Predicting the suitability of photocatalysts for water splitting using Koopmans spectral functionals: The case of TiO$_2$ polymorphs

This paper demonstrates that a computationally efficient workflow combining DFT interface calculations with Koopmans spectral functionals can accurately predict the band structures and level alignments of rutile, anatase, and brookite TiO$_2$, offering a promising strategy for screening novel photocatalysts for water splitting.

The Gist

The Big Picture: The Solar-Powered Water Splitter

Imagine you have a magical machine that can take a glass of water, shine a flashlight on it, and instantly turn it into hydrogen fuel (like the fuel in a rocket) and oxygen (the air we breathe). This is called photocatalytic water splitting. It's the "Holy Grail" of renewable energy because it uses sunlight to create clean fuel without pollution.

The problem? We don't have a perfect machine yet. We need a material (a catalyst) that acts like a solar-powered gatekeeper. It needs to:

1. Catch the light (absorb sunlight).
2. Let the energy through (have the right "gap" to let electrons jump).
3. Push the water apart (have the right "height" to push electrons up to make hydrogen and pull them down to make oxygen).

For decades, scientists have been trying to find the perfect material. The most famous candidate is Titanium Dioxide (TiO₂), a white powder found in paint and sunscreen. But TiO₂ comes in three different "flavors" (crystal structures): Rutile, Anatase, and Brookite.

The big question is: Which flavor is the best gatekeeper?

The Problem: The "Crystal Ball" is Broken

To answer this, scientists use supercomputers to simulate these materials. They usually use a standard method called DFT (Density Functional Theory). Think of DFT as a crystal ball that predicts how materials behave.

However, this crystal ball has a flaw. It's great at predicting how atoms move, but it's terrible at predicting the energy levels (the "height" of the gates) needed for water splitting. It's like trying to measure the height of a building with a ruler that stretches and shrinks randomly. If the ruler is wrong, you might think a building is tall enough to reach the clouds, when in reality, it's too short.

Because of this, scientists often have to guess or run incredibly expensive, slow simulations (like the "GW" method) to get the right answer. These expensive methods are like hiring a team of 100 architects to measure one building—it's accurate, but it takes forever and costs a fortune.

The Solution: The "Koopmans" GPS

This paper introduces a new, smarter tool called Koopmans spectral functionals.

Think of the old DFT method as a GPS that gives you the wrong address because it doesn't account for traffic. The Koopmans method is like a smart GPS that knows exactly where the traffic is and recalculates the route in real-time.

How does it work?
Instead of just looking at the average crowd (the total electron density), Koopmans looks at individual people (individual electrons). It forces the math to follow a strict rule: "If you add or remove one person, the energy must change in a perfectly straight line." This rule fixes the stretching ruler problem.

The Experiment: Testing the Three Flavors

The researchers used this new "Koopmans GPS" to test the three TiO₂ flavors:

1. Rutile: The most common and stable form.
2. Anatase: The form that usually works best in real life.
3. Brookite: The rare, less-studied form.

What they found:

• The Ruler is Fixed: Koopmans predicted the energy gaps and "gate heights" with amazing accuracy, matching real-world experiments much better than the old methods.
• The Winner: The calculations confirmed that Anatase is the best candidate.
  • Why? Its "gates" are perfectly positioned. The top gate is high enough to split water, and the bottom gate is low enough to catch the hydrogen.
  • Rutile is close, but its bottom gate is a tiny bit too high, making it slightly less efficient.
  • Brookite has great gates, but its "window" (band gap) is too small, meaning it misses out on some of the sunlight energy.

The Best Part: Speed and Efficiency

Here is the real magic of this paper. Usually, to get accurate results, you have to do a massive, slow simulation for every single surface of the material.

The Koopmans method is a shortcut.

• Old Way: You have to build a full model of the material and the air around it, then run a super-expensive calculation on the whole thing.
• Koopmans Way: You only need to do the expensive calculation on the bulk material (the inside). For the surface (the part touching the air/water), you can use the fast, cheap standard method.

It's like checking the engine of a car (the expensive part) to know how fast it can go, and then just looking at the tires (the cheap part) to see how it handles the road. You get the accurate speed without rebuilding the whole car.

The Takeaway

This paper proves that we have a new, fast, and accurate way to screen materials for clean energy. Instead of guessing or waiting months for supercomputers to finish, we can now quickly test thousands of new materials to see if they are the "perfect gatekeeper" for water splitting.

In short: We found a better ruler to measure the energy of solar materials, and it confirmed that Anatase TiO₂ is still the champion, but now we have a fast, cheap way to find even better champions in the future.

Technical Summary

1. Problem Statement

Photocatalytic water splitting (PWS) is a promising route for renewable hydrogen production. For a semiconductor to be an effective photocatalyst, it must satisfy two critical electronic criteria:

1. Band Gap: Must be large enough to overcome the thermodynamic potential of water splitting (1.23 eV) plus kinetic overpotentials (typically requiring 1.6–1.8 eV), but small enough to absorb visible light.
2. Band Alignment: The Valence Band Maximum (VBM) must be higher than the water oxidation potential ($O_2/H_2O$), and the Conduction Band Minimum (CBM) must be lower than the hydrogen reduction potential ($H^+/H_2$).

The Challenge: Standard computational methods struggle to predict these properties accurately.

• Kohn-Sham DFT (KS-DFT): While computationally efficient, standard semi-local functionals (like PBE) suffer from self-interaction errors and lack of piecewise linearity, leading to severe underestimation of band gaps and unreliable band edge positions.
• High-Accuracy Methods: Hybrid functionals (e.g., HSE06) and many-body perturbation theory (e.g., GW) offer better accuracy but are computationally expensive. Furthermore, calculating band alignment (relative to vacuum) typically requires modeling explicit slab/vacuum interfaces, which multiplies the computational cost when combined with high-level theories like GW.

2. Methodology

The authors propose a computationally efficient workflow using Koopmans spectral functionals to predict band structures and band alignments for three $TiO_2$ polymorphs: Rutile, Anatase, and Brookite.

A. Theoretical Framework: Koopmans Functionals

Koopmans functionals are "orbital-density-dependent" (ODD) functionals designed to restore the equivalence between Kohn-Sham eigenvalues and total energy differences ($\Delta$SCF). They enforce the Generalized Piecewise Linearity (GPWL) condition, ensuring that orbital energies are independent of their occupation numbers.

• KI (Koopmans Integral): Corrects the base DFT functional to match $\Delta$SCF energies. For insulating systems, the ground-state density of KI matches the base functional (PBE), allowing the use of PBE densities as a starting point.
• pKIPZ (Perturbative Koopmans-Perdew-Zunger): Adds a screened self-interaction correction to KI. While KIPZ modifies the ground-state density (requiring full minimization), the authors utilize a perturbative approach (pKIPZ) where the Hamiltonian is evaluated on the KI ground state to reduce cost.

B. The Hybrid Workflow for Band Alignment

The paper introduces a novel strategy to decouple the expensive spectral calculations from the surface modeling:

1. Slab Calculation (DFT Level): A slab model of the photocatalyst interfaced with vacuum is calculated using standard semi-local DFT (PBE). This provides the macroscopic average electrostatic potential difference ($\Delta V$) between the bulk and the vacuum. Since the ground-state density of KI/pKIPZ matches PBE for insulators, this potential shift is valid for the corrected functionals.
2. Bulk Calculation (Koopmans Level): The band gap and band edge shifts ($\Delta \epsilon$) are calculated for the bulk primitive cell using Koopmans functionals (KI or pKIPZ).
3. Alignment: The Ionization Potential (IP) and Electron Affinity (EA) are derived by combining the DFT vacuum reference ($\Delta V$) with the Koopmans-corrected band edges:
   $$IP = \Delta V_{DFT} - (\epsilon^{DFT}_{VBM} + \Delta \epsilon_{VBM})$$
   $$EA = \Delta V_{DFT} - (\epsilon^{DFT}_{CBM} + \Delta \epsilon_{CBM})$$

C. Computational Details

• Software: Quantum ESPRESSO with the koopmans package.
• Parameters: Norm-conserving pseudopotentials, 80 Ry cutoff.
• Screening Parameters: Calculated via $\Delta$SCF in supercells (with image charge corrections) or Density Functional Perturbation Theory (DFPT).
• Surfaces: Thermodynamically stable facets were modeled: Rutile (110), Anatase (101), and Brookite (210).

3. Key Contributions

1. Efficient Band Alignment Protocol: Demonstrated that accurate band alignment can be achieved by combining low-cost DFT slab calculations (for vacuum referencing) with high-accuracy Koopmans bulk calculations. This avoids the prohibitive cost of performing full Koopmans or GW calculations on large slab models.
2. Benchmarking $TiO_2$ Polymorphs: Provided a comprehensive comparison of Rutile, Anatase, and Brookite using Koopmans functionals, a method rarely applied to Brookite in this context.
3. Validation Against Experiment: Showed that Koopmans functionals (KI and pKIPZ) predict band gaps and band edge positions with accuracy comparable to or better than Hybrid functionals and GW methods, but with a different error profile and computational scaling.

4. Key Results

A. Band Gaps

• Accuracy: KI@PBE and pKIPZ@PBE predicted band gaps within ~0.36 eV of experimental values (after accounting for zero-point renormalization, ZPR).
• Comparison:
  • Rutile: KI predicted ~4.04 eV (Exp: 3.1–4.1 eV).
  • Anatase: KI predicted ~4.51 eV (Exp: 3.4–3.9 eV).
  • Brookite: KI predicted ~4.63 eV.
• Trends: Unlike HSE06 and $G_0W_0@PBE$, which often predict Brookite to have the smallest gap, Koopmans functionals predict Brookite to have the largest gap among the three polymorphs. The authors note that the "correct" ordering remains ambiguous due to wide experimental scatter and lack of ZPR data for Brookite.

B. Band Alignment (Suitability for Water Splitting)

• Rutile: The CBM lies only marginally above the hydrogen reduction potential ($H^+/H_2$). While thermodynamically possible, this borderline alignment may explain its lower experimental efficiency compared to Anatase.
• Anatase: Shows the most favorable alignment. The VBM is sufficiently high for water oxidation, and the CBM is well-positioned for hydrogen reduction. This aligns with experimental consensus that Anatase is the superior photocatalyst.
• Brookite: Shows favorable band edge positions but a larger band gap (predicted ~4.6 eV) which would limit visible light absorption, potentially requiring band-gap engineering.

C. Excitonic Effects

The paper notes that while excitonic binding energies differ significantly (Anatase has strongly bound excitons ~160 meV vs. Rutile's weak ~4–26 meV), these effects do not significantly shift the band-edge positions relative to redox potentials. Therefore, the single-particle band alignment results remain valid for assessing thermodynamic suitability.

5. Significance and Conclusion

• Predictive Power: The study validates Koopmans spectral functionals as a robust tool for predicting the suitability of photocatalysts. The method achieves accuracy comparable to GW and Hybrid functionals but offers a more systematic approach to spectral properties by enforcing piecewise linearity.
• Computational Efficiency: The proposed workflow (DFT slabs + Koopmans bulk) significantly reduces the computational barrier for screening new materials. It allows researchers to assess band alignment without the massive cost of explicit interface calculations at the GW or full Koopmans level.
• Future Applications: This strategy is scalable and can be deployed to screen novel, less-studied photocatalyst candidates, accelerating the discovery of materials for renewable energy production.

In summary, the paper establishes that Anatase is the most promising $TiO_2$ polymorph for water splitting based on Koopmans predictions, and it provides a streamlined, high-accuracy computational protocol for future materials discovery.