Magnetism: from Stoner to Hubbard - cond
Transcription
Magnetism: from Stoner to Hubbard - cond
Magnetism: from Stoner to Hubbard Alexander Lichtenstein University of Hamburg In collaboration with A. Rubtsov, M. Katsnelson Outline Heisenberg, Stoner, and Hubbard Many-body approach: D(M)FT functionals Correlation effects in electronic structure Magnetism of correlated systems physicsworld.com Itinerant ferromagnetism John Hubbard: the man behind the model Those who knew John Hubbard describe him as a very shy man – to the point that others, who did not know him so well, may have perceived him as somewhat aloof. Born on 27 October 1931, Hubbard was educated first at Hampton Grammar school and then at Imperial College, London, where he obtained his PhD in 1958 under Stanley Raimes. Unusually for his time and social context, he lived with his parents in Teddington throughout his university education. At the end of his PhD, Hubbard was recruited to the Atomic Energy Research Establishment in Harwell, Oxfordshire, by Brian Flowers, who was then heading the theory division. An anecdote from this period of Hubbard’s career illustrates his retiring personality. While at Imperial, Hubbard had dealt with the project assigned to him for his PhD fairly quickly, and had then looked for a more challenging problem. At the time, quantum-field-theory methods, particularly Feynman diagrams, were being applied to problems in many-body theory. However, it was difficult to bring the same methods to bear on the many-electron problem – relevant to solid-state systems – because the Coulomb interaction between electrons made quantities like the total energy diverge. Hubbard realized that these divergences could be controlled: the trick was to sum up an infinite series of a particular class of Feynman Stoner T=0 T<Tc T>Tc he saw an article by other researchers who Heisenbergithadupintroduced aHubbard different method to solve the same problem. Hubbard had found their method physically appealing, checked privately that their results coincided with his, and concluded there was no need for an additional publication on the topic. Flowers then issued an explicit order that Hubbard should publish his groundbreaking work. Hubbard’s most famous papers are the series he wrote on his eponymous model, starting in 1963. He was not the only one working on the strong-correlations problem: some months earlier, Takeo Izuyama, working at Nagoya University, and Duk-Joo Kim and Ryogo Kubo, at the University of Tokyo, both in Japan, had argued that a proper description of correlations in metals with strong electron–electron k k k From Stoner to Hubbard k o discuss the different models ofband magnetic materials (Fig. 1) awith theasimplest where " is the energy band spectrum and I is Stone k where " is the energy spectrum and I is Stoner interat k is the energy band spectrum and I is a Stoner interatomic exchange parameter. ange interaction in athe two-site quantumdependent model. magnetic properties er Hamiltonian this case temperature this the temperature magnetic properties areSton rel the temperature dependent magneticdependent properties and are related the so-called are thecase hopping parameters U with the characteri X + s from occupied ”spin-up” band to unoccupied ”spin-down” band, reduce th excitations from occupied ”spin-up” band towhich unoccupied H = (" + I < n >)c excitations from occupied ”spin-up” band unoccupied ”spin-d s k k ck to Stoner mode easily realize Stoner model is just E k that ean of the magnetization, soofthat finally atthe the Curie point the itinerant system becom F Cu magnitude the magnetization, so that finally at the e themagnitude hopping parameters and U the characteristic Coulomb of the magnetization, so that finallyHubbard at the Curie poin netic Mean Field is themetal. energy band spectrum and I is a Stoner interatomic exchange paramete model. In the correlated the only po a nonmagnetic metal. n easily realize thatweakly the Stoner model is just case, a mean-field approxim a nonmagnetic metal. I m Hami mpare the a standardproperties Hubbard approach withwith the the following ΔhσStoner = U < nmodel he temperature dependent are related so-called S −σ > nσwith magnetic Ifcompare we compare the Stoner model with a standard Hubbar del. IfInand the weakly correlated case, the onlya possible magnetic excitat we the Stoner model with standard Hubbard appro nes, the corresponding energy is of the order s from occupied ”spin-up”X band to unoccupied ”spin-down” band, which reducI X + is of the order I · M with M =< n Hubbard mode Hh = energy tonian: s, and the corresponding t c +the Curie U ni" npoint "bec( tonian: ij i# i cj at e of the magnetization, so that finally the itinerant systemX X X X +For the o arger then realistic Curie temperatures. ij i + t limit ger then realistic Curie temperatures. For the opposite ofU+ strongl H = c c U h ij j netic metal. H = t c c + n n i h ij j i" i# i U >> t ij with the i H 2 del theat half-filling casea[1], onecase can derive anone effective m pareatthe Stoner with standard Hubbard approach ij ifollowing model the half-filling [1], canHeisenberg derive an tmodel J = −2 U Hh = Heisenberg exchange X Hije = exchange interactions Jij = X X+ U ni" ni# ~ ~j . JijiSi · S tij c+ i cj ij 2tij tji U He = X ij ~i · S ~j . Jij S are of the order of magnetic (Nee Magnetism of H2: Heisenberg vs. Slater Exchange interaction: F i j F U t U AF i AF e2 J =< ij | | ji > r12 j Exchange interactions: Anderson U t U ED in subspace N- =1,N+ =1 U>>t Anderson kinetic exchange Exchange: Local force approach θ U t Spectrum: Exchange energy: U Spin-polarized LSDA From Atom to Solids Electrons in solids: -Effective potential -Bloch states -Pauli principle Density Functional Theory (DFT) Effective one-particle states Local Density Approximation (LDA) ⇥r )] => n0 (⌅ <= ⇥ E[n(⌅r)] = ⇥ |H| ⇤ =Vextn(⌅ r)Vext (⌅rr))d⌅ r 0+ VFext [n(⌅ 1 ⇥1 2 0 0 2 r <E⇥1 <2 |H | |H |(H H12|)|H⇤+⇥ =2E +E2n r)|n )[V (⌅rext )E 0(⌅ V+ (⌅ rext )]d⌅ ⇥ 1 |2 |H = ⇥ |H | ⇤+⇥ )| ⇤ = + (⌅ r )[V r ) V (⌅ r 0 (⌅ 2⇤ 2 |H 22 2 ⇤+⇥ 2= 1 |(H 22⇤ 2 ext ext 0 1 |= 2⇥0⇤ < 2 2 2 E ⇥ | ⇤ ⇥ |H |(H H ⇤ = n (⌅ r ) 0 2 1 2 2 2 2 2 1 2 2 1 2 0 0 1 T [n] + U [n] )| 2 ⇤ = E2 + n0 (⌅r)[Vext (⌅r E1 < ⇥ 2 |H1 | 2 ⇤ = ⇥ 2 |HF2 |[n]2= ⇤+⇥ 2 |(H1 H2 ⇥ ⇥ ⇥ F [n] =1 T [n] + U [n] 0 02 2 ⇥2 1 0 0 1 r 0 0 E2 <E⇥2 <1 |H | |H |(H H21n|)|0H =1E +E1n r)|n )[V (⌅rext )E (⌅ V+ (⌅ rext )]d⌅ V (⌅ r <= V ⇥ 2 |1 |H = ⇥ |H | ⇤+⇥ |(H )| ⇤ = + (⌅ r )[V r ) V 1⇤ 1 |H 11 1 ⇤+⇥ 1= 2=> 11)⇤ 0 (⌅ 1 ext ext 2 |= 1⇥⇤ < 1 1 1 0 ext ext E ⇥ | ⇤ ⇥ |H ⇤+⇥ |(H H ⇤ = n r(⌅ )r 1 2 11 1 1 1 2 1 2 1 10 02(⌅ 2 1 0 Vext r1 ⇤+⇥ ) <= 1V|(H E2 < ⇥ 1 |H2 | 1 ⇤ = ⇥ => r)[Vext (⌅r 1 |H1n|0 (⌅ ext 2 H1 )| 1 ⇤ = E⇥ 1 + n0 (⌅ 0 0 0 0 0 0 1 0 < 0⇤+⇥ 0 |(H ⇥ E0 + n (⌅ 0 E1 0 H 02 )| 20⇤ = E1⇥ +E E + E E < ⇥ |H | r )[V (⌅ r) Effective one-electron equation: + E < E + 0 2 |H1 | Schrödinger-like 2⇤ = 2E 2 2 2 2 2 1 1 2 ext 1 2 0 2 E01 +1 E20 < E020 + E1 0 1 2 E1 < ⇥ 2 |H1 | 2 ⇤ = ⇥ 2 2 |H2 | 2 ⇤+⇥ 2 E |(H H22< )| E22⇤ + =E E12 + n0 (⌅r⇥)[Vext (⌅r) Vext (⌅r 1 E 1 + 2 2 h̄ h̄2 0 2 h̄ 2 ⇥ E 0 + n (⌅ 2 2r⇤)i 2 ⌅ V (⌅ r ))⇥ (⌅ r )⌅ ⇥ ri)⇥ ( E < ⇥ | |(H H )| ⇤ = r )[V (⌅ r) V (⌅ r (⌅ = ⇤ (⌅ r ) ef f1h̄ i ))⇥ i (⌅ 1 |H2 | (1 ⇤ = ⇥ ⌅1 |H 1f⇤+⇥ 1= 1 1 0 ef i i 2 1 ext V (⌅ r ))⇥ (⌅ r ) = ⇤ ⇥ (⌅ r ) ( ef f i i 2 0 0i 1 2 2m 2m E2 < ⇥ 1 |H2 | 1 ⇤ = ⇥ 1 |H1 | (1 ⇤+⇥ ⌅ Hf1(⌅ )|r))⇥ =r)E=1 + ⇤i ⇥in(⌅r0)(⌅r)[Vext (⌅r) Vext (⌅r 2m2 Vef 1 |(H 1 ⇤i (⌅ 2m 0 0 0 N 0 N N1 E + E < E + E 1 2 2 2 20 0= 0|⇥ (⌅ 0 N n(⌅rE ) n(⌅ r )| r ) = |⇥ (⌅ r )| i Charge density: + E < E + E i n(⌅ r |⇥ r)|2 1 2 2 2 1) = 2 i (⌅ h̄ i 2 n(⌅ |⇥i (⌅ri )| i r) = 2( ⌅⇥ ⇥Vef f (⌅r))⇥ r) = r) h̄ i i (⌅ ⇥ ⇤i ⇥i (⌅ 22m Vef f+ (⌅rr))⇥ (⌅r)V )r)d⌅ = ⇤r+ ⇥iE (⌅rxc )[n] ( T+[n] ⇥ r(⌅ ir i)d⌅ E[n]E[n] = Ts= [n] VH⌅ n(⌅ )V (⌅ +[n] VH+[n] n(⌅ r + Er)V ext s2m ext xc [n] E[n] = T [n] + V [n] + n(⌅ r)d⌅r + Exc [n] s ext (⌅ N H Energy Functional: E[n] = Ts [n] + VH [n] + n(⌅r2)Vext (⌅r)d⌅r + Exc [n] N |⇥ (⌅ e2 ⇥ e2 ⇥⇥n(⌅r)⇥= n(⌅r)n(⌅ r2i)n(⌅ )⇥r)|r )⇥ n(⌅ 2 e n(⌅r)n(⌅r ) n(⌅rd⌅ ) r= d⌅ i )| VH [n] d⌅ r|⇥ ⇥[n] i (⌅ 2 d⌅ N VH=[n] = rH d⌅ rr= V d⌅ r r 2 2 |⌅r ⇥ |⌅ ⌅r2r | ⌅r | h̄ 2 |⌅ i T [n] = d⌅ r ⇥ (⌅ r )( ⌅ )⇥ri (⌅r)⌅r | s i KS-kinetic energy: E[n] = Ts [n] + VH⇥[n] + n(⌅r)Vext2m (⌅ r )d⌅ r + Exc [n] i E[n] = Ts [n] + VH [n] r⇥)Vext⇥(⌅r)d⌅r⇥ + Exc [n] ⇥ + ⇥ n(⌅ n(⌅ [n] 2 ⇥ Erxc [n] ⇥ ⇥ r ) Excn(⌅ 2 2 e r )n(⌅ n(⌅ r) ) Exc [n] )n(⌅ r 2 2 ⇥ e n(⌅ r )n(⌅ r ) V (⌅ r ) = V (⌅ r ) + e d⌅ r + V (⌅ r ) = V (⌅ r ) + e d⌅ r + ef f ext ef f ext ⇥ ⇥ V (⌅ r ) = V (⌅ r ) + e d⌅ r + V [n] = d⌅ r d⌅ r Hartree potential: 2Hf = ef VH e[n] d⌅ r⌅rn(⌅ |⌅r2ext | rd⌅ )rn(⌅ |⌅ ⌅rr| rn(⌅ rr⌅)r⇥ | ⌅r | )n(⌅ )r|⌅ |⌅ n(⌅r) 2 |⌅ r ⌅ r | VH [n] = d⌅r d⌅r E10 DFT: KS-equation (1965) Effective potential: 2 |⌅r ⇥ ⇥ ⌅r | n(⌅ r⇥ ) ExcE[n] xc [n] ⇥ n(⌅ r ) 2 ⇥ V (⌅ r ) = V (⌅ r ) + e d⌅ r + efrf) = Vext (⌅ ext Vef f (⌅ r) + e n(⌅ d⌅ +⇥[n] rr) |⌅r |⌅r⌅rE| xc ⌅ r | n(⌅rn(⌅ 2 ) r) Vef f (⌅r) = Vext (⌅r) + e d⌅r + |⌅r ⌅r | n(⌅r) 2 Computational Material Science: DFT CERAN-plate DFT-theory: LiAlSiO4 A.L, R. O. Jones, H. Xu,P. J. Heaney, Phys. Rev. B 58, 6219 (1998 Correlation driven MIT photoemission spectra (DOS) A. Fujimori et al. U/W Spectral function: Correlations effects ARPES Free electrons Correlated electrons Strongly Correlated Electron Systems 3d - 4f Charge fluct. open shells materials I H Li Na K Rb Cs Fr II Be Mg Ca Sr Ba Ra • Kondo • Mott-Hubbard • Heavy Fermions • High-Tc SC • Spin-charge order • Colossal MR U<<W U>>W Spin fluct. IIIb IVb Vb Sc Y La * Ac** VIb VIIb IIb III Cu Ag Au Zn Cd Hg B C Al Si Ga Ge In Sn Tl Pb Pm Sm Eu Gd Tb Np Pu Am Cm Bk Dy Cf Ho Es Ti Zr Hf Rf V Nb Ta Db Cr Mn Mo Tc W Re Sg Bh Lanthanides * Ce Th Actinides ** Pr Pa Nd U VIIIb Fe Ru Os Hs Co Rh Ir Mt Ib Ni Pd Pt Control parameters VII N P As Sb Bi O S Se Te Po F Cl Br I At Er Tm Yb Fm Md No Lu Lr • Orbital • Lattice La1-xCaxMnO3 Nd2-xCexCuO4 La2-xSrxCuO4 'Normal' Metal 200 eu Ps 100 AF p AF ga do Temperature (K) 300 0 0.3 Dopant Concentration x VI • Charge / Spin • Band filling • Dimensionality FM V Degrees of freedom • Bandwidth (U/W) CMR IV SC 0.2 0.1 0.0 0.1 SC 0.2 Dopant Concentration x 0.3 0 He Ne Ar Kr Xe Rn Hubbard model for correlated electrons H = ∑ t ijci+σc jσ + U∑ n i↑ n i↓ ij i U t The Theory of Everything Hamiltonian for multi-fermionic system in field-operators: Atomic Units: Coilomb interaction: Second quantisation operators in orthonormal basis: Wannier Basis: with site, orbital andspins quantum numbers where'0 e Hoehnberg and ~~i*(r)~~i (r)dr, Kohn: DFT(2) PHYSICAL REVIEW NOVEMEBR 1964 I N HOMOGENEOUS 136, VOLUM E 2 NUM B ER 3 8 EL EleCtrOn Gaa* InhOmOgeIIeouS P. 9 HOHENBERGt ELECTRON HOMOGENEOUS whe and GAS potential an external of v(r) C under the V= influence v(r)i(*(r)P(r)dr, W. KonNt num hasis a universa st where Pfn] Coulomb potential Hamiltonian an external The the mutual Ecole of the influencerepulsion. v(r) and under I'aris, France number of particles" and the mutual Coulomb repulsion. The Hamiltonian has func the form at La San Jolla, of functional plays a central the form 18 June 1964) H= With its aid we define, W Hamiltonian in T+V+U, H= v T+V+U, where'0 state of P*(r)P*(r') It energy functional This v(r). drdr' (r') P (r) that that of v(r), a F I (r) where'0 enea = fs(r)n (r)dr+Ft I (r) j E— its correct Field Operators: I NFrance Ecole Xornzale Superzeure, I'aris, AND Prance and I'aculte des Sciences, Orsay, Xonnale Superieure, and University Calzfo&nia Diego, (Received Calzfornia f an interacting electron gas in an external potential is paper deals with the ground the exthere exists universal functional of the density, such t g, independent has as minimum value the ground-state energy associated with s(r). The functional FLn(r)j is then discussed for two situations: (1) n(r) @san(r), 8/ao((1, and ~~i*(r)~~i ~. In both cases F can be expressed entirely in terms (2) of the cor(2) a(r) = q (r/ra) with p arbitrary and 1'p 2 order electronic polarizabilities of a uniform electron gas. This approach relation energy and linear and higher also sheds some light on generalized Thomas-Fermi methods and their limitations. Some new extensions of these methods are presented. proved pression ~ (r)dr, E„gn]=— ELECTRON GAS MOGENEOUS that assume for simplicity follows We shall in all that ~~i*(r)~~i for the correct (r)dr, (2) Clearly, V= v(r)i(*(r)P(r)dr, E. state in which situations ground INTRODUCTION universal valid forthe functional, 2dealing only ntial v(r)weandare where any is a description theoretical considerations ofenergy this Pfn] is awith ' v ( is( We shall now show that functional. Once known, it is relatively easy to deterthe last decade there has been considerable &~ IJRINGnumber particles" This and of potential. miltonian state has anythe external electronic the density denote in a given ground-state external energy is nondegenerate. value for the correct n(r) the properties We in understanding of a mine progress drdr' (r') P*(r)P*(r') (r) P ' f potential. The of homogeneous functional interacting electron gas. point are restricted by the condi central role in the present paper. plays aelectrons In Part II, we obtain an expression for FLnj when tr in general, to state as has been, regard the 0' inviewthe ground by Cle i.e. , the deviates from uniformity, n(r)=1'cp slightly for a only aid we define, collection potential similar to a V= noninteracting particles With ofits given v(r)i(*(r)P(r)dr, v(r), —0; In this that case FLej is entirely assume with for follows+ts(r), ts/tss simplicity in allofthat We shall collective with the important additional concept — =en stat Ãfm] functional the ground exact ground-state energy expressible in excitations. energy we are only dealing with situations in terms whichof the ' & the exact electronic polarizability n(g) of a uniform On the other hand, there has been in existence sinceWeand electronic the This density denote gas. state is nondegenerate. electron procedure will describe correctly 7920's the the a different approach, represented by 0' ell known Thomas-Fermi method' in in whichby the long-range Friedel charge oscillations' set isupv by its ground and the re6nements,state All refinements localized a perturbation. previous of the and in the electronic density n(r) plays a central role=— energy functional v which the system of electrons is pictured more like a Thomas-Fermi method have failed to include these. In Part III we consider the case of a slowly varying, classical liquid. This approach has been useful, up to functionalbut of+ofv(r). which is clearly ofa inhomonecessarily almost constant density, tr (r) W that fo It of functional a is which clearly E„gn] v(r). 4' (2) of valu F dr+ I LN]. (10) (r) (r) drdr' P*(r)P*(r') (r') P (r) is a unique are We shall now show that conversely v(r) f Path Integrals for Fermions Short introduction from Alexei Kamenev “Field Theory of Non-Equilibrium Systems” (Cambridge, 2011) Fermions second-quantization operators (Pauli principle) Algebra of Grassmann anti-commuting numbers: Grassmann numbers anticommute with fermionic operators Grassmann calculus Differentiation: N.B. order: Example: Integration: (equivalent to differetition) Coherent State Eigenstate of annihilation operator Diefinition of coherent states Proof Left Coherent State: is just another Grassman number (NOT a complex conjugate) Unity operator in coherent states Overlap of Coherent States (non-orthogonal) Resolution of Unity Proof Trace of Fermionic Operators Matrix elements of normally ordered operators Trace-formula ''Minus'' due to commutation Left and Right coherent state Gaussian Path Integrals Only one analytical path integral: Short notation Proof - '‘det‘‘: expand the exponent only N-th oder is non-zero Examples: N=1 N=2 Correlation Function: U=0 Change of variables Using: Single-particle correlation function: Two-particle correlation function: Path Integral for Everything Euclidean action One- and two-electron matrix elements: Shot notation: One- and Two-particle Green Functions One-particle Green function 1 2 Two-particle Green function (generalized susceptibilities) Vertex function: Baym-Kadanoff functional Source term Partition function and Free-energy: Legendre transforming from J to G: Decomposition into the single particle part and correlated part = Functional Family Exact representation of Φ: Vαee=α Vee Different Functionals and constrained field J: G=ρ G=G(iω) G=G(k,iω) G. Kotliar et. al. RMP (2006) J=V=Vh+Vxc J=Σloc(iω) J=Σ(k,iω) DFT LDA+DMFT GW++ Dynamical Mean Field Theory BZ 1 ˆ k , iω Gˆ (iωn ) = ∑ G n Ω k ( G0 (τ − τ ʹ′) Σ Σ Σ U Σ Σ Gˆ0−1 (iωn ) = Gˆ −1 (iωn ) + Σˆ (iωn ) G0 (τ − τ ʹ′) Σ QMC Σ Σ ) Σ W. Metzner and D. Vollhardt (1987) A. Georges and G. Kotliar (1992) ED Single Impurity Solver DMRG IPT FLEX Σˆ new (iωn ) = Gˆ0−1 (iωn ) − Gˆ −1 (iωn ) DMFT: Charge+Spin+Orbital Fluctuations DMFT self-consistensy DMFT Impurity solver Analogy with conventional MFT Dieter Vollhardt Metal-Insulator Transition Dynamical Mean-Field Approach for Strongly Correlated Materials 1 1.18 Fig. 9: Mott-Hubbard MIT phase diagram showing the metallic phase and the insulating pha respectively, at temperatures below the critical end point, as well as a coexistence region; fr Ref. [54]. than linearly with the temperature, the difference ∆S = Smet −Sins eventually becomes posit whereby the slope also becomes positive at lower temperatures;8 this is indeed observed cluster DMFT calculations [60]. Since ∆S = 0 at T = 0 the phase boundary must terminat T = 0 with infinite slope. At half filling and for bipartite lattices in dimensions d > 2 (in d = 2 only at T = 0), paramagnetic phase is unstable against antiferromagnetic long-range order. The metal-insula transition is then completely hidden by the antiferromagnetic insulating phase, as shown Fig. 10. G. Kotliar and D. Vollhardt, correlations in materials Physics Today 3, 53 (2004) spectral function (“density of states”)6 ofElectronic the Hubbard model in the Strong correlation limit and Magnetism PM AFM V. Janis, et al, EPL 24, 287 (1993) U>W/2=4 P. Werner, et al, PRB 86, 205101 (2012) Jττ ' Sτ ⋅ Sτ' Formation of Local Moments and AFM correlations What is the Mott transition? a correlation driven metal-insulator transition Mott ’49 σ a n ~ 1/a3 ~ t cannot be obtained in band theory: t not due to AF (weak coupling effect): 2a n Comparison of LDA and realistic DMFT arison of LDA and realistic DMFT schemes LDA LDA+DMFT Density functional Baym-Kadanoff functional Density ⇢(r) Green-Function G(r, r0 , !) Potential Vxc (r) Self-energy ⌃i (!) Etot = Esp Edc P Esp = k<kF "k R Edc = EH + ⇢Vxc dr ⌦ = ⌦sp ⌦sp = ⌦dc T r ln[ G 1 ] Exc ⌦dc = T r⌃G LW DMFT model of ferromagnetism DOS-peaks D. Vollhardt, et. al., In:Bandferromagnetism, Springer, 2000 Band degeneracy W=4 U =6 V0=4 F0=2 Orbital degrees of freedom • e g orbitals • t 2g orbitals Mn (3+) = 3d4 3d-ion in cubic crystal field 5x d 2x eg 3x t2g J. van der Brink, D. Khomskii Charge transfer TMO insulators Zaanen-Sawatzky-Allen (ZSA) phase diagram Phys. Rev. Lett. 55, 418 (1985) LDA+U: static mean-filed approximation LDA+U functional: U U E = E LDA + ∑ n i n j - n d (n d -1) 2 ij 2 ∂E 1 εi = =ε + U( − ni ) One-electron energies: LDA ∂n i 2 U n i = 1 ⇒ ε i = ε LDA − Occupied states: Mott2 Hubbard U gap Empty states: n i = 0 ⇒ ε i = ε LDA + 2 LDA U≡ V. Anisimov et al, PRB, 44, 943 (1991) review: J. Phys. Cond. Mat. 9, 767 (1997) ∂ε LDA ∂n d Rotationally invariant LDA+U LDA+U functional Local screened Coulomb correlations (Orbital Polarization!) LDA-double counting term (nσ =Tr(nmm0 σ ) and n=n- +n+ ): Occupation matrix for correlated electrons: Spin and Orbital moments in CoO LDA+U+SO+non-collinear S L h h L I. Solovyev, A. L., and K. Terakura, PRL 80, 5758 (1998) Electronic structure of TMO: LDA+U DOS MnO MnO 12 8 NiO NiO LSDA LSDA U= 5e V U= 5e V Spin-wave Spectrum NiO I. Solovyev 0 8 4 8 U= 9e V U= 9e V 4 0 8 U= 1 3e V U= 1 3e V 3d 0 4 0 -‐12 -‐8 -‐4 0 Energ y (eV) O2p 4 -‐12 3d -‐8 -‐4 0 Energ y (eV) U= 1 3 LDA 5 400 0 w(q), meV 100 200 300 Density of Sta tes (sta tes/eV formula unit) 4 4 8 G 7 9 11 13 exp Z F G L Slater: Magnetic Transition State DFT: Janak theorem DFT: Transition State ∂E[n] εi = ∂ni ∂E[n] ΔE = E[ni = 1]− E[ni = 0] = Δni = ε (ni = 1 / 2) ∂ni n =1/2 i Exchange interaction: J = E[AF]− E[F] = Δε (n↑ = n↓ = 1 / 2) ε↓ − ε↑ Local Force Theorem: Functionals here = 2 ⌃i ± ⌃i , ⌃i = ⌃i ei ,and with ei being the unit vector in0the dire hence i 1 is⌃an ch analog of the ”local force theorem” inis an the density fu = 2 T r (Gij œ). We assume that the⇥ bare Green function G does which analog of the ”loc c ⇤ ⇤ ⇤ ⌦ = ⌦ = .ctive Spin excitations with low energies are connected with the rT spin-dependent potential on site i, = ( , , ) are Pauli matrices, G = sp x y z s with low energies are connected with ⇤ ⇤ 1 the rotations ofijve ⌦ = scheme, ⌦sp = the T rself ln including ⌃energy G0 the the LDA+DMFT t and terms Hartree-Fock term s all 1the spin-dependent 0 e LDA+DMFT isIn local, i.e. isscheme, diago Magnetic Torque nd Gij = 2 T r (Gij œ). We assume that the bare Green function G does not depe which is an analogthe of spin-matrix the ”local force theorem” structure of thin energy. Spin excitations low are connected with the rections and all structure the spin-dependent including the Hartree-Fock terms ini the LDA+DMFT scheme, the self energyare is loc pin-matrix ofwith theinterms self energy and Green function log of the ”local force theorem” theInenergies density functional theory [57]. ' e = eare the self energy. Spin excitations with low energies connected with rotation i i structure ⌃G the spin-matrix of the self the energy iis⇥diagonal i ' ei the = self ienergy ⇥ ei is local, i.e. MFT scheme, in site indices. Letandus : ⇣ s ⌘ s c c c s (c,s) " # structure of the self energy and Green function in the following form 1 ' e⌃ e,i where ⌃ ==⌃i ++ ⌃i iG ,⌃Gi ij i = ⌃= ⌃ ±of i ⇥ ⌃ + G i i G ”local force theorem” ⌃ (47) corresponding variation 2 i =the ij i i ij ij th ei = ' i ⇥ ei orem” (47) the corresponding variation of the ⇣fective ⌘ thermodyn spin-dependent potenti (c,s) " # 1 s s s c c s as⌃i ”local where ⌃G = 2 and ⌃i G ±s⌃= ⌃ ⌃iœ). ei , with gwritten to the ”local the corresponding of i = i i ,1variation =force ⌃i force +theorem” ⌃ itheorem” , G = G +corresponding ⇣ ⌘ij (47) T r (G We t ij ccording to the (47) the variation of the therm ij ij ij 2 (c,s) " # 1 spin-dependent potential on =( x s ⇤ sfective directions and allsite thei,the spin-dep re ⌃ = ⌃ ± ⌃ , ⌃ = ⌃ e , with e being un an⇣ bei can written as otential be⌘written as i ⌦ = ⌦ = 1i i ' i i i s V i i sp 2 and Gij = 2 T r (G œ).self Weenergy. assumeSpin that excitat the bar inijthe " # ⇤ s 1 s ⇤ ⌦ = ,⌃ ⌦isp '⌦i⌦e=i=being Theorem: ⇤⌦spthe ⌃Magnetic == ⌃i V ei ,iwith unit vector in the direction o == V 'i ithe i ± ⌃i Force i all directions and spin-dependent terms inclu 2 ⌦ V ' e : sp ve spin-dependent potential on site i, =i (i x , y , c z ) are P 1 in, the self energy. Spin excitations with=lowT energ indextorque is equal to endent potential on site i, = ( , ) are Pauli matrices, G r x y z ij 2 1 s here the torque indextorque is We equalassume to e : that the bare Green functi G = T r (G œ). i ij equal to ijijMagnetic torque is equal to Green 2We Torque: (G œ).indextorque assume that the bare function According G0 doestonot depend on the ”localeiforce = ' s s s ctions and all the spin-dependent Hartr ll the spin-dependent terms Hartree-Fock terms are incorpo ⌃⇥ V 2T rthe [⌃ ⇥G Gincluding ]] potential can bethe written as !L ⌃terms Vi =including 2Ti =r!L [⌃ i ii i ii s s force theorem” (47) the c According to the ”local slow energies s= 2T ⌃ V r [⌃ ⇥ G y. Spin excitations with are connected with of ve i !L i energies ii ] the rotations e self Spin with can low are connec ⌃ iexcitations Vi energy. = 2T r!L [⌃ ⇥ Gii ] potential be written as where the torque indextorque ⌦ = ⇤ ⌦sp ei = ' i ⇥ ei where the torque indextorque is equal to center. near magnetic structures in the following form In the nonrelativistic case one can rewrite the spin Hamiltonian for small spin de ⇣ ⌘ # Functional s "from s X Exchange interactions collinear magneticJstructures in the following form T r!L ⌦ ⌃i Gij= ⌃j Gji J e · e ij = spin ij i j X ⌦spin = generalize Jijijei · ethe Heisenberg exchagne: j e exchange parameters. This formula LDA expression ij related re wheresystems. ⇣ ⌘ ⇣ ⌘ " #both direc exspinExchange wave spectrum in ferromagnets can be considered s "⌃ ssG# ⌃ s G J = T r Jijij= T r!L !L ⌃i Gij ⌃ij Gjiij j ji interactions: meters or by the consideration of the energy of corresponding spiral s are the effective exchangeparameters. parameters. This formula generalize the LDA expressio he effective exchange This formula generalize the LD onrelativistic case when the anisotropy is absent one has theSpin case ofwave correlated systems. spectrum: caseSpin of wave correlated systems. indexspin wave spectrum in ferromagnets X 4 4 can be considered both dire !indexspin Jor0jbyspectrum 1the consideration cos qR ⌘the energy [J(0)of can J(q)] n wave wave inj ferromagnets be consider exchange of corresponding spiral q =parameters M j M Ref.parameters [57]). In nonrelativistic when the anisotropy is absent oneof hascorrespo hange or by thecase consideration of the energy Non-collinear excitation: X 4 4 is ion. [57]). In nonrelativistic case when the anisotropy e magnetic moment (in Bohr magnetons) per magnetic !q = J0j 1 cos qRj ⌘ [J(0) absent J(q)] one ha M j M ted that the expression for4 spin Xstiffness tensor D↵ defined 4 by the re !q = moment cos qR ⌘as the[J(0) J M is of the exchange magnetic (inJ0j Bohr1has magnetons) per ion. 0)where in terms parameters to be exact conseque j magnetic M j for spin stiffness tensor D Mdefined by the r It should be noted that the expression ↵ M. Katsnelson and A. L. , which Phys. Rev. 8906 (2000) LandauLifshitz equations are61, definitely correct in the long- Exchange interactions and Band structure 3d in Ni Fe, Co, Ni Ni J0 Co Fe Green Function calculations : J0=Σ Jij V. Antropov, et. Al. PRB (1988) LDA+Exchange Interactions Spin-waves T<Tc εF J ij = π1 i i ij j j ji ( V − V ) G ( V − V ) G ∫ ↑ ↓ ↑ ↑ ↓ ↑ dε −∞ Curie temperature Tc = 2 3 ∑J 0j j S. Halilov, et. al., PRB 58, 293 (1998) Ni Fe Co LDA+Disordered Local Moments DLM The best first-principle Spin-fluctuation model with classical moments EXP DLM J. Staunton and B. Gyorffy PRL69, 371 (1992) EXP Orbital order: KCuF3 In KCuF3 Cu+2 ion has d9 configuration with a single hole in eg doubly degenerate subshell. Experimental crystal structure LDA+U calculations for undistorted perovskite structure hole density of the same symmetry antiferro-orbital order A.L. V. Anisimov and J. Zaanen, Phys. Rev.B 52, R5467 (1995) 1d-AFM in KCuF3 J(K) Jc Jab Theory -240 +6 Exp. -202 +3 KCuF3 Superexchange interaction LSDA gave cubic perovskite crystal structure stable in respect to JahnTeller distortion of CuF6 octahedra LDA+U produces total energy minimum for distorted structure Quadrupolar distortion in KCuF3 Magnetism 5.25 Exchange in Iron: LSDA++ 500 Fe spin-wave Energy, meV 400 300 LSDA LDA+ (E) 200 100 Exp 0 0,0 0,5 Wave vector (001) 1,0 H Quantum Impurity Solver Σ Σ For a general N × N super-site impurity model (simp) the partition function can b functional integral over the 2N-component spin and site-dependent spinor Grassm and c : ' Z = D[c∗, c]e−Ssimp , Σ where Σ Σ U Σ Σ Σ Ssimp = − N ' ( I,J=0 0 + β dτ ' β dτ # c∗Iσ (τ) 0 N ' ( ) −1 * # Gσ (τ − τ ) IJ cJσ (τ # ) β dτUnI,↑ (τ)nI,↓(τ), I=1 0 τ’ U where G is the N × N matrix of effective bath Green’s function for a spin-colline The main problem of all cluster extension of DMFT is to find an optimal self-cons best scheme? obtain the bath Green’s functionWhat matrixisinaimaginary time GIJ (τ − τ # ) or in Ma Quantum Monte Carlo ! GIJ (iω). In the free-cluster version of the CDMFT scheme [6] which is equivalent DMFT method [8] or to the molecular CPA scheme in alloy theory [9] we can use prescription. First, we need to integrate out the superlattice degrees of freedom, si G(τ−τ’) τ Continuous Time Quantum Monte Carlo Partition function: Continuous Time Quantum Monte Carlo (CT-QMC) E. Gull, A. Millis, A.L., A. Rubtsov, M. Troyer, Ph. Werner, Rev. Mod. Phys. 83, 349 (2011) Weak coupling QMC: CT-INT A. Rubtsov, 2004 Random walks in the k-space Z=… Zk-1 + Zk + Zk+1+ …. k-1 k+1 Acceptance ratio increase decrease k D k −1 k w D Step k+1 D is tribution Step k-1 0 0 20 Maximum at βUN 2 40 k 60 w D k +1 k + 1 Dk Convergence with Temperature: CT-INT Maximum: βUN 2 Strong-Coupling Expansion CT-HYB P. Werner, 2006 Strong-Coupling Expansion CT-HYB P. Werner, 2006 Comparison of different CT-QMC Σ Σ Σ Σ U Σ Σ Σ Σ U τ τ’ G(τ−τ’) Ch. Jung, unpublished CT-QMC review:E. Gull et al. RMP (2011) Magnetism vs. Kondo resonance #1 Three impurity atoms with Hubbard #2 repulsion and exchange interaction U=2.4, J=-0.2 and J=0, β=64 0.6 DOS 0.4 0.2 0.0 -4 -2 0 2 4 Energy M. Crommie, PRL(2001) CT-QMC: single vs. trimer V. Savkin, et al, PRL 94, 026402 (2005) Equilateral and Isosceles Trimers Density of states at geometry modification of the trimer Equilateral (ET) and isosceles (IT) trimers J23=J, J12=J13=J/3 AFM V. Savkin et al, PRL 94, 026402 (2005) 0,7 0,7 0,6 0,6 ET 1-2-3 IT 1 IT 2-3 0,4 0,3 1 0,2 2 0,1 ET 1-2-3 IT 1 IT 2-3 0,5 DOS DOS 0,5 FM 1 ET 0,4 0,3 1 1 ET IT 0,2 3 2 2 0,1 3 3 2 3 0,0 0,0 -4 -3 -2 -1 0 Energy 1 2 3 4 -4 -3 -2 -1 0 1 Energy One can see a reconstruction of the Kondo resonance for isosceles trimer at antiferromagnetic exchange interaction 2 3 4 Hybridization function Co on/in Cu(111) • Hybridization of Co in bulk twice stronger than on surface • Hybridization in energy range of Cu-d orbitals more anisotropic on surface • Co-d occupancy: n= 7-8 B. Surer, et al PRB (2012) Orbitally resolved Co DOS from QMC Orbitally resolved DOS of the Co impurities in bulk Cu and on Co (111) obtained from QMC simulations at temperature. T = 0.025 eV and chemical potential μ = 27 eV and μ = 28 eV, respectively. All Co d-orbitals contribute to LDOS peak near EF=0 B. Surer, et al, PRB (2012). Magnetic susceptibility: nanosystems Bethe-Salpeter Equation: Susceptibility: Local correlated nano-system: U Spin and Charge susceptibility near impurity K. Patton, H. Hafermann,et.al PRB (2009) From Atom to Solid Atomic physics Bands effects (LDA) N (E) N (E) E n d | SL> F d n+ 1 E EF N(E) LHB QP EF UHB E LDA+DMFT E Spectral Function Fe: ARPES vs. DMFT SP-ARPES (BESY) J. Sánchez-Barriga, et al, PRL (2010) Magnetism of metals: LDA+DMFT l Exchange l Finite interactions in metals temperature 3d-metal magnetism 1,2 1,2 M(T) and χ(T): LDA+DMFT 1,0 M(T) 0,8 χ(T) Fe 0,6 0,8 0,6 Ni 0,4 0,4 0,2 0,2 0,0 0,0 0,0 0,2 0,4 0,6 0,8 1,0 1,2 1,4 1,6 1,8 2,0 T/Tc A. L., M. Katsnelson and G. Kotliar, PRL87, 067205 (2001) 2,2 χ-1Meff2/3Tc Global spin flip M(T)/M(0) 1,0 Interaction of electrons with collective excitations Magnon Plasmon Orbiton Non-local Coulomb interactions General non-local action for solids: Atomic action with local Hubbard-like interaction Bosonic charge and spin variables: A. Rubtsov et al, Annals of Physics 327, 1320 (2012) Efficient DB-perturbation theory Separate local and non-local effective actions: Imuprity action with fermionic and bosonic bathes (CT-QMC) F B Dual boson-fermion transformation: EDMFT Diagrams: Dual Boson: General Idea 2 + 1 ⌫ HTSC + ⇤! tk ] (14) ~ Vk ] Jττ ' Sτ ⋅ Sτ' g⌫ 1 1 (15) X⇣ ⌫ ⇤ † ! ⌘! f⌫+! f⌫ ! !0 ⌦ ⌫ ! 1 X + 4 0 † † ! f f 0 ⌫⌫ ⌫+! ⌫ 0 ! f⌫ f⌫ 0 DB-diagrammatic scheme ⌫⌫ ⌦ + ⌫⇤ † ⌦ ⌘! f⌫ f⌫+! Bosonic Selfenergy = ⌘ < c!+⌦ c!0 † † c ⌦ ! c! 0 >imp g!+⌦ g!0 < c⌫+! c⌫ a† ⇢! >imp = Renormalized vertex: g⌫ g⌫+! g! g! 0 ( ⌦ g! g! 0 + ... Fermionic Selfenergy ⌦) ⌦+! ! 0 < ⇢ >imp g⌫ ! ! ˜ k⌫ g⌫ ) 1 + ⌫ tk ] 1 Gk⌫ = [(g⌫ + g⌫ ⌃ ˜ k⌫ g⌫ ) 1 + ⌫ tk ] 1 Gk⌫ =![(g⌫ + g⌫ ⌃ ˜ q! ! ) 1 + ⇤! Vk ] 1 X = [( + ⇧ X q! ! ! 0 0 ⌫ 1 0 1 gX ! = ! ⌫⌫ 0 ! g⌫ 0 s ⌫ 0q!!= [( ! + ! ⇧ ˜ q! ! ) 1 + ⇤! Vk ] 1 0 0 X Fermionic and Bosonic⌫ Green Functions SCF-condition Gk⌫ = g⌫ X k 1 1 1 ˜ k⌫ g⌫ ) + ⌫ tG = )gg⌫⌫ )1 + ˜gk⌫ ˜gG⌫k⌫ Gk⌫ = [(g⌫ + g⌫ ⌃ ] = [(g+⌫ g+ ⌃ ⌃ + ⌫ k G= k⌫ [(g k⌫ Xq! = [( ! + ˜ ! ⇧q! ! ) 1 ⌫ ⌫ k⌫ ⌫ X Xq! = k ⌫ 1 1 X˜q! ⇧ ˜=q!!)! )1 + + ⇤! XVq! ]= kX = [( + + + ⇧ ⇤!⇤ q! [( ! ! ! X ! q! ! q q A. Rubtsov, M.I. Katsnelson, A. L., Annals of Phys. 327, 1320 (2012) XX ! X̃ ⇥ G̃ G̃ ⌃k ⇤+⌃,K+k † ⇤K − (0) < cω+Ω cω ρΩ >imp − < ρ> g δ imp ω Ω ⌃k⇤ λωΩ = X̃⇤K g g χ lattice , Simple Test: Hubbard (0) usceptibility” of the dual 1 system. X̃ ⇤ . ω ω+Ω Ω Next, we use t m in the typically equals zero as of < the ρ> imp vanishes. ding the nominator triangle vertices to the both ends ladder and usin " ... † U>>t # # # d λ are related, since (0) ρΩ = ω # sσσ cω⇤# σ cω −Ω,σ , ⌅ 1 ⇥ X̃⇤K ⇥ G̃⌃k G̃⇤+⌃,K+k 1 (0) ⌥⌥⌥↵↵ % ' X̃ ⌃k⇤ ⇤K ⌥⌥⌥↵↵↵⇥A triangle represents the ⇥ vertex⇥and square represents Figure A.7. Bosoinc dual self-energy in the⌅ladder approximation. the verte & ⇥ = + ⇧ Note that t ⇤in −1 ⇤K ⌥ ⌦ (0) ⇧ ⌃ # # # # λ = χ 1 − γ g g s Ωω σσ Ω 1ω,ω ,Ω⇤ X̃ω⇤Kω −Ω e dual system. Let us cons ω# ertices to the both ends of the ladder and using the exact relationof(1t behaviour Fermionic Selfenergy substitute formulas (26,28) for ⇧ and ⇥. in cancella 13 This results in this casebu sh rtant note that for⇤ the Gaussian impurity problem γ vanishes, ⌅ 1 ⇥ 1 (0) ⌥ pression ⌥⌥⌥↵↵↵ fermionic X̃⇤K and bosonic Green’s functions of ori t relations between ⌅ ⇤K ⇥⇤K = ⌥⌥⌥⇧↵↵⇥ 1 Gωk ⇥⌦ + ⇧⇥⇤⌅⌃ = . ⇥⇤ X̃ (0) ⇥⇤ . ⇤K ⇤K (0) X̃ = (∆ω − εk )⇤−1⇤K gω−1 G̃ωk gω−1 (∆ω − $k )−1 + (∆ω − $k ) om Eq.(28) has a meaning of an e⌅ective local Stoner parameter 2 s (26,28) for ⇧ and ⇥. This results of −1out⌅ of the tex −1 in cancellation −1 −1 JijΩ =− Generalization of = Anderson case Xcontribution (ΛΩ −superexchange Vto χΩnon-local X̃toΩkfrequency-dependent χΩ (Λ + (Λ Vk Ωk k ) the Ω − Vk ) with fluctuation self-energy ⇥ is desc U ⇤K not reproduce (0) 8 ⌅ A. Rubtsov, et al, Annals ⇥ Phys. = 327, ⇥ X̃1320⇥(2012) . ⇤K ⇤ ⇤K ⇤ from a single Introduction Plasmon mode in ladder DB Beyond DMFT Dual Bosons How Mott transition affect plasmon mode? Dual fermion susceptibility: Plasmon mode U=1.5 V=0.4 t=0.25 Beta=10 Square 64x64 lattice 6 5 E 4 3 2 Γ1 X M Γ 0 Γ X M Susceptibility using Dual Fermion: h⇢⇢iK !0 6= 0 Γ 0 0 0 0 0 0 0 0 0 0 0 Summary Magnetism of correlation systems can be well described in the LDA+DMFT scheme Local correlations efficiently included in CT-QMC impurity solver Imaginary Time and Matsubara space Constrain GW calculations of U Polarisation F. Aryasetiawanan et al PRB(2004) Double-Bethe Lattice: exact C-DMFT A. Ruckenstein PRB (1999) Self-consistent condition: CDMFT AF-between plane AF-plane Finite temperature phase diagram • order-disorder transition at t? / t=p2 for large U • MIT for intermediate U H. Hafermann, et al. EPL, 85, 37006 (2009) Density of States: large U Slater parametrization of U Multipole expansion: Coulomb matrix elements in Ylm basis: Angular part – 3j symbols Slater integrals: CT-HYB: General Interaction CT-HYB: Krylov code CT-QMC-Krylov: performance Satellite structure in Ni Ni: LDA+DMFT (T=0.9 Tc) 2.5 <S(τ)S(0)> -1 2.0 Density of states, eV nd in final states d-orbital spectral function n 9 full U 1.5 EF LDA 0.5 1.5 0.0 0 2 4 6 -1 band τ, eV 1.0 PES E =that6(?) eV satfind We one-particle 0.5 DMFT 0.0 8 (LDA) Eex =PES 0.3(0.6) eV WE = 0.3(0.6) 3(4) eV eV band = ex E 6(?) eV Eex= == 0.3(0.6) eV B. Valence-ba Wsat 3(4) eV 1.0 -8 -6 -4 -2 Energy, eV full U 1 G1 3 1 S 9 d →d 8 D P 3 F 0 2 s double-counting potential UH i LDA+DMFT+QMC only barely distinguishable. Fig ! PRL (2001) A. L., M. Katsnelson and G.P Kotliar, tral function Im m Gm! .E ! displayed result is relatively di main band (" 4 eV) as well as th identical to those obtained with thus share the same poor agree symmetry-resolved exchange s given directly by the selfenergy †eg " .EF / ! †eg T-Lanczos (5d+10k) † t2g " .EF / ! † t2g J. Kolorenc et al PRB (2012)