Bayesian sparse modeling for interpretable prediction of hydroxide ion conductivity in anion-conductive polymer membranes

Ryo Murakami a , Kenji Miyatake b,c , Ahmed Mohamed Ahmed Mahmoud b,d , Hideki Yoshikawa a , Kenji Nagata e · 2025

Read Bayesian sparse modeling for interpretable prediction of hydroxide ion conductivity in anion-conductive polymer membranes by Ryo Murakami a , Kenji Miyatake b,c , Ahmed Mohamed Ahmed Mahmoud b,d , Hideki Yoshikawa a , Kenji Nagata e (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-19044.txt.

Introduction

Ryo Murakami and co-authors use sparse Bayesian modeling to predict how well fuel cell membranes carry hydroxide ions, in this 2025 paper. These membranes serve in alkaline fuel cells and electrolysis units. Their performance depends on chemical makeup in ways that were mostly described in words, not numbers.

The team counted atoms in the water-loving and water-shy parts of each polymer recipe. A sparse model then picked which counts truly drive the flow of ions. The winning features predicted performance well and can be read and understood by chemists. Clear rules like these help design better membranes without endless trial and error.

Catalog record

CREATORRyo Murakami a , Kenji Miyatake b,c , Ahmed Mohamed Ahmed Mahmoud b,d , Hideki Yoshikawa a , Kenji Nagata e
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-19044

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Bayesian sparse modeling for interpretable prediction of hydroxide ion conductivity in anion-conductive polymer membranes Ryo Murakami a , Kenji Miyatake b,c , Ahmed Mohamed Ahmed Mahmoud b,d , Hideki Yoshikawa a , Kenji Nagata e Abstract Anion-conductive polymer membranes have attracted considerable attention as solid electrolytes for alkaline fuel cells and electrolysis cells. Their hydroxide ion conductivity varies depending on factors such as the type and distribution of quaternary ammonium groups, as well as the structure and connectivity of hydrophilic and hydrophobic domains. In particular, the size and connectivity of hydrophilic domains significantly influence the mobility of hydroxide ions; however, this relationship has remained largely qualitative. In this study, we calculated the number of key constituent elements in the hydrophilic and hydrophobic units based on the copolymer composition, and investigated their relationship with hydroxide ion conductivity by using Bayesian sparse modeling. As a result, we successfully identified composition-derived features that are critical for accurately predicting hydroxide ion conductivity. keywords anion-conductive polymer membranes; Materials informatics; Data-driven science; Sparse modeling; Bayesian inference † † articletype: Regular article † † affiliation: a Research Network and Facility Services Division, National Institute for Materials Science, Tsukuba 305-0044, Ibaraki, Japan; b Clean Energy Research Center, University of Yamanashi, Kofu 400-8510, Yamanashi, Japan; c Department of Applied Chemistry, Waseda University, Tokyo 169-8555, Japan; d Office of Institutional Advancement and Communications, Kyoto University, Kyoto 6068501, Japan; e Center for Basic Research on Materials, National Institute for Materials Science, Tsukuba 305-0044, Ibaraki, Japan 1 Introduction Anion-conductive polymer membranes are promising candidates for use as solid electrolytes in alkaline energy devices, such as fuel cells and water electrolysis cells. In particular, anion exchange membrane water electrolysis systems, which can produce green hydrogen efficiently by utilizing renewable energy sources, are being actively investigated worldwide as a core technology for realizing a carbon-neutral hydrogen society. For such applications, desirable properties of anion-conductive polymers include anion conductivity comparable to that of alkaline aqueous electrolytes, the ability to form thin membranes (thickness <50\mu m ) with sufficient mechanical strength, gas barrier properties against hydrogen and oxygen, and water permeability and alkaline stability under low-humidity conditions. Among these, hydroxide ion conductivity is the most critical property, as it directly determines the voltage efficiency of anion exchange membrane water electrolysis systems. Anion-conductive polymers are generally composed of hydrophobic polymer backbones functionalized with cationic groups. While a variety of backbones—including aliphatic and aromatic types—are available, it is known that structures without hetero-bonds such as ether linkages in the main chain are preferable from the viewpoint of alkaline stability [ 1 ] . Onium salts are commonly employed as cationic functional groups due to their synthetic accessibility, ease of incorporation into polymers, and adequate hydroxide ion dissociation; among them, quaternary ammonium groups are known to provide high anion conductivity. In general, the ion exchange capacity (IEC) of quaternary ammonium groups has a major impact on conductivity—higher IEC values typically correspond to higher hydroxide ion conductivity in thin membranes. However, even at similar IEC values, the conductivity can vary depending on the polymer backbone, the type of ammonium group, and their combination, as well as the connectivity of hydrophilic and hydrophobic domains. This is because quaternary ammonium groups and water molecules can aggregate to form hydrophilic domains, and the size, connectivity, and tortuosity of these domains significantly affect hydroxide ion mobility. Hydrophilic domain structures can be analyzed by various experimental techniques. For instance, small-angle X-ray scattering (SAXS) provides information on the shape and periodicity of hydrophilic domains, while transmission electron microscopy (TEM) allows direct observation of domain size and connectivity [ 2 ] . Atomic force microscopy (AFM) can distinguish hydrophilic and hydrophobic domains on the membrane surface based on phase contrast, and electrochemical AFM can visualize ion-conductive pathways as current responses during oxidation or reduction reactions. Although these techniques provide valuable insights into the phase-separated morphology of the membrane, most discussions to date have remained qualitative, and a quantitative understanding has not been established. Moreover, cases where membranes exhibit similar morphologies and IECs but differ in hydroxide ion conductivity remain unresolved. In this study, we aimed to quantitatively understand the variation in conductivity among a series of anion-conductive polymer membranes previously developed by our group [ 3 , 4 , 5 , 6 ] . We calculated the number of key constituent elements in the hydrophilic and hydrophobic units, taking into account the copolymer composition, and examined their potential as predictive features for hydroxide ion conductivity. Our goal was to construct an interpretable and low-dimensional predictive model using a linear analytical framework. Since the set of important features is not known a priori, we generated multiple candidate features and performed both conductivity prediction and feature selection simultaneously, as illustrated in Figure 1 . Figure 1: Overview diagram of this study. We design controllable and interpretable candidate features from the chemical structures shown on the left side. Then, as shown on the right side, we explore equations that describe material properties (conductivity) by using model selection based on the marginal log-likelihood in Bayesian estimation framework. 2 Dataset outline In this study, the target variable was defined as the hydroxide ion conductivity of anion-conductive polymer membranes measured in water at 30 {}^{\circ}\mathrm{C} , along with the structural characteristics of their hydrophilic and hydrophobic domains. It is well known that anion conductivity correlates with several physical parameters of the polymer membranes, including ion exchange capacity (IEC), water uptake, and the size and connectivity of hydrophilic clusters. Generally, a higher IEC leads to an increased carrier concentration and, consequently, higher conductivity. However, excessive IEC can result in increased water uptake—particularly in terms of the number of water molecules adsorbed per ionic group—thereby reducing the effective carrier density and leading to lower conductivity. The hydrophilic clusters in anion-conductive polymer membranes can be visualized from transmission electron microscopy (TEM) images of samples in which the ammonium groups were selectively stained by ion exchange with \mathrm{PtCl_{4}^{2-}} ions. It is expected that larger hydrophilic clusters provide more continuous conduction pathways, leading to higher ion mobility and thus higher conductivity. Figure 2 shows scatter plots of hydroxide ion conductivity versus the average diameter of hydrophilic and hydrophobic clusters estimated from TEM images. However, no clear correlation was observed in either case, indicating that cluster size alone is insufficient to describe conductivity, especially when the polymer structures differ. We hypothesize that the number of carbon (C) and fluorine (F) atoms in the hydrophilic and hydrophobic units influences the spatial occupation, domain size, and connectivity within the polymer structure, thereby functioning as key structural descriptors governing hydroxide ion conductivity. Therefore, in this study, we calculated the number of C and F atoms based on the copolymer composition for each segment and investigated whether these values could serve as predictive features for conductivity. The dataset used in this study consisted of 18 polymer samples [ 3 , 4 , 5 , 6 ] , and the number of explanatory variables (features) was 15. We refer you to the subsection 3.2 for details on features. Figure 2: Scatter plot of conductivity against the average diameter of (left) hydrophobic cluster size and (right) hydrophilic cluster size obtained from TEM images [ 3 , 4 , 5 , 6 ] . 3 Method

Read the complete text