Yuya Murakami

Prof. Yuya Murakami

orcid

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
7 April 2023 in Gels
Verified
Journal Article
Open access
PDF file
YMYuya Murakami
Yuya Murakami
TUTaisei Uchiyama
Taisei Uchiyama
ASAtsushi Shono
Atsushi Shono
The Hansen solubility parameter (HSP) is a useful index for reasoning the gelation behavior of low-molecular-weight gelators (LMWGs). However, the conventional HSP-based methods only “classify” solvents that can and cannot form gels and require many trials to achieve this. For engineering purposes, quantitative estimation of gel properties using the HSP is highly desired. In this study, we measured critical gelation concentrations based on three distinct definitions, mechanical strength, and light transmittance of organogels prepared with 12-hydroxystearic acid (12HSA) and correlated them with the HSP of solvents. The results demonstrated that the mechanical strength, in particular, strongly correlated with the distance of 12HSA and solvent in the HSP space. Additionally, the results indicated that the constant volume-based concentration should be used when comparing the properties of organogels to a different solvent. These findings are helpful in efficiently determining the gelation sphere of new LMWGs in HSP space and contribute to designing organogels with tunable physical properties.

14 May 2025 in Chemical Engineering Research and Design
Verified
Journal Article
Closed access
MAMasao Aida
Masao Aida
YMYuya Murakami
Yuya Murakami
ASAtsushi Shono
Atsushi Shono

3 September 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
AHAtsushi Hatoyama
Atsushi Hatoyama
YMYuya Murakami
Yuya Murakami
Reliable design of multicomponent fluids requires property-prediction methods applicable to unmeasured mixture systems. Kinematic viscosity is a challenging prediction target because of its sensitivity to intermolecular interactions and non-ideal composition effects. We developed a physics-guided, McAllister-inspired neural network that combines explicit, permutation-invariant n-body composition aggregation with a shared nonlinear mapping from molecular descriptors and pure-component properties to interaction contributions. We evaluated extrapolative performance on ternary mixtures spanning regular solutions and strongly non-ideal water-containing systems using nested leave-one-component-out (LOCO) validation that excluded each of 20 organic components from model development. Diagnostic fits showed that the McAllister n-body form accurately represented composition dependence, whereas a compact shared parameterization constrained the generalized McAllister model's expressiveness. The best-performing 4-body excess-log-viscosity neural network achieved component-macro average absolute relative deviations of 6.47% overall under LOCO and 9.98% for challenging water-containing systems. Under identical data and validation conditions, its errors were approximately half those of two representative general-purpose machine-learning baselines. Ablation analysis identified permutation invariance as the largest source of improvement, with an additional gain from explicit McAllister-type n-body aggregation and a suggested benefit of 4-body interactions in water-containing systems. Combining nonlinear interaction representations learned from molecular structure and pure-component information with physically structured composition aggregation enabled accurate, broadly applicable modeling of non-ideal mixtures. The architecture can accommodate additional components without increasing its trainable parameter count and may provide a general design principle for data-driven prediction of non-ideal mixture propertie

22 July 2026 in ChemRxiv
Verified
Preprint
Open access
YMYuya Murakami
Yuya Murakami
Data-driven property models support chemical process and molecular design, yet reliable prediction remains difficult beyond sampled chemical and operating-condition domains. Using pure-component viscosity, we diagnosed structural-and temperature-extrapolation errors and tested axis-specific, physics-guided derivative regularization. We analyzed 21,616 measurements for 2,501 compounds in a leakage-controlled comparison across linear and nonlinear model classes, treating descriptor selection as part of nested hyperparameter optimization without access to outer test data. When similar compounds were distributed across training and test sets, a standard neural network matched kernel ridge regression; however, its error increased more sharply than those of kernel and tree models as structural similarity decreased. Analysis of its descriptor-space prediction surface suggested that off-domain curvature contributed to this deterioration. Hessian regularization reduced median pathwise curvature by 43.2% and achieved the lowest compound-weighted mean absolute error among all evaluated models. For temperature extrapolation, a neural network with a soft penalty on deviations from inverse-temperature linearity provided the strongest performance across the 20 and 50 K tasks among the standard neural network, kernel ridge regression, and a network enforcing exact Arrhenius behavior. Notably, it yielded lower aggregate errors than compound-specific local Arrhenius fits, suggesting that shared molecular information enabled data-supported corrections to strict Arrhenius extrapolation. Combining the structural and temperature regularizers yielded the lowest mean integrated score. Together, these results show that axis-specific derivative regularization can translate neural-network flexibility into robust extrapolation, offering a transferable strategy for property prediction across chemical and operating-condition domains.

20 July 2021 in KAGAKU KOGAKU RONBUNSHU
Verified
Journal Article
Closed access
PDF file
MAMasao Aida
Masao Aida
YMYuya Murakami
Yuya Murakami
ASAtsushi Shono
Atsushi Shono
乱流拡散モデルを基に新たな混合時間の相関方法を提案した.無次元数X1, X3は動力数Np,回転数n,撹拌レイノルズ数Re,混合時間tm, d/D, H/D(D; 槽径,d; 翼径,H; 液深)の関数であり,αはアスペクト比の影響を表す実験パラメータである. αは汎用小型翼に対して0.25, 大型翼に対して0.15であった.無次元数X1, X3を用いた相関方法を用いることで,邪魔板の条件,偏心の有無によらず,遷移域および乱流域(Re≧100)の混合時間を精度良く相関できることが明らかとなった.