Formally, I prove a theorem (\emph{The Law of the Stack}) showing that adaptability at higher layers is bottlenecked by adaptability at lower layers. I further show that, under standard viability assumptions, maximising adaptability is equivalent to minimising variational free energy, implying that delegation is necessary for free-energy minimisation. Generalising bioelectric accounts of cancer as isolation from collective informational structures
Three-body recombination, or ternary association, is a termolecular reaction in which three particles collide, forming a bound state between two, whereas the third escapes freely. Three-body recombination reactions play a significant role in many systems relevant to physics and chemistry. In particular, they are relevant in cold and ultracold chemistry, quantum gases, astrochemistry, atmospheric physics, physical chemistry, and plasma physics. As a result, three-body recombination has been the subject of extensive work during the last 50 years, although primarily from an experimental perspective. Indeed, a general theory for three-body recombination remains elusive despite the available experimental information. Our group recently developed a direct approach based on classical trajectory calculations in hyperspherical coordinates for three-body recombination to amend this situation, leading to a first principle explanation of ion-atom-atom and atom-atom-atom three-body recombination processes. This review aims to summarize our findings on three-body recombination reactions and identify the remaining challenges in the field.
We present two open-source implementations of the Locally Optimal Block Preconditioned Conjugate Gradient (LOBPCG) algorithm to find a few eigenvalues and eigenvectors of large, possibly sparse matrices. We then test LOBPCG for various quantum chemistry problems, encompassing medium to large, dense to sparse, wellbehaved to ill-conditioned ones, where the standard method typically used is Davidson's diagonalization. Numerical tests show that, while Davidson's method remains the best choice for most applications in quantum chemistry, LOBPCG represents a competitive alternative, especially when memory is an issue, and can even outperform Davidson for ill-conditioned, non diagonally dominant problems.
Kohn-Sham density functional theory is in principle an exact formulation of quantum mechanical electronic structure theory, but in practice we have to rely on approximate exchange-correlation (xc) functionals. The objective of our work has been to design an xc functional with broad accuracy across as wide an expanse of chemistry and physics as possible, leading-as a long-range goal-to a functional with good accuracy for all problems, i.e., a universal functional. To guide our path toward that goal and to measure our progress, we have developed-building on earlier work in our group-a set of databases of reference data for a variety of energetic and structural properties in chemistry and physics. These databases include energies of molecular processes such as atomization, complexation, proton addition, and ionization; they also include molecular geometries and solid-state lattice constants, chemical reaction barrier heights, and cohesive energies and band gaps of solids. For the present paper we gather many of these databases into four comprehensive databases, two with 384 energetic data for chemistry and solid-state physics and another two with 68 structural data for chemistry and s
Biological systems, from a cell to the human brain, are inherently complex. A powerful representation of such systems, described by an intricate web of relationships across multiple scales, is provided by complex networks. Recently, several studies are highlighting how simple networks -- obtained by aggregating or neglecting temporal or categorical description of biological data -- are not able to account for the richness of information characterizing biological systems. More complex models, namely multilayer networks, are needed to account for interdependencies, often varying across time, of biological interacting units within a cell, a tissue or parts of an organism.
Many biological functions are executed by molecular machines, which consume energy and convert it into mechanical work. Biological machines have evolved to transport cargo, facilitate folding of proteins and RNA, remodel chromatin and replicate DNA. A common aspect of these machines is that their functions are driven out of equilibrium. It is a challenge to provide a general framework for understanding the functions of biological machines, such as molecular motors, molecular chaperones, and helicases. Using these machines as prototypical examples, we describe a few general theoretical methods providing insights into their functions. Although the theories rely on coarse-graining of these complex systems they have proven useful in not only accounting for many in vitro experiments but also addressing questions such as how the trade-off between precision, energetic costs and optimal performances are balanced. However, many complexities associated with biological machines will require one to go beyond current theoretical methods. Simple point mutations in the enzyme could drastically alter functions, making the motors bi-directional or result in unexpected diseases or dramatically restr
Provided that there is no theoretical frame for complex engineered systems (CES) as yet, this paper claims that bio-inspired engineering can help provide such a frame. Within CES bio-inspired systems play a key role. The disclosure from bio-inspired systems and biological computation has not been sufficiently worked out, however. Biological computation is to be taken as the processing of information by living systems that is carried out in polynomial time, i.e., efficiently; such processing however is grasped by current science and research as an intractable problem (for instance, the protein folding problem). A remark is needed here: P versus NP problems should be well defined and delimited but biological computation problems are not. The shift from conventional engineering to bio-inspired engineering needs bring the subject (or problem) of computability to a new level. Within the frame of computation, so far, the prevailing paradigm is still the Turing-Church thesis. In other words, conventional engineering is still ruled by the Church-Turing thesis (CTt). However, CES is ruled by CTt, too. Contrarily to the above, we shall argue here that biological computation demands a more ca
Monocular depth estimation (MDE) aims to transform an RGB image of a scene into a pixelwise depth map from the same camera view. It is fundamentally ill-posed due to missing information: any single image can have been taken from many possible 3D scenes. Part of the MDE task is, therefore, to learn which visual cues in the image can be used for depth estimation, and how. With training data limited by cost of annotation or network capacity limited by computational power, this is challenging. In this work we demonstrate that explicitly injecting visual cue information into the model is beneficial for depth estimation. Following research into biological vision systems, we focus on semantic information and prior knowledge of object sizes and their relations, to emulate the biological cues of relative size, familiar size, and absolute size. We use state-of-the-art semantic and instance segmentation models to provide external information, and exploit language embeddings to encode relational information between classes. We also provide a prior on the average real-world size of objects. This external information overcomes the limitation in data availability, and ensures that the limited cap