Eric Boittier, Ph.D.

I am a scientist and software engineer working at the interface of physical chemistry and machine learning.

Profiles

	GitHub > Public code contributions
	Google Scholar > Research and publications
	LinkedIn > Network, experience and education

Below you can find some of my recent publications and blog posts.

Paper Highlights

Figure P1. KernelMDCM publication visual.

• Kernel-based Minimal Distributed Charges: A Conformationally Dependent ESP-Model for Molecular Simulations
  This study introduces a kernel-based framework for predicting conformationally dependent distributed charges, enabling more physically faithful electrostatic representations in molecular simulations.¹¹ The method is designed to improve transferability across molecular conformations while preserving simulation efficiency.

Figure P2. MLCCSD(T) publication visual.

• Transfer Learning to CCSD(T): Accurate Anharmonic Frequencies from Machine Learning Models
  This work applies transfer learning to bridge lower-cost quantum data and CCSD(T)-level targets, delivering highly accurate anharmonic vibrational frequencies at reduced computational cost.²² It demonstrates a practical route to near high-level quantum accuracy without prohibitive scaling.

Figure P3. MLDATABASE publication visual.

• Impact of the Characteristics of Quantum Chemical Databases on Machine Learning Prediction of Tautomerization Energies
  This paper evaluates how database composition, chemical diversity, and sampling strategy influence machine-learning predictions of tautomerization energies.³³ The analysis highlights how data curation decisions directly affect generalization and model reliability.

Blog Highlights

Figure B1. Fortran logo used for the profiling post.

• Fortran is fast. Profile your code to make it faster!
  A practical guide to performance engineering in scientific codebases, covering robust benchmarking, profiler-driven diagnosis, and targeted optimization strategies for Fortran workflows.⁴⁴ Emphasis is placed on reproducible timing methodology and actionable bottleneck identification.

Figure B2. Jupyter logo for notebook-to-blog workflows.

• From Jupyter to Blog
  A workflow for converting exploratory Jupyter notebooks into maintainable, publication-ready technical articles while preserving clarity, reproducibility, and narrative structure.⁵⁵ The approach separates experimentation from presentation to improve long-term maintainability.

Figure B3. Gaussian process bands visualization.

• Notes on Bayesian Optimization
  Concise notes on Bayesian optimization for scientific machine learning, with a focus on acquisition design, uncertainty-aware exploration, and efficient hyperparameter search under limited budgets.⁶⁶ The post emphasizes practical heuristics for balancing exploration and exploitation in real experiments.