In the famous 1996 paper of Burke, Perdew and Ernzerhof in the very first paragraph it is stated that $E_{XC} = E_X + E_C$. Then, a GGA functional is derived starting from the energy density of uniform electron gas with an attenuation function. In the final step a formula for the XC-energy is abruptly given but the correlation energy is seemingly left out with no explanation.

The starting point for the correlation energy is $E_C = \int \text{d}^3 r\, n \cdot (\epsilon_C^{\text{unif}} + H)$, where $H$ is dependent on $\nabla n/n$ and is chosen such that $E_C$ is sensible for certain conditions of the particle density $n$.

Furthermore, the exchange energy is given by $E_X = \int \text{d}^3r\, n \epsilon_X^{\text{unif}} F_X$, where in almost the same way $F_X$ is dependent on $\nabla n / n$ and is constructed to satisfy certain conditions of $n$.

Thinking back to the definition of $E_{XC}$, one would expect that these two energies are taken as a sum to give the final result. However, to my surprise a new coefficient is given

$$ E_{XC} = \int \text{d}^3 r\, n \epsilon_X^{\text{unif}}F_{XC}. $$

Moreover, the actual functional form of $F_{XC}$ is omitted, which leaves me in confusion.

Where are any of the parts of $E_C$ in the final XC-energy? Are they contained in $F_{XC}$? Why did they not write the form of $F_{XC}$ into the paper? What am I missing here?


The final equation you quoted merely serves to define $F_{XC}$, which is plotted in Fig. 1 of the article to succinctly illustrate the behavior of the PBE functional as a function of density and density gradient. Without removing the $\epsilon_X^{unif}(n)$ factor from $E_{XC}$, one cannot plot the $r_s=0$ and $r_s=\infty$ curves anymore, that's why the authors have to define $F_{XC}$ and plot it, rather than plotting $E_{XC}$ itself. The equation is not meant to be an alternative definition of $E_{XC}$.

  • $\begingroup$ I think I understand please correct me if I'm wrong. Since $\epsilon_X^{\text{unif}} \propto 1/r_s$ at $r_s \rightarrow 0$ the exchange energy is diverging and for $r_s \rightarrow \infty$ the exchange term is zero which would make for an uninteresting case. Indeed, in Ref. [4] it says that for LDA $F_{XC} = \epsilon_{XC}/\epsilon_X$. So just to be sure, $E_{XC}^{PBE} = E_X + E_C$ as defined in my question? $\endgroup$
    – A-V Labs
    Jul 7 at 11:20
  • $\begingroup$ Yes, your understanding is right. And $E_{XC}$ is always defined as $E_X + E_C$. $\endgroup$
    – wzkchem5
    Jul 7 at 12:29
  • 1
    $\begingroup$ Not always; there are also non-separable forms e.g. several Minnesota functionals. Most empirical functionals don't divide in exchange and correlation; after all, the whole concept of "exchange" and "correlation" boils down to our inability of solving the Schrödinger equation. Feynman said we should call it "stupidity energy" instead! $\endgroup$ Jul 12 at 15:05
  • $\begingroup$ @SusiLehtola Yes you're right. I forgot about these "NGA" functionals $\endgroup$
    – wzkchem5
    Jul 12 at 15:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.