Screening in the finite-temperature reduced Hartree-Fock model (Q2192400): Difference between revisions

The author justifies the electronic screening effect of defects in periodic crystals at finite temperature within the reduced Hartree-Fock (rHF) model, i.e., the Hartree-Fock model without exchange term. The finite temperature assumption has the effect that the electrons in the crystal behave as in a metal, which ultimately leads to a perfect screening of small impurities. The effective model describes infinitely many electrons and has the following form in absence of a defect: let \(\mathcal{R}\) be the crystal lattice, \(\Gamma\) a unit cell, and \(W_{\mathrm{nucl}}\) the \(\mathcal{R}\)-periodic potential created by the nuclei. Then the periodic rHF model for the total potential \(W\) is \begin{align*} \tag{1} \begin{cases} W = W_\mathrm{nucl} + v_\mathrm{per} F_{\epsilon_F}(W)\\ \int_\Gamma F_{\epsilon_F}(W) = N_\mathrm{el}\,. \end{cases} \end{align*} Here \(N_\mathrm{el}\) is the number of electrons per unit cell, \(v_\mathrm{per}\rho\) is the unique periodic solution of \begin{align*} \begin{cases} -\Delta(v_\mathrm{per}\rho) = \rho - \frac{1}{|\Gamma|}\int_\Gamma\rho\\ \int_\Gamma(v_\mathrm{per}\rho) = 0 \end{cases}\,, \end{align*} and \(F_{\epsilon_F}(W)=f_{\epsilon+F}(-\Delta+W)(x,x)\) denotes the potential-to-density-mapping from periodic potentials to periodic densities, i.e., the integral kernel of the operator \(f_{\epsilon+F}(-\Delta+W)\) with the Fermi--Dirac distribution \(f_{\epsilon_F}(\epsilon) = \left[1+\exp\left(\frac{\epsilon-\epsilon_F}{k_B T}\right)\right]^{-1}\) at temperature \(T\) with Boltzmann constant \(k_B\). For \(T=0\) this model was derived as a thermodynamic limit and studied in [\textit{I. Catto} et al., Ann. Inst. Henri Poincaré, Anal. Non Linéaire 18, No. 6, 687--760 (2001; Zbl 0994.35115)] and [\textit{F. Nier}, Commun. Partial Differ. Equations 18, No. 7--8, 1125--1147 (1993; Zbl 0785.35086)]. In particular, existence and uniqueness of solutions of (1) was proved in the latter work. The first main result, Theorem 1.1, states that the fixed-point iteration \(W_{n+1}=W_n+\alpha(W_\mathrm{nucl}+v_\mathrm{per} F(W_n)-W_n)\) converges for small \(\alpha\) to the unique solution of (1). The theorem and its proof are crucial for the investigation of defects later, but are also of independent interest, e.g., for the study of other, more complex, algorithms like the Anderson acceleration. The second part of the paper concerns the situation when the crystal is perturbed by a small defect and extends previous results for the zero temperature, insulating case in [\textit{É. Cancès} et al., Commun. Math. Phys. 281, No. 1, 129--177 (2008; Zbl 1157.82042)] and [\textit{É. Cancès} and \textit{M. Lewin}, Arch. Ration. Mech. Anal. 197, No. 1, 139--177 (2010; Zbl 1197.82113)], where only partial screening occurs. The author notes that the analysis of screening effects for small impurities in a metal at zero temperature, where \(V\) is expected to oscillate (Friedel oscillations), is open. For finite temperatures the model is introduced as follows: for a fixed solution \(W_\mathrm{per}\) of the periodic model (1), a Fermi level \(\epsilon_F\), and a given defect potential \(V_\mathrm{deff}\), the rHF equation for the total change in the potential \(V\) generated by the inclusion of the defect \(V_\mathrm{deff}\) reads \begin{align*} \tag{2} V = V_\mathrm{deff} + v_c G(V)\,. \end{align*} Here \((v_c\rho)(x)=(4\pi)^{-1}(\rho\ast|\cdot|^{-1})(x)\) denotes the convolution with the Coulomb potential, and \(G(V)=\left(f_{\epsilon_F}(-\Delta+W_\mathrm{per}+V)-f_{\epsilon_F}(-\Delta+W_\mathrm{per})\right)(x,x)\) denotes the renormalized potential-to-density mapping. The author notes that -- in contrast to the zero temperature case -- neither (1) nor (2) have been derived from a thermodynamic limit for the rHF model at finite temperature, but refers to [\textit{H. Chen} et al., Arch. Ration. Mech. Anal. 230, No. 2, 701--733 (2018; Zbl 1412.35279)] for a related work in a simpler (tight binding) model. The previous fixed point ansatz in absence of the defect is not well defined in this situation due to the long range of the Coulomb operator \(v_c\). The author considers the Newton iteration \(V_{n+1} = V_n + J(V_n)^{-1}(V_\mathrm{deff} + v_c G(V_n)-V_n)\) with the Jacobian \(J(V)=1-v_c G'(V)\) instead. Due to the complicated nature of \(G(V)\), the Jacobian \(J(V)\) is approximated with the help of the Thomas--Fermi potential-to-density mapping \(G_{TF}(V)=(\epsilon_{F}-V)_+^{3/2}\), which is known to describe large Coulomb systems correctly to leading order [\textit{E. H. Lieb} and \textit{B. Simon}, Adv. Math. 23, 22--116 (1977; Zbl 0938.81568)]. In this case \(J_{TF}(0)=1-v_c G_{TF}'(0)\) whose kernel in Fourier space is \((J_{TF}(0))(q)=(|q|^2-\chi_{0,TF})|q|^{-2}\). Setting \(\chi_{0,TF}=-1\) cures the long-range disease of the Coulomb potential and leads to the new (preconditioned) fixed-point iteration \begin{align*} \tag{3} V_{n+1}=V_n+\alpha\mathcal{K}(V_\mathrm{deff}+v_c G(V_n)-V_n) \end{align*} for (2) with small \(\alpha\) and \(\mathcal{K}=-\Delta(1-\Delta)^{-1}\) whose kernel decays exponentially. The second main result of the paper, Theorem 1.2, states the convergence of the iteration (3) to the unique solution of (2), and thereby the existence of equilibrium states in the presence of a defect. In particular, Theorem 1.2 admits defect potentials of the form \(V_\mathrm{deff}(x)=Q/|x|\). In this case, the theorem states in particular that, when \(Q\) is sufficiently small, then the screened potential \(V\) is in \(L^2\) and therefore decays faster than \(V_\mathrm{deff}\). The third main result, Theorem 1.3, makes this more precise and states that, if \(V_\mathrm{deff}\in v_c H_N^{-2}\) (with \(H_N^n=\{(1+|x|^2)^{N/2}f\in H^n\}\)) and if \(V_\mathrm{deff}\) is small enough in \(v_c H_1^{-2}\), then \(V(V_\mathrm{deff})\in L_N^2\). In particular, if \(V_\mathrm{deff}=Q/|x|\) and \(Q\) is small enough, then \(V(V_\mathrm{deff})\) decays faster than any polynomial which shows the total screening of the impurity at finite temperature. The author notes that it is an open problem to analyze the screening of the finite temperature rHF model in the non-perturbative regime \(Q\gg1\), where he expects that energy methods, in place of fixed-point iterations, should be applicable to prove existence of solutions to the defect equations. This approach may also shed light on the convergence rate of the fixed-point algorithms. Moreover, it is open to analyze screening in the presence of an exchange term, i.e., when the defect equation (2) is replaced by \(V=V_\mathrm{deff}+v_c G(V) + V_{\mathrm{xc}}(G(V))\) with the exchange-correlation potential \(V_{\mathrm{xc}}(\rho)\). The proofs essentially rely on a version of Banach's fixed point theorem (Theorem 3.3), which only requires the spectral radius of the Jacobian of the iteration operator to be \(<1\), and, more importantly, a detailed analysis of the potential-to-density-maps (Lemmas 4.1, 4.2, 5.1, 5.2). To that end, the author exploits the holomorphic functional calculus similarly as in [\textit{F. Nier}, Commun. Partial Differ. Equations 18, No. 7--8, 1125--1147 (1993; Zbl 0785.35086)]. Most importantly, these maps are analytic on \(L^2\) and their derivatives are non-positive operators.