Valence and Rydberg excited state bond dissociation curves of CO2 from orbital-optimized density functional calculations

This study demonstrates that orbital-optimized density functional theory, particularly when using the PBE functional with complex orbitals, provides a computationally efficient and highly accurate method for modeling the valence and Rydberg excited state bond dissociation curves of CO2, significantly outperforming linear-response time-dependent DFT and offering a promising approach for simulating photorelaxation processes in condensed-phase environments.

The Gist

The Big Picture: Taking a Snapshot of a Shaking Molecule

Imagine the Carbon Dioxide ($CO_2$) molecule as a tiny, three-bead necklace floating in space. Usually, it sits calmly. But if you hit it with a burst of energy (like a cosmic ray or a laser), it gets "excited." It starts vibrating, stretching, and potentially breaking apart.

Scientists want to predict exactly how this necklace behaves when it's excited. They need to draw a map (a "potential energy curve") that shows how much energy the molecule has as its bonds stretch. This map is crucial for understanding things like how ice on distant planets reacts to space radiation.

However, drawing this map is incredibly hard. The molecule doesn't just vibrate; its electrons jump into weird, fuzzy, high-energy orbits called Rydberg states. These are like electrons that have decided to live in a cloud far away from the nucleus, making them very hard to track with standard math tools.

The Problem: The Old Mapmakers Were Getting Lost

For a long time, scientists used a popular method called TD-DFT (Time-Dependent Density Functional Theory) to draw these maps. Think of TD-DFT as a GPS that works great for driving on a highway (normal chemical reactions) but gets completely confused when you try to drive off-road into a swamp (excited, fuzzy electron states).

• The Issue: When the old GPS tried to map the "fuzzy" Rydberg states, it often gave the wrong location. Sometimes it was off by a huge margin (up to 1.9 electron-volts, which is a lot in the atomic world). It was like telling a hiker they were at the base of a mountain when they were actually halfway up the peak.
• The Consequence: If your map is wrong, your simulation of how the molecule breaks apart or interacts with neighbors will be wrong, too.

The New Solution: The "Orbital-Optimized" GPS

The authors of this paper tested a different, smarter approach called Orbital-Optimized (OO) Density Functional Theory.

Instead of using a generic GPS that tries to guess the path based on average traffic, this new method customizes the map for the specific journey.

• The Analogy: Imagine you are trying to find a specific person in a crowded room.
  • The Old Way (TD-DFT): You ask a security guard, "Where is the person in the red shirt?" The guard gives you a general area based on where red shirts usually stand. You might miss them.
  • The New Way (OO): You walk into the room, spot the person in the red shirt, and optimize your path directly to them. You adjust your route specifically for that person's location.

In the paper, they found that this "customized" approach works much better. Even when they used a simple, standard mathematical tool (the PBE functional), the results were incredibly accurate—within 0.3 eV of the "gold standard" reference values.

Key Discoveries in Simple Terms

1. The "Fuzzy" Clouds Need Special Glasses
The Rydberg electrons are so spread out (diffuse) that they are hard to capture with standard computer grids. The authors found that using complex numbers (a specific type of math that handles rotation and symmetry better) was like putting on 3D glasses. It allowed the computer to see the perfect, round symmetry of the electron cloud, whereas the old method (using real numbers) made the cloud look lopsided and broken.

2. The "Gold Standard" Comparison
To prove their new method works, they compared it against the most expensive, high-precision methods available (like EOM-CCSD and MRCI).

• The Result: The new method was almost as accurate as the expensive "gold standard" but cost a tiny fraction of the computer power. It's like getting a luxury car ride for the price of a bus ticket.

3. The "Trap" in the Ice
One of the most interesting findings concerns the 3pσ state.

• The Story: When $CO_2$ gets hit by radiation, it usually breaks apart quickly into Carbon Monoxide (CO) and Oxygen (O).
• The Twist: The new maps showed that the 3pσ state acts like a trap. Instead of breaking immediately, the molecule gets stuck in this excited state for a tiny moment (about 150 femtoseconds).
• Why it matters: In the freezing cold of space (interstellar ice), this tiny pause is huge. It gives the excited molecule time to bump into its neighbors and transfer energy, potentially triggering chemical reactions that wouldn't happen otherwise. This could explain how complex molecules form in space.

Why Should We Care?

This research is a game-changer for astrochemistry (the study of chemistry in space).

• Current Models: Scientists currently use rough estimates to guess how cosmic rays affect ice on comets or planets.
• The Future: With this new, fast, and accurate method, scientists can now simulate exactly how $CO_2$ ice reacts to radiation in space. They can model how these tiny "traps" help build the building blocks of life in the universe.

The Bottom Line

The authors developed a way to calculate the behavior of excited $CO_2$ molecules that is fast, cheap, and surprisingly accurate. By treating each excited state individually and using the right mathematical "glasses" (complex orbitals), they created reliable maps of how these molecules stretch and break. This opens the door to understanding the chemical evolution of the universe, from the ice on distant moons to the clouds of interstellar space.

Technical Summary

1. Problem Statement

The accurate simulation of excited-state dynamics in molecules, particularly those involving Rydberg states (highly diffuse electron distributions) and valence states with mixed character, remains a significant challenge in computational chemistry.

• Limitations of TD-DFT: Standard Time-Dependent Density Functional Theory (TD-DFT), relying on the adiabatic approximation and linear response, often fails to describe excitations involving large electron density rearrangements, such as Rydberg and charge-transfer states. It frequently underestimates excitation energies, with errors varying significantly depending on the specific functional and the character of the excitation.
• Limitations of High-Level Methods: While high-accuracy methods like Multireference Configuration Interaction (MRCI) and Equation-of-Motion Coupled-Cluster (EOM-CCSD) exist, they are computationally expensive, scaling poorly with system size. This restricts their application to small molecules and prevents their use in modeling condensed-phase systems (e.g., interstellar ices).
• Specific Challenge: The CO2 molecule presents a complex test case due to its rich manifold of excited states, including crossings between Rydberg (3s, 3pσ) and valence (π*) states near the equilibrium geometry. Existing high-level reference curves for CO2 dissociation often suffer from spurious oscillations (in MRCI without Rydberg orbitals) or incorrect dissociation limits (in EOM-CCSD).

2. Methodology

The authors employ Orbital-Optimized (OO) Density Functional Theory, a state-specific approach where orbitals are variationally optimized for the specific excited state of interest, rather than relying on linear response.

• Direct Orbital Optimization (DO): The method finds saddle points on the electronic energy surface corresponding to excited states. It uses an L-SR1 quasi-Newton algorithm to optimize the unitary transformation of an initial set of orbitals (derived from a ground-state reference with non-aufbau occupation).
• Complex vs. Real Orbitals: To preserve the cylindrical symmetry of the electron density in linear CO2 (D∞h symmetry), the authors utilize complex-valued orbitals ($\pi_{\pm}$) rather than real orbitals ($\pi_x, \pi_y$). This is crucial for correctly describing excitations from degenerate $\pi$ orbitals to $\pi^*$ or Rydberg orbitals.
• Spin Purification: Since the calculations use a single Slater determinant, open-shell singlet states are spin-mixed. The authors apply spin purification ($E_s = 2E_{sm} - E_t$) to correct the energy.
• Software & Basis Sets:
  • OO Calculations: Performed using GPAW (Grid-based Projector Augmented Wave) with the PBE functional and d-aug-cc-pVDZ basis sets (or real-space grids).
  • TD-DFT Comparisons: Performed using ORCA with various functionals (PBE, B3LYP, PBE0, BHHLYP, CAM-B3LYP) and real orbitals.
  • Reference Data: Compared against hybrid EOM-CCSD/MRCI curves from Triana et al.

3. Key Contributions

• Validation of OO-DFT for Rydberg States: The study demonstrates that OO-DFT, even with the simple PBE functional, provides remarkably accurate results for Rydberg excitations, outperforming standard TD-DFT.
• Importance of Complex Orbitals: The paper establishes that using complex orbitals is essential for maintaining symmetry and accuracy in linear molecules with degenerate orbitals, preventing symmetry breaking artifacts seen with real orbitals.
• Basis Set Assessment: A rigorous assessment shows that the d-aug-cc-pVDZ basis set is necessary to accurately describe the diffuse nature of Rydberg orbitals, whereas standard aug-cc-pVDZ leads to significant confinement errors (~0.5 eV for 3pσ).
• Dissociation Curves: The authors generate full C-O bond dissociation curves for the 3s, 3pσ, and π* states, comparing them to high-level reference data.

4. Key Results

A. Vertical Excitation Energies

• TD-DFT Performance: Exhibits large errors (up to 1.9 eV) for the diffuse 3pσ state. Errors are highly dependent on the functional; for example, B3LYP is accurate for 3s but poor for 3pσ, while BHHLYP is better for 3pσ but overestimates 3s.
• OO-DFT Performance:
  • Systematically underestimates excitation energies but keeps absolute errors below 0.5 eV across all tested functionals.
  • PBE + Complex Orbitals: Achieves an error within 0.3 eV of MRCI reference values for the 3s state.
  • Hybrid Functionals: Further improve accuracy, reducing errors for both 3s and 3pσ states.
  • Stability: OO results show significantly less dependence on the choice of functional and excitation character compared to TD-DFT.

B. Dissociation Energy Curves

• 3s Rydberg State: The OO/PBE curve reproduces the shape of the reference curve well, including the double crossing with the valence π* state. However, at large bond separations (>4.2 Bohr), the spin-purified OO curve deviates, suggesting spin purification may become unreliable at dissociation limits.
• 3pσ Rydberg State: The state remains bound across the entire dissociation range, consistent with high-level references. The OO curve reproduces the shape accurately, with a slight energy shift.
• π Valence State:*
  • The OO curve follows a diabatic path, leading to dissociation into triplet fragments (CO($a^3\Pi$) + O($^3P$)).
  • The reference curve (constructed via pseudo-diabatization) shows an avoided crossing, leading to singlet fragments (CO($X^1\Sigma^+$) + O($^1S$)).
  • The divergence is attributed to the reference curve's handling of the avoided crossing with a higher-lying state, which the single-determinant OO method naturally avoids. Despite this, the OO curve matches the shape of the adiabatic MRCI curves (without diabatization) remarkably well.

5. Significance and Implications

• Computational Efficiency: OO-DFT offers a favorable balance between accuracy and cost, requiring only a fraction of the computational effort of EOM-CCSD/MRCI. This makes it feasible to model condensed-phase systems, such as CO2 in interstellar ices.
• Astrochemical Applications: The ability to model high-energy Rydberg states is critical for understanding cosmic-ray-driven processes in space. These states can trap excited populations (~150 fs), potentially delaying dissociation and influencing neighboring molecules in ice matrices.
• Future Directions: The method enables the calculation of analytical forces, paving the way for non-adiabatic molecular dynamics simulations of photoexcited molecules in solids and clusters.
• Limitations: The authors note that for systems with multiple CO2 molecules, the self-interaction error in GGA functionals (like PBE) may lead to over-delocalization of the hole. Explicit self-interaction corrections may be required for accurate modeling of clusters.

In conclusion, the paper validates orbital-optimized DFT with complex orbitals as a robust, efficient, and accurate tool for modeling both valence and Rydberg excited states, particularly for applications in astrochemistry and condensed-phase photochemistry where high-level wavefunction methods are computationally prohibitive.