In the following, suppose $n$ is the electron density and $n^\uparrow$ and $n^\downarrow$ denote the density of spin up and spin down electrons respectively. Therefore $n = n^\uparrow + n^\downarrow$.

I am following NIST in their calculation of exchange and correlation for the DFT LDA approximation. The authors state that $$ V_{xc} = \frac{d}{dn} n\cdot\big(\varepsilon_x(n) + \varepsilon_c(n)\big), \tag{1} $$ but with the definition $\varepsilon_x \equiv \varepsilon_x^P + f(\zeta)(\varepsilon_x^F - \varepsilon_x^P)$ where $\zeta = (n^\uparrow - n^\downarrow)/n$, I will need the total derivative $dn^\uparrow/dn$ (by the chain rule). To me, it seems that this is impossible to determine, because I do not have enough equations to solve $n^\uparrow$ as a function of $n$ only, and I am therefore stuck in the cycle $$ \frac{d n^\uparrow}{dn} = \frac{\partial n^\uparrow}{n} + \frac{\partial n^\uparrow}{\partial n^\downarrow} \frac{d n^\downarrow}{d n} = 1 - \frac{d n^\downarrow}{d n} = 1 - \left[ \frac{\partial n^\downarrow}{n} + \frac{\partial n^\downarrow}{\partial n^\uparrow} \frac{d n^\uparrow}{d n} \right] = 1 - \left[1 - \frac{d n^\uparrow}{dn} \right] = \frac{d n^\uparrow}{dn}. $$ This is just using the definition of the total derivative twice. This leads me to these three questions:

  1. Which principle/definition does equation $(1)$ derive from? Is this some functional derivative magic?

  2. Could it be that the authors mean the partial derivative of some kind instead, i.e. $$ V_{xc} = \frac{\partial}{\partial n} n\cdot\big(\varepsilon_x(n) + \varepsilon_c(n)\big) \quad ? $$ This would solve my problem.

  3. Can I use the answer to 1. to determine $V_{xc}^\sigma$, where $\sigma \in \{\uparrow, \downarrow\}$?

PS: For reference, I believe we have the following relation $$ E_{xc}[n^\uparrow,n^\downarrow] = \int d^3r n(r) (\varepsilon_x(n^\uparrow(r), n^\downarrow(r)) + \varepsilon_c(n(r))), $$ according to my textbook by Richard M Martin on Electronic structure theory.


To arrive at the exchange correlation potential starting from the exchange correlation energy, one indeed does not take a total derivative with respect to the density. The exchange correlation energy is a functional of the density, and the exchange correlation potential is given by the functional derivative of the exchange correlation energy $$E_{\rm xc} = \int_{\rm Volume} {\rm d}^3r n(\vec r)(\epsilon_{\rm x}[n(\vec r)]+\epsilon_{\rm c}[n(\vec r)])$$ ($\epsilon_{\rm x}$ and $\epsilon_{\rm c}$ are energy densities per particle, hence the above multiplication by the density $n$) with respect to $n(\vec r)$. Laxly speaking, the functional derivative will here basically correspond taking the partial derivative of the integral kernel with respect to $n$. If $n_{\uparrow} \neq n_{\downarrow}$, the exchange and correlation energy densities will also depend on the polarization $$\zeta = \frac{n_{\uparrow} - n_{\downarrow}}{n_{\uparrow} + n_{\downarrow}},$$ leading to an additional functional derivative term in the potential with respect to $\zeta(\vec r)$: $$ V_{\rm xc}[n_{\uparrow,\downarrow}(\vec r)] = \frac{\delta E_{\rm xc}}{\delta n(\vec r)} + \frac{\delta E_{\rm xc}}{\delta \zeta(\vec r)} \frac{\partial \zeta}{\partial n_{\uparrow,\downarrow} } \,\,\,{\rm with} $$

$$ \frac{\partial \zeta}{\partial n_{\uparrow,\downarrow} } = \frac{1}{n} (\pm 1 - \zeta) $$

  • $\begingroup$ Would it be possible to show what the result is in the present case, i.e. when $n^\uparrow \neq n^\downarrow$? Thanks Pete $\endgroup$ – Mikkel Rev Nov 10 '19 at 22:00
  • 1
    $\begingroup$ For the exchange part this is "trivial", since simple spin scaling applies here: Ex(nup,ndown) = 1/2 * ( Ex_spinpaired(2*nup) + Ex_spinpaired(2*ndown), and the same spin average applies to Vx as well. The spin-dependence of Ec is much more complicated, and I would refer to the following article: Perdew and Wang, Phys. Rev. B 43, 8911 (1991) [the famous PW91 parametrization of Quantum Monte Carlo simulations of the homogeneous electron gas]. $\endgroup$ – Pete Nov 10 '19 at 22:10

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.