Effects of electronic correlations in iron and iron pnictides A. A. Katanin In collaboration with: A. Poteryaev, P. Igoshev, A. Efremov, S. Skornyakov, V. Anisimov Institute of Metal Physics, Ekaterinburg, Russia Special thanks to Yu. N. Gornostyrev for stimulating discussions Iron properties Mikhaylushkin, PRL 99, 165505 (2007) a - bcc, g - fcc, e - hcp ag Ts =1185 K a-iron: TC = 1043K, meff =3.13mB g-iron: qCW =-3450K, meff =7.47mB Arajs. J.Appl.Phys. 31, 986 (1960) Parsons, Phil.Mag. 3, 1174 (1959) • Itinerant approach (Stoner theory) 𝐼𝑁 𝐸𝐹 = 1 Large DOS implies ferromagnetism, provided that other magnetic or charge instabilities are less important - Too large magnetic transition temperatures, no CW-law • Moriya theory: paramagnons Reasonable magnetic transition temperatures, CW law • Local moment approaches (e.g. Heisenberg model) CW law Rhodes-Wollfarth diagram a-Iron (almost) fulfills Rhodes-Wollfarth criterion (pc/ps 1) Proposals for iron: • Local moments are formed by eg electrons (Goodenough, 1960) • 95% d-electron localization (Stearns, 1973) • Local moments are formed from the vH singularity eg states (Irkhin, Katsnelson, Trefilov, 1993) The magnetism of iron a-Iron shows features of both, itinerant (fractional magnetic moment) and localized (Curie-Weiss law with large Curie constant) systems • Can one decide unbiasely (ab-initio), which states are localized (if any) ? • What is the correct physical picture for decribing local magnetic moments in an itinerant system? Itinerant (Stoner and Moriya theory) Local moments (Heisenberg model) Mixed (Shubin s-d(f) = FM Kondo model) How the local moments (if they exist) influence magnetic properties? What is the similarity and differences between magnetism of a- and g- iron? Dynamical Mean Field Theory Energy-dependent effective medium theory ( ) The self-energy of the embedded atom coincides with that of the solid (lattice model), which is approximated as a k-independent quantity -1 ( ) = -1 ( ) - Gloc ( ) Gloc ( ) = k 1 - e k m - ( ) A. Georges et al., RMP 68, 13 (1996) Spin-polarized LDA+DMFT U = 2.3 eV, J = 0.9 eV Magnetic moment 3.09 Critical temperature 1900 K Lichtenstein, Katsnelson, Kotliar, PRL 87, 67205 (2001) (3.13) (1043K) a (Bcc) iron: band structure eg A. Katanin et al., PRB 81, 045117 (2010) t2g t2g и eg states are qualitatively different and weakly hybridized Correlations can “decide”, which of them become local a-iron: orbitally-resolved self-energy Imaginary frequencies t2g states - quasi-particles eg states - non-quasiparticle! A. Katanin et al., PRB 81, 045117 (2010) Comparison to MIT: Linear for the Fermi liquid Divergent for an insulator Bulla et al., PRB 64, 45103 (2001) Self-energy and spectral functions at the real frequency axis Real frequencies a-Fe Comparison to MIT: From: Bulla et al., PRB 64, 45103 (2001) How to see local moments: local spin correlation function Local moments are stable when S(0)S() S z ( ) S z (0) const Fulfilled at the conventional Mott transition. Can it be fulfilled in the metallic phase ? J=0.9 J=0 A. Katanin et al., PRB 81, 045117 (2010) Fourier transform of spin correlation function ( ) = meff2 3T f ( / T ) Fourier transform of spin correlation function ( ) = meff2 3T Local moments formed out of eg states do exist in iron! f ( / T ) Which form of 𝜒𝑙𝑜𝑐 𝜔 one can expect for the system with local moments? S (0) S ( ) const( ) (in ) = d S z (0) S z ( ) ei S 2 n ,0 / 3T n 0 Broaden delta-symbol: (in ) = meff2 g 3T | n | g ( ) = meff2 ig 3T ig g is the damping of local collective excitations Re ( ) = Im ( ) = meff2 g2 3T g 2 m 2 g 3T 2 g 2 2 eff For a-iron: 𝜇eff = 3.3𝜇𝐵 𝛾 ≈ 𝑇/2 (𝜔 ≪ 𝐽) Curie law for local susceptibility eg t2g Total local moment = g m B p( p 1) / (3T ) 2 2 p(eg) = 0.56 p(t2g) = 0.45 p(total)=1.22 agrees with the experimental data (known also after A.Liechtenstein, M. Katsnelson, and G. Kotliar, PRL 2001) Effective model The local moments are coupled via RKKY-type of exchange: J H effd = H eg H t2 g H t2 g -eg (U - ) N i nim - 2 J Si sim 2 i , ,mt2 g ' i , mt2 g ' RKKY type (similar to s-d Shubin-Vonsovskii model). The theoretical approaches, similar to those for s-d model can be used g-(fcc) iron TN≈100K Which physical picture (local moment, itinerant) is suitable to describe g-iron ? What is the prefered magnetic state for the g iron at low T (and why)? LDA DOS The peak in eg band is shifted by 0.5eV downwards with respect to the Fermi level g-(fcc) iron P. A. Igoshev et al., PRB 88, 155120 (2013) More itinerant than a-iron ? DOS with correlations Static local susceptibility P. A. Igoshev, A. Efremov, A. Poteryaev, A. K., and V. Anisimov, PRB 88, 155120 (2013) Dynamic local susceptibility Size of local moment Magnetic state: Itinerant picture QX=(0,0,2) Comparison of energies in LDA approach Shallcross et al., PRB 73, 104443 (2006) SDW2 Magnetic state: Heisenberg model picture Heisenberg model A. N. Ignatenko, A.A. Katanin, V.Yu.Irkhin, JETP Letters 87, 555 (2008) For stability of (0,0,2) state one needs J1>0, J2<0. The polarization bubble, low T LDA LDA+DMFT m k k+q T=290К 2(1,0,0) 2(1/2,1/2,1/2) 2(1,1/2,0) m' Experimental magnetic structure q = (2/a) (1, 0.127, 0) Tsunoda, J.Phys.: Cond.Matt. 1, 10427 (1989) Naono and Tsunoda, J.Phys.: Cond.Matt. 16, 7723 (2004) Fermi surface nesting Colorcoding: red – eg, green – t2g, blue – s+p (0,x,2) state is supported by the Fermi surface geometry – an evidence for itinerant nature of magnetism The polarization bubble, high T LDA LDA+DMFT T=1290К Uniform susceptibility m k 1/ k From high-temperature part: 𝜇𝑒𝑓𝑓 = 4.07𝜇𝐵 (𝑝𝐶 ≅ 3/2) m' g-(fcc) iron 𝜇𝐶𝑊 = 7. 7𝜇𝐵 exp -1500 K DMFT -2700 K TN 100 K (small particles) Strong frustration! Nonlocal correlations are important The experimental value of the Curie constant is reproduced by the theory, although the absolute value of paramagnetic Curie temperature appears too large Magnetic exchange in g-iron 𝜒0 𝜒𝐪 = 1 − 𝐽𝐪 𝜒0 # 𝜒irr (𝐪) 𝜒𝐪 = 1 − Γ 𝜒irr (𝐪) 1 J z/2,K -669 2 3 4 5 6 173 -449 17 -25 -123 -116 −1 𝐽𝐪 = −[𝜒irr (𝐪)] 7 8 29 +𝐶 𝐽𝟎 = −2500𝐾 𝐽𝐐 = 1200𝐾 The Neel temperature is much larger than the experimental one, similar to the result of the Stoner theory: o Paramagnons o Frustration, i.e. degeneracy of spin susceptibility in different directions Local spin susceptibility of Ni A. S. Belozerov, I. A. Leonov, and V. I. Anisimov, PRB 2013 Iron pnictide LaFeAsO Antiferromagnetic fluctuations Superconductivity Itinerant system in the normal state Effect of electronic correlations? Possibility of local moment formation? Density of states Electronic correlations Damped qp states qp states No qp states 387K 580К 1160К xy 0.142 0.242 0.454 xz, yz 0.131 0.163 0.306 3z2-r2 0.054 0.092 0.228 x2-y2 0.053 0.101 0.334 dxz, dyz, dxy states can be more localized Local susceptibility 387K 580К 1160К xy -0.142 -0.242 -0.454 xz, yz -0.131 -0.163 -0.306 3z2-r2 -0.054 -0.092 -0.228 x2-y2 -0.053 -0.101 -0.334 Spin correlation functions The situation is similar to g-iron, i.e. local moments may exist at large T only, and, therefore, seem to have no effect on superconductivity Orbital-selective uniform susceptibility Local fluctuations are responsible for the part of linear-dependent term in (T) S. L. Skornyakov, A. Katanin, and V. I. Anisimov, PRL ’ 2011 Summary The peculiarities of electronic properties (flat bands, peaks of density of states) near the FL may lead to the formation of local moments; Analysis of orbitally-resolved static and dynamic local susceptibilities proves to be helpful in classification of different substances regarding the degree of local moment formation In alfa-iron: The existence of local moments is observed within the LDA+DMFT approach The formation of local moments is governed by Hund interaction In gamma-iron: Local moments are formed at high T>1000K, where this substance exist in nature, but not at low-T (in contrast to alfairon); the low-temperature magnetism appears to be more itinerant Antiferromagnetism is provided by nesting of the Fermi surface Conclusions In the iron pnictide: Electronic correlations are important, but, similarly to g-iron, local moments may be formed at large T only Different orbitals give diverse contribution to magnetic properties Linear behavior of uniform susceptibility is (at least partly) due to peaks of density of states near the Fermi level Thank you for attention ! Spin correlation functions Spectral functions Damped qp states qp states No qp states Effective model and diagram technique H effd = H t2 g H eg - 2 I Ss i im i ,mt2 g I (U - ) nim nim 2 i , ,meg ,mt2 g (similar to s-d Shubin-Vonsovskii model). Treat eg electrons within DMFT and t2g electrons perturbatively Simplest way is to decouple an interaction and integrate out t2g electrons [R L = Leg -1 m m mm q - q t t - mm q ( t 2 ISq )( t m q m -q 2 IS- q )] q ,mm “bare” quadratic term mm m m m mm ( t 2 I S ) ( t 2 I S ) ( t q1q2 q3 ,abcd q1 q1 a q2 q2 b q3 2 IS q3 ) c qi ,mi ( t m 2 IS ) ... Diagram technique: perspective mm q = mm mm q1q2 q3 ,abcd = q0,e = g abcd a b c d (4), = S q1q2q3 q1 Sq2 Sq3 S- q1 -q2 -q3 c ,eg = The dynamic susceptibility ( R q ,t2 g I ) R q R R q q ,eg “Moriya” correction bare ( q0,t2 g )-1 t02 g 4 I 2 e0g = -2 I q ( q0,eg )-1 - 4 I 2 q 4 I 2 (4) t02 g bare RKKY -2 I q -1 Influence of itinerant electrons on local moment degrees of freedom • Two different approaches to magnetism of transition metals (and explaining Curie-Weiss behavior): - Itinerant (Stoner, Moriya, …) - Local moment (Heisenberg, …) Local moments in transition metals Since they are (good) metals, at first glance no ‘true’ local moments are formed However, under some conditions the formation of (orbital-selective) local moments is possible: -Weak hybridization between different states (e.g. t2g and eg) -Presence of Hund exchange interaction - Specific shape of the density of states Can one unify these approaches (one band: Moriya, degenerate bands: Local moments in transition metals Since they are (good) metals, at first glance no ‘true’ local moments are formed However, under some conditions the formation of (orbital-selective) local moments is possible: -Weak hybridization between different states (e.g. t2g and eg) -Presence of Hund exchange interaction - Specific shape of the density of states Dependence on imaginary frequency Paramagnetic LDA+DMFT U = 2.3 eV, J = 0.9 eV, T = 1120 K t2g states eg states Weakly correlated compound ?!?!?!? t2g и eg состояния качественно различны и слабо гибридизованы Важно учесть влияние электронных корреляций U dependence J = 0.9 eV, = 10 eV-1 Stability with temperature Weak itinerant magnets Saturation magnetic moment is small The thermodynamic properties are detrmined by paramagnons; Hertz-Moriya-Millis theory: for ferromagnets (d=3, z=3) the bosonic mean-field (Moriya) theory is sufficient to describe qualitatively thermodynamic properties even close to QCP. k0,i “paramagnon” 1 - U k0,i n n 0abab 0 k ,i 4/3 T 0abab T 0 b =1...3 k ,i 1 - U k ,i n n n Curie-Weiss-like susceptibility Frustration in Heisenberg FCC model Polarization bubble m m' eg t2g-e2g t2g G. Stollhoff, 2007 Diagram technique: perspective mm q = mm mm q1q2 q3 ,abcd = q0,e = g abcd a b c d (4), = S q1q2q3 q1 Sq2 Sq3 S- q1 -q2 -q3 c ,eg = S(0)S() Spin correlation function at different U eg almost flat ! t2g Weak itinerant magnets Saturation magnetic moment is small The thermodynamic properties are detrmined by paramagnons; Hertz-Moriya-Millis theory: for ferromagnets (d=3, z=3) the bosonic mean-field (Moriya) theory is sufficient to describe qualitatively thermodynamic properties even close to QCP. k0,i “paramagnon” 1 - U k0,i n n 0abab 0 k ,i 4/3 T 0abab T 0 b =1...3 k ,i 1 - U k ,i n n n Curie-Weiss-like susceptibility Effective model H effd = H t2 g H eg - 2 I it loc i , mt2 g Sisim I (U - ) N i nim 2 i , ,mt2 g (similar to s-d Shubin-Vonsovskii model). Treat itinerant electrons perturbatively: introduce effective bosons for an interaction between itinerant electrons and integrate out itinerant fermionic degrees of freedom L = Lloc m m [ Rmm-1 t mq t m- q - mm q (t q 2 IS q )(t - q 2 IS - q )] q , mm “bare” quadratic term mm m m q1q2 q3 , abcd m q2 (t 2 IS q1 ) a (t 2 IS q2 )b (t m q1 m q3 2 IS q3 )c qi , mi (t m- q1 -q2 -q3 2 IS - q1 -q2 -q3 ) quartic interaction ... The dynamic susceptibility ( R q ,it I ) R q R q R q ,loc “Moriya” correction bare 0 ( q0,it ) -1 it0 4 I 2 loc = -2 I q -2 I q ( q0,loc ) -1 - 4 I 2 q 4 I 2 (4) it0 bare -1 RKKY Influence of itinerant electrons on local moment degrees of freedom Exchange integrals and magnetic properties can be extracted Return to a-iron Return to a-iron How do we recover RKKY exchange for a-iron? Assume: 𝜒irr (𝐪) = 1/𝐼 + 𝜒 ′ irr (𝐪) −1 𝐽𝐪 = −(𝜒irr ) 𝜒′irr ≪ 1/𝐼 + 𝐶 = 𝐶 ′ + 𝐼 2𝜒 ′ irr (𝐪) I ~ 1 eV – extracted in this way, in agreement with performed analysis and band structure calculations Size of local moment Orbitally-resolved DOS LDA U = 4 eV, = 10 eV-1 a-Iron can be viewed as a system in the vicinity of an orbital-selective Mott transition (OSMT) Ratio of moments The size of the instantaneous and effective moment Magnetic exchange: r L(S)DA formula: J (r , r , ) = f m - f 1 m (r ) Bxc (r ) (r ) (r ) 4 m - e m e Bxc (r ) m (r ) (A. I. Liechtenstein, M. I. Katsnelson, et al.) Requires a ‘reference magnetic state’ to calculate exchange integrals: Reference state is needed to introduce magnetic moment in an itinerant approach In which cases one can avoid use of the ‘reference state’ ? Example: (one-band) Hubbard model at half filling due to metal-insulator transition the electrons are localized, Jij=4t2/U r' (2,0,0) NM FM (0,0,) FM, bcc