arXiv subjects

Maciej Haranczyk

Publications and source records attributed to Maciej Haranczyk.

10 recordsLinked to original sources

arXiv2026

Bayesian Optimization of Genetic Algorithm Hyperparameters in a Multi-Fidelity Framework for Efficient Lattice Material Design

This study presents a multi-fidelity framework for the systematic optimization of genetic algorithm (GA) hyperparameters. The framework integrates three fidelity levels: high-fidelity Fast Fourier Transform (FFT) homogenization for validation, a medium-fidelity 3D convolutional neural network surrogate for rapid property evaluation, and a low-fidelity Gaussian process (GP) surrogate within a Bayesian optimization (BO) framework to guide the hyperparameter search. Various acquisition functions are evaluated, with logNEI achieving the best performance by effectively accounting for the noise inherent in GA evaluations. The proposed framework identifies hyperparameter configurations that enable a 25-generation GA run to achieve elastic modulus values comparable to those obtained in a full 75-generation optimization. Furthermore, introducing a penalized BO objective significantly reduces the number of required lattices with only minor decreases in absolute achieved elastic modulus, revealing a practical trade-off between performance and the number of structures that must be evaluated. High-fidelity FFT validation verifies the effectiveness of the surrogate-driven optimization strategy. The optimized hyperparameters allow for rapid convergence, eliminate the need for lattice mutation, and reduce the overall computational cost by 24% (from 225 to 171 hours) while preserving mechanical performance. These results demonstrate the potential of multi-fidelity optimization as an efficient and practical approach for GA hyperparameter tuning and future experimental lattice design studies.

cond-mat.mtrl-sci↗

arXiv2025

Noise-Aware Optimization in Nominally Identical Manufacturing and Measuring Systems for High-Throughput Parallel Workflows

Device-to-device variability in experimental noise critically impacts reproducibility, especially in automated, high-throughput systems like additive manufacturing farms. While manageable in small labs, such variability can escalate into serious risks at larger scales, such as architectural 3D printing, where noise may cause structural or economic failures. This contribution presents a noise-aware decision-making algorithm that quantifies and models device-specific noise profiles to manage variability adaptively. It uses distributional analysis and pairwise divergence metrics with clustering to choose between single-device and robust multi-device Bayesian optimization strategies. Unlike conventional methods that assume homogeneous devices or generic robustness, this framework explicitly leverages inter-device differences to enhance performance, reproducibility, and efficiency. An experimental case study involving three nominally identical 3D printers (same brand, model, and close serial numbers) demonstrates reduced redundancy, lower resource usage, and improved reliability. Overall, this framework establishes a paradigm for precision- and resource-aware optimization in scalable, automated experimental platforms.

cs.DC↗

arXiv2025

Descriptor and Graph-based Molecular Representations in Prediction of Copolymer Properties Using Machine Learning

Copolymers are highly versatile materials with a vast range of possible chemical compositions. By using computational methods for property prediction, the design of copolymers can be accelerated, allowing for the prioritization of candidates with favorable properties. In this study, we utilized two distinct representations of molecular ensembles to predict the seven different physical polymer properties copolymers using machine learning: we used a random forest (RF) model to predict polymer properties from molecular descriptors, and a graph neural network (GNN) to predict the same properties from 2D polymer graphs under both a single- and multi-task setting. To train and evaluate the models, we constructed a data set from molecular dynamic simulations for 140 binary copolymers with varying monomer compositions and configurations. Our results demonstrate that descriptors-based RFs excel at predicting density and specific heat capacities at constant pressure (Cp) and volume (Cv) because these properties are strongly tied to specific molecular features captured by molecular descriptors. In contrast, graph representations better predict expansion coefficients (γ, α) and bulk modulus (K), which depend more on complex structural interactions better captured by graph-based models. This study underscores the importance of choosing appropriate representations for predicting molecular properties. Our findings demonstrate how machine learning models can expedite copolymer discovery with learnable structure-property relationships, streamlining polymer design and advancing the development of high-performance materials for diverse applications.

cond-mat.mtrl-sci↗

arXiv2025

tda-segmentor: A tool to extract and analyze local structure and porosity features in porous materials

Local geometrical features of a porous material such as the shape and size of a pore or the curvature of a solid ligament often affect the macroscopic properties of the material, and their characterization is necessary to fully understand the structure-property relationships.In this contribution, we present an approach to automatically segment large porous structures into such local features. Our work takes inspiration from techniques available in Topological Data Analysis(TDA).In particular, using Morse theory, we generate Morse-Smale Complexes(MSC) of our structures that segment the structure, and/or its porosity into individual features that can then be compared. We develop a tool that is built on the topology toolkit (TTK) library, an open source platform for the topological analysis of scalar data, with which we can perform segmentation of these structures. Our tool takes a volumetric grid representation as an input, which can be generated from atomistic or mesh structure models and any function defined on such grid, e.g. the distance to the surface or the interaction energy with a probe. We demonstrate the applicability of the tool by two examples related with analysis of porosity in zeolite materials as well as analysis of ligaments in a porous metal structure. Specifically, by segmenting the pores in the structure we demonstrate some applications to zeolites such as assessing pore-similarity between structures or evaluating the accessible volume to a target molecule such as methane that can be adsorbed to its surface. Moreover, once the MSC's are generated, we can construct graph representations of the void space, replacing the entire pore structure by a simply connected graph. Similarly, the same tool is used to segment and generate graphs representing the solid structure and we show how they can be used to correlate structure and mechanical properties of the material.

cond-mat.mtrl-sci↗

arXiv2025

A Novel Constrained Sampling Method for Efficient Exploration in Materials and Chemical Mixture Design

Efficient exploration of multicomponent material composition spaces is often limited by time and financial constraints, particularly when mixture and synthesis constraints exist. Traditional methods like Latin hypercube sampling (LHS) struggle with constrained problems especially in high dimensions, while emerging approaches like Bayesian optimization (BO) face challenges in early-stage exploration. This article introduces ConstrAined Sequential laTin hypeRcube sampling methOd (CASTRO), an open-source tool designed to address these challenges. CASTRO is optimized for uniform sampling in constrained small- to moderate-dimensional spaces, with scalability to higher dimensions through future adaptations. CASTRO uses a divide-and-conquer strategy to decompose problems into parallel subproblems, improving efficiency and scalability. It effectively handles equality-mixture constraints, ensuring comprehensive design space coverage and leveraging LHS and LHS with multidimensional uniformity (LHSMDU). It also integrates prior experimental knowledge, making it well-suited for efficient exploration within limited budgets. Validation through two material design case studies, a four-dimensional problem with near-uniform distributions and a nine-dimensional problem with additional synthesis constraints, demonstrates CASTRO's effectiveness in exploring constrained design spaces for materials science, pharmaceuticals and chemicals. The software and case studies are available on GitHub.

stat.CO↗

arXiv2024

Model-Based Reinforcement Learning Control of Reaction-Diffusion Problems

Mathematical and computational tools have proven to be reliable in decision-making processes. In recent times, in particular, machine learning-based methods are becoming increasingly popular as advanced support tools. When dealing with control problems, reinforcement learning has been applied to decision-making in several applications, most notably in games. The success of these methods in finding solutions to complex problems motivates the exploration of new areas where they can be employed to overcome current difficulties. In this paper, we explore the use of automatic control strategies to initial boundary value problems in thermal and disease transport. Specifically, in this work, we adapt an existing reinforcement learning algorithm using a stochastic policy gradient method and we introduce two novel reward functions to drive the flow of the transported field. The new model-based framework exploits the interactions between a reaction-diffusion model and the modified agent. The results show that certain controls can be implemented successfully in these applications, although model simplifications had to be assumed.

math.OC↗

arXiv2024

Toward Diverse Polymer Property Prediction Using Transfer Learning

The prediction of mechanical and thermal properties of polymers is a critical aspect for polymer development. Herein, we discuss the use of transfer learning approach to predict multiple properties of linear polymers. The neural network model is initially trained to predict the heat capacity in constant pressure (Cp) of linear polymers. Once, the pre-trained model is transferred to predict four additional properties of polymers: specific heat capacity (Cv), shear modulus, flexural stress strength at yield, and tensile creep compliance. They represent a diverse set of mechanical, thermal, and rheological properties. We demonstrate the effectiveness of the approach by achieving high accuracy in predicting the four additional properties using relatively small datasets of 13 to 18 samples. Also, the performance of the base model is examined using five different loss functions. Our results suggest that the combined loss function had better performance compared to the individual loss functions.

cond-mat.soft↗

arXiv2021

Machine learning with persistent homology and chemical word embeddings improves prediction accuracy and interpretability in metal-organic frameworks

Machine learning has emerged as a powerful approach in materials discovery. Its major challenge is selecting features that create interpretable representations of materials, useful across multiple prediction tasks. We introduce an end-to-end machine learning model that automatically generates descriptors that capture a complex representation of a material's structure and chemistry. This approach builds on computational topology techniques (namely, persistent homology) and word embeddings from natural language processing. It automatically encapsulates geometric and chemical information directly from the material system. We demonstrate our approach on multiple nanoporous metal-organic framework datasets by predicting methane and carbon dioxide adsorption across different conditions. Our results show considerable improvement in both accuracy and transferability across targets compared to models constructed from the commonly-used, manually-curated features, consistently achieving an average 25-30% decrease in root-mean-squared-deviation and an average increase of 40-50% in R2 scores. A key advantage of our approach is interpretability: Our model identifies the pores that correlate best to adsorption at different pressures, which contributes to understanding atomic-level structure--property relationships for materials design.

cond-mat.mtrl-sci↗

arXiv2020

Topological Descriptors Help Predict Guest Adsorption in Nanoporous Materials

Machine learning has emerged as an attractive alternative to experiments and simulations for predicting material properties. Usually, such an approach relies on specific domain knowledge for feature design: each learning target requires careful selection of features that an expert recognizes as important for the specific task. The major drawback of this approach is that computation of only a few structural features has been implemented so far, and it is difficult to tell a priori which features are important for a particular application. The latter problem has been empirically observed for predictors of guest uptake in nanoporous materials: local and global porosity features become dominant descriptors at low and high pressures, respectively. We investigate a feature representation of materials using tools from topological data analysis. Specifically, we use persistent homology to describe the geometry of nanoporous materials at various scales. We combine our topological descriptor with traditional structural features and investigate the relative importance of each to the prediction tasks. We demonstrate an application of this feature representation by predicting methane adsorption in zeolites, for pressures in the range of 1-200 bar. Our results not only show a considerable improvement compared to the baseline, but they also highlight that topological features capture information complementary to the structural features: this is especially important for the adsorption at low pressure, a task particularly difficult for the traditional features. Furthermore, by investigation of the importance of individual topological features in the adsorption model, we are able to pinpoint the location of the pores that correlate best to adsorption at different pressure, contributing to our atom-level understanding of structure-property relationships.

cond-mat.mtrl-sci↗

arXiv2017

PyCDT: A Python toolkit for modeling point defects in semiconductors and insulators

Point defects have a strong impact on the performance of semiconductor and insulator materials used in technological applications, spanning microelectronics to energy conversion and storage. The nature of the dominant defect types, how they vary with processing conditions, and their impact on materials properties are central aspects that determine the performance of a material in a certain application. This information is, however, difficult to access directly from experimental measurements. Consequently, computational methods, based on electronic density functional theory (DFT), have found widespread use in the calculation of point-defect properties. Here we have developed the Python Charged Defect Toolkit (PyCDT) to expedite the setup and post-processing of defect calculations with widely used DFT software. PyCDT has a user-friendly command-line interface and provides a direct interface with the Materials Project database. This allows for setting up many charged defect calculations for any material of interest, as well as post-processing and applying state-of-the-art electrostatic correction terms. Our paper serves as a documentation for PyCDT, and demonstrates its use in an application to the well-studied GaAs compound semiconductor. We anticipate that the PyCDT code will be useful as a framework for undertaking readily reproducible calculations of charged point-defect properties, and that it will provide a foundation for automated, high-throughput calculations.

cond-mat.mtrl-sci↗