Crystal Orbital Guided Iteration to Atomic Orbitals: A Pathway to Chemically Adaptive Atomic Orbitals from DFT

arXiv:2511.19725 · cond-mat.mtrl-sci · Submitted 2025-11-24 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.

Kai: We're looking at the paper: "Crystal Orbital Guided Iteration to Atomic Orbitals".

Mira: Here's the gist of the paper. Crystal Orbital Guided Iteration to Atomic Orbitals: A Pathway to Chemically Adaptive Atomic Orbitals from DFT shows how an iterative framework can adapt underlying atomic orbitals to achieve accurate tight-binding interpolation while maintaining chemical character The deviation from atomic character in projected Wannier constructions arises from two intrinsic mathematical obstacles: uncontrolled orbital mixing and a fixed-overlap constraint The authors introduce Crystal Orbital Guided Iteration To atomic-Orbitals (COGITO), a framework that iteratively adapts the underlying atomic orbitals such that the nonorthogonal Wannier representation achieves accurate tight-binding interpolation while maintaining atomic character The Core Problem with Existing Methods Existing methods for constructing local orbitals often fail to simultaneously achieve completeness, locality, and chemical interpretability Early approaches searched for an optimized orbital basis by restricting the orbital to a functional form with tunable parameters, but this limits reliability and adaptability Enforcing orbital orthogonality forces the resulting orbitals to mix with atomic orbitals on neighboring atoms, making orthogonal orbitals non-transferable between systems The COGITO Solution Framework COGITO is an iterative strategy that alternates between constructing projected Wannier functions and using them to fit a new atomic basis, allowing the Wannier and atomic representations to co-evolve The full COGITO process—including the Bloch orbital update, coefficient refinement, and atomic orbital fitting—is iterated such that the atomic orbital basis and its overlap matrix converge to a chemically faithful representation of the KS electronic structure Mechanism of Adaptation The iterative procedure involves three key steps: 1) Restrict the updated Bloch orbitals to have no orbital mixing and the correct complex phase of reciprocal-space coefficients, 2) Modify the orbital coefficients to span the KS wavefunctions as desired, and 3) Extract numerical local orbitals from the numerical Bloch orbitals at k = 0 and fit to analytical atomic orbitals in a flexible multi-Gaussian form Step 1 restricts the formulation of the projected Bloch orbital to remove orbital mixing, nudging the Wannier representation towards atomic character Validation and Results The COGITO basis demonstrates high quality by meeting four criteria across a test set of 200 semiconductors and metals It shows a tenfold reduction in sensitivity to basis initialization (Criterion 2) and achieves near completeness with a median charge spilling of only 0.27% and median orbital mixing of only 0.56% The COGITO tight-binding model achieves a median band distance error of 1.3 meV when coefficient optimization is included, which is a significant improvement over the projection onto PAW pseudo-orbitals Chemical Insight and Interpretation COGITO enables rigorous analysis of covalent bonding by decomposing the Kohn–Sham wavefunctions into local atomic contributions (COHP) and visualizing short-range versus long-range interactions The method successfully captures chemical intuition regarding the covalency versus ionicity of GaN polymorphs, aligning with physical expectations while contradicting previous results from LOBSTER Finally, COGITO effectively distinguishes between fundamentally different bonding regimes in the 200 compounds by separating short-range and long-range interactions based on atom composition COGITO establishes a new foundation for understanding and modeling the quantum mechanics of bonding in materials by providing a minimal and localized yet complete basis that bridges the gap between atomic and plane-wave representations COGITO successfully captures covalency, ionicity, and metallicity across materials with clarity, consistency, and predictive power Pruning redundant reasoning tokens The deviation from atomic character in projected Wannier constructions arises from two intrinsic mathematical obstacles: uncontrolled orbital mixing and a fixed-overlap constraint The authors introduce Crystal Orbital Guided Iteration To atomic Orbitals (COGITO), a framework that iteratively adapts the underlying atomic orbitals such that the nonorthogonal Wannier representation achieves accurate tight-binding interpolation while maintaining atomic character How the training signal is built The COGITO procedure alternates between refining atomic-like Bloch orbitals (Steps 1 and 2) and fitting corresponding strictly atomic orbitals (Step 3) Step 2 ensures the Wannier representation spans the desired KS states by proposing a combined Gram-Schmidt and Lowdin orthogonalization that enforces completeness for the valence bands while hierarchically decoupling bands by energy and projectability Where the gains come from COGITO achieves an orbital mixing 9× better than the PAW pseudo-orbitals, which is a result of non-negligible PAW pseudo-orbitals projections on high-energy bands not included in the calculated KS bands The improved projection leads to better band interpolation of valence band and lowest conduction bands, as demonstrated by the second two columns in Fig.

Kai: 8 The final coefficient optimization hugely reduces the error, achieving a 65-fold reduction in median error compared to the unmodified projections Where the gains come from COGITO achieves an orbital mixing 9× better than the PAW pseudo-orbitals, which is a result of non-negligible PAW pseudo-orbitals projections on high-energy bands not included in the calculated KS bands The improved projection leads to better band interpolation of valence band and lowest conduction bands, as demonstrated by the second two columns in Fig.

Title and authors: Kai: So we're looking at this paper today, "Crystal Orbital Guided Iteration to Atomic Orbitals: A Pathway to Chemically Adaptive Atomic Orbitals from DFT." It’s about how to get local orbitals that actually match what atoms are doing in a material.

Mira: Yeah, it tackles the problem that when you try to make these local orbitals, they end up mixing with their neighbors instead of staying localized on the atom itself.

Lev: I mean, from an error correction standpoint, if the basis isn't localized properly, then trying to build a reliable tight-binding model on top of it is going to be like trying to correct noise in a signal that's already smeared out across the whole system.

Kai: Exactly. The paper says they introduce this Crystal Orbital Guided Iteration, or COGITO, as an iterative framework to adapt those underlying atomic orbitals. It’s not just one static way of doing it; it’s a process where the orbitals and the Wannier functions co-evolve.

Mira: What interests me is how they address those two specific mathematical hurdles: uncontrolled orbital mixing and that fixed-overlap constraint that forces delocalization onto neighboring atoms.

Lev: Because in real hardware, you can't just assume a fixed overlap; the system parameters change, and if your basis isn't adaptive, the whole calculation becomes unstable.

Kai: They show this iterative process—the Bloch orbital update, coefficient refinement, and fitting to analytical atomic orbitals—is repeated until the atomic orbital basis and its overlap matrix converge to something chemically faithful.

Mira: So, instead of picking a single functional form upfront and hoping for the best, they have this dynamic mechanism that constantly nudges the representation toward actual chemical character.

Lev: It sounds like a lot of computational overhead, though I wonder how much time that iteration actually takes before you hit convergence on a real system.

Kai: Well, they validate this approach by testing it across two hundred semiconductors and metals to show it meets four criteria for quality in terms of chemical interpretability and adaptability.

Mira: And the results are pretty compelling. They report a tenfold reduction in sensitivity to the initial basis choice, which is significant because that sensitivity usually tells you how fragile your method is.

Lev: A tenfold reduction is good, but I’m looking at the error numbers; how much actual physical error are we talking about when you do this?

Kai: When they include coefficient optimization in their tight-binding model, they achieve a median band distance of one point three meV across their test set, which they say is a big improvement over projecting onto PAW pseudo-orbitals.

Mira: That comparison to PAW pseudo-orbitals is important because those projections can be quite messy; this method seems to clean that up by achieving an orbital mixing of only zero point five six percent in their best results.

Lev: Nine times better than the PAW pseudo-orbitals, you mentioned earlier? What does that mean for the actual computational cost when running error correction schemes on a system where the basis is this much cleaner?

Kai: It means the improved projection leads to better interpolation of both valence band and lowest conduction bands, as shown in their figures, suggesting more accurate physics in those regions.

Mira: They also use this framework to decompose Kohn–Sham wavefunctions into solely local atomic contributions, which gives us a new way to visualize short-range versus long-range interactions within the crystal structure.

Lev: That decomposition sounds useful for analyzing material stability, but what about the limitations? What is this method explicitly saying it doesn't handle well?

Kai: They do mention that they are working to achieve better band interpolation, and they show how this iterative approach can help distinguish between different bonding regimes based on atom composition.

Mira: So, to wrap up these results, the main implication is that COGITO provides a minimal yet localized basis that successfully bridges the gap between atomic and plane-wave representations.

Lev: For someone building real hardware, the key thing here is that this iterative process offers a pathway to create bases that are robust enough to handle the complexities of electronic structure without relying on overly simplistic fixed constraints.

Kai: So, in summary, Crystal Orbital Guided Iteration to Atomic Orbitals: A Pathway to Chemically Adaptive Atomic Orbitals from DFT gives us a method for constructing chemically adaptive atomic orbitals by iteratively adapting underlying atomic orbitals such that the nonorthogonal Wannier representation achieves accurate tight-binding interpolation while maintaining atomic character.

Mira: It moves us away from methods where we just restrict the orbital form with tunable parameters because that limits how reliable and adaptable the basis can really be.

Lev: And for those of us building real hardware, it shows that by allowing those representations to co-evolve, you can enforce a level of chemical fidelity that standard projections often miss.

Kai: We're going to take a quick break and then talk about how this iterative idea connects to other methods we’re exploring in the realm of quantum information.

Mira: Stick around, because I want to show you what this kind of basis adaptation looks like when you start looking at how AI can use it for generating more interpretable models of electronic structure.

Lev: We'll be right back with more on the practicalities and the error scaling.

The paper's summary: Kai: So, we just looked at how Crystal Orbital Guided Iteration to Atomic Orbitals, or COGITO, builds these new atomic orbitals by fixing the way they mix with neighbors and keeping them chemically correct.

Mira: Right, it’s about this iterative process where you alternate between building a Wannier function and then using that function to refine the actual atomic orbitals themselves.

Lev: From my side, what I'm hearing is that this isn't just one calculation; it’s a whole loop—updating the Bloch orbitals, refining the coefficients, and then fitting them back to these analytical atomic orbital forms.

Kai: Exactly. It’s a feedback system designed to make sure the atomic basis actually reflects the real physics of the Kohn-Sham electronic structure rather than just being an arbitrary mathematical starting point.

Mira: The core idea is tackling those two big problems we talked about before, which are uncontrolled mixing and that fixed overlap constraint that forces things to stay on one atom.

Lev: So, instead of trying to force a single, perfect basis all at once, this method uses the iteration itself—the process—to break those constraints step by step until they converge on something chemically faithful.

Kai: That convergence leads to a tight-binding model that is much more accurate when you include coefficient optimization. They're talking about a median band distance of one point three meV there.

Mira: And that accuracy comes from the fact that this process can capture what’s happening at different length scales, allowing them to decompose the Kohn–Sham wavefunctions into purely local atomic contributions.

Lev: That decomposition is useful because it helps visualize exactly where the covalent bonding and the ionicity are coming from in a material. It gives you a clearer picture of those short-range versus long-range interactions.

Kai: The real world implication for us, especially with AI, is that this method gives us a way to build local bases that are both minimal and complete enough to bridge the gap between the atomic description and the plane-wave description.

Mira: It means an AI system could use these orbitals to generate more interpretable models of electronic structure by visualizing covalent bonds and charge transfer directly.

Lev: And for error correction, if you can start with a basis that’s already this localized and chemically informed, it should make running those high-fidelity simulations on real hardware much easier because the initial setup is less sensitive.

Kai: But they do have limitations, Mira mentioned earlier—they are focused on capturing these local contributions accurately within the DFT framework, and they're still working on pushing the limits of how far this can generalize across different material classes.

Mira: It’s a tool that shows how to get a basis that is adaptable—capturing shifts in orbital shape under different charge states or crystal environments—which is crucial if you want your AI models to stay accurate when simulating materials under varying conditions.

Lev: So, the next thing we need to look at is how this iterative setup compares to other ways of achieving localization, like those symmetry-based methods we discussed on lattice thermal conductivity.

The paper's improvements: Tom: So, we’re looking at how Crystal Orbital Guided Iteration to Atomic Orbitals suggests specific ways to make this approach even better than just running the base iteration loop.

Kai: It seems they focus on refining the fitting process, specifically using a combined Gram-Schmidt and Lowdin orthogonalization method when proposing those new Wannier functions.

Mira: Right, that’s the training signal part. They aren't just letting the system run; they are actively modifying how the Wannier representation spans those Kohn–Sham states by enforcing completeness for valence bands while also hierarchically decoupling different energy bands.

Lev: That sounds like a clever way to manage complexity. It means they aren’t trying to solve everything at once, but structuring the problem so that you get better band separation as you go.

Kai: And this is where it gets interesting for hardware, because this structured approach should lead to more stable results when we try to map these concepts onto superconducting qubit architectures or any finite-basis system.

Mira: If the underlying atomic orbitals are built this way, the resulting tight-binding model should be much more robust against small perturbations in the input data or crystal environment.

Lev: I’m thinking about the error scaling here—if you do all that iterative refinement, does it keep everything in constant time territory, or does it start hitting something like big O of n squared?

Kai: Well, they suggest that with coefficient optimization included in the tight-binding model, you see a six-five fold reduction in median error compared to just doing the basic projection onto PAW pseudo-orbitals.

Mira: That improvement comes from the fact that this better projection leads to better band interpolation for both the valence and lowest conduction bands. It’s about getting a cleaner picture of those key electronic states.

Lev: So, when you run this on real hardware, it means we might need less aggressive error correction overhead because the underlying basis itself is already much closer to the true physical situation.

Kai: It really does change how we think about the fidelity required for these local descriptions when we try to build things that are meant to mimic real materials.

Mira: This also gives us a better way to analyze bonding—through that decomposition into purely local atomic contributions—which is great for understanding those short-range and long-range interactions in materials.

Lev: The caveat is still that you’re relying on the DFT input being good, because if the initial Kohn–Sham wavefunctions are flawed, you’re just refining a flawed picture.

Kai: Right, so it’s an iterative refinement of a calculation that assumes a certain level of accuracy from the start.

Mira: Exactly. It shows how this framework can bridge that gap by allowing the Wannier representation and the atomic representation to co-evolve for chemical faithfulness.

Lev: So, if we look at how this compares to other techniques, like those symmetry-based methods for thermal conductivity or sequential retrieval in memory, it seems COGITO offers a path toward creating local descriptions that are truly adaptable and chemically meaningful.

Conclusion: Kai: So to wrap up, Crystal Orbital Guided Iteration to Atomic Orbitals gives us a way to build chemically adaptive atomic orbitals by iteratively adapting those underlying orbitals so the nonorthogonal Wannier representation achieves accurate tight-binding interpolation while maintaining atomic character.

Mira: It shows that you can move past methods that just restrict the orbital form with fixed parameters because this iterative process allows the representations to co-evolve toward chemical faithfulness.

Lev: I think what’s really important is how this method tackles those fixed overlap constraints by using iterative, chemically guided modifications to break them step by step.

Kai: The results show that when you include coefficient optimization in the tight-binding model, you get a median band distance of one point three meV, which is a big improvement over just projecting onto PAW pseudo-orbitals.

Mira: That better projection leads to improved interpolation for both the valence and lowest conduction bands, which is exactly what we need for accurate electronic structure modeling.

Lev: For running this on real hardware, it means the initial setup is less sensitive because the basis itself is already much closer to the true physical situation.

Kai: So in short, COGITO provides a minimal yet localized basis that bridges the gap between atomic and plane-wave representations with high precision.

Mira: It’s a tool for rigorous chemical analysis because it lets you decompose Kohn–Sham wavefunctions into solely local atomic contributions to visualize bonding trends.

Lev: The main limitation is still that you rely on the DFT input being good, because if the initial Kohn–Sham wavefunctions are flawed, you’re just refining a flawed picture.

Kai: And that’s fair—it’s a refinement of a calculation, not a magic fix for any initial error in the ground state calculation.

Mira: But it does show how to get local orbitals that are adaptable, capturing shifts in orbital shape under different charge states or crystal environments.

Lev: So looking ahead, we need to see how this iterative idea connects to other methods we’re exploring, like those symmetry-based techniques for lattice thermal conductivity.

Kai: Exactly. We have a lot more work on scaling this up and seeing how it performs when you push it onto larger systems or materials with more complex bonding.

Emily Oliphant, Emmanouil Kioupakis, Wenhao Sun

University of Michigan · Massachusetts Institute of Technology

cond-mat.mtrl-sci

Submitted: 2025-11-24

Updated: 2026-10-28

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 100/100

The gist: Crystal Orbital Guided Iteration to Atomic Orbitals: A Pathway to Chemically Adaptive Atomic Orbitals from DFT shows how an iterative framework can adapt underlying atomic orbitals to achieve…

Key concepts

Orbital Mixing
This occurs when the constructed local orbitals are not strictly localized to single atoms but instead overlap significantly with neighboring atoms. COGITO addresses this by iteratively modifying the representation to ensure orbitals maintain their distinct atomic character while still accurately describing the electronic structure.
Fixed-Overlap Constraint
Traditional methods often force Wannier functions to be orthogonal, which prevents them from accurately representing chemical shifts between different environments. COGITO breaks this constraint through iterative adaptation, allowing the basis to change flexibly and maintain fidelity across varying chemical conditions.
Tight-Binding Interpolation
This refers to the ability of the resulting local orbitals to accurately reconstruct the electronic structure using a simple tight-binding model. COGITO ensures that even when using nonorthogonal Wannier representations, this interpolation remains precise, leading to accurate energy band predictions.
COHP (Kohn–Sham Hamilton Population)
This is a tool used by COGITO to rigorously analyze chemical bonding by decomposing the Kohn-Sham wavefunctions into contributions solely from individual atoms. This allows for a clear visualization of short-range versus long-range interactions, helping to distinguish between different types of bonding regimes.

Terminology

Summary

Crystal Orbital Guided Iteration to Atomic Orbitals: A Pathway to Chemically Adaptive Atomic Orbitals from DFT shows how an iterative framework can adapt underlying atomic orbitals to achieve accurate tight-binding interpolation while maintaining chemical character

The deviation from atomic character in projected Wannier constructions arises from two intrinsic mathematical obstacles: uncontrolled orbital mixing and a fixed-overlap constraint The authors introduce Crystal Orbital Guided Iteration To atomic-Orbitals (COGITO), a framework that iteratively adapts the underlying atomic orbitals such that the nonorthogonal Wannier representation achieves accurate tight-binding interpolation while maintaining atomic character

The Core Problem with Existing Methods

Existing methods for constructing local orbitals often fail to simultaneously achieve completeness, locality, and chemical interpretability Early approaches searched for an optimized orbital basis by restricting the orbital to a functional form with tunable parameters, but this limits reliability and adaptability Enforcing orbital orthogonality forces the resulting orbitals to mix with atomic orbitals on neighboring atoms, making orthogonal orbitals non-transferable between systems

The COGITO Solution Framework

COGITO is an iterative strategy that alternates between constructing projected Wannier functions and using them to fit a new atomic basis, allowing the Wannier and atomic representations to co-evolve The full COGITO process—including the Bloch orbital update, coefficient refinement, and atomic orbital fitting—is iterated such that the atomic orbital basis and its overlap matrix converge to a chemically faithful representation of the KS electronic structure

Mechanism of Adaptation

The iterative procedure involves three key steps: 1) Restrict the updated Bloch orbitals to have no orbital mixing and the correct complex phase of reciprocal-space coefficients, 2) Modify the orbital coefficients to span the KS wavefunctions as desired, and 3) Extract numerical local orbitals from the numerical Bloch orbitals at k = 0 and fit to analytical atomic orbitals in a flexible multi-Gaussian form Step 1 restricts the formulation of the projected Bloch orbital to remove orbital mixing, nudging the Wannier representation towards atomic character

Validation and Results

The COGITO basis demonstrates high quality by meeting four criteria across a test set of 200 semiconductors and metals It shows a tenfold reduction in sensitivity to basis initialization (Criterion 2) and achieves near completeness with a median charge spilling of only 0.27% and median orbital mixing of only 0. p. 0

Where the paper says it

Although Maximally localized Wannier functions (MLWFs) can form a complete orbital basis, the resulting Wannier orbitals smear onto neighboring atoms, obscuring their real-space atomic character. Here, we show that the deviation from atomic character in projected Wannier constructions arises from two intrinsic mathematical obstacles: uncontrolled orbital mixing and a fixed-overlap constraint.

56% The COGITO tight-binding model achieves a median band distance error of 1. p. 1
Where the paper says it

IV outlines the construction of a nonorthogonal tight-binding model from COGITO. V demonstrates that the COGITO basis accurately spans the Kohn-Sham wavefunctions (Criterion 3) and yields high-quality tight-binding interpolations of band structure (Criterion 4).

3 meV when coefficient optimization is included, which is a significant improvement over the projection onto PAW pseudo-orbitals

Chemical Insight and Interpretation

COGITO enables rigorous analysis of covalent bonding by decomposing the Kohn–Sham wavefunctions into local atomic contributions (COHP) and visualizing short-range versus long-range interactions The method successfully captures chemical intuition regarding the covalency versus ionicity of GaN polymorphs, aligning with physical expectations while contradicting previous results from LOBSTER Finally, COGITO effectively distinguishes between fundamentally different bonding regimes in the 200 compounds by separating short-range and long-range interactions based on atom composition

Conclusion

COGITO establishes a new foundation for understanding and modeling the quantum mechanics of bonding in materials by providing a minimal and localized yet complete basis that bridges the gap between atomic and plane-wave representations COGITO successfully captures covalency, ionicity, and metallicity across materials with clarity, consistency, and predictive power

Pruning redundant reasoning tokens

The deviation from atomic character in projected Wannier constructions arises from two intrinsic mathematical obstacles: uncontrolled orbital mixing and a fixed-overlap constraint The authors introduce Crystal Orbital Guided Iteration To atomic Orbitals (COGITO), a framework that iteratively adapts the underlying atomic orbitals such that the nonorthogonal Wannier representation achieves accurate tight-binding interpolation while maintaining atomic character

How the training signal is built

The COGITO procedure alternates between refining atomic-like Bloch orbitals (Steps 1 and 2) and fitting corresponding strictly atomic orbitals (Step 3) Step 2 ensures the Wannier representation spans the desired KS states by proposing a combined Gram-Schmidt and Lowdin orthogonalization that enforces completeness for the valence bands while hierarchically decoupling bands by energy and projectability

Where the gains come from

COGITO achieves an orbital mixing 9× better than the PAW pseudo-orbitals, which is a result of non-negligible PAW pseudo-orbitals projections on high-energy bands not included in the calculated KS bands The improved projection leads to better band interpolation of valence band and lowest conduction bands, as demonstrated by the second two columns in Fig. 8 The final coefficient optimization hugely reduces the error, achieving a 65-fold reduction in median error compared to the unmodified projections

COGITO achieves an orbital mixing 9× better than the PAW pseudo-orbitals, which is a result of non-negligible PAW pseudo-orbitals projections on high-energy bands not included in the calculated KS bands The improved projection leads to better band interpolation of valence band and lowest conduction bands, as demonstrated by the second two columns in Fig.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Crystal Orbital Guided Iteration To atomic-Orbitals (COGITO) can be used to construct a chemically adaptive basis by iteratively adapting underlying atomic orbitals such that the nonorthogonal Wannier representation achieves accurate tight-binding interpolation while maintaining atomic character. This allows AI systems to generate more chemically interpretable models of electronic structure through orbital-resolved covalent bonds and charge transfer.

  2. COGITO enables the creation of a highly reliable local basis by satisfying four criteria: The basis should be chemically interpretable, carrying the form of atomic orbitals without distortion from the angular function of spherical harmonics, and The basis should be adaptable and unique, capturing shifts in orbital radial function under different charge states and crystal environments. This means AI systems can maintain accuracy when simulating materials under varying chemical environments.

  3. The iterative nature of COGITO allows the system to overcome limitations like orbital mixing and the implicit fixed-overlap constraint by performing iterative, chemically guided modifications to the Wannier representation that break the fixed-overlap constraint. This capability means AI systems can generate local orbitals that are strictly atomic while still accurately spanning Kohn-Sham wavefunctions.

  4. The COGITO framework produces a tight-binding model with high accuracy, achieving a median band distance of 1.3 meV when coefficient optimization is included, and demonstrating an ability to reproduce DFT calculated bands with improved accuracy compared to standard projections like PAW pseudo-orbitals. This allows AI systems to perform physics or chemistry based applications that rely on a faithful description of local electronic structure with high precision.

  5. COGITO provides a tool for rigorous chemical bonding analysis by decomposing the Kohn-Sham Hamilton Population into solely local atomic contributions and enabling a new algorithmic visualization of bonds within the crystal structure. This allows AI systems to analyze covalent bond densities and bond energies (COHP) with enhanced resolution, distinguishing between short-range and long-range bonding trends.

Abstract

Atomic orbitals underpin our understanding of electronic structure, providing intuitive descriptions of bonding, charge transfer, magnetism, and correlation effects. Despite their utility, an atomic basis that is adaptable, strictly localized on atomic centers, and enables accurate tight-binding interpolation has remained elusive. Here, we introduce Crystal Orbital Guided Iteration To atomic-Orbitals (COGITO), a framework that constructs an optimal atomic orbital basis by identifying and resolving key mathematical obstacles inherent to nonorthogonal bases -- particularly uncontrolled orbital mixing, and the fixed-overlap constraint between orbitals. We demonstrate that COGITO enables tight-binding models as accurate as MLWF-based approaches, while preserving the ability of tight-binding parameters to represent the projected atomic basis -- an essential feature lost in schemes that enforce orbital orthogonality or maximal localization. By creating accurate and chemically interpretable models of electronic structure, COGITO reveals the orbital-resolved covalent bonds and charge transfer that is encoded in the Kohn-Sham wavefunctions of DFT. Our method thus offers a powerful tool for any physics- or chemistry-based application that relies on a faithful description of local electronic structure.

Sources

Related papers