Lecture9: Polarization th. II
Transcript
Lecture9: Polarization th. II
Macroscopic Polarization from Electronic Wavefunctions Raffaele Resta Dipartimento di Fisica Teorica, Università di Trieste <[email protected]> Sanibel Symposium 1999 The dipole of a microscopic system (atom, molecule....) Neglecting (irrelevant) spin variables: n(r) = N Z dr2dr3 . . . drN |Ψ(r, r2, · · · rN )|2; d= Z dr r n(r). Alternatively: d = hΨ|R̂|Ψi; R̂ ≡ N X i=1 R. Resta ri (a one-body multiplicative operator). – Sanibel Symposium 1999 – 1 The polarization of a macroscopic solid (The Feynman Lectures in Physics, Vol.2, Ch.11) R. Resta – Sanibel Symposium 1999 – 2 Incorrect (and unusable) definitions • Kittel, Introduction to Solid State Physics: A ferroelectric crystal exhibits an electric dipole moment even in the absence of an external electric field. In the ferroelectric state the center of positive charge does not coincide with the center of negative charge. • Ashcroft & Mermin, Solid State Physics: Crystal whose natural primitive cells have a nonvanishing dipole moment p0 are called pyroelectric. • Lines & Glass, Principles and Applications of Ferroelectrics and Related Materials (1977): If and when good electron-density maps become available for ferroelectrics, expressing charge density ρ(r) as a function of position vector r throughout the unit cell, more quantitative estimates of spontaneous polarization might be envisaged as 1 rρ(r) dr. Ps = V V Z R. Resta – Sanibel Symposium 1999 – (6.1.19) 3 Outline • What makes the polarization problem difficult for theorists? • How is macroscopic polarization measured? • Theory of polarization (& calculations) after 1993: • A novel view: the ultimate solution: Very general: either periodic or disordered either independent–electron or correlated. • Relationship to the (discretized) Berry phase. • Some comments about polarization within DFT. • Latest development: wavefunction localization in insulators. R. Resta – Sanibel Symposium 1999 – 4 Back to basics: the QM position operator • It is the simple multiplicative operator which maps: ψ(x) → x ψ(x). • The expectation value is: hxi = dx x |ψ(x)|2. R R. Resta – Sanibel Symposium 1999 – 5 Fundamental issues • What goes wrong? • What is the correct definition of bulk macroscopic polarization? • How to compute the polarization from the wavefunction? R. Resta – Sanibel Symposium 1999 – 6 The Hilbert space of condensed matter BvK boundary conditions: ψ(x + L) = ψ(x) (L large with respect to atomic dimensions). The multiplicative operator x is not a legitimate operator in this Hilbert space. x ψ(x) does not belong to the Hilbert space whenever ψ(x) does. R. Resta – Sanibel Symposium 1999 – 7 Reconsidering some basic facts • The “absolute” macroscopic polarization of a solid has never been measured (as a bulk quantity). • The experimental quantities are invariably derivatives (piezolelectricity, pyroelectricity, dielectric permittivity.....) or finite differences (ferroelectricity measured through polarization reversal). R. Resta – Sanibel Symposium 1999 – 8 Basic facts cont’d • These derivatives/differences are obtained by measuring currents. • The current transported by a Bloch orbital (within BvK) is a well defined quantity. • The current is essentially related to the phase of the wavefunction. R. Resta – Sanibel Symposium 1999 – 9 The modern theory (since 1993) Macroscopic polarization is a Berry phase of the Bloch orbitals (in the uncorrelated case). In general, a Berry phase is an observable which cannot be cast as the expectation value of any operator. Instead it is a gauge–invariant phase of the wavefunction. Basic review: Rev. Mod. Phys. 66, 899 (1994). Nontechnical review: Europhys. News 28, 18 (1997). R. Resta – Sanibel Symposium 1999 – 10 Berry phase: Actual calculations • DFT-LDA, HF, tight-binding, Hubbard model. • PW (norm-conserving pseudo, ultrasoft pseudo), FLAPW, LCAO (contracted Gaussians). • Properties: piezoelectricity, dynamical charges, ferroelectricity. • Permittivity ε∞ not accessible. R. Resta – Sanibel Symposium 1999 – 11 A precursor: “Elettrone in brodo” (Selloni, Carnevali, Car, Parrinello, PRL 1987) A lone electron in a large periodic cell (rari nantes in gurgite vasto) R. Resta – Sanibel Symposium 1999 – 12 Position defined modulo L: The wrong approach: Any variation on this theme. ξ(x) = x hxi = R. Resta Z L 0 mod L dx ξ(x)|ψ(x)|2 – Sanibel Symposium 1999 – 13 Position defined modulo L 0 L hxi L hxi = Im log 2π Z L 0 dx e i 2π Lx |ψ(x)|2 Example: |ψ(x)|2 = ∞ X δ(x − x0 − mL) m=−∞ L i 2π Im log e L x0 ≡ x0 (mod L) hxi = 2π R. Resta – Sanibel Symposium 1999 – 14 Q: Why the “electron–in–broth” formula is the correct one? A: Because it provides the macroscopic current (for L → ∞): d hx(t)i = hj(t)i dt R. Resta – Sanibel Symposium 1999 – 15 N electrons in a box of size L Naı̈f generalization: |ψ(x)|2 → n(x): Z L L i 2π Im log dx e L x n(x) hXi = 0 2π Wrong: The Fourier coefficients of the periodic density do not carry any information about polarization. R. Resta – Sanibel Symposium 1999 – 16 Many–body BvK boundary conditions Ψ(x1, . . . , xi, . . . , xN ) = Ψ(x1, . . . , xi +L, . . . , xN ). (periodic on each electronic variable separately). Thermodynamic limit: N → ∞, L → ∞, N/L = n0 constant. The operator X̂ ≡ R. Resta PN i=1 xi becomes meaningless. – Sanibel Symposium 1999 – 17 The ultimate solution [PRL 80, 1800 (1998)] L i 2π hXi = Im log hΨ|e L X̂ |Ψi 2π e i 2π Im log hΨ|e L X̂ |Ψi Pel = lim L→∞ 2π A tricky limit. hXi is defined modulo L. A Berry phase in disguise. R. Resta – Sanibel Symposium 1999 – 18 Within poor man’s first quantization: L hXi = Im log zN 2π zN = i 2π L X̂ e Z L 0 PN 2π dx1dx2 . . . dxN |Ψ(x1, x2, . . . xN )|2 ei L i=1 xi . is a genuine many–body (multiplicative) operator. R. Resta – Sanibel Symposium 1999 – 19 Crystalline case, independent electrons Born-von-Kàrmàn boundary conditions; M crystal cells of size a; total length L = M a. Allowed Bloch vectors in the reciprocal cell [0,2π/a): 2π 2π s = s, qs = a M L R. Resta s = 0, 1, . . . , M −1 – Sanibel Symposium 1999 – 20 The one-particle orbitals may be chosen in the Bloch form: ψqs,m(x + τ ) = eiqsτ ψqs,m(x) N/M occupied bands: | Ψi = A N/M −1 Y MY m=1 s=0 zN = i 2π hΨ|e L X̂ |Ψi = MY −1 s=0 Sm,m′ (qs, qs+1) = Z L 0 dx ψqs,m. det S(qs, qs+1) −i 2π ∗ ψqs,m(x)e L xψqs+1,m′ (x) −→ King-Smith & Vanderbilt discrete Berry phase. R. Resta – Sanibel Symposium 1999 – 21 Berry phase via an expectation value: Is it “legally accurate” or “improper”? γ = Im log Z L 0 dx1 . . . dxN |Ψ(x1, x2, . . . xN i 2π 2 )| e L X̂ . Berry phase expressed in terms of the modulus of the wavefunction. The modulus of the many–body wavefunction contains interference terms due to the relative phases of the Bloch orbitals: these are “extracted” in a gauge–invariant manner. R. Resta – Sanibel Symposium 1999 – 22 What about DFT? The HK theorem implicitly assumes a finite system of N electrons, with a square–integrable (L2) wavefunction Ψ(r1, r2, · · · rN ). The same HK proof carries over with no change assuming instead BvK boundary conditions for the finite system of N electrons. Then one can proceed as in KS. But...... R. Resta – Sanibel Symposium 1999 – 23 Using L2 boundary conditions the dipole of the system is a function of the density. Caveat: Of the whole density, surface and bulk. Using instead BvK boundary conditions: (i) There is no surface by construction; (ii) Macroscopic polarization has nothing to do with the density of the polarized solid: polarization = phase of Ψ; density = modulus of Ψ. Ergo...... R. Resta – Sanibel Symposium 1999 – 24 The dipole of a finite (possibly large) correlated system coincides with the dipole of the corresponding noninteracting KS system. The macroscopic polarization of a correlated solid does not coincide with the polarization of the corresponding noninteracting KS system (both evaluated within BvK). R. Resta – Sanibel Symposium 1999 – 25 Hot topic these days: “Polarization DFT” • “Ultranonlocality” of the KS exchange–correlation potential? • Solution of the “permittivity problem”? • Recipes for practical calculations in periodic solids? • Related to Current Density Functional Theory?. Several papers continue to appear.... R. Resta – Sanibel Symposium 1999 – 26 Latest developments: Localization 0 L hxi What about hx2i − hxi2 ? Marzari & Vanderbilt localization (alias Boys localization). R. Resta & S. Sorella, PRL 82, 370 (1999). R. Resta – Sanibel Symposium 1999 – 27 z= Z L 0 dx e i 2π Lx |ψ(x)|2 0 ≤ |z| ≤ 1 , Complete localization: |ψ(x)|2 = ∞ X 2π δ(x − x0 − mL) −→ z = ei L x0 m=−∞ Complete delocalization: 1 |ψ(x)| = L 2 hxi = −→ z=0 L Im log z 2π 2 L hx2i − hxi2 = − log |z|2 2π R. Resta – Sanibel Symposium 1999 – 28