Stijn De Baerdemacker
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Stijn De Baerdemacker has a total of 108 co-authors. Here are the top co-authors by number of shared publications:
Dimitri Van Neck
Center for Molecular Modeling Ghent University Technologiepark 903 9052 Zwijnaarde Belgium
27 shared publications- Patrick Bultinck
Department of Chemistry Ghent University Ghent Belgium
19 shared publications K. Heyde
Universiteit Gent, Vakgroep Subatomaire- en Stralingsfysica, Proeftuinstraat 86, B-9000 Gent, Belgium
15 shared publicationsV. Hellemans
Université Libre de Bruxelles, Service de Physique Nucléaire Théorique, B-1050 Bruxelles, Belgium
13 shared publicationsAlexis De Vos
10 shared publications- Pieter W. Claeys
Max Planck Institute for the Physics of Complex Systems
9 shared publications
Latest Publications
- Open Access
- 22 October 2026
- Published by American Physical Society (APS) in Physical Review Research
- Vol. 8 (4)
Max-Cut graph-driven quantum circuit design for geometrically frustrated planar spin systems with spin-glass-like energy landscapes
Finding the ground state of geometrically frustrated spin systems is a challenging problem with broad implications. Many hard optimization problems, including NP-complete problems, can be mapped, for instance, to frustrated Ising models, where competing interactions produce rugged, spin-glass-like energy landscapes. The difficulty is particularly pronounced in the weak-field regime, where geometrical frustration dominates, and the spectral gap becomes exponentially small, making it hard to identify the true ground state. In this work, we consider planar frustrated lattices constructed from the triangular motif, the minimal unit of geometrical frustration. We present a graph-based approach that allows for accurate state initialization of a frustrated triangular spin lattice with up to 20 sites while avoiding barren plateaus. To optimize circuit efficiency and trainability, we employ a clustering strategy that organizes qubits into distinct groups based on the maximum cut technique, which divides the lattice into two maximally disconnected subsets. We provide evidence that this Max-Cut-based lattice division offers a robust framework for optimizing circuit design and effectively modeling frustrated systems at polynomial cost. All simulations are performed within the variational quantum eigensolver formalism, the current paradigm for noisy intermediate-scale quantum devices, but can be extended beyond. Our results underscore the potential of hybrid quantum-classical methods in addressing complex optimization problems.
- 22 October 2026
- Published by American Physical Society (APS) in Physical Review Research
- Vol. 8 (4)
Open AccessMax-Cut graph-driven quantum circuit design for geometrically frustrated planar spin systems with spin-glass-like energy landscapes
Finding the ground state of geometrically frustrated spin systems is a challenging problem with broad implications. Many hard optimization problems, including NP-complete problems, can be mapped, for instance, to frustrated Ising models, where competing interactions produce rugged, spin-glass-like energy landscapes. The difficulty is particularly pronounced in the weak-field regime, where geometrical frustration dominates, and the spectral gap becomes exponentially small, making it hard to identify the true ground state. In this work, we consider planar frustrated lattices constructed from the triangular motif, the minimal unit of geometrical frustration. We present a graph-based approach that allows for accurate state initialization of a frustrated triangular spin lattice with up to 20 sites while avoiding barren plateaus. To optimize circuit efficiency and trainability, we employ a clustering strategy that organizes qubits into distinct groups based on the maximum cut technique, which divides the lattice into two maximally disconnected subsets. We provide evidence that this Max-Cut-based lattice division offers a robust framework for optimizing circuit design and effectively modeling frustrated systems at polynomial cost. All simulations are performed within the variational quantum eigensolver formalism, the current paradigm for noisy intermediate-scale quantum devices, but can be extended beyond. Our results underscore the potential of hybrid quantum-classical methods in addressing complex optimization problems.
- Open Access
- 17 July 2026
- Published by American Chemical Society (ACS) in ACS Omega
- Vol. 11 (30) , 44761-44774
From Dipeptide Systems to Polypeptides: Evolution of Mutual Information
Understanding the electronic structure of amino acids is crucial to understanding protein stability, folding mechanisms, and molecular interactions. In this study, we introduce fragment-wise mutual information (FMI) as a quantum information-based tool to quantify interatomic correlations in peptides. By extending mutual information (MI) analysis to amino acid fragments, FMI provides a detailed map of electronic interactions beyond classical descriptors, such as van der Waals forces. We first validated FMI on 400 dipeptides, demonstrating a correlation with the atomization and bonding energies. Expanding this approach to the 10-mer Neh2 peptide, we analyze molecular dynamics (MD) simulations and reveal how interatomic correlations evolve during folding. Our results show that FMI distinguishes stabilizing interactions such as salt bridges and variable hydrogen-bond strengths, providing deeper insight into peptide stability. These findings suggest that FMI could enhance molecular modeling and force-field development by incorporating quantum electronic effects into biomolecular analysis.
- 17 July 2026
- Published by American Chemical Society (ACS) in ACS Omega
- Vol. 11 (30) , 44761-44774
Open AccessFrom Dipeptide Systems to Polypeptides: Evolution of Mutual Information
Understanding the electronic structure of amino acids is crucial to understanding protein stability, folding mechanisms, and molecular interactions. In this study, we introduce fragment-wise mutual information (FMI) as a quantum information-based tool to quantify interatomic correlations in peptides. By extending mutual information (MI) analysis to amino acid fragments, FMI provides a detailed map of electronic interactions beyond classical descriptors, such as van der Waals forces. We first validated FMI on 400 dipeptides, demonstrating a correlation with the atomization and bonding energies. Expanding this approach to the 10-mer Neh2 peptide, we analyze molecular dynamics (MD) simulations and reveal how interatomic correlations evolve during folding. Our results show that FMI distinguishes stabilizing interactions such as salt bridges and variable hydrogen-bond strengths, providing deeper insight into peptide stability. These findings suggest that FMI could enhance molecular modeling and force-field development by incorporating quantum electronic effects into biomolecular analysis.
- Open Access
- 5 February 2026
- Published by American Chemical Society (ACS) in ChemRxiv
Machine learning-based screening of high-entropy alloy AlCoCrFeNi particles produced via high-energy mechanical alloying method
There is a growing interest in exploring high entropy alloys (HEAs) due to their superior mechanical and thermal properties. Although numerous routes exist for producing HEA particles, mechanical alloying (MA) remains the most conductive method at the laboratory level due to its simplicity and cost-effectiveness. In powder metallurgy, the degree of alloying plays a crucial role in achieving the desired quality of final manufactured parts. However, in mechanically milled HEA powders, segregation often arises from localized energy concentrations, leading to significant compositional inhomogeneity. Currently, researchers rely on conventional energy-dispersive spectroscopy (EDS) to assess alloying state. SEM alone is less time-consuming than with EDS integration. Yet, EDS maps can be inaccurate, have low reproducibility, and be misleading for HEA powder, indicating alloyed composition, even when the opposite is true. This study proposes a deep learning regression approach to predict compositional homogeneity, which can be thresholded to classify alloyed and segregated particles directly from scanning electron microscopy (SEM) images. The approach utilizes compositional/alloying uniformity derived from per-particle EDS composition vectors of the AlCoCrFeNi HEA system, serving as a ground truth. Different regression models were trained to directly predict Shannon entropy (H), a nondimensional measure of compositional uniformity, with DenseNet121 achieving an R² of 0.918 and an MAE of 0.03, corresponding to an average deviation of approximately 3% of the full alloying uniformity scale. This approach enables rapid, objective, reproducible, and automated screening of HEA powder, offering a scalable alternative to conventional microanalysis techniques. The proposed model can be integrated into existing laboratory workflows for mechanically alloyed AlCoCrFeNi systems, enabling quality assessment during the intermediate and final stages of milling using only bench-top SEMs without requiring an EDS detector.
- 5 February 2026
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessMachine learning-based screening of high-entropy alloy AlCoCrFeNi particles produced via high-energy mechanical alloying method
There is a growing interest in exploring high entropy alloys (HEAs) due to their superior mechanical and thermal properties. Although numerous routes exist for producing HEA particles, mechanical alloying (MA) remains the most conductive method at the laboratory level due to its simplicity and cost-effectiveness. In powder metallurgy, the degree of alloying plays a crucial role in achieving the desired quality of final manufactured parts. However, in mechanically milled HEA powders, segregation often arises from localized energy concentrations, leading to significant compositional inhomogeneity. Currently, researchers rely on conventional energy-dispersive spectroscopy (EDS) to assess alloying state. SEM alone is less time-consuming than with EDS integration. Yet, EDS maps can be inaccurate, have low reproducibility, and be misleading for HEA powder, indicating alloyed composition, even when the opposite is true. This study proposes a deep learning regression approach to predict compositional homogeneity, which can be thresholded to classify alloyed and segregated particles directly from scanning electron microscopy (SEM) images. The approach utilizes compositional/alloying uniformity derived from per-particle EDS composition vectors of the AlCoCrFeNi HEA system, serving as a ground truth. Different regression models were trained to directly predict Shannon entropy (H), a nondimensional measure of compositional uniformity, with DenseNet121 achieving an R² of 0.918 and an MAE of 0.03, corresponding to an average deviation of approximately 3% of the full alloying uniformity scale. This approach enables rapid, objective, reproducible, and automated screening of HEA powder, offering a scalable alternative to conventional microanalysis techniques. The proposed model can be integrated into existing laboratory workflows for mechanically alloyed AlCoCrFeNi systems, enabling quality assessment during the intermediate and final stages of milling using only bench-top SEMs without requiring an EDS detector.
- Open Access
- 20 January 2026
- Published by Royal Society of Chemistry (RSC) in Chemical Communications
- Vol. 62 (11) , 3642-3645
Divide and correlate: mapping electronic correlations in proteins via local cut-wise reconstruction
We introduce a scalable method to quantify electronic correlations in insulin using mutual information (MI), characterizing interatomic and inter-residue interactions.
- 20 January 2026
- Published by Royal Society of Chemistry (RSC) in Chemical Communications
- Vol. 62 (11) , 3642-3645
Open AccessDivide and correlate: mapping electronic correlations in proteins via local cut-wise reconstruction
We introduce a scalable method to quantify electronic correlations in insulin using mutual information (MI), characterizing interatomic and inter-residue interactions.
- 8 January 2026
- Published by Springer Nature in Theoretical Chemistry Accounts
- Vol. 145 (1) , 16
Occupancies vs embeddings: internal representations of Ab Initio and graph probabilistic models.
- 8 January 2026
- Published by Springer Nature in Theoretical Chemistry Accounts
- Vol. 145 (1) , 16
Occupancies vs embeddings: internal representations of Ab Initio and graph probabilistic models.
- 13 October 2025
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 163 (14)
Seniority-zero states are mean-field wavefunctions
We use symbolic computation, numerical experiments, and mathematical arguments to establish that every seniority-zero state can be expressed as a (number-symmetry-broken) antisymmetric product of interacting geminals (APIG). Moreover, for bosons, the configuration interaction coefficients are exactly parameterized by matrix permanents. This establishes that bosonic wavefunctions can be exactly described by a mean field, specifically the (number-symmetry-broken) symmetric product of boson orbitals. While exact parameterization is extremely costly, we present numerical evidence showing that the APIG wavefunction approaches the true wavefunction exponentially quickly as the number of interacting geminals increases.
- 13 October 2025
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 163 (14)
Seniority-zero states are mean-field wavefunctions
We use symbolic computation, numerical experiments, and mathematical arguments to establish that every seniority-zero state can be expressed as a (number-symmetry-broken) antisymmetric product of interacting geminals (APIG). Moreover, for bosons, the configuration interaction coefficients are exactly parameterized by matrix permanents. This establishes that bosonic wavefunctions can be exactly described by a mean field, specifically the (number-symmetry-broken) symmetric product of boson orbitals. While exact parameterization is extremely costly, we present numerical evidence showing that the APIG wavefunction approaches the true wavefunction exponentially quickly as the number of interacting geminals increases.
- Open Access
- 15 September 2025
- Published by Springer Nature in Research Square
Occupancies vs Embeddings: Internal representations of Ab Initio and Graph Probabilistic Models*
Chemical modeling has traditionally been separated into two fairly non-overlapping frameworks: the exact framework of \textit{ab initio} models and the data-driven framework of statistical models. However, recent advances in machine learning approaches have intercepted the exact \textit{ab initio} frameworks by using an \textit{end-to-end} probabilistic model of the Schr\"odinger equation in the Born-Oppenheimer approximation---from raw nuclear configurations to quantum observables such as the energy. While probabilistic modelling offers promising advancements to speed up computational predictions, the internal representation and its connection to the underlying quantum data---the wave function or electron density---which they implicitly rely on for training, remains relatively poorly understood. This study seeks to make a first comparison of the internal representation from a probabilistic end-to-end models with those from pure \textit{ab initio} frameworks such as density functional theory. We do this by comparing the internal atomwise representation of graph models, also called embeddings, to their most natural \textit{ab initio} counterpart, electron occupancy values. Our findings show that the embedding representations can be employed to transfer learn atomic occupancy values whereas the reverse mapping appears to be less accurate. We discuss the assumption that the probabilistic model could infer the existence of an underlying electron density underpinning the computational \textit{ab-initio} data.
- 15 September 2025
- Published by Springer Nature in Research Square
Open AccessOccupancies vs Embeddings: Internal representations of Ab Initio and Graph Probabilistic Models*
Chemical modeling has traditionally been separated into two fairly non-overlapping frameworks: the exact framework of \textit{ab initio} models and the data-driven framework of statistical models. However, recent advances in machine learning approaches have intercepted the exact \textit{ab initio} frameworks by using an \textit{end-to-end} probabilistic model of the Schr\"odinger equation in the Born-Oppenheimer approximation---from raw nuclear configurations to quantum observables such as the energy. While probabilistic modelling offers promising advancements to speed up computational predictions, the internal representation and its connection to the underlying quantum data---the wave function or electron density---which they implicitly rely on for training, remains relatively poorly understood. This study seeks to make a first comparison of the internal representation from a probabilistic end-to-end models with those from pure \textit{ab initio} frameworks such as density functional theory. We do this by comparing the internal atomwise representation of graph models, also called embeddings, to their most natural \textit{ab initio} counterpart, electron occupancy values. Our findings show that the embedding representations can be employed to transfer learn atomic occupancy values whereas the reverse mapping appears to be less accurate. We discuss the assumption that the probabilistic model could infer the existence of an underlying electron density underpinning the computational \textit{ab-initio} data.
- Open Access
- 30 July 2025
- Published by American Chemical Society (ACS) in ChemRxiv
From dipeptide systems to polypeptides: evolution of mutual information.
Understanding the electronic structure of amino acids is crucial for understanding protein stability, folding mechanisms, and molecular interactions. In this study, we introduce Fragment-wise Mutual Information (FMI) as a quantum information-based tool to quantify inter-atomic correlations in peptides. By extending mutual information (MI) analysis to amino acid fragments, FMI provides a detailed map of electronic interactions beyond classical descriptors such as van der Waals forces. We first validated FMI on 400 dipeptides, demonstrating a correlation with atomization and bonding energies. Expanding this approach to the 10-mer Neh2 peptide, we analyze molecular dynamics (MD) simulations and reveal how inter-atomic correlations evolve during folding. Our results show that FMI distinguishes stabilizing interactions such as salt bridges and variable hydrogen bond strengths, providing deeper insight into peptide stability. These findings suggest that FMI could enhance molecular modeling and force field development by incorporating quantum electronic effects into biomolecular analysis.
- 30 July 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessFrom dipeptide systems to polypeptides: evolution of mutual information.
Understanding the electronic structure of amino acids is crucial for understanding protein stability, folding mechanisms, and molecular interactions. In this study, we introduce Fragment-wise Mutual Information (FMI) as a quantum information-based tool to quantify inter-atomic correlations in peptides. By extending mutual information (MI) analysis to amino acid fragments, FMI provides a detailed map of electronic interactions beyond classical descriptors such as van der Waals forces. We first validated FMI on 400 dipeptides, demonstrating a correlation with atomization and bonding energies. Expanding this approach to the 10-mer Neh2 peptide, we analyze molecular dynamics (MD) simulations and reveal how inter-atomic correlations evolve during folding. Our results show that FMI distinguishes stabilizing interactions such as salt bridges and variable hydrogen bond strengths, providing deeper insight into peptide stability. These findings suggest that FMI could enhance molecular modeling and force field development by incorporating quantum electronic effects into biomolecular analysis.
- 23 July 2025
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 163 (4)
Spin-generator coordinate method for electronic structure
We present a new application of the generator coordinate method (GCM) as an electronic structure method for strong electron correlation in molecular systems. We identify spin fluctuations as an important generator coordinate responsible for strong static electron correlation that is associated with bond-breaking processes. Spin-constrained unrestricted HF (c-UHF) states are used to define a manifold of basis states for the Hill–Wheeler equations, which are discretized and solved as a non-orthogonal configuration interaction expansion. The method was tested on two-electron systems that are dominated by static and/or dynamic correlations. In a minimal basis set for H2, the resulting GCM quickly captures the ground-state full configuration interaction energy with just a few c-UHF states, whereas second-order perturbation theory on top of the GCM is needed to recover over 90% of the correlation energy in the cc-pVDZ basis set.
- 23 July 2025
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 163 (4)
Spin-generator coordinate method for electronic structure
We present a new application of the generator coordinate method (GCM) as an electronic structure method for strong electron correlation in molecular systems. We identify spin fluctuations as an important generator coordinate responsible for strong static electron correlation that is associated with bond-breaking processes. Spin-constrained unrestricted HF (c-UHF) states are used to define a manifold of basis states for the Hill–Wheeler equations, which are discretized and solved as a non-orthogonal configuration interaction expansion. The method was tested on two-electron systems that are dominated by static and/or dynamic correlations. In a minimal basis set for H2, the resulting GCM quickly captures the ground-state full configuration interaction energy with just a few c-UHF states, whereas second-order perturbation theory on top of the GCM is needed to recover over 90% of the correlation energy in the cc-pVDZ basis set.
- Open Access
- 23 July 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Divide and Correlate: Mapping Electronic Correlations in Proteins via Local Cut-Wise Reconstruction
This study quantifies electronic correlations in the insulin protein using mutual information (MI) to evaluate interatomic and inter-residue interaction strength. More specifically, we introduced and analyzed a cut-wise approach rooted in the principle of quantum locality, assuming that electronic correlations decay rapidly with distance. Through localized DFT calculations on 49 overlapping spherical cuts, we reconstruct global mutual informa- tion matrices at both the atomic and residue levels. This method captures biochemically meaningful features while closely matching results from full-protein DFT calculations, demonstrating the scalability of quantum correlation analysis for large biomolecular systems. The framework sets the stage for future applications in protein–ligand interaction modeling, pharmacophore development, and quantum-enhanced drug discovery.
- 23 July 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessDivide and Correlate: Mapping Electronic Correlations in Proteins via Local Cut-Wise Reconstruction
This study quantifies electronic correlations in the insulin protein using mutual information (MI) to evaluate interatomic and inter-residue interaction strength. More specifically, we introduced and analyzed a cut-wise approach rooted in the principle of quantum locality, assuming that electronic correlations decay rapidly with distance. Through localized DFT calculations on 49 overlapping spherical cuts, we reconstruct global mutual informa- tion matrices at both the atomic and residue levels. This method captures biochemically meaningful features while closely matching results from full-protein DFT calculations, demonstrating the scalability of quantum correlation analysis for large biomolecular systems. The framework sets the stage for future applications in protein–ligand interaction modeling, pharmacophore development, and quantum-enhanced drug discovery.