Linear response time-dependent density functional theory

Transcription

1 Linear response time-dependent density functional theory Emmanuel Fromager Laboratoire de Chimie Quantique, Université de Strasbourg, France Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 1 / 12

2 Runge and Gross theorem Let us consider the time-dependent Hamiltonian Ĥ(t) = ˆT + Ŵee + N i=1 v(r i, t) and a fixed initial electronic wavefunction Ψ(t = 0). By varying the time-dependent local potential v(r, t) we obtain a map of time-dependent densities nv](r, t) = n Ψv](t) (r) where Ψv](t) is the solution to the time-dependent Schrödinger equation with potential v(r, t) and Ψv](t = 0) = Ψ(t = 0). If two potentials v(r, t) and v (r, t) differ by a time-dependent function, then Ψv](t) and Ψv ](t) will have the same density at any time. According to the Runge and Gross theorem, the map of time-dependent densities can be inverted (up to an additive time-dependent function in the potential). This is nothing but the extension of the first Hohenberg Kohn theorem to the time-dependent regime. In the following, we will consider that v(r, t) = v ne(r) + V(r, t) where V(r, t) is a local time-dependent perturbation. We will also assume that the system is in the ground state at time t = 0. E. Runge and E. K. U. Gross, Phys. Rev. Lett. 52, 997 (1984). Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 2 / 12

3 Time-dependent KS equations Like in the static case, the non-interacting N-electron time-dependent Schrödinger equation can be simplified into one-electron equations. For example, in the case of two electrons, the exact solution can be written as where r + v(r, t) Φ(r 1, r 2, t) = ϕ(r 1, t)ϕ(r 2, t) ] ϕ(r, t) = i d ϕ(r, t). dt The basic idea in standard TD-DFT is to make a mapping in the time-dependent regime between the real interacting (and therefore difficult to describe) system with Hamiltonian Ĥ(t) and a non-interacting system, in complete analogy with KS-DFT. ] Therefore, by solving ˆT + N i=1 v KS (r, t) Φ KS (t) = i d dt ΦKS (t), where Φ KS (t) is a time-dependent Slater determinant, we should be able, in principle, to reproduce the exact time-dependent electron density n(r, t) of the molecule. Therefore, for two electrons, we should have n(r, t) = 2 ϕ KS (r, t) 2. Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 3 / 12

4 Linear response regime and adiabatic approximation Consequently, if a time-dependent perturbation is applied to the molecule, the linear (and higher-order) response of the KS and real densities should be the same In particular, the poles of the KS density (which correspond to the poles of the KS orbitals), should be the exact excitation energies of true molecule (!). The exact time-dependent KS potential is in general unknown and difficult to model (memory effects). In the standard adiabatic approximation, it is simplified as follows, v KS (r, t) v ne(r) + V(r, t) + δehxcn Φ KS (t)], where E Hxcn] is the (time-independent) ground-state Hxc functional. Note that, within the adiabatic approximation, the KS potential is local in time (no memory effects are taken into account). Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 4 / 12

5 Therefore, the (approximate) time-dependent KS equation to be solved for two electrons is ĥ ϕ KS] (t) ϕ KS (t) = i d ϕ KS (t) dt where ĥ ϕ KS] (t) 1 δe Hxc 2 ϕ KS (r, t) 2] 2 2 r + v ne(r) + V(r, t) + Note that, in contrast to conventional wavefunction-based response theory, we have to solve a self-consistent time-dependent equation (time-dependent density functional perturbation theory). In the following, the time-dependent local potential perturbation will be written as V(r, t) = 2ɛ cos(ωt)v(r). In the presence of a uniform electric field along the z axis, V(r) = z. Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 5 / 12

6 Let us first consider the KS system in the absence of perturbation (ɛ = 0). The time-dependent KS orbital simply reads ϕ KS (r, t) = e iε 0t ϕ 0(r) where ϕ 0(r) is the ground-state KS orbital with energy ε 0: ( r + v ne(r) + δehxc 2 ϕ0(r) 2] ] ) ϕ 0(r) = ε 0ϕ 0(r) }{{} ĥ Consequently, in TD-DFT, the unperturbed Hamiltonian is the KS Hamiltonian ĥ. The perturbation in the KS system will be V(r, t) + δe Hxc 2 ϕ KS (r, t) 2] δehxc 2 ϕ0(r) 2]. Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 6 / 12

7 In the following, we will use the ground (occupied) and excited (virtuals) KS orbitals {ϕ i (r)} i=0,1,2,... with energies {ε i } i=0,1,2,... as a basis for expanding the time-dependent KS orbital in the presence of the perturbation. Note that ĥϕ i (r) = ε i ϕ i (r). By analogy with exact response theory, we therefore obtain ϕ KS (t) = i e iε i t C i (t) ϕ i. The derivation of the linear response TD-DFT equation (known as Casida equation) requires the expansion of the Hxc potential through first order in the perturbation strength ɛ. By definition, the variation of the Hxc potential around a given density n 0(r) equals, through first order in δn(r ), δe Hxc n 0 + δn] δehxc n0] = dr δ 2 E Hxcn 0] δn(r ) δn(r ) }{{} kernel! f Hxc(r, r) Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 7 / 12

8 In our case, n 0(r) = 2 ϕ 0(r) 2 and, by using C i (t) = δ i0, we obtain δn(r ) = 2 ϕ KS (r, t) 2 2 ϕ 0(r ) 2 = 2 ij = 2ɛ i e i(ε i ε j )t C i (t)c j (t)ϕ i (r )ϕ j (r ) 2 ϕ 0(r ) 2 ( e i ε i t where ε i = ε i ε 0. ) C i (t) ɛ ϕ i (r )ϕ 0 (r ) + e i ε i t Ci (t) ɛ ϕ 0(r )ϕ i (r ) Note that the two-electron KS wavefunction reads through first order ϕ KS (r 1, t)ϕ KS (r 2, t) = e 2iε 0t ϕ 0(r 1)ϕ 0(r 2) +ɛ i +... e i(ε i +ε 0 )t C i (t) ɛ ( ϕ 0(r 1)ϕ i (r 2) + ϕ i (r 1)ϕ 0(r 2) ) }{{} single electron excitation! Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 8 / 12

9 Finally, the (local) perturbation in the KS system equals through first order 2ɛ cos(ωt)v(r) +2ɛ i +2ɛ i e i ε i t e i ε i t C i (t) ɛ dr f Hxc(r, r)ϕ i (r )ϕ 0 (r ) Ci (t) ɛ dr f Hxc(r, r)ϕ 0(r )ϕ i (r ) By analogy with exact response theory, we obtain d dt C j (t) ɛ ] = iv j0 e i(ω+ ε j )t + e i(ω ε j )t 2iɛ i 2iɛ i e i(ε i ε j )t e i(ε i +ε j 2ε 0 )t C i (t) ɛ j0 f Hxc 0i Ci (t) ɛ ji f Hxc 00 where ij f Hxc kl = dr dr ϕ i (r)ϕ j (r )f Hxc(r, r)ϕ k (r)ϕ l (r ). Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 9 / 12

10 When the Hxc kernel is neglected, the TD-DFT linear response equation reduces to the exact linear response equation for the KS-DFT system. In this case, we obtain (approximate) excitation energies which are simply equal to the KS excitation energies ε j ε 0 with j > 0. In a sense, KS excitation energies can be viewed as zeroth order excitation energies if the Hxc kernel is considered as a perturbation. We immediately see that, within the adiabatic approximation, multiple electron excitations (double excitations in particular) cannot be described. They are simply absent from the TD-DFT spectrum. The Hxc kernel couples the time-dependent linear response coefficients. Consequently, there is no simple analytical solution to the problem. Casida has shown that a linear response equation can be written in matrix form for TD-DFT, exactly like in exact response theory. An important difference though is that the hessian matrix is not diagonal anymore. Moreover, within the adiabatic approximation, it is frequency-independent, which explains once again why multiple excitations cannot be described. M. Casida in Recent Advances in Density Functional Methods, edited by D. P. Chong (World Scientific, Singapore, 1995). Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 10 / 12

11 The only way to describe multiple excitations in TD-DFT is to use a frequency-dependent kernel, thus leading to a frequency-dependent hessian matrix. This would basically consists in introducing memory effects into the time-dependent KS potential. Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 11 / 12

12 Excitation energy (atomic units) H 2 molecule (2 1 g, cc pvqz) exact (single excitation) exact (double excitation) approximate TD DFT Interatomic distance (atomic units) Emmanuel Fromager (UdS) Cours RFCT, Strasbourg, France 12 / 12