Density of States in the Magnetic Ground State of the Friedel-Anderson Impurity 9 0 0 2 Gerd Bergmann n Department of Physics a J University of Southern California 1 3 Los Angeles, California 90089-0484 e-mail: [email protected] ] l l a February 3, 2009 h - s e m Abstract . t a By applying a magnetic field whose Zeeman energy exceeds the Kondo energy by m an order of magnitude the ground state of the Friedel-Anderson impurity is a magnetic - d state. In recent years the author introduced the Friedel Artificially Inserted Resonance n (FAIR) method to investigate impurity properties. Within this FAIR approach the o c magnetic ground state is derived. Its full excitation spectrum and the composition of [ the excitations is calculated and numerically evaluated. From the excitation spectrum 1 the electron density of states is calculated. Majority and minority d-resonances are v obtained. The width of the resonances is about twice as wide as the mean field theory 3 predicts. This broadening is due to the fact that any change of the occupation of the 3 0 d-state in one spin band changes the eigenstates in the opposite spin band and causes 0 transitions in both spin bands. This broadening reduces the height of the resonance . 2 curve and therefore the density of states by a factor of two. This yields an intuitive 0 understanding for a previous result of the FAIR approach that the critical value of the 9 0 Coulomb interaction for the formation of a magnetic moment is twice as large as the : mean field theory predicts. v i PACS: 75.20.Hr, 71.23.An, 71.27.+a X r a 1 1 Introduction The properties of magnetic impurities in a metal is one of the most intensively studied prob- lems in solid state physics. The work of Friedel [1] and Anderson [2] laid the foundation to understand why some transition-metal impurities form a local magnetic moment while oth- ers don’t. Kondo [3] showed that multiple scattering of conduction electrons by a magnetic impurity yields a divergent contribution to the resistance in perturbation theory. Yoshida [4] introduced the concept that the (spin 1/2) magnetic impurity forms a singlet state with the conduction electrons and is non-magnetic at zero temperature. These new insights stim- ulated a large body of theoretical and experimental work (see for example [4], [5], [6], [7], [8], [9], [10], [11], [12], [13]). The majority of experimental and theoretical work has focussed on the singlet Kondo ground state. However, the ”magnetic state” of the impurity is of equal or even greater importance because magnetic impurities are always present, including in micro-chips and nanostructures, and influence the thermodynamic and transport properties of the hosts. Since many experiments and almost all technical applications are not performed at low temperatures the magnetic impurities are generally far above their Kondo temperature T K and show their full magnetic behavior. The theoretical investigation of the magnetic state has been explored in much less detail than the Kondo ground state for spin 1/2 impurities. In many cases the Kondo temperature is very low, in the range of liquid helium tem- perature. In this case the impurity is in the magnetic state at relatively low temperature. (The word impurity is in this paper reserved to impurities which possess - at sufficiently high temperature - a magnetic moment). When the temperature is several times the Kondo temperature one is sufficiently above T to destroy the Kondo ground state. On the other K hand one may expect that the properties of the magnetic state are not yet influenced by the thermal excitations due to the finite temperature. Therefore a number of theoretical investigations treat the magnetic state at zero temperature, i.e. as a magnetic ground state. This approach is probably justified but it leaves the work always vulnerable to the criticism that there is no magnetic moment at zero temperature. Therefore in this paper I prefer to use the effect of a magnetic field on the Kondo state. A magnetic field which is an order of magnitude larger than k T /µ (µ =Bohr magneton) B K B B destroys the Kondo singlet state as well and yields the magnetic state. Its side effects are that it changes the energy of the d-states by µ B and shifts the conduction bands by B ± µ B. The latter yields the Pauli susceptibility but has otherwise only a negligible effect B ± on the interaction between the impurity and the conduction electrons because the Fermi level for spin-up and down electrons readjusts to the same height (as before). Friedel [1] and Anderson [2] derived a criterion for the instability of the paramagnetic state, i.e. the formation of a magnetic moment: Take the density of states N (ε ) of the d F d-resonance at the Fermi energy (in the paramagnetic state) and multiply it by the Coulomb repulsion energy U. If the product N U > 1 then a magnetic moment is formed. Within d 2 mean field theory the d-density of states is given by a Lorentz function 1 Γ mf N (ε) = d,σ π(ε E )2 +Γ2 − d,σ mf where E is an effective energy of the d-electrons in the spin-up or down state, E = d,σ d,σ E + U n while n is the average occupation of the d-electron with the opposite d d,−σ d,−σ h i h i spin) and the resonance width Γ is given in mean field theory by mf Γ = π V 2N mf sd s | | Here V is the s-d-hopping matrix element between a conduction electron and the d-state sd at the impurity and N is the density of states of the conduction electrons. In the mean field s † † theory an occupied d electron state can only make transitions into c -states. (Throughout ↑ k↑ this paper I express electron states by their creation operators). It is well known that themean field theoryhas anumber ofshortcomings. During thelast few years the group of the author has developed a new approach to the impurity problem, in particular the Friedel-Anderson and the Kondo impurity. In this FAIR method a Friedel state is Artificially built from each conduction band and Inserted as a Resonance state into the conduction or s-band of spin-up and spin-down electrons. In the appendix a short review of the FAIR solution for the Friedel impurity is sketched. The FAIR solution for the magnetic state yields a considerably lower energy for the ”magnetic ground state” and requires a much larger critical Coulomb interaction to form a magnetic state. This is of some practical importance since the mean field approximation is used in a number of numerical spin-density functional theory calculations for the magnetic moment of impurities in an (s,p) metal host [14], [15], [16], [17], [18]. In addition to the size of the magnetic moment one would like to know the density of states in the magnetic state. The answer of the mean field theory has been discussed above. But there have been a number of suggestions that the d-resonance is broader than the mean field suggests (see for example Logan [19]). The mean field theory decouples the spin-up d-electron from the spin-down d-electron, but in reality the d-electrons are coupled through † the Coulomb energy. A transition in the d electron state changes the energy and the state ↑ of the d† electron as well. Therefore it has been suggested in the past that the d-resonances ↓ in the Friedel-Anderson impurity are larger than the mean field theory predicts. A wider d-resonance in the Friedel-Anderson impurity together with the condition N U > 1 would d require a larger Coulomb energy to formamagnetic moment. In thisconnection theprevious result of the author that the FAIR solution requires a (two times) larger Coulomb energy to form a magnetic moment would find a simple physical interpretation. Itisthegoalofthispapertocalculatethedensityofstatesofthe”magneticgroundstate” in the FAIR solution and compare it with the mean field density of states. In section II the theoretical background of the magnetic state of the Friedel-Anderson impurity is sketched. In section III electrons and holes are introduced into the magnetic ground state. Their interactions and excitation energies are derived. In section IV the results of the numerical 3 calculations are presented. Finally in section V and VI the results are discussed together with the conclusion. In the appendix A the basic idea of the FAIR method is sketched. 2 Theoretical Background The simplified Hamiltonian for a magnetic impurity is generally described by the Friedel- Anderson (FA) Hamiltonian N−1 N−1 H = ε c† c +E d†d + Vsd[d†c +c† d ] +Un n (1) FA σ ν νσ νσ d σ σ ν σ νσ νσ σ d↑ d↓ ( ) ν=0 ν=0 P X X Here the operators c† represent s-electrons, i.e. the conduction band. νσ 2.1 The FAIR method In the Friedel-Anderson Hamiltonian in equ. (1) the d-state for each spin interacts with every electron in the conduction band. Imagine how much easier the task would be if the d-electron would interact only with a single electron state (in each spin band). All other conduction band states would represent just a background or quasi-vacuum. This is the FAIR approach. During the last few years the author introduced such a solution to the Friedel-Anderson impurity problem in which only four electron states, the spin-up and spin-down d-states d† ↑ † † † and d and two FAIR states, a and b interact through the Coulomb and s-d-hopping ↓ 0↑ 0↓ potential. These states a† and b† are composed of the spin-up and spin-down conduction 0↑ 0↓ band states. They are the Friedel Artificially Inserted Resonance states or FAIR states. The interaction of the remaining conduction electron states with the d-states is insignificant; they just yield a background. This yields very good ground-state properties. The FAIR states are composed of the corresponding conduction bands a = N−1ανc b = N−1βνc 0↑ ν=0 0 ν↑ 0↓ ν=0 0 ν↓ The remaining (N 1) states inPeach spin band arePconstructed orthogonal to the corre- − sponding FAIR state, orthonormal to each other and sub-diagonal with respect to the band energy Hamiltonian N−1 H = ε c† c 0 ν νσ νσ ν=0 X This yields new bases for the conduction bands a† and b† with 1 i (N 1). i,↑ i,↓ ≤ ≤ − These new bases are uniquely determined by the ntwo FoAIR stnates.o 4 Within the new bases the FA-Hamiltonian (1) can be expressed as H = H′ +H′ FA 0 1 with H′ = H′ +H′ +Un n 0 0,↑ 0,↓ d↑ d↓ H′ = H′ +H′ 1 1,↑ 1,↓ where N−1 H′ = E(a)a† a +E(a)a† a +E d†d +V(a)sd a† d +d†a (2) 0,↑ i i,↑ i,↑ 0 0,↑ 0,↑ d ↑ ↑ 0 0,↑ ↑ ↑ 0,↑ Xi=1 h i N−1 N−1 H′ = V(a)fr a† a +a† a + V(a)sd d†a +a† d (3) 1,↑ i 0,↑ i,↑ i,↑ 0,↑ i ↑ i,↑ i,↑ ↑ Xi=1 h i Xi=1 h i and the spin-down Hamiltonians are obtained by replacing by and the a†-states by † ↑ ↓ i b -states. i 2.1.1 Nest states The Hamiltonian H′ is diagonal in the band states a† and b† for 0 < i < N 1. The 0 i,↑ i,↓ − only interaction takes place between the states a† , d , b and d . I call these states the 0,↑ ↑ 0,↓ ↓ nest states. The ground state of the Hamiltonian H′ is straight forward. It consists of the 0 coupled state between the nest states and a partially occupied spin-up and down band. I occupy each spin component with N/2 electrons, putting n = N/2 1 electrons into each − conduction band states and one spin-up and one spin-down electron into the nest. This yields the magnetic ground state as described in equ. (4). † † † † † † † † Ψ = A a b +A a d +A d b +A d d 0 0 (4) MS a,b 0↑ 0↓ a,d 0↑ ↓ d,b ↑ 0↓ d,d ↑ ↓ | a,↑ b,↓i h i where 0 0 = n−1a† n−1b† Φ represents a kind of quasi-vacuum (n = N/2). | a,↑ b,↓i j=1 j↑ j=1 j↓| 0i The calculation of the coefficients A ,.. yields a secular Hamiltonian Hnst which I call Q Q a,b 1/1 the nest-Hamiltonian and which has the form E(a) +E(b) Vsd Vsd 0 0 0 b a Vsd E(a) +E 0 Vsd Hnst =  b 0 d a  1/1 Vsd 0 E +E(b) Vsd  a d 0 b   0 Vsd Vsd 2E +U   a b d    Heretheabbreviationsareused: Vsd = V(a)sd andVsd = V(b)sd (seeequ.(2)). Thesuperscript a 0 b 0 nst stands for nest and the subscript 1/1 gives the number of nest electrons in the spin-up and spin-down state. (The energy of the occupied band states is not included. It yields 5 the same contribution to each component). Hnst has four eigenvalues and eigenstates. The 1/1 lowest eigenvalue yields the ground state. The components of the ground state, nest plus the band states, are shown in Fig.1. The first order correction to the energy, i.e. the expectation value of H′ is zero. But in 1 addition the second order perturbation of H′ is extremely small. This is demonstrated in 1 appendix B. It may appear remarkable that the neglect of the interactions between the d-electron and all the band states a† and b† yields a realistic ground state. But it is not unheard off j,↑ j,↓ that onecan obtainnaneoxcellennt grooundstate while neglecting a major part of the interaction in the system. The BCS theory is a good example because it only includes the electron- phononinteractionbetweenCooperpairsoftime-reversedelectrons. Theinteractionbetween all the other electrons is neglected although their number is much larger. One major part of the numerical calculation is, of course, the optimization of the two FAIR states a† and b† so that the expectation value of the energy E = Ψ H′ Ψ 0,↑ 0,↓ 00 h MS| 0| MSi of the Hamiltonian H′ has a minimum. The optimization procedure is described at length 0 in previous papers [20], [21], [22] and is taken for granted in this paper and will not be described here. (The FAIR states are rotated in Hilbert space). Since we don’t count the FAIR states any more as band states the number of band states is reduced by one and their energy is slightly shifted (by less than the original energy spacing). The band states enter in the energy E only through the kinetic (band) energy of the occupied band states. In a 00 way they just prepare the nest for the states a† ,d ,b ,d . 0,↑ ↑ 0,↓ ↓ h i + + + = (a,b) (a,d) (d,b) (d,d) Fig.1: The composition of the magnetic state Ψ in the nest is shown. MS It consists of four Slater states. Each Slater state has a half-full spin-up and down band, two FAIR states (circles in within the bands) and two d-states (circles on the left and right of the band). Full black circles represent occupied states and light grey represent empty states. The band at the right with the half-filled circles symbolizes the magnetic solution with four Slater states. In the numerical calculation we will present the results for two examples with the pa- rameters U = 1.0, E = 0.5 and V0 2 = 0.05 and V0 2 = 0.025. The smaller value of d − | sd| | sd| the s-d-matrix element permits a better fit of the resulting resonance curve with a Lorentz curve since the effect of the finite width of the band is smaller. 6 2.1.2 Self-consistent perturbation In the construction of the magnetic ground state Ψ the Hamiltonian H′ has been com- MS 1 pletely neglected. Below we will derive the excitation energies by introducing an additional electron (hole) into an empty (occupied) states. For this calculation it is important to know whether the empty state is really empty or whether transitions from the ground state into the state due to H′ have partially occupied this state. (This problem is well known from 1 the calculation of the electron-phonon mass enhancement. In the calculation of the electron- phonon self-energy one injects an electron into an ”empty state” k above the Fermi energy. The transitions of this electron via the electron-phonon interaction into other empty states k′ contribute to the self-energy Σ. However, the state k was not really empty because tran- sitions from the ground state into k already created a finite occupation of k. One has to correct the self-energy due to these processes). In the appendix I show that transitions from the ground state into empty band states (due to H′) are practically zero. The interference between transitions from the d-state and 1 fromthecorresponding FAIRstatealmost perfectlycanceleachother. Thetotalweight inall perturbation states is only of the order of 10−4 and can be completely neglected. Therefore the band states are either completely empty or fully occupied. 3 Calculation of Excitations 3.1 Injection of an electron In Fig.2 a spin-up electron is injected into one of the empty states a† of the a† -band. j↑ ↑ This yields the Slater states (α). The Slater state (β) is obtained by injectingnan oelectron into the nest, either into the state a† or d†. 0↑ ↑ a b g d Fig.2: An electron has been injected into the spin-up band and the spin up nest. This induces electron or hole transitions in the spin-down band. The resulting Slater states are shown as (γ) and (δ). (Each band with half circles consists of two Slater states.) 7 The injection into a† or d† yields 0↑ ↑ a† Ψ = A a†d†b† +A a†d†d† 0 0 for a† 0↑ MS d,b 0 ↑ 0↓ d,d 0 ↑ ↓ | a,↑ b,↓i 0 (5) d†Ψ = (cid:16) A a† d†b† +A a† d(cid:17)†d† 0 0 for d† ↑ MS − a,b 0↑ ↑ 0↓ a,d 0↑ ↑ ↓ | a,↑ b,↓i ↑ (cid:16) (cid:17) Both final states yield a double occupancy of the spin up nest states. Furthermore these states are not eigenstates of the nest. With respect to the (basis) states a†d†b† 0 0 and 0 ↑ 0↓| a,↑ b,↓i a†d†d† 0 0 the nest Hamiltonian takes the form 0 ↑ ↓| a,↑ b,↓i E +E(a) +E(b) Vsd Hnst = d 0 0 b (6) 2/1  (cid:16) Vsd(cid:17) E +E(a) +E +U  b d 0 d  (cid:16) (cid:17)  (1) (2) By diagonalization one obtains the eigenstates Ψ and Ψ with 2/1 2/1 Ψ(α) = a† d† B(α)b† +B(α)d† 0 0 (7) 2/1 0↑ ↑ b↓ 0↓ d↓ ↓ | a,↑ b,↓i (cid:16) (cid:17) The states (α) and (β) in Fig.2 are the initial states which one obtains through injection of a spin-up electron into the ground state. Due to the perturbation Hamiltonian H′ these 1 states interact with each other and (β) interacts with the states (γ) and (δ). In table I the possible states which can be obtained through the injection of a spin-up electron plus linear coupling through H′ are collected. These states are (α) the two-electron 1 nest ground state plus one electron, (β) a nest with two spin-up and one spin-down electron, (γ) a full spin-up and empty spin-down nest plus one spin down electron and (δ) a full spin- up and full spin-down nest and one spin down hole. In table I these states, their number and their energies are listed. Ψ number energy f † (a) a Ψ N/2 E j↑ MS j Ψ(α) 2,α = 1,2 E +E(a) +Eα E 2/1 d 0 2/1 − 00 a† d†b† 0 0 N/2 Ed+E(a) +E(b) E 0↑ ↑ j↓| a,↑ b,↓i 0 j − 00 (a) (b) 2E +U +E +E b a† d†b† d† 0 0 N/2 1 d 0 0 k↓ 0↑ ↑ 0↓ ↓| a,↑ b,↓i − E(b) E − k − 00 Table I: This table describes the states of Fig.2 which are generated by the injection of one spin-up electron into the magnetic ground state and transition from the resulting states through H′. The state Ψ is 1 MS (α) given by equ. (4) and Ψ is given by (7). The 2/1 energy is measured from the ground-state energy E . 00 Fig.2 and table I show all the spin-up electron excitations which interact linearly in H′. 1 This is a total of 3N +1 states. It is straight forward to construct the secular matrix (i.e. 2 (cid:0) (cid:1) 8 the excitation Hamiltonian Hxct) between the excitations in table I. One may put the two nest states Ψ(1) and Ψ(2) at the positions one and two, followed by the 3N 1 additional 2/1 2/1 2 − single particle excitations. We denote these 3N +1 states as ϕ . The diagonal of the 2 ν (cid:0) (cid:1) Hamiltonian is given by the energies in table I. The off-diagonal elements of Hxct are the (cid:0) (cid:1) matrix elements of H′ between the states (α,β,γ,δ) in Fig.2. The single particle excitations 1 interact only with the first two nest states through H′ but not among each other. (In 1 appendix B the corresponding matrix elements are shown in table III for similar transitions from the ground state.) This Hamiltonian Hxct is diagonalized and yields a set of 3N +1 new eigenstates ψ 2 µ with eigenenergies Exct. The components of the eigenstates ψ in terms of ϕ are given as µ (cid:0)µ (cid:1) ν columns ψν . The νth row of the matrix ψν yields the amplitude of our νth original state µ µ † ϕ in terms of the new eigenstates ψ . If this νth original state is, for example, a Ψ then ν (cid:0) (cid:1) µ (cid:0) (cid:1) j↑ MS it can be expressed in the new eigenstates as ϕ = a† Ψ = ψνψ ν j↑ MS µ µ µ P Its density of states is then N (ε) = ψν 2δ ε Exct ν µ µ − µ P (cid:12) (cid:12) (cid:0) (cid:1) Since electron injection creates only the s(cid:12)tat(cid:12)es (α) and (β) one obtains the full (spin-up) excitation spectrum by summing over these 1N +2 states. The weight of states a† Ψ 2 j↑ MS is one, however, the weight of a† Ψ is only A 2 + A 2 since 0↑ MS (cid:0)| d,b| (cid:1) | d,d| a† A a† b† +A a† d† +A d†b† +A d†d† 0 0 0↑ a,b 0↑ 0↓ a,d 0↑ ↓ d,b ↑ 0↓ d,d ↑ ↓ | a,↑ b,↓i h i = A a† d†b† +A a† d†d† 0 0 d,b 0↑ ↑ 0↓ d,d 0↑ ↑ ↓ | a,↑ b,↓i h i A similar result is found for the weight of d†Ψ which is A 2 + A 2. ↑ MS | a,b| | a,d| Since a† Ψ and d†Ψ are not eigenstatet of H′ they represent a combination of the 0↑ MS ↑ MS 0 (α) two eigenstates Ψ . In Fig.3 is sketched what happens when an electron is injected into 2/1 either the state a† or d†. The electron injection yields a superposition of the two eigenstates 0↑ ↑ (α) Ψ . From these states the electron can make a transition into any of the (γ) states via 2/1 H′. The two transition amplitudes interfere in this transition. This interference has to 1 be included in the calculation of the spectral weight density of a† Ψ and d†Ψ (which 0↑ MS ↑ MS requires just the scalar products between a† Ψ and Ψ(α) (or d†Ψ and Ψ(α)). This is 0↑ MS 2/1 ↑ MS 2/1 discussed in more detail in appendix C. 9 E=0.5712 E j E=0.2349 Fig.3: An electron has been injected into the a† state (or d† state, 0↑ ↑ dashed arrow). The resulting state is a superposition of two nest states. From these nest states the electron makes (as one possibility) a transition into the state a† where the two amplitudes interfere. The energies of the j↑ two nest states are shown. The different thickness of the arrows shows different probabilities for the two paths. 3.2 Injection of a hole For the full spectrum of excitations one has to include the injection of holes into the occupied states. This is shown in Fig.4. The hole can be injected into the occupied states a† yielding j↑ a Ψ or into the nest. In the latter case the spin-up part of the nest is emptied. This j↑ MS yields for the secular matrix of the nest in analogy to equ. (6) E(b) Vsd Hnst = 0 b (8) 0/1 Vsd E (cid:18) b d (cid:19) aa bb gg dd Fig.4: An hole has been injected into the spin-up band and the spin up nest. This induces electron or hole transitions in the spin-down band. The resulting Slater states are shown as (γ) and (δ). 10