Optical excitations in nanographenes from the Bethe-Salpeter equation and time-dependent density functional theory: absorption spectra and spatial descriptors

This paper presents a validated implementation of the GW-BSE formalism in the CP2K code to accurately predict the optical spectra and excitation sizes of nanographenes, demonstrating its superiority over time-dependent density functional theory for describing electronic excitations in nanostructures.

The Gist

Imagine a nanographene as a tiny, flat, rectangular piece of a honeycomb made of carbon atoms. It's so small that it's measured in nanometers, but it's big enough to act like a miniature semiconductor. When light hits this tiny sheet, it can kick an electron loose, leaving behind a "hole" (a spot where an electron used to be). Because opposite charges attract, the electron and the hole don't just run away; they hold hands and dance around each other, forming a bound pair called an exciton.

This paper is about figuring out exactly how these electron-hole pairs dance, how much energy it takes to start the dance, and how big the dance floor is.

The Problem: Guessing the Dance Moves

Scientists have two main ways to predict how these particles behave:

1. The "Local" Guess (TDDFT): This is like trying to predict a dance by only looking at the dancers' immediate neighbors. It's fast and easy to compute, but it often misses the fact that the electron and hole are pulling on each other from a distance. It's like trying to predict a long-distance phone call by only listening to the people in the same room.
2. The "Full Picture" Method (GW-BSE): This is the gold standard. It's like having a super-accurate map of the entire ballroom, including the invisible magnetic forces pulling the dancers together. It's much more computationally expensive (it takes a lot of computer power), but it's supposed to be the most accurate.

What the Authors Did

The researchers, Maximilian Graml and Jan Wilhelm, built a new tool inside a popular computer program called CP2K. They implemented the "Full Picture" method (GW-BSE) to study these nanographenes.

Think of it as upgrading a video game engine. Before, the game could only simulate simple physics. Now, they added a high-fidelity physics engine that can simulate the complex "electron-hole dance" accurately.

The Results: A Perfect Match

They tested their new tool on a standard set of organic molecules first. It was like a driving test: the car (their code) performed perfectly, matching the reference data with an error so small it's barely noticeable (less than the width of a single atom).

Then, they applied it to nanographenes of increasing lengths.

• The Spectrum: They calculated the "absorption spectrum," which is essentially the color of light the material absorbs. When they compared their computer predictions to real-world experiments, the colors matched almost perfectly.
• The Size: They measured the "size of the excitation." Imagine the electron and hole are holding a stretchy rubber band. How long is that band?
  • For short nanographenes, the band stretches as the molecule gets longer.
  • But once the molecule gets big enough (around 4 nanometers long), the band stops stretching. It settles at a fixed size of about 7.6 Angstroms (roughly the width of a few atoms). This proves that the electron and hole are tightly bound, like a couple dancing in a small circle, regardless of how big the room gets.

The Comparison: Why the "Local" Guess Fails

The authors then asked: Can the faster, cheaper method (TDDFT) do the same job if we just tweak the settings?

They tried different "recipes" (mathematical functions) for the TDDFT method, changing how much "exact exchange" (a specific type of mathematical correction) was included.

• The Result: No matter which recipe they used, the cheaper method failed to get both the energy and the size right at the same time.
  • Some recipes got the energy right but predicted the electron and hole were too far apart (the rubber band was too loose).
  • Others got the size right but the energy was wrong.
  • One recipe even created "ghost peaks" in the data—predicting colors of light that shouldn't exist.

The Conclusion

The paper concludes that while the cheaper methods are useful for quick guesses, they are fundamentally flawed for describing these specific nanostructures. They miss the long-range "hand-holding" (Coulomb attraction) between the electron and hole.

To get a truly accurate picture of how these tiny carbon sheets interact with light—both the energy they absorb and the physical size of the electron-hole pair—you need the heavy-duty, many-body physics approach (GW-BSE). The authors have successfully put this powerful tool into the CP2K software, making it available for others to use to study these tiny, light-harvesting materials.

Technical Summary

Technical Summary: Optical Excitations in Nanographenes from the Bethe-Salpeter Equation and Time-Dependent Density Functional Theory

Problem Statement
The accurate description of electronic excitations in nanometer-sized materials, such as nanographenes, presents a significant challenge. While Time-Dependent Density Functional Theory (TDDFT) is widely used, standard local or semi-local exchange-correlation functionals fail to capture the long-range electron-hole attraction mediated by the Coulomb interaction. Consequently, these approximations often miss bound excitons or yield inaccurate excitation energies and spatial descriptors. Hybrid functionals partially restore this interaction via exact exchange but introduce a sensitivity to the chosen exact-exchange fraction and remain computationally expensive. The GW plus Bethe-Salpeter equation (GW-BSE) formalism offers a rigorous many-body approach that explicitly includes screened Coulomb interactions, yet its application to finite nanostructures requires robust implementation and validation against both reference data and experimental spectra. Furthermore, the extent to which TDDFT can reproduce the spatial characteristics of excitations (e.g., exciton size) compared to GW-BSE remains an open question for finite systems.

Methodology
The authors implemented the GW-BSE formalism within the CP2K software package. The workflow begins with a Kohn-Sham (KS) DFT calculation to establish the ground-state electronic structure. This is followed by an $evGW_0$ calculation, where quasiparticle energies are obtained through eigenvalue self-consistency in the Green's function ($G$) while keeping the screened Coulomb interaction ($W_0$) fixed from the DFT starting point. The Bethe-Salpeter equation (BSE) is then solved in the product space of occupied and unoccupied orbitals to determine excitation energies and wavefunctions.

To manage computational scaling ($O(N^6)$ for the eigenvalue problem), the authors employ energy cutoffs for both occupied and unoccupied states. The implementation utilizes Gaussian basis functions and the resolution-of-the-identity (RI) technique for efficient evaluation of Coulomb integrals. The optical absorption spectra are derived from the dynamical dipole polarizability tensor $\alpha_{\mu\mu'}(\omega)$.

Spatial descriptors are computed directly from the BSE excitation wavefunction $\Psi^{(n)}_{exc}(r_e, r_h)$. These include the excitation size ($d_{exc}$), electron and hole spreads ($\sigma_e, \sigma_h$), the electron-hole separation ($d_{e\to h}$), and the correlation coefficient ($R_{eh}$). The study focuses on rectangular nanographenes with a fixed width (seven carbon atoms in the zigzag direction) and varying lengths ($L$) in the armchair direction.

Key Contributions and Results

1. Implementation and Validation: The GW-BSE implementation in CP2K was validated against the FHI-aims code using Thiel's set of organic molecules. The results showed excellent agreement, with a mean absolute error (MAE) in excitation energies of 2.7 meV, confirming the numerical precision of the new implementation.
2. Optical Spectra of Nanographenes: The authors calculated absorption spectra for nanographenes of increasing length ($L=5, 10, 13$). The imaginary part of the polarizability was dominated by the longitudinal component ($\alpha_{xx}$). By extrapolating the peak frequencies of the first two dominant longitudinal excitations to an experimental length of approximately 20 nm ($L \approx 46$), the authors obtained excitation energies of 2.09 eV and 2.26 eV. These values align closely with experimental measurements (2.05 eV and 2.31 eV), demonstrating the capability of the BSE formalism to accurately predict optical properties in this intermediate regime between molecules and crystals.
3. Spatial Descriptors and Exciton Convergence: The study computed the size of the lowest optically active excitation ($d_{exc}$). As the nanographene length increased, the excitation size converged to approximately 7.6 Å. This saturation indicates the formation of a bound exciton with a well-defined spatial extent (exciton radius) that is intrinsic to the material and smaller than the system size. The electron-hole correlation coefficient approached unity for large $L$, confirming strongly correlated electron-hole motion.
4. Comparison with TDDFT: A systematic comparison with TDDFT using various functionals (PBE, BLYP, PBE0, B3LYP, HSE06, CAM-B3LYP) revealed significant discrepancies.
   • Excitation Energies: Semi-local functionals (GGA) substantially underestimated excitation energies. Hybrid functionals improved agreement but showed sensitivity to the exact-exchange fraction; for instance, CAM-B3LYP overestimated energies due to excessive long-range exact exchange.
   • Spatial Descriptors: While some hybrid functionals could reproduce the excitation energy, they failed to consistently reproduce the spatial size of the excitation. GGAs predicted overly delocalized excitations where size grew linearly with system length, lacking the bound exciton character. Hybrid functionals with short-range exact exchange (e.g., HSE06) showed some confinement but still exhibited size growth with length.
   • Spectral Shape: TDDFT spectra, even with hybrid functionals, displayed qualitative differences compared to GW-BSE, including incorrect peak intensities and the appearance of spurious sidebands.

Significance
The paper establishes that the GW-BSE formalism, implemented in CP2K, provides a consistent and accurate description of both the spectral and spatial properties of electronic excitations in nanographenes. The work highlights that while TDDFT with hybrid functionals can offer qualitative trends or match specific descriptors (like excitation energy) through parameter tuning, it fails to simultaneously reproduce the correct excitation size, optical spectra, and spatial correlations without the explicit many-body treatment of the electron-hole interaction provided by GW-BSE. The convergence of the excitation size to a finite value (~7.6 Å) in nanographenes underscores the bound nature of excitons in these finite nanostructures, a feature that standard TDDFT approximations struggle to capture accurately. The study underscores the necessity of many-body methods for the precise characterization of electronic excitations in nanoscale optoelectronic materials.