Derivative Discontinuity in Many-Body Perturbation Theory and Chemical Potentials in Random Phase Approximation

This paper derives analytical expressions for chemical potentials within the Random Phase Approximation (RPA) and demonstrates that the GW correlation self-energy exhibits a derivative discontinuity at integer particle numbers, thereby resolving the apparent inconsistency between accurate quasiparticle energies and large delocalization errors in RPA total energies.

The Gist

The Big Picture: The "Price Tag" of Electrons

Imagine you are running a store where you sell electrons.

• Selling an electron is like removing a customer from your store (Ionization).
• Buying an electron is like adding a new customer (Electron Affinity).

In the world of chemistry and physics, the "Chemical Potential" is essentially the price tag for these transactions. It tells you how much energy it costs to add or remove an electron from a molecule. If you know this price, you can predict how a material will behave, how it conducts electricity, or how it reacts with other materials.

For decades, scientists have had two different ways of calculating this price tag, and they usually agreed. But when they looked at a specific, very popular method called RPA (Random Phase Approximation), the two methods gave completely different answers. It was like one app saying a coffee costs $2, and another saying it costs $20.

This paper solves that mystery.

---

The Mystery: Two Roads, One Destination?

Scientists usually calculate the "price" (chemical potential) in two ways:

1. The Direct Method (The Scale): You weigh the system with $N$ electrons, then weigh it again with $N+1$ electrons. The difference in weight is the price. This is the "gold standard" and is very accurate.
2. The Functional Method (The Recipe): Instead of weighing, you look at the "recipe" (the mathematical formula) used to calculate the energy. You take the derivative (the slope) of that recipe to see how the price changes.

The Problem:
When scientists used the Direct Method on the RPA recipe, they got a price that made sense. But when they used the Functional Method (the standard mathematical shortcut used for almost everything else), they got a wildly wrong price.

It was as if the recipe said, "To add an electron, you need to pay $0.05," but the scale said, "No, it actually costs $5.00."

Why was the standard math failing?

The Solution: The "Cliff" in the Math

The authors of this paper discovered that the RPA recipe has a hidden cliff or a jump right at the point where you have a whole number of electrons (like 10 electrons, 20 electrons, etc.).

The Analogy: The Staircase vs. The Ramp

Imagine you are walking up a hill to get to a store.

• Normal Math (like MP2 or DFT): The hill is a smooth ramp. If you take a tiny step forward, the slope changes smoothly. You can calculate the steepness at any point easily.
• The RPA Math: The hill is actually a staircase.
  • If you are standing on a step (an integer number of electrons), the ground is flat.
  • But the moment you try to take a tiny step forward (adding an electron), you have to step up onto the next level.
  • The moment you try to take a tiny step backward (removing an electron), you have to step down.

The "slope" (the chemical potential) is different depending on whether you are stepping up or stepping down. There is a sudden jump (a discontinuity) right at the integer number.

The Mistake:
Previous scientists tried to calculate the slope right at the top of the step (the integer number) using the standard formula. But because the formula assumes a smooth ramp, it calculated the slope of the flat step itself, which is zero (or very small). It missed the fact that there is a huge jump to the next step.

The Fix:
The authors showed that to get the right answer, you can't just look at the integer number. You have to look at the "fractional" steps just before and just after the integer.

• If you look at the slope just before the integer (removing an electron), you get the correct "removal price."
• If you look at the slope just after the integer (adding an electron), you get the correct "addition price."

When they did this, the "Recipe Method" finally matched the "Scale Method."

Why Does This Matter?

1. It Fixes a Broken Tool: For a long time, people thought the RPA method was great for calculating total energies but terrible for predicting chemical reactions (because the "price tags" were wrong). Now we know why it was wrong: they were using the wrong math to find the price. If you use the new math (accounting for the jump), RPA works much better.
2. It's a Fundamental Rule: The authors show that this "jump" isn't just a glitch in RPA. It's a fundamental feature of how nature works. Even the most perfect, exact theories of chemistry have these jumps. It's like a law of physics: You cannot smoothly transition from having 10 electrons to 10.0001 electrons without a sudden change in energy cost.
3. Better Materials: By understanding these jumps, scientists can design better batteries, solar cells, and computer chips. If you get the "price tag" wrong, you might think a material is a good conductor when it's actually an insulator.

The Takeaway

Think of the RPA method as a very powerful but slightly tricky calculator.

• Old belief: "The calculator is broken because it gives weird prices for adding electrons."
• New discovery: "The calculator isn't broken; we were just pressing the wrong button. We were trying to measure the slope on a flat step, but we needed to measure the height of the jump to the next step."

Once they realized there was a jump (a derivative discontinuity) in the math, everything made sense. The "expensive" price tag they saw on the scale was real, and the "cheap" price tag from the old math was just an illusion caused by ignoring the jump.

This paper essentially tells us: In the quantum world, the path from "10 electrons" to "11 electrons" isn't a smooth slide; it's a ladder with a big step, and we finally figured out how to measure the height of that step correctly.

Technical Summary

1. Problem Statement

The paper addresses a fundamental inconsistency in the application of the Random Phase Approximation (RPA) and Many-Body Perturbation Theory (MBPT) to calculating chemical potentials (derivatives of total energy with respect to particle number, $N$).

• The Paradox: Standard GW quasiparticle energies (derived from the GW self-energy $\Sigma_{GW}$) provide accurate predictions for Ionization Potentials (IP) and Electron Affinities (EA). However, RPA total energies are known to suffer from massive delocalization errors, leading to poor predictions for IPs and EAs when calculated via finite differences ($\Delta$-RPA).
• The Contradiction: It was previously assumed that the GW self-energy evaluated at integer electron numbers ($\Sigma_{GW}(\omega, 0)$) represents the functional derivative of the RPA correlation energy with respect to the non-interacting Green's function ($G_s$). If this were true, the RPA chemical potential derived via the chain rule should match the accurate GW quasiparticle energies.
• The Discrepancy: As shown in Figure 1 of the paper, the chemical potentials derived from the GW self-energy at integer $N$ differ significantly from the finite-difference derivatives of the RPA energy. The GW self-energy at integer $N$ fails to capture the correct chemical potential, while the RPA energy itself exhibits large errors for fractional charges. The authors seek to resolve why the functional derivative approach fails for RPA compared to other methods like MP2.

2. Methodology

The authors employ two distinct theoretical approaches to derive the RPA chemical potential and compare them against numerical finite-difference benchmarks.

A. Direct Derivative Approach

• Framework: Uses the asymmetric RPA matrix formulation extended to systems with fractional electron numbers ($N \pm \delta$).
• Mechanism: The chemical potential is calculated as the direct derivative of the RPA correlation energy ($E_c^{RPA}$) with respect to the occupation number of the frontier orbital ($n_f$).
• Key Steps:
  1. Solve the asymmetric RPA eigenvalue problem for systems with infinitesimal fractional charges ($\delta < 0$ for electron removal, $\delta > 0$ for addition).
  2. Differentiate the excitation energies ($\Omega_m$) and the trace of the RPA matrix ($A^{asym}$) with respect to $n_f$.
  3. Include orbital relaxation effects via coupled-perturbed Hartree-Fock (CPHF) or Generalized Kohn-Sham (GKS) equations.

B. Functional Derivative Approach (Chain Rule)

• Framework: Treats RPA correlation energy as a functional of the non-interacting Green's function ($G_s$).
• Mechanism: Applies the chain rule: $\mu = \text{Tr} \left[ \frac{\delta E_c^{RPA}}{\delta G_s} \frac{d G_s}{d n_f} \right]$.
• Critical Investigation: The authors investigate the functional derivative $\frac{\delta E_c^{RPA}}{\delta G_s}$ (the GW correlation self-energy, $\Sigma_c^{GW}$) in the limit as the fractional charge $\delta \to 0$.
  • They derive explicit expressions for $\Sigma_c^{GW}$ for $\delta < 0$ and $\delta > 0$.
  • They compare these limits against the standard GW self-energy calculated at integer $N$ ($\Sigma_c^{GW}(\omega, 0)$).

3. Key Contributions and Findings

A. Discovery of Derivative Discontinuity in RPA

The central finding is that the GW correlation self-energy is discontinuous at integer electron numbers when viewed as a functional of $G_s$.

• Two Distinct Limits: As the fractional charge $\delta$ approaches zero, the self-energy has two different limits:
  • $\Sigma_c^{GW}(\omega, -)$: The limit from the electron removal side ($\delta < 0$).
  • $\Sigma_c^{GW}(\omega, +)$: The limit from the electron addition side ($\delta > 0$).
• Inequality: Neither of these limits is equal to the standard self-energy $\Sigma_c^{GW}(\omega, 0)$ calculated strictly at the integer system.
• Consequence: The standard GW self-energy used in quasiparticle calculations ($\Sigma_c^{GW}(\omega, 0)$) is not the correct functional derivative for determining the chemical potential (left or right derivative) of the RPA energy functional.

B. Resolution of the Delocalization Error Paradox

• The large delocalization error in RPA total energies arises because the standard approximation neglects this derivative discontinuity.
• The "accurate" GW quasiparticle energies correspond to the specific limits ($\Sigma_c^{GW}(\omega, \pm)$) required for the chemical potential, but the RPA total energy functional itself, when treated as a continuous function of $G_s$ using $\Sigma_c^{GW}(\omega, 0)$, fails to reproduce the correct slopes (chemical potentials) at integer $N$.
• This explains why RPA energies show large errors for fractional charges and charged states, despite GW quasiparticle energies being accurate for IPs/EAs.

C. Numerical Validation

• Benchmarking: The authors tested 16 molecular systems (e.g., $H_2O$, $C_2H_4$, $N_2$).
• Results:
  • Both the Direct Derivative and the corrected Functional Derivative (using $\Sigma_c^{GW}(\omega, \pm)$) show excellent agreement with finite-difference benchmarks (Mean Absolute Error $\approx 0.01$ eV).
  • Using the standard integer-system self-energy $\Sigma_c^{GW}(\omega, 0)$ yields massive errors (MAE $\approx 5.66$ eV for HOMO derivatives), confirming the discontinuity.

D. Comparison with Other Functionals

The paper categorizes the continuity of functional derivatives across different theories (Table III):

• LDA/GGA: Continuous derivatives with respect to density ($\rho_s$), density matrix ($\gamma_s$), and $G_s$.
• Hybrid Functionals: Continuous with respect to $\gamma_s$ and $G_s$, but discontinuous with respect to $\rho_s$.
• MP2: Continuous derivative with respect to $G_s$ (no discontinuity).
• RPA & Exact XC: Discontinuous with respect to $G_s$, $\gamma_s$, and $\rho_s$.

4. Significance and Implications

1. Theoretical Foundation: The work establishes that derivative discontinuity is a fundamental feature of correlation energy functionals in MBPT, analogous to the known discontinuity in exact DFT exchange-correlation functionals with respect to electron density.
2. Correction of Methodology: It corrects the common misconception that the GW self-energy at integer $N$ is the functional derivative of the RPA energy. Future implementations of RPA chemical potentials must explicitly account for the left and right limits of the self-energy.
3. Fundamental Gap Interpretation: The RPA fundamental gap is shown to be the difference between eigenvalues of two different Hamiltonians (one derived from $\Sigma_c^{GW}(\omega, -)$ and one from $\Sigma_c^{GW}(\omega, +)$), rather than a single Hamiltonian's HOMO-LUMO gap. This aligns with the equation:
   $$\frac{\partial E}{\partial N}\bigg|_+ - \frac{\partial E}{\partial N}\bigg|_- = \Delta \epsilon_{GKS} + \Delta_{xc}$$
   where $\Delta_{xc}$ represents the derivative discontinuity.
4. Practical Impact: The study highlights that minimizing delocalization error is crucial for accurate IP/EA predictions. While RPAE (RPA with exchange) reduces delocalization error compared to pure RPA, the theoretical framework provided here is essential for developing next-generation correlation functionals that correctly handle these discontinuities.

In conclusion, Li and Yang resolve the inconsistency between GW quasiparticle accuracy and RPA delocalization errors by proving the existence of a derivative discontinuity in the RPA correlation energy functional with respect to the Green's function, necessitating a reformulation of how chemical potentials are derived in many-body perturbation theory.