Atomic forces from correlation energy functionals based on the adiabatic-connection fluctuation-dissipation theorem

This paper presents the implementation of analytical atomic forces for correlation energy functionals based on the adiabatic-connection fluctuation-dissipation theorem within the random phase approximation (RPA) and RPAx frameworks, demonstrating their high numerical accuracy and systematic improvement over standard DFT methods for predicting geometries, vibrational frequencies, and anharmonic phonon shifts in molecules and solids.

The Gist

Imagine you are trying to build the perfect model of a molecule or a crystal, like a tiny, invisible Lego structure made of atoms. To do this, you need a computer program that can predict exactly where every atom should sit and how it will vibrate.

For decades, scientists have used a set of rules called Density Functional Theory (DFT) to do this. Think of DFT as a "good enough" map. It's fast and usually gets you to the right neighborhood, but it has some known flaws. It often misses the subtle "glue" that holds things together (like the weak van der Waals forces) and sometimes gets the distance between atoms slightly wrong. It's like using a map that says "New York is here," but it's actually 5 miles off.

This paper introduces a much more precise, high-definition map called the Random Phase Approximation (RPA), and the authors have just added a crucial new feature to it: Atomic Forces.

Here is the breakdown of what they did, using some everyday analogies:

1. The Problem: The "Blind" Map

In the past, the RPA method was like a super-accurate GPS for calculating the energy of a system (how stable the structure is). However, it couldn't easily calculate forces.

• The Analogy: Imagine you are trying to find the bottom of a valley (the most stable position for atoms). The old RPA method could tell you the altitude of the ground at any point, but it couldn't tell you which way is "downhill." Without knowing which way is downhill, the computer has to guess its way down, step by tiny step, which is incredibly slow and inefficient.

2. The Solution: Adding the "Compass"

The authors of this paper figured out how to calculate the forces (the "downhill" direction) directly and analytically within the RPA method.

• The Analogy: They didn't just give you the altitude; they gave you a compass that points exactly where the atoms need to move to find the most stable spot. Now, instead of stumbling around in the dark, the computer can glide straight to the bottom of the valley.

3. Two Ways to Drive the Car

The paper describes two ways to use this new compass:

• The "Self-Consistent" Way (The Perfect Driver): The computer recalculates the map and the compass at the same time, over and over, until everything is perfectly aligned. This is the most accurate method but takes a long time to compute.
• The "Non-Self-Consistent" Way (The Smart Shortcut): The computer starts with a rough, fast map (using a simpler method called PBE) and then applies the high-precision RPA compass on top of it.
  • The Result: The authors found that for most molecules and solids, the "shortcut" works almost as well as the "perfect driver." It's like using a high-end GPS on top of a basic map; you get 99% of the accuracy for a fraction of the time.

4. Why Does This Matter?

With this new tool, scientists can now:

• Predict Shapes Better: They can tell you exactly how long a chemical bond is or what angle a molecule makes, with much higher precision than before.
• Predict Vibrations: Atoms aren't static; they jiggle. This method can predict exactly how fast they vibrate (which determines things like the color of a diamond or how heat moves through silicon).
• Fix Old Mistakes: The authors tested this on Diamond, Silicon, and Germanium (the stuff computer chips are made of). They found that their new method corrected errors that previous methods had made, bringing the theoretical predictions almost perfectly in line with real-world experiments.

5. The "Secret Sauce" (RPAx)

The paper also touches on an even more advanced version called RPAx, which includes an extra layer of "exact exchange" (a very specific type of quantum interaction).

• The Analogy: If RPA is a high-definition map, RPAx is a 3D holographic map. It's even more accurate. The authors couldn't calculate the forces for this version analytically yet (it's too complex), so they used a "brute force" method (checking every step manually) to prove that RPAx is indeed the gold standard, matching the results of the most expensive and complex supercomputer simulations available.

The Bottom Line

This paper is a major upgrade for the "engine" that drives modern materials science. By teaching the computer how to feel the "push and pull" between atoms using the most accurate physics available (RPA), the authors have made it possible to design new materials, drugs, and electronics with a level of precision that was previously too slow or too difficult to achieve.

In short: They took a super-accurate but slow calculator, gave it a steering wheel, and showed that it can drive almost as fast as the old, less accurate cars while arriving at a much better destination.

Technical Summary

1. Problem Statement

While the Adiabatic-Connection Fluctuation-Dissipation Theorem (ACFDT) and its approximations, specifically the Random Phase Approximation (RPA) and RPA + exchange (RPAx), have established themselves as highly accurate methods for calculating electronic correlation energies (often rivaling advanced wavefunction methods like CCSD(T)), their application to structural and vibrational properties has been limited.

• The Gap: Most studies using RPA rely on "non-self-consistent" approaches where geometries are optimized using cheaper semi-local functionals (like PBE) and then refined using RPA energies. This creates a mismatch, as semi-local functionals often overestimate bond distances and underestimate vibrational frequencies.
• The Challenge: Calculating analytical atomic forces (first-order derivatives of the energy with respect to atomic positions) for RPA is mathematically complex, particularly when:
  1. Achieving self-consistency (solving the Optimized Effective Potential or OEP equation).
  2. Handling nonlocal pseudopotentials (common in plane-wave codes), which introduce additional terms in the force expression that are often neglected.
  3. Extending these calculations to periodic solids and large molecular systems efficiently.

2. Methodology

The authors implemented analytical atomic forces for RPA and RPAx within the Quantum ESPRESSO (QE) software package, utilizing plane-wave basis sets and norm-conserving pseudopotentials.

A. Theoretical Framework

• Self-Consistent (SCF) Forces:

  • The authors derived the analytical gradient of the RPA energy at full self-consistency using the Hellmann-Feynman theorem (HFT) combined with the OEP method.
  • Key Derivation: They identified that when using nonlocal pseudopotentials, the standard HFT expression is incomplete. They derived a crucial correction term ($\Delta F_{xc}$) arising from the variation of the nonlocal potential with respect to atomic positions. This term involves the variation of the RPA correlation energy with respect to the effective potential.
  • The implementation utilizes an eigenvalue decomposition of the linear density response function ($\chi_0$) to avoid explicit summation over unoccupied states, solving an eigenvalue problem iteratively via Density Functional Perturbation Theory (DFPT).

• Non-Self-Consistent (NSCF) Forces:

  • To reduce computational cost, they also implemented forces starting from a PBE reference (denoted RPA[PBE]).
  • These forces are calculated using DFPT, treating the PBE orbitals as the starting point and evaluating the RPA correlation contribution as a perturbation.

• RPAx Implementation:

  • Analytical forces for RPAx (which includes the exact-exchange kernel) were not derived analytically in this work due to complexity. Instead, forces for RPAx were estimated using finite-difference total energy calculations.

B. Computational Setup

• Code: Implemented in the ACFDT external package of Quantum ESPRESSO.
• Basis: Plane-waves with Optimized Norm-Conserving Vanderbilt (ONCV) pseudopotentials.
• Validation: Numerical accuracy was verified by comparing analytical forces against finite-difference total energy derivatives for systems like HF, H$_2$O, and $\omega$-BN.

3. Key Contributions

1. First Analytical RPA Forces in Plane-Waves: This is the first implementation of analytical RPA forces within a plane-wave/pseudopotential framework that handles both self-consistent and non-self-consistent cases.
2. Correction for Nonlocal Pseudopotentials: The authors rigorously derived and implemented the missing force correction term required when using nonlocal pseudopotentials within the OEP formalism. They demonstrated that omitting this term leads to significant errors in bond lengths (up to 0.03 Å for HF).
3. Validation of Numerical Quality: The implementation was shown to achieve high numerical precision, with errors in force calculations well below the convergence thresholds required for geometry relaxation (BFGS algorithm).
4. Systematic Benchmarking: The authors provided a comprehensive benchmark of RPA and RPAx against CCSD(T), Diffusion Monte Carlo (DMC), and experimental data for a diverse set of molecules and solids.

4. Results

A. Numerical Accuracy

• Force Convergence: The analytical forces match finite-difference energy derivatives with high precision (errors $< 10^{-5}$ Ry/bohr for SCF RPA).
• Geometry Relaxation: The BFGS algorithm successfully relaxed molecular geometries (e.g., HF) using the new forces, converging to the same equilibrium geometries as total energy minimization.
• Importance of Correction: Removing the nonlocal pseudopotential correction term resulted in large errors (0.085 Ry/a$_0$ in total force), confirming the necessity of the derived term.

B. Molecular Systems (H$_2$, LiH, HF, N$_2$, H$_2$O, NH$_3$, CH$_4$)

• SCF vs. NSCF: Self-consistent RPA and non-self-consistent RPA[PBE] yielded nearly identical structural parameters and vibrational frequencies. This suggests that the cheaper RPA[PBE] approach is sufficient for most applications.
• Accuracy vs. PBE: RPA systematically improved upon PBE.
  • Bond Distances: RPA reduced the Mean Absolute Percentage Error (MAPE) from 2.72% (PBE) to 2.11%.
  • Vibrational Frequencies: RPA significantly improved "bending" modes (MAPE reduced from 4.14% to 0.75%) but slightly overestimated "stretching" modes.
• RPAx Performance: Including the exact-exchange kernel (RPAx) further improved results, achieving MAPEs of 0.42% for bond distances and 0.85% for stretching frequencies, comparable to CCSD(T).

C. Solids (Diamond, Silicon, Germanium)

• Lattice Parameters: RPA predicted lattice constants in excellent agreement with experiment (after correcting for Zero-Point Energy). RPAx[PBE] showed the best agreement with experimental lattice parameters for diamond.
• Optical Phonons ($\Gamma$-point):
  • Diamond: RPA predicted a harmonic frequency of 1342 cm$^{-1}$ (vs. 1334 cm$^{-1}$ exp), a massive improvement over PBE (1290 cm$^{-1}$). RPAx(1) predicted 1348 cm$^{-1}$, nearly perfect agreement.
  • Silicon & Germanium: Similar trends were observed. RPA corrected the PBE underestimation of phonon frequencies. RPAx(1) provided the most accurate results, often matching experimental values within a few cm$^{-1}$.
• Anharmonicity: The authors estimated anharmonic shifts using finite-difference methods with RPAx-optimized hybrid functionals, providing refined theoretical references for the harmonic frequencies of these materials.

5. Significance and Outlook

• Reliability: The work establishes RPA and RPAx as reliable, first-principles tools for predicting structural and vibrational properties, bridging the gap between semi-local DFT and expensive wavefunction methods.
• Efficiency: The demonstration that non-self-consistent RPA forces (RPA[PBE]) are nearly as accurate as full SCF forces makes these high-accuracy methods computationally feasible for larger systems.
• Future Directions:
  • The implementation of analytical forces for RPAx (currently done via finite differences) is the next logical step to enable efficient geometry optimization with the exact-exchange kernel.
  • The framework paves the way for calculating second-order derivatives (stress tensors for cell relaxation) and electron-phonon couplings, which are critical for understanding superconductivity and thermal transport.

In conclusion, this paper removes a major barrier to the practical application of ACFDT-based functionals in materials science by providing robust, accurate, and efficient analytical forces, enabling the study of lattice dynamics and structural stability with unprecedented accuracy.