Engineering Papers⌕ Search

Engineering topics

Whitmer, Jonathan

Publications and source records attributed to Whitmer, Jonathan.

Assessment of fine-tuned large language models for real-world chemistry and material science applications

The current generation of large language models (LLMs) has limited chemical knowledge. Recently, it has been shown that these LLMs can learn and predict chemical properties through fine-tuning. Using natural language to train machine learning models opens doors to a wider chemical audience, as field-specific featurization techniques can be omitted. In this work, we explore the potential and limitations of this approach. We studied the performance of fine-tuning three open-source LLMs (GPT-J-6B, Llama-3.1-8B, and Mistral-7B) for a range of different chemical questions. We benchmark their performances against “traditional” machine learning models and find that, in most cases, the fine-tuning approach is superior for a simple classification problem. Depending on the size of the dataset and the type of questions, we also successfully address more sophisticated problems. The most important conclusions of this work are that, for all datasets considered, their conversion into an LLM fine-tuning training set is straightforward and that fine-tuning with even relatively small datasets leads to predictive models. These results suggest that the systematic use of LLMs to guide experiments and simulations will be a powerful technique in any research study, significantly reducing unnecessary experiments or computations.

Van Herck, Joren↗

Code interoperability extends the scope of quantum simulations

The functionality of many materials is critically dependent on the integration of dissimilar components and on the interfaces that arise between them. The description of such heterogeneous components requires the development and deployment of first principles methods, coupled to appropriate dynamical descriptions of matter and advanced sampling techniques, in order to capture all the relevant length and time scales of importance to the materials’ performance. It is thus essential to build simple, streamlined computational schemes for the prediction and design of multiple properties of broad classes of materials, by developing interoperable codes which can be efficiently coupled to each other to perform complex tasks. We discuss the use of interoperable codes to simulate the structural and spectroscopic characterization of materials, including chemical reactions for catalysis, the description of defects for quantum information science, and heat and charge transport.

36 MATERIALS SCIENCE↗

Interfacial Tension of Polyelectrolyte Complex Coacervate Phases

We consider polyelectrolyte solutions which, under suitable conditions, phase separate into a liquid-like coacervate phase and a coexisting supernatant phase that exhibit an extremely low interfacial tension. Such interfacial tension provides the basis for most coacervate-based applications, but little is known about it, including its dependence on molecular weight, charge density, and salt concentration. By combining a Debye-Huckel treatment for electrostatic interactions with the Cahn-Hilliard theory, we derive explicit expressions for this interfacial tension. In the absence of added salts, we find that the interfacial tension scales as N-3/2(eta/eta(c)-1)(3/2) near the critical point of the demixing transition, and that it scales as eta(1/2) far away from it, where N is the chain length and eta measures the electrostatic interaction strength as a function of temperature, dielectric constant, and charge density of the polyelectrolytes. For the case with added salts, we find that the interfacial tension scales with the salt concentration psi as N-1/4(1-psi/psi(c))(3/2) near the critical salt concentration psi(c). Our predictions are shown to be in quantitative agreement with experiments and provide a means to design new materials based on polyelectrolyte complexation.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗