This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Thermoelectric transport and current noise through a multilevel Anderson impurity: Three-body Fermi-liquid corrections in quantum dots and magnetic alloys

Yoshimichi Teratani Department of Physics, Osaka City University, Sumiyoshi-ku, Osaka 558-8585, Japan Nambu Yoichiro Institute of Theoretical and Experimental Physics, Osaka Metropolitan University, Osaka 558-8585, Japan    Kazuhiko Tsutsumi Department of Physics, Osaka City University, Sumiyoshi-ku, Osaka 558-8585, Japan Nambu Yoichiro Institute of Theoretical and Experimental Physics, Osaka Metropolitan University, Osaka 558-8585, Japan    Kaiji Motoyama Department of Physics, Osaka City University, Sumiyoshi-ku, Osaka 558-8585, Japan Department of Physics, Osaka Metropolitan University, Sumiyoshi-ku, Osaka 558-8585, Japan    Rui Sakano Department of Physics, Keio University, 3-14-1 Hiyoshi, Kohoku-ku, Yokohama, Kanagawa 223-8522, Japan    Akira Oguri Department of Physics, Osaka City University, Sumiyoshi-ku, Osaka 558-8585, Japan Nambu Yoichiro Institute of Theoretical and Experimental Physics, Osaka Metropolitan University, Osaka 558-8585, Japan
Abstract

We present a comprehensive Fermi liquid description for thermoelectric transport and current noise, applicable to multilevel quantum dots (QD) and magnetic alloys (MA) without electron-hole or time-reversal symmetry. Our formulation for the low-energy transport is based on an Anderson model with NN discrete impurity levels, and is asymptotically exact at low energies, up to the next-leading order terms in power expansions with respect to temperature TT and bias voltage eVeV. The expansion coefficients can be expressed in terms of the Fermi liquid parameters, which include the three-body correlation functions defined with respect to the equilibrium ground state in addition to the linear susceptibilities and the occupation number NdN_{d} of impurity electrons. We apply this formulation to the SU(NN) symmetric QD and MA, and calculate the correlation functions for N=4N=4 and 66, using numerical renormalization group approach. The three-body correlations are shown to be determined by a single parameter over a wide range of electron fillings 1NdN11\lesssim N_{d}\lesssim N-1 for strong Coulomb interactions UU, and they also exhibit the plateau structures due to the SU(NN) Kondo effects at integer values of NdN_{d}. We find that the Lorenz number L=κ/(Tσ)L=\kappa/(T\sigma) for QD and MA, defined as the ratio of the thermal conductivity κ\kappa to the electrical conductivity σ\sigma, deviates from the universal Wiedemann-Franz value π2/(3e2)\pi^{2}/(3e^{2}) as the temperature increases from T=0T=0, showing the T2T^{2} dependence, the coefficient for which depends on the three-body correlations away from half filling. Furthermore, we find that the current noise for the SU(4) quantum dots and that for SU(6) show a pronounced difference at the quarter Nd/N=1/4N_{d}/N=1/4 and 3/43/4 fillings. In particular, the linear noise for N=4N=4 exhibits a flat peak while the peak for N=6N=6 shows a round shape, reflecting the fact that, at these filling points, the SU(NN) Kondo effects occur for N0N\equiv 0 (mod 44), whereas the intermediate-valence fluctuations occur for N2N\equiv 2 (mod 44). We also demonstrate the role of three-body correlations on the nonlinear current noise and the other transport coefficients.

pacs:
71.10.Ay, 71.27.+a, 72.15.Qm

I Introduction

The Kondo effect is a fascinating many-body effect[1, 2], taking place in dilute magnetic alloys (MA), quantum dots (QD), and other novel quantum systems such as ultracold atomic gases [3] and quark matter [4]. It was shown in the 1970s that the low-energy behavior of the Kondo systems can be described by a quantum impurity version of the Fermi liquid (FL) theory [5, 6, 7, 8, 9]. In particular, using the numerical renormalization group (NRG) approach [10, 11, 12], Wilson et al. demonstrated that the low-lying excited states of the Kondo and the Anderson models exhibit a one-to-one correspondence with the excitations of the renormalized quasiparticles.

The quasiparticles are asymptotically free in the equilibrium ground state, where the perturbations from the environment or reservoirs, which may depend on external parameters such as frequency ω\omega, temperature TT, bias voltages eVeV, etc., are absent. As these perturbations are adiabatically switched on, the quasiparticles capture the damping rate of order ω2\omega^{2}, T2T^{2}, and (eV)2(eV)^{2} through the residual interactions and this significantly affects the transport properties [5, 6, 7, 8, 9, 13, 14]. When the electron-hole or time-reversal symmetry is broken by a potential or external fields, the quasiparticles also capture the energy shift of order ω2\omega^{2}, T2T^{2}, and (eV)2(eV)^{2}, i.e., the corrections in the same order as the ones due to the damping rate. The contributions of these higher-order energy shifts, however, had not been fully understood until very recently.

It has recently been clarified that these higher-order energy shifts of the quasiparticles can be described exactly in terms of the three-body correlations between impurity electrons. The proof was given in two different ways, which complement each other. One is given by Mora et al. and Filippone et al. [15, 16, 17, 18], extending Nozières’ description [5] that is based on an invariance against the “floating of Kondo resonance on the Fermi sea”. The other is based on the higher-order Fermi liquid relations [19, 20, 21], which can be derived from the Ward identities for the second derivatives of the self-energy, extending Yamada-Yosida’s field-theoretical approach [6, 7, 8, 9]. These proofs enabled one to express the next-leading order terms of the transport coefficients in terms of three-body correlation functions, and these formulations have been applied to the nonlinear currents, current noise, and thermocurrent through quantum impurities without electron-hole or time-reversal symmetry [22, 23, 24, 25, 26, 27, 28, 29].

The purpose of this paper is twofold. The first one is to extend the latest version of the FL theory for treating the thermoelectric transport coefficients of multilevel quantum dots and magnetic alloys described by the Anderson model with NN arbitrary impurity levels. The second one is to demonstrate how the next-leading order terms of various transport coefficients vary with the impurity level ϵd\epsilon_{d} and Coulomb interaction UU. Specifically, for the second purpose, we consider quantum dots and magnetic alloys having an SU(NN) symmetry, and calculate the three-body correlation functions for N=4N=4 and 66 using the NRG approach. There have been a number of intensive works which theoretically studied low-energy transport: the nonlinear current [30, 31, 13, 32, 33, 34, 35, 14, 36, 37, 18], nonlinear current noise [38, 39, 40, 41, 16, 42, 23], and thermoelectric transport[43, 15, 44, 21]. However, the parameter space for quantum impurities is so huge that many parts are still left unexplored. In this paper, we explore the whole region of the electron fillings, 0NdN0\leq N_{d}\leq N, in which the occupation number NdN_{d} of electrons in the impurity levels varies continuously across the various SU(NN) Kondo and intermediate valence regimes.

One of the advantages of the Kondo systems realized in quantum dots is that the information about the many-body quantum states can be probed in such a highly tunable way [45, 46, 47, 48]. For instance, recent experiments have succeeded in directly probing the Kondo screening cloud [49]. Furthermore, low-energy Fermi liquid behaviors have experimentally been confirmed for nonequilibrium currents [50, 51], current noise [52, 53, 54, 55, 56], and themocurrent [57, 58]. Internal degrees of freedoms of quantum impurities also bring an interesting variety to the Kondo effect. The systems having the SU(44) symmetry have been realized, for instance, in the multiorbital semiconductor quantum dots and carbon nanotube (CNT) quantum dots [59], and have intensively been investigated theoretically [34, 60, 61, 62, 63, 64, 65, 15, 16, 66, 67, 26, 68, 22] and experimentally [69, 70, 71, 72, 73, 55]. Realization of the SU(NN) Kondo effects for various N>2N>2 has also been proposed by several authors, using triple quantum dots [74, 75] and CNT [76]. In this paper, we provide a comprehensive FL view of the low-energy transport for SU(4) and SU(6) symmetric QD and MA.

Our results reveal the fact that the three-body correlations exhibit plateau structures, caused by the SU(NN) Kondo effects occurring at integer fillings Nd=1N_{d}=1, 22, ,N1\ldots,N-1. We also calculate the Lorenz number L=κ/(Tσ)L=\kappa/(T\sigma) for QD and MA, defined as the ratio of the thermal conductivity κ\kappa to the electrical conductivity σ\sigma, and show how it deviates from the universal Wiedemann-Franz value π2/(3e2)\pi^{2}/(3e^{2}) as the temperature increases from T=0T=0. Furthermore, we demonstrate how the three-body correlations affect the order T2T^{2} electrical conductivity, the order T3T^{3} thermal conductivity, and the order (eV)3(eV)^{3} nonlinear current and current noise. At quarter Nd/N=1/4N_{d}/N=1/4 and three-quarters Nd/N=3/4N_{d}/N=3/4 fillings, the SU(NN) Kondo effects occur for N0N\equiv 0 (mod 44) while the intermediate valence fluctuations occur for N2N\equiv 2 (mod 44). We show that this dependence on (mod 44) causes a pronounced difference appearing in the peak structures of linear noise for N=4N=4 and 66.

This paper is organized as follows. In Sec. II, we describe the multilevel Anderson impurity model for quantum dots and magnetic alloys. Sections III and IV are devoted to the FL descriptions for the next-leading order terms of electrical and thermoelectric transport coefficients for quantum dots, applicable to arbitrary NN impurity-level structures. In Sec. V, the low-energy transport formulas for the SU(NN) quantum dots are described in terms of the five FL parameters. We present the NRG results for the three-body correlation functions in Sec. VI. The results for nonlinear current, current noise, and thermoelectric transport through quantum dots are discussed in Secs. VII and VIII. Section IX is devoted to the FL description of the three-body correlations in thermoelectric transport of dilute magnetic alloys. In Sec. X, we discuss the results for the electrical and thermal resistivities of the SU(4) and SU(6) magnetic alloys. Summary is given in Sec. XI. In Appendix, we provide details of the derivations for the transport formulas and additional NRG results for the FL parameters in the SU(NN) cases for N=2N=2, 44, and 66, for comparison.

II Model

We consider a multi-orbital Anderson impurity coupled to two noninteracting leads on the left (LL) and right (RR): =d+c+T\mathcal{H}=\mathcal{H}_{d}+\mathcal{H}_{c}+\mathcal{H}_{\mathrm{T}},

d=\displaystyle\mathcal{H}_{d}= σ=1Nϵdσndσ+12σσUσσndσndσ,\displaystyle\ \sum_{\sigma=1}^{N}\epsilon_{d\sigma}\,n_{d\sigma}+\frac{1}{2}\sum_{\sigma\neq\sigma^{\prime}}U_{\sigma\sigma^{\prime}}\,n_{d\sigma}n_{d\sigma^{\prime}}\;, (1)
c=\displaystyle\mathcal{H}_{c}= j=L,Rσ=1NDD𝑑ϵϵcϵjσcϵjσ,\displaystyle\ \sum_{j=L,R}\sum_{\sigma=1}^{N}\int_{-D}^{D}\!\!d\epsilon\,\epsilon\,c^{\dagger}_{\epsilon j\sigma}c_{\epsilon j\sigma}, (2)
T=\displaystyle\mathcal{H}_{\mathrm{T}}= j=L,Rσ=1Nvj(ψj,σdσ+dσψj,σ).\displaystyle\ -\sum_{j=L,R}\sum_{\sigma=1}^{N}v_{j}\left(\psi_{j,\sigma}^{\dagger}d_{\sigma}+d_{\sigma}^{{\dagger}}\psi_{j,\sigma}\right). (3)

Here, the level index runs over σ=1,2,,N\sigma=1,2,\ldots,N. The inter-electron interaction UσσU_{\sigma\sigma^{\prime}} generally depends on σ\sigma and σ\sigma^{\prime}, with the requirements Uσσ=UσσU_{\sigma^{\prime}\sigma}=U_{\sigma\sigma^{\prime}} for σσ\sigma^{\prime}\neq\sigma. For N=2N=2, it describes the usual single-orbital Anderson model for spin 1/21/2 fermions. The operator dσd^{{\dagger}}_{\sigma} creates an impurity electron with spin σ\sigma in the impurity level of energy ϵdσ\epsilon_{d\sigma}, and ndσdσdσn_{d\sigma}\equiv d^{{\dagger}}_{\sigma}d_{\sigma}. Conduction electrons in the two lead at j=Lj=L and RR obey the anti-commutation relation {cϵjσ,cϵjσ}=δjjδσσδ(ϵϵ)\{c^{\phantom{\dagger}}_{\epsilon j\sigma},c^{\dagger}_{\epsilon^{\prime}j^{\prime}\sigma^{\prime}}\}=\delta_{jj^{\prime}}\,\delta_{\sigma\sigma^{\prime}}\delta(\epsilon-\epsilon^{\prime}). The linear combination of the conduction electrons, ψj,σDD𝑑ϵρccϵjσ\psi_{j,\sigma}\equiv\int_{-D}^{D}d\epsilon\sqrt{\rho_{c}}\,c^{\phantom{\dagger}}_{\epsilon j\sigma} with ρc=1/(2D)\rho_{c}=1/(2D), couples to the impurity level. The bare level width due to the tunnel couplings is given by ΔΓL+ΓR\Delta\equiv\Gamma_{L}+\Gamma_{R} with Γj=πρcvj2\Gamma_{j}=\pi\rho_{c}v_{j}^{2}. We consider the parameter region where the half band-width DD is much grater than the other energy scales, Dmax(Uσσ,Δ,|ϵdσ|,T,|eV|)D\gg\max(U_{\sigma\sigma^{\prime}},\Delta,|\epsilon_{d\sigma}|,T,|eV|). We use a unit kB=1k_{B}=1 throughout this paper.

Low-energy properties of the Anderson model can be described in terms of a set of Fermi liquid parameters defined with respect to the equilibrium ground state, i.e., the phase shift δσ\delta_{\sigma}, the linear susceptibilities χσσ\chi_{\sigma\sigma^{\prime}} and the three-body correlation functions χσσσ′′[3]\chi_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]} between impurity electrons; see Appendix A for details. The phase shift is a primary parameter that reflects the charge distribution of impurity levels, through the Friedel sum rule: ndσT0δσ/π\langle n_{d\sigma}\rangle\xrightarrow{\,T\to 0\,}\delta_{\sigma}/\pi. In this paper, we explore the low-energy transport of quantum dots and magnetic alloys in the whole region of the electron fillings,

Ndσ=1Nndσ.\displaystyle N_{d}\,\equiv\,\sum_{\sigma=1}^{N}\langle n_{d\sigma}\rangle\,. (4)

The current conservation, which follows from the Heisenberg equation of motion for the occupation number ndσn_{d\sigma},

ndσt+J^R,σJ^L,σ= 0\displaystyle\frac{\partial n_{d\sigma}}{\partial t}+\widehat{J}_{R,\sigma}-\widehat{J}_{L,\sigma}\,=\,0\, (5)

also plays an essential role in the Fermi liquid description, through the Ward identities. Here, J^L,σ\widehat{J}_{L,\sigma} represents the current flowing from the left lead to the dot, and J^R,σ\widehat{J}_{R,\sigma} the current from the dot to the right lead:

J^L,σ=\displaystyle\widehat{J}_{L,\sigma}= ivL(ψLσdσdσψLσ),\displaystyle\ -i\,v_{L}\left(\psi^{\dagger}_{L\sigma}d_{\sigma}-d^{\dagger}_{\sigma}\psi_{L\sigma}\right), (6)
J^R,σ=\displaystyle\widehat{J}_{R,\sigma}= +ivR(ψRσdσdσψRσ).\displaystyle\ +i\,v_{R}\left(\psi^{\dagger}_{R\sigma}d_{\sigma}-d^{\dagger}_{\sigma}\psi_{R\sigma}\right). (7)

In the next section, we will give a brief overview of the low-energy expansion formulas for nonlinear current and current noise, obtained previously in Refs.  27, 29, to show how the formulas can be expressed in terms of the FL parameters, including the three-body correlations. This is for comparison with the formulas for thermal electric transport coefficients, which we extend to multilevel quantum dots and magnetic alloys in this paper. Specifically, we will describe the derivation of the formulas for multilevel quantum dots and magnetic alloys in Sec. IV and Appendix G, respectively.

III Fermi liquid description for nonlinear current noise of QD

We consider the nonequilibrium steady state under a finite bias voltage eVμLμReV\equiv\mu_{L}-\mu_{R}, applied between the two leads by setting the chemical potentials of the left and right leads to be μL\mu_{L} and μR\mu_{R}, respectively. The retarded Green’s function plays a central role in the microscopic description of the Fermi liquid transport:

Gσr(ω)=\displaystyle G_{\sigma}^{r}(\omega)= i0𝑑tei(ω+i0+)t{dσ(t),dσ}V,\displaystyle\ -i\int_{0}^{\infty}dt\,e^{i(\omega+i0^{+})t}\,\Bigl{\langle}\,\Bigl{\{}d_{\sigma}(t),\,d_{\sigma}^{\dagger}\Bigr{\}}\Bigr{\rangle}_{V}\,, (8)
Aσ(ω)\displaystyle A_{\sigma}(\omega)\,\equiv 1πImGσr(ω).\displaystyle\ -\frac{1}{\pi}\,\mathrm{Im}\,G_{\sigma}^{r}(\omega)\;. (9)

Here, V\langle\cdots\rangle_{V} represents a nonequilibrium steady-state average taken with the statistical density matrix, which is constructed at finite bias voltages eVeV and temperatures TT, using the Keldysh formalism [33, 13, 77].

III.1 Differential conductance dJ/dVdJ/dV

The nonequilibrium current JJ through quantum dots can be expressed in terms of the spectral function Aσ(ω)A_{\sigma}(\omega) [33, 13]:

J=\displaystyle J\,= ehσ𝑑ω[fL(ω)fR(ω)]𝒯σ(ω),\displaystyle\ \frac{e}{h}\sum_{\sigma}\int_{-\infty}^{\infty}\!d\omega\,\bigl{[}\,f_{L}(\omega)-f_{R}(\omega)\,\bigr{]}\,\mathcal{T}_{\sigma}(\omega), (10)
𝒯σ(ω)=\displaystyle\mathcal{T}_{\sigma}(\omega)\,= 4ΓLΓRΓL+ΓRπAσ(ω).\displaystyle\ \frac{4\Gamma_{L}\Gamma_{R}}{\Gamma_{L}+\Gamma_{R}}\,\pi A_{\sigma}(\omega)\,. (11)

Here, fj(ω)f(ωμj)f_{j}(\omega)\equiv f(\omega-\mu_{j}) for j=Lj=L, RR, with f(ω)=[eω/T+1]1f(\omega)=[e^{\omega/T}+1]^{-1} the Fermi function. Specifically, in this paper, we consider the case where the chemical potentials are applied in a symmetric way:

μL=μR12eV,\displaystyle\mu_{L}\,=\,-\mu_{R}\ \equiv\ \frac{1}{2}\,eV\,, (12)

choosing the Fermi level at equilibrium to be the origin of one-particle energies EF=0E_{F}=0. Note that the role of bias and tunneling asymmetries, (μL+μR)/2EF(\mu_{L}+\mu_{R})/2\neq E_{F} and ΓLΓR\Gamma_{L}\neq\Gamma_{R}, has been precisely discussed in Refs. 28, 29.

The nonlinear current JJ can be expanded up to the next-leading order terms, using the low-energy asymptotic form of the spectral function Aσ(ω)A_{\sigma}(\omega), which has been obtained up to terms of order ω2\omega^{2}, (eV)2(eV)^{2}, and T2T^{2} [27], as shown in Appendix B. In particular, for symmetric junctions with ΓL=ΓR\Gamma_{L}=\Gamma_{R} and μL=μR=eV/2\mu_{L}=-\mu_{R}=eV/2, the low-energy expansion of the differential conductance takes the form,

dJdV=e2hσ=1N[sin2δσcT,σ(πT)2cV,σ(eV)2+].\displaystyle\frac{dJ}{dV}=\frac{e^{2}}{h}\sum_{\sigma=1}^{N}\left[\,\sin^{2}\delta_{\sigma}-c_{T,\sigma}\left(\pi T\right)^{2}-c_{V,\sigma}\left(eV\right)^{2}\,+\,\cdots\right]. (13)

Here, the coefficients cT,σc_{T,\sigma} and cV,σc_{V,\sigma} of the next-leading order terms can be expressed in terms of the phase shift δσ\delta_{\sigma}, the linear susceptibilities χσσ\chi_{\sigma\sigma^{\prime}}, and the static three-body correlation functions χσσσ′′[3]\chi_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]} defined in Appendix A:

cT,σ=π23[(χσσ2+2σ(σ)χσσ2)cos2δσ\displaystyle\!\!\!\!c_{T,\sigma}=\frac{\pi^{2}}{3}\Biggl{[}\,-\biggl{(}\chi_{\sigma\sigma}^{2}+2\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}}^{2}\biggr{)}\cos 2\delta_{\sigma}
+(χσσσ[3]+σ(σ)χσσσ[3])sin2δσ2π],\displaystyle\qquad\qquad\,+\biggl{(}\chi_{\sigma\sigma\sigma}^{[3]}+\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\biggr{)}\frac{\sin 2\delta_{\sigma}}{2\pi}\,\Biggr{]}, (14)
cV,σ=π24[(χσσ2+5σ(σ)χσσ2)cos2δσ\displaystyle\!\!\!\!c_{V,\sigma}=\frac{\pi^{2}}{4}\Biggl{[}\,-\biggl{(}\chi_{\sigma\sigma}^{2}+5\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}}^{2}\biggr{)}\cos 2\delta_{\sigma}
+(χσσσ[3]+3σ(σ)χσσσ[3])sin2δσ2π].\displaystyle\qquad\qquad+\biggl{(}\chi_{\sigma\sigma\sigma}^{[3]}+3\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\biggr{)}\frac{\sin 2\delta_{\sigma}}{2\pi}\,\Biggr{]}. (15)

III.2 Nonlinear current noise SnoiseQDS_{\mathrm{noise}}^{\mathrm{QD}}

We also study the current noise [78, 26, 27], defined by

SnoiseQD=e2σσ𝑑tδJ^σ(t)δJ^σ(0)+δJ^σ(0)δJ^σ(t)V.\displaystyle S_{\mathrm{noise}}^{\mathrm{QD}}=\,e^{2}\sum_{\sigma\sigma^{\prime}}\int_{-\infty}^{\infty}\!\!dt\,\left\langle\delta\widehat{J}_{\sigma}(t)\,\delta\widehat{J}_{\sigma^{\prime}}(0)+\delta\widehat{J}_{\sigma^{\prime}}(0)\,\delta\widehat{J}_{\sigma}(t)\right\rangle_{V}. (16)

Here, δJ^σ(t)J^σ(t)J^σ(0)V\delta\widehat{J}_{\sigma}(t)\equiv\widehat{J}_{\sigma}(t)-\bigl{\langle}\widehat{J}_{\sigma}(0)\bigr{\rangle}_{V} represents fluctuations of the symmetrized current:

J^σ\displaystyle\widehat{J}_{\sigma}\equiv ΓLJ^R,σ+ΓRJ^L,σΓL+ΓR.\displaystyle\ \frac{\Gamma_{L}\widehat{J}_{R,\sigma}+\Gamma_{R}\widehat{J}_{L,\sigma}}{\Gamma_{L}+\Gamma_{R}}\;. (17)

Behavior of the current noise in the low-energy Fermi liquid regime has been studied by several authors, taking into account the three-body correlations [17, 16]. In a previous work, we have derived a general formula for the current noise through the multilevel Anderson impurity model up to terms of order |eV|3|eV|^{3} for symmetric junctions ΓL=ΓR\Gamma_{L}=\Gamma_{R} and μL=μR=eV/2\mu_{L}=-\mu_{R}=eV/2 [26, 27]:

SnoiseQD=2e2h|eV|σ=1N[sin22δσ4+cS,σ(eV)2+].\displaystyle S_{\mathrm{noise}}^{\mathrm{QD}}=\frac{2e^{2}}{h}|eV|\sum_{\sigma=1}^{N}\left[\frac{\sin^{2}2\delta_{\sigma}}{4}\,+c_{S,\sigma}\left(eV\right)^{2}+\cdots\right]. (18)

The coefficient cS,σc_{S,\sigma} for the next-leading order term has been calculated by taking into account all components of the Keldysh vertex function together with the low-energy asymptotic form of Aσ(ω)A_{\sigma}(\omega). It has been shown to be expressed in the following form,

cS,σ=\displaystyle\!\!c_{S,\sigma}= π212[cos4δσχσσ2+(2+3cos4δσ)σ(σ)χσσ2\displaystyle\ \frac{\pi^{2}}{12}\Biggl{[}\,\cos 4\delta_{\sigma}\,\chi_{\sigma\sigma}^{2}+\bigl{(}2+3\cos 4\delta_{\sigma}\bigr{)}\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}}^{2}
+ 4σ(σ)cos2δσcos2δσχσσ2\displaystyle\quad\ \ +\,4\sum_{\sigma^{\prime}(\neq\sigma)}\cos 2\delta_{\sigma}\cos 2\delta_{\sigma^{\prime}}\,\chi_{\sigma\sigma^{\prime}}^{2}
+ 3σ(σ)σ′′(σ,σ)sin2δσsin2δσχσσ′′χσσ′′\displaystyle\quad\ \ +\,3\sum_{\sigma^{\prime}(\neq\sigma)}\sum_{\sigma^{\prime\prime}(\neq\sigma,\sigma^{\prime})}\sin 2\delta_{\sigma}\,\sin 2\delta_{\sigma^{\prime}}\chi_{\sigma\sigma^{\prime\prime}}\chi_{\sigma^{\prime}\sigma^{\prime\prime}}
(χσσσ[3]+3σ(σ)χσσσ[3])sin4δσ4π].\displaystyle\quad\ \ -\,\biggl{(}\chi_{\sigma\sigma\sigma}^{[3]}+3\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\biggr{)}\,\frac{\sin 4\delta_{\sigma}}{4\pi}\,\Biggr{]}. (19)

IV Fermi liquid description for thermoelectric transport of QD

In this section, we discuss thermoelectric transport through multilevel quantum dots in the linear-response regime, the low-energy behaviors of which can be deduced from the asymptotic form of the spectral function Aσ(ω)A_{\sigma}(\omega) given in Appendix B.

The linear conductance gg, thermopower 𝒮QD\mathcal{S}_{\mathrm{QD}} and thermal conductance κQD\kappa_{\mathrm{QD}} of a quantum dot can be expressed in the form [79, 80, 81, 22, 82]:

g\displaystyle g\,\equiv dJdV|eV=0=e2hσ0,σQD,\displaystyle\ \left.\frac{dJ}{dV}\right|_{eV=0}\ =\ \frac{e^{2}}{h}\sum_{\sigma}\mathcal{L}_{0,\sigma}^{\mathrm{QD}}\,, (20)
𝒮QD=\displaystyle\mathcal{S}_{\mathrm{QD}}\,= 1|e|Tσ1,σQDσ0,σQD,\displaystyle\ \frac{-1}{|e|T}\frac{\sum_{\sigma}\mathcal{L}_{1,\sigma}^{\mathrm{QD}}}{\sum_{\sigma}\mathcal{L}_{0,\sigma}^{\mathrm{QD}}}\,, (21)
κQD=\displaystyle\kappa_{\mathrm{QD}}\,= 1hT[σ2,σQD(σ1,σQD)2σ0,σQD].\displaystyle\ \frac{1}{h\,T}\left[\,\sum_{\sigma}\mathcal{L}_{2,\sigma}^{\mathrm{QD}}-\frac{\left(\sum_{\sigma}\mathcal{L}_{1,\sigma}^{\mathrm{QD}}\right)^{2}}{\sum_{\sigma}\mathcal{L}_{0,\sigma}^{\mathrm{QD}}}\,\right]\,. (22)

Here, n,σQD\mathcal{L}_{n,\sigma}^{\mathrm{QD}} for n=0,1n=0,1, and 22 is defined at eV=0eV=0 with respect to the thermal equilibrium, as

n,σQD=𝑑ωωn𝒯σ(ω)(f(ω)ω),\displaystyle\mathcal{L}_{n,\sigma}^{\mathrm{QD}}=\int_{-\infty}^{\infty}d\omega\,\omega^{n}\,\mathcal{T}_{\sigma}(\omega)\,\left(-\frac{\partial f(\omega)}{\partial\omega}\right), (23)

with 𝒯σ(ω)\mathcal{T}_{\sigma}(\omega), the transmission probability defined in Eq. (11). Note that thermal conductance κQD\kappa_{\mathrm{QD}} is the linear-response coefficient of the heat current JQ=κQDδTJ_{Q}=\kappa_{\mathrm{QD}}\,\delta T, flowing from the high-temperature side toward the low-temperature side, with δT\delta T the temperature difference between the two sides.

The thermoelectric coefficients n,σQD\mathcal{L}_{n,\sigma}^{\mathrm{QD}} can be calculated, by substituting the low-energy asymptotic form of Aσ(ω)A_{\sigma}(\omega), given in Eq. (111) into 𝒯σ(ω)\mathcal{T}_{\sigma}(\omega). At low temperatures, the component 0,σQD\mathcal{L}_{0,\sigma}^{\mathrm{QD}} for n=0n=0, which determines gg, is given by

0,σQD=\displaystyle\mathcal{L}_{0,\sigma}^{\mathrm{QD}}\,= sin2δσcT,σ(πT)2+O(T4).\displaystyle\ \sin^{2}\delta_{\sigma}\,-\,c_{T,\sigma}\,\left(\pi T\right)^{2}\,+\,O(T^{4})\,. (24)

Here, cT,σc_{T,\sigma} is the coefficient that we have already described in Eq. (14). The next component, 1,σQD\mathcal{L}_{1,\sigma}^{\mathrm{QD}} for n=1n=1, takes the form

1,σQD=\displaystyle\mathcal{L}_{1,\sigma}^{\mathrm{QD}}\,= πΔ3ρdσ(πT)2+O(T4).\displaystyle\ \frac{\pi\Delta}{3}\rho_{d\sigma}^{\prime}\left(\pi T\right)^{2}\,+\,O(T^{4})\,. (25)

Here, ρdσ\rho_{d\sigma}^{\prime} is the derivative of the density of states with respect to the frequency ω\omega, which can also be written in terms of the phase shift δσ\delta_{\sigma} and diagonal linear susceptibility χσσ\chi_{\sigma\sigma}, as shown in Eq. (101). Thus, the leading-order term of thermopower for quantum dots is given by

𝒮QD=\displaystyle\mathcal{S}_{\mathrm{QD}}\,= π23σρdσσρdσT|e|+O(T3).\displaystyle\ -\frac{\pi^{2}}{3}\,\frac{\sum_{\sigma}\rho_{d\sigma}^{\prime}}{\sum_{\sigma}\rho_{d\sigma}}\,\frac{T}{|e|}\,+\,O(T^{3})\,. (26)

The thermal conductance κQD\kappa_{\mathrm{QD}} depends on the other component 2,σQD\mathcal{L}_{2,\sigma}^{\mathrm{QD}} for n=2n=2, the low-energy asymptotic form of which is given by

2,σQD=\displaystyle\!\!\mathcal{L}_{2,\sigma}^{\mathrm{QD}}\,= (πT)23[sin2δσ+a2,σQD(πT)2+O(T4)],\displaystyle\ \frac{\left(\pi T\right)^{2}}{3}\Bigl{[}\,\sin^{2}\delta_{\sigma}\,\,+\,a_{2,\sigma}^{\mathrm{QD}}\left(\pi T\right)^{2}\,+\,O(T^{4})\,\Bigr{]},
a2,σQD\displaystyle\!\!a_{2,\sigma}^{\mathrm{QD}}\,\equiv 7π25[cos2δσ(χσσ2+67σ(σ)χσσ2)\displaystyle\ \frac{7\pi^{2}}{5}\Biggl{[}\,\cos 2\delta_{\sigma}\left(\chi_{\sigma\sigma}^{2}+\frac{6}{7}\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}}^{2}\right)
sin2δσ2π(χσσσ[3]+521σ(σ)χσσσ[3])].\displaystyle-\frac{\sin 2\delta_{\sigma}}{2\pi}\,\left(\chi_{\sigma\sigma\sigma}^{[3]}+\frac{5}{21}\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\right)\,\Biggr{]}. (27)

Therefore, the thermal conductance can be calculated up to terms of order T3T^{3}, by substituting these asymptotic forms into Eq. (22):

κQD=π2T3hσ=1N[sin2δσcκ,σQD(πT)2+],\displaystyle\!\!\!\kappa_{\mathrm{QD}}\,=\,\frac{\pi^{2}\,T}{3\,h}\,\sum_{\sigma=1}^{N}\left[\,\sin^{2}\delta_{\sigma}\,-\,c_{\kappa,\sigma}^{\mathrm{QD}}\,\left(\pi T\right)^{2}\,+\,\cdots\right], (28)
cκ,σQD=a2,σQD+π23(1Nσ′′χσ′′σ′′sin 2δσ′′)2(sin2δ¯)AM.\displaystyle\!\!\!c_{\kappa,\sigma}^{\mathrm{QD}}\,=\,-a_{2,\sigma}^{\mathrm{QD}}\,+\,\frac{\pi^{2}}{3}\,\frac{\left(\displaystyle\mathstrut\frac{1}{N}\sum_{\sigma^{\prime\prime}}\,\chi_{\sigma^{\prime\prime}\sigma^{\prime\prime}}\,\sin\,2\delta_{\sigma^{\prime\prime}}\right)^{2}}{\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{AM}}}\,. (29)

Here, the arithmetic mean (AM) of the phase shifts is defined by

(sin2δ¯)AM1Nσsin2δσ.\displaystyle\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{AM}}\,\equiv\,\frac{1}{N}\,\sum_{\sigma}\,\sin^{2}\delta_{\sigma}\,. (30)

Furthermore, the Lorenz number LQDκQD/(gT)L_{\mathrm{QD}}\equiv\kappa_{\mathrm{QD}}/(g\,T) can be calculated up to terms of order T2T^{2}:

LQD=\displaystyle L_{\mathrm{QD}}\,= π23e2[ 1cLQD(sin2δ¯)AM(πT)2+O(T4)],\displaystyle\ \frac{\pi^{2}}{3\,e^{2}}\,\left[\,1\,-\,\frac{c_{L}^{\mathrm{QD}}}{\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{AM}}}\,\left(\pi T\right)^{2}\ +\ O(T^{4})\,\right], (31)
cLQD=\displaystyle c_{L}^{\mathrm{QD}}\,= 1Nσ(cκ,σQDcT,σ).\displaystyle\ \frac{1}{N}\,\sum_{\sigma}\left(c_{\kappa,\sigma}^{\mathrm{QD}}\,-\,c_{T,\sigma}\right)\,. (32)

The Wiedemann-Franz law holds between the leading-order terms of the linear conductance gg and the thermal conductance κQD\kappa_{\mathrm{QD}}, such that the Lorenz number approaches the universal value LQDT0π2/(3e2)L_{\mathrm{QD}}\xrightarrow{T\to 0}\pi^{2}/(3e^{2}) in the low-temperature limit. The Lorenz number deviates from this universal value as temperature increases, exhibiting the T2T^{2} dependence.

V Three-body correlations in the SU(NN) symmetric Fermi liquid

This Hamiltonian \mathcal{H}, defined in Sec. II, has an SU(NN) symmetry in the case at which the impurity states become degenerate ϵdσϵd\epsilon_{d\sigma}\equiv\epsilon_{d} for all σ\sigma and the Coulomb interaction is isotropic UσσUU_{\sigma\sigma^{\prime}}\equiv U for all σ\sigma and σ\sigma^{\prime}.

In the atomic limit vj0v_{j}\to 0 of the SU(NN) symmetric case, the total number of impurity electrons NdN_{d} takes an integer value MM and exhibits the Coulomb-stair case behavior as a function of ϵd\epsilon_{d}. It consists of a series of plateaus of the width UU and the height Nd=MN_{d}=M for M=1M=1, 22, \ldots, N1N-1, emerging at MU<ϵd<(M1)U-MU<\epsilon_{d}<-(M-1)U around the midpoint ϵd,Mmid(M12)U\epsilon_{d,M}^{\mathrm{mid}}\equiv-(M-\frac{1}{2})U. We will use the following notation for the shifted impurity level ξd\xi_{d}, which includes the Hartree-Fock energy shift defined with respect to the half filling in such a way that

ξdϵd+N12U.\displaystyle\xi_{d}\,\equiv\,\epsilon_{d}\,+\,\frac{N-1}{2}\,U\,. (33)

The system has an electron-hole symmetry at ξd=0\xi_{d}=0. When the tunneling couplings are switched on, the stair-case structure emerges for strong interactions UΔU\gg\Delta.

In the SU(NN) symmetric case, the linear susceptibilities have two linearly independent components, i.e., the diagonal one χσσ\chi_{\sigma\sigma} and the off-diagonal one χσσ\chi_{\sigma\sigma^{\prime}} for σσ.\sigma\neq\sigma^{\prime}. These two components determine the essential properties of quasiparticles:

T14χσσ,R 1χσσχσσ.\displaystyle T^{\ast}\,\equiv\,\frac{1}{4\chi_{\sigma\sigma}}\,,\qquad\quad R\,\equiv\,1-\frac{\chi_{\sigma\sigma^{\prime}}}{\chi_{\sigma\sigma}}\,. (34)

Here, TT^{\ast} is a characteristic energy scale of the SU(NN) Fermi liquid, by which the TT-linear specific heat of impurity electrons can be expressed in the form 𝒞imp=Nπ212T/T\mathcal{C}_{\mathrm{imp}}=\frac{N\pi^{2}}{12}T/T^{*} [6, 8, 9]. The Wilson ratio RR corresponds to a dimensionless residual interaction between quasiparticles [83]; we will also use the rescaled ratio,

K~(N1)(R1),\displaystyle\widetilde{K}\,\equiv\,(N-1)(R-1)\,, (35)

which is bounded in the range 0K~10\leq\widetilde{K}\leq 1.

V.1 Charge and spin susceptibilities

The charge susceptibility is given by a linear combination of the two-body correlations,

χ¯C1N2Ωϵd2=1Nσ1σ2χσ1σ2\displaystyle\!\!\!\!\!\!\overline{\chi}_{C}\,\equiv\,-\frac{1}{N}\frac{\partial^{2}\Omega}{\partial\epsilon_{d}^{2}}\ =\ \frac{1}{N}\sum_{\sigma_{1}\sigma_{2}}\chi_{\sigma_{1}\sigma_{2}} (36)
SU(N)χσσ+(N1)χσσ=1K~4T.\displaystyle\xrightarrow{\,\mathrm{SU}(N)\,}\,\chi_{\sigma\sigma}+(N-1)\chi_{\sigma\sigma^{\prime}}\ =\ \frac{1-\widetilde{K}}{4T^{*}}\,\,. (37)

Here, ΩTlog[Tre/T]\Omega\equiv-T\log\left[\mathrm{Tr}\,e^{-\mathcal{H}/T}\right] is the free energy.

Next, we consider the spin susceptibility for the SU(NN) symmetric case, using the notation in which the internal degrees freedom are separated into two parts σ=(m,s)\sigma=(m,s), where m=1,2,,N/2m=1,2,\ldots,N/2 and s=,s=\uparrow,\downarrow, assuming NN to be even (extension to odd NN is straightforward). The Zeeman splitting is induced by an external field bb, which couples to the impurity spin ss:

ϵd,m,=ϵdb,ϵd,m,=ϵd+b.\displaystyle\epsilon_{d,m,\uparrow}=\epsilon_{d}\,-\,b\,,\qquad\epsilon_{d,m,\downarrow}=\epsilon_{d}\,+\,b\,. (38)

The magnetization \mathcal{M} of the impurity spin is given by

\displaystyle\mathcal{M}\,\equiv 1NΩb=1Nm=1N2nd,mnd,m.\displaystyle\ -\frac{1}{N}\frac{\partial\Omega}{\partial b}\,=\,\frac{1}{N}\,\sum_{m=1}^{\frac{N}{2}}\left\langle n_{d,m\uparrow}-n_{d,m\downarrow}\right\rangle\,. (39)

Note that Ω\Omega is an even function of bb. The spin susceptibility χ¯S(1/N)2Ωb2|b=0\overline{\chi}_{S}\equiv-(1/N)\frac{\partial^{2}\Omega}{\partial b^{2}}\big{|}_{b=0} can be expressed in the following form,

χ¯S=1Nm=1N2m=1N2[χm,m+χm,m\displaystyle\!\!\overline{\chi}_{S}\,\,=\,\frac{1}{N}\sum_{m=1}^{\frac{N}{2}}\sum_{m^{\prime}=1}^{\frac{N}{2}}\biggl{[}\,\chi_{m\uparrow,m^{\prime}\uparrow}+\chi_{m\downarrow,m^{\prime}\downarrow}
χm,mχm,m]\displaystyle\qquad\qquad\qquad\qquad-\chi_{m\uparrow,m^{\prime}\downarrow}-\chi_{m\downarrow,m^{\prime}\uparrow}\,\biggr{]} (40)
SU(N)χσσχσσ=14T(1+K~N1),\displaystyle\xrightarrow{\,\mathrm{SU}(N)\,}\,\chi_{\sigma\sigma}-\chi_{\sigma\sigma^{\prime}}\ =\ \frac{1}{4T^{*}}\left(1+\frac{\widetilde{K}}{N-1}\right), (41)

where σσ\sigma\neq\sigma^{\prime}.

V.2 Three-body correlation functions

Among N3N^{3} components of the three-body correlation χσ1σ2σ3[3]\chi_{\sigma_{1}\sigma_{2}\sigma_{3}}^{[3]}, only three components become linearly independent in the SU(NN) symmetric case. They can be expressed in terms of the derivatives of the linear susceptibilities, using Eqs. (113), (114), and (117) given in Appendix C:

χσσσ[3]=\displaystyle\chi_{\sigma\sigma\sigma}^{[3]}\,= 1NχσσϵdN1NχB[3],\displaystyle\ \frac{1}{N}\frac{\partial\chi_{\sigma\sigma}}{\partial\epsilon_{d}}-\frac{N-1}{N}\,\chi_{B}^{[3]}\;, (42)
χ~σσσ[3]=\displaystyle\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\,= N1Nχσσϵd+N1NχB[3],\displaystyle\ \frac{N-1}{N}\frac{\partial\chi_{\sigma\sigma}}{\partial\epsilon_{d}}+\frac{N-1}{N}\,\chi_{B}^{[3]}\;, (43)
χ~σσσ′′[3]=\displaystyle\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]}\,= N1Nχσσϵd+N12χσσϵdN1NχB[3],\displaystyle\ -\frac{N-1}{N}\frac{\partial\chi_{\sigma\sigma}}{\partial\epsilon_{d}}\,+\,\frac{N-1}{2}\frac{\partial\chi_{\sigma\sigma^{\prime}}}{\partial\epsilon_{d}}\,-\,\frac{N-1}{N}\,\chi_{B}^{[3]}, (44)

for σσσ′′σ\sigma\neq\sigma^{\prime}\neq\sigma^{\prime\prime}\neq\sigma. Here,

χB[3]\displaystyle\chi_{B}^{[3]}\,\equiv b(χm,mχm,m2)|b=0\displaystyle\ \frac{\partial}{\partial b}\left.\left(\frac{\chi_{m\uparrow,m\uparrow}-\chi_{m\downarrow,m\downarrow}}{2}\right)\right|_{b=0}
=\displaystyle= χσσσ[3]+χσσσ[3],\displaystyle\ -\chi_{\sigma\sigma\sigma}^{[3]}+\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\,,\rule{0.0pt}{17.07182pt} (45)

and the scale factors (N1)(N-1) and (N1)(N2)/2(N-1)(N-2)/2 have been introduced for the off-diagonal three-body components in such a way that

χ~σσσ[3]\displaystyle\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\, (N 1)χσσσ[3],\displaystyle\equiv\,\left(N\,-\,1\right)\,\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\,, (46)
χ~σσσ′′[3]\displaystyle\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]}\, (N 1)(N 2)2χσσσ′′[3].\displaystyle\equiv\,\frac{\left(N\,-\,1\right)\left(N\,-\,2\right)}{2}\,\chi_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]}\,. (47)

We will also use the dimensionless three-body correlations, defined by

ΘI\displaystyle\!\!\!\!\!\!\!\Theta_{\mathrm{I}}\,\equiv sin2δ2πχσσσ[3]χσσ2,Θ~IIsin2δ2πχ~σσσ[3]χσσ2,\displaystyle\ \frac{\sin 2\delta}{2\pi}\,\frac{\chi_{\sigma\sigma\sigma}^{[3]}}{\chi_{\sigma\sigma}^{2}}\,,\qquad\widetilde{\Theta}_{\mathrm{II}}\,\equiv\ \frac{\sin 2\delta}{2\pi}\,\frac{\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}}{\chi_{\sigma\sigma}^{2}}\,, (48)
Θ~III\displaystyle\widetilde{\Theta}_{\mathrm{III}}\,\equiv sin2δ2πχ~σσσ′′[3]χσσ2.\displaystyle\ \frac{\sin 2\delta}{2\pi}\,\frac{\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]}}{\chi_{\sigma\sigma}^{2}}\,. (49)

In this work, we calculate the right-hand side of Eqs.  (42)–(44) with the NRG to obtain the three-body correlations for the SU(NN) case. Note that, for noninteracting electrons at U=0U=0, only the diagonal components of the three-body correlation ΘI0\Theta_{\mathrm{I}}^{0} and the susceptibility χσσ0\chi_{\sigma\sigma}^{0} remain finite due to the Pauli exclusion principle,

ΘI0=2ϵd2ϵd2+Δ2,χσσ0=1πΔϵd2+Δ2,\displaystyle\Theta_{\mathrm{I}}^{0}\,=\,\frac{-2\epsilon_{d}^{2}}{\epsilon_{d}^{2}+\Delta^{2}}\,,\qquad\qquad\chi_{\sigma\sigma}^{0}\,=\,\frac{1}{\pi}\frac{\Delta}{\epsilon_{d}^{2}+\Delta^{2}}, (50)

and TU0πΔ[ 1+(ϵd/Δ)2]/4T^{*}\xrightarrow{\,U\to 0\,}\pi\Delta\bigl{[}\,1+(\epsilon_{d}/\Delta)^{2}\,\bigr{]}/4.

Table 1: Low-energy expansion of transport coefficients through SU(NN) quantum dots (QD) and magnetic alloys (MA), described in Eqs. (51)–(54) and Eqs. (81)–(83). Here, T1/(4χσσ)T^{*}\equiv 1/(4\chi_{\sigma\sigma}) is a characteristic FL energy scale.
    dJdV=Ne2h[sin2δCT(πTT)2CV(eVT)2+]\frac{dJ}{dV}\,=\,\frac{Ne^{2}}{h}\left[\,\sin^{2}\delta-C_{T}\left(\frac{\pi T}{T^{*}}\right)^{2}-C_{V}\left(\frac{eV}{T^{*}}\right)^{2}+\cdots\right]
    SnoiseQD=2Ne2|eV|h[sin2δ(1sin2δ)+CS(eVT)2+]S_{\mathrm{noise}}^{\mathrm{QD}}=\frac{2Ne^{2}|eV|}{h}\left[\,\sin^{2}\delta\,\bigl{(}1-\sin^{2}\delta\bigr{)}\,+C_{S}\left(\frac{eV}{T^{*}}\right)^{2}+\cdots\right]
    κQD=Nπ2T3h[sin2δCκQD(πTT)2+]\kappa_{\mathrm{QD}}\,=\,\frac{N\pi^{2}T}{3h}\,\left[\,\sin^{2}\delta\,-C_{\kappa}^{\mathrm{QD}}\,\left(\frac{\pi T}{T^{*}}\right)^{2}+\cdots\right]
    LQDκQDgT=π23e2[1CLQDsin2δ(πTT)2+]L_{\mathrm{QD}}\,\equiv\,\frac{\kappa_{\mathrm{QD}}}{g\,T}\,=\,\frac{\pi^{2}}{3e^{2}}\,\left[1\,-\,\frac{C_{L}^{\mathrm{QD}}}{\sin^{2}\delta}\,\left(\frac{\pi T}{T^{*}}\right)^{2}\,+\,\cdots\right]
    ϱMA1σMA=1σMAunit[sin2δCϱMA(πTT)2+]\varrho_{\mathrm{MA}}\,\equiv\,\frac{1}{\sigma_{\mathrm{MA}}}\,=\,\frac{1}{\sigma_{\mathrm{MA}}^{\mathrm{unit}}}\,\left[\,\sin^{2}\delta\,-\,C_{\varrho}^{\mathrm{MA}}\left(\frac{\pi T}{T^{*}}\right)^{2}+\cdots\,\right]
    1κMA=3e2π2σMAunitT[sin2δCκMA(πTT)2+]\frac{1}{\kappa_{\mathrm{MA}}}\,=\,\frac{3\,e^{2}}{\pi^{2}\,\sigma_{\mathrm{MA}}^{\mathrm{unit}}\,T}\left[\,\sin^{2}\delta-C_{\kappa}^{\mathrm{MA}}\,\left(\frac{\pi T}{T^{*}}\right)^{2}+\cdots\,\right]
    LMAκMAσMAT=π23e2[ 1CLMAsin2δ(πTT)2+]L_{\mathrm{MA}}\,\equiv\,\frac{\kappa_{\mathrm{MA}}}{\sigma_{\mathrm{MA}}T}\,=\,\frac{\pi^{2}}{3\,e^{2}}\,\left[\,1\,-\,\frac{C_{L}^{\mathrm{MA}}}{\sin^{2}\delta}\,\left(\frac{\pi T}{T^{*}}\right)^{2}\ +\ \cdots\,\right]
Table 2: Coefficients CC’s for the next-leading order terms of SU(NN) quantum dots and magnetic alloys (MA), summarized in Table 1. In these formulas, WW’s represent the contributions determined by the phase shift δ\delta and the rescaled Wilson ratio K~=(N1)(R1)\widetilde{K}=(N-1)(R-1). Three-body correlations enter through ΘIsin2δ2πχσσσ[3]χσσ2\Theta_{\mathrm{I}}\equiv\frac{\sin 2\delta}{2\pi}\,\frac{\chi_{\sigma\sigma\sigma}^{[3]}}{\chi_{\sigma\sigma}^{2}} and Θ~IIsin2δ2πχ~σσσ[3]χσσ2\widetilde{\Theta}_{\mathrm{II}}\equiv\frac{\sin 2\delta}{2\pi}\,\frac{\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}}{\chi_{\sigma\sigma}^{2}}.
CT=π248[WT+ΘI+Θ~II]C_{T}\ =\,\frac{\pi^{2}}{48}\,\bigl{[}\,W_{T}\,+\,\Theta_{\mathrm{I}}+\widetilde{\Theta}_{\mathrm{II}}\,\bigr{]}\quad      WT[ 1+2K~2N1]cos2δW_{T}\ \equiv\,-\left[\,1+\frac{2\widetilde{K}^{2}}{N-1}\,\right]\cos 2\delta
CV=π264[WV+ΘI+3Θ~II]C_{V}\ =\,\frac{\pi^{2}}{64}\,\bigl{[}\,W_{V}\,+\,\Theta_{\mathrm{I}}+3\,\widetilde{\Theta}_{\mathrm{II}}\,\bigr{]}       WV[ 1+5K~2N1]cos2δW_{V}\ \equiv\,-\left[\,1+\frac{5\widetilde{K}^{2}}{N-1}\,\right]\cos 2\delta
CS=π2192[WScos2δ{ΘI+3Θ~II}]C_{S}\ =\,\frac{\pi^{2}}{192}\left[\,W_{S}-\cos 2\delta\,\Bigl{\{}\Theta_{\mathrm{I}}+3\widetilde{\Theta}_{\mathrm{II}}\Bigr{\}}\,\right]      WScos4δ+[ 4+5cos4δ+32(1cos4δ)(N2)]K~2N1W_{S}\ \equiv\,\cos 4\delta\,+\Bigl{[}\,4+5\cos 4\delta+\frac{3}{2}\bigl{(}1-\cos 4\delta\bigr{)}\,(N-2)\,\Bigr{]}\frac{\widetilde{K}^{2}}{N-1}
CκQD=7π280[WκQD+ΘI+521Θ~II]C_{\kappa}^{\mathrm{QD}}=\,\frac{7\pi^{2}}{80}\,\bigl{[}\,W_{\kappa}^{\mathrm{QD}}\,+\,\Theta_{\mathrm{I}}+\frac{5}{21}\,\widetilde{\Theta}_{\mathrm{II}}\,\bigr{]}      WκQD1011cos2δ2167K~2N1cos2δW_{\kappa}^{\mathrm{QD}}\,\equiv\,\frac{10-11\cos 2\delta}{21}-\frac{6}{7}\frac{\widetilde{K}^{2}}{N-1}\,\cos 2\delta
CLQD=π2240[WLQD+ 16ΘI]=CκQDCTC_{L}^{\mathrm{QD}}=\,\frac{\pi^{2}}{240}\,\bigl{[}\,W_{L}^{\mathrm{QD}}\,+\,16\,\Theta_{\mathrm{I}}\,\bigr{]}\ =\ C_{\kappa}^{\mathrm{QD}}-C_{T}      WLQD 106cos2δ8K~2N1cos2δW_{L}^{\mathrm{QD}}\,\equiv\,10-6\cos 2\delta-\frac{8\widetilde{K}^{2}}{N-1}\,\cos 2\delta
CϱMA=π248[WϱMA+ΘI+Θ~II]C_{\varrho}^{\mathrm{MA}}=\,\frac{\pi^{2}}{48}\,\bigl{[}\,W_{\mathcal{\varrho}}^{\mathrm{MA}}\,+\,\Theta_{\mathrm{I}}+\widetilde{\Theta}_{\mathrm{II}}\,\bigr{]}\quad      WϱMA 2+cos2δ2K~2N1cos2δ= 4cos2δ+WTW_{\mathcal{\varrho}}^{\mathrm{MA}}\equiv\,2+\cos 2\delta-\frac{2\widetilde{K}^{2}}{N-1}\,\cos 2\delta\ \,=\ 4\cos^{2}\delta+W_{T}
CκMA=7π280[WκMA+ΘI+521Θ~II]C_{\kappa}^{\mathrm{MA}}=\,\frac{7\pi^{2}}{80}\,\bigl{[}\,W_{\kappa}^{\mathrm{MA}}\,+\,\Theta_{\mathrm{I}}+\frac{5}{21}\,\widetilde{\Theta}_{\mathrm{II}}\,\bigr{]}      WκMA32+11cos2δ2167K~2N1cos2δ=4421cos2δ+WκQDW_{\kappa}^{\mathrm{MA}}\equiv\,\frac{32+11\cos 2\delta}{21}-\frac{6}{7}\frac{\widetilde{K}^{2}}{N-1}\cos 2\delta\ \,=\ \frac{44}{21}\cos^{2}\delta+W_{\kappa}^{\mathrm{QD}}
CLMA=π2240[WLMA 16ΘI]=CϱMACκMAC_{L}^{\mathrm{MA}}=\,\frac{\pi^{2}}{240}\,\bigl{[}\,W_{L}^{\mathrm{MA}}\,-\,16\,\Theta_{\mathrm{I}}\,\bigr{]}\ =\ C_{\varrho}^{\mathrm{MA}}-C_{\kappa}^{\mathrm{MA}}      WLMA226cos2δ+8K~2N1cos2δ=24cos2δWLQDW_{L}^{\mathrm{MA}}\equiv\,-22-6\cos 2\delta+\frac{8\widetilde{K}^{2}}{N-1}\,\cos 2\delta\ \,=\ \ -24\cos^{2}\delta-W_{L}^{\mathrm{QD}}

V.3 Transport formulas for quantum dots in SU(NN) Fermi liquid regime

We next consider the low-energy expansion of transport coefficients dJ/dVdJ/dV, SnoiseQDS_{\mathrm{noise}}^{\mathrm{QD}}, κQD\kappa_{\mathrm{QD}}, and the Lorenz number LQDL_{\mathrm{QD}}. Specifically, for the junctions having tunneling and bias symmetries ΓL=ΓR\Gamma_{L}=\Gamma_{R} and μL=μR=eV/2\mu_{L}=-\mu_{R}=eV/2, these transport coefficients take the following form in the SU(NN) symmetric case,

dJdV=Ne2h[sin2δCT(πTT)2CV(eVT)2+],\displaystyle\!\!\!\frac{dJ}{dV}=\frac{Ne^{2}}{h}\left[\,\sin^{2}\delta\,-C_{T}\left(\frac{\pi T}{T^{*}}\right)^{2}-C_{V}\left(\frac{eV}{T^{*}}\right)^{2}+\cdots\right], (51)
SnoiseQD=2Ne2|eV|h[sin22δ4+CS(eVT)2+],\displaystyle\!\!\!S_{\mathrm{noise}}^{\mathrm{QD}}=\frac{2Ne^{2}|eV|}{h}\left[\,\frac{\sin^{2}2\delta}{4}\,+C_{S}\left(\frac{eV}{T^{*}}\right)^{2}+\cdots\right], (52)
κQD=Nπ2T3h[sin2δCκQD(πTT)2+],\displaystyle\!\!\!\kappa_{\mathrm{QD}}\,=\,\frac{N\pi^{2}T}{3h}\,\left[\,\sin^{2}\delta\,-C_{\kappa}^{\mathrm{QD}}\,\left(\frac{\pi T}{T^{*}}\right)^{2}+\cdots\right], (53)
LQD=π23e2[ 1CLQDsin2δ(πTT)2+].\displaystyle\!\!\!L_{\mathrm{QD}}\,=\,\frac{\pi^{2}}{3e^{2}}\,\left[\,1\,-\,\frac{C_{L}^{\mathrm{QD}}}{\sin^{2}\delta}\,\left(\frac{\pi T}{T^{*}}\right)^{2}\,+\,\cdots\right]\,. (54)

The formulas for the coefficients CTC_{T}, CVC_{V}, CSC_{S}, CκQDC_{\kappa}^{\mathrm{QD}}, and CLQDC_{L}^{\mathrm{QD}} of the next-leading order terms are summarized in Tables 1 and 2. Each of these CC’s can be decomposed into two parts, denoted as WW’s and Θ\Theta’s. The WW part, defined in the right column of Table 2, represents the two-body contributions determined by K~\widetilde{K} and δ\delta. The Θ\Theta part represents the three-body contributions which can be described in terms of the dimensionless parameters defined in Eq. (48).

These transport formulas in the SU(NN) Fermi liquid regime clarify the fact that the next-leading order terms for the symmetric tunnel junctions are completely determined by five parameters: δ\delta, TT^{*}, K~\widetilde{K}, ΘI\Theta_{\mathrm{I}}, and Θ~II\widetilde{\Theta}_{\mathrm{II}}. The three-body correlations can be experimentally deduced through the measurements of the coefficients CC’s. The other three-body component, Θ~III\widetilde{\Theta}_{\mathrm{III}}, defined with respect to three different levels, couples to the tunnel and bias asymmetries, i.e., ΓLΓR\Gamma_{L}\neq\Gamma_{R} and μLμR\mu_{L}\neq-\mu_{R}, and contributes to the nonlinear current [28, 29]. The behavior of CC’s depend significantly on the electron filling NdN_{d} of the impurity levels. For instance, in the noninteracting case at U=0U=0, these coefficients vary with the level position ϵd\epsilon_{d} as shown in Fig. 1.

Refer to caption
Figure 1: Coefficients CVC_{V}, CTC_{T}, CSC_{S}, and CκQDC_{\kappa}^{\mathrm{QD}} for noninteracting U=0U=0 quantum dots plotted vs ϵd\epsilon_{d}.

In the rest of this paper, we will demonstrate the behavior of the next-leading order terms of the transport coefficients for N=4N=4 and 66. To this end, we calculate the correlation functions δ\delta, χσ1σ2\chi_{\sigma_{1}\sigma_{2}}, and χσ1σ2σ3[3]\chi_{\sigma_{1}\sigma_{2}\sigma_{3}}^{[3]} using the NRG approach [84, 26], with parameter settings described in Appendix D. Specifically, Eqs. (42)–(45) are used for obtaining the three-body correlations. We have reported part of the results for coefficient CVC_{V} in a previous paper, studying the role of bias and tunneling asymmetries on the nonlinear terms of dJ/dVdJ/dV at T=0T=0 [29]. In this paper, we provide a comprehensive view of the three-body Fermi liquid effect through a systematic analysis of the next-leading order terms CSC_{S}, CTC_{T}, and CκQDC_{\kappa}^{\mathrm{QD}} for quantum dots, and through the related coefficients for magnetic alloys, CϱMAC_{\varrho}^{\mathrm{MA}}, and CκMAC_{\kappa}^{\mathrm{MA}}. In order to quickly grasp the underlying physics derived from quasiparticle properties in the SU(4) and SU(6) cases, we provide a brief review of the key characteristics of the two-body correlation functions in Appendix E, extending the interaction range up to U/(πΔ)=6.0U/(\pi\Delta)=6.0. Additionally, we also include some new results for the renormalized impurity level ϵ~dσ\widetilde{\epsilon}_{d\sigma} there.

VI Three-body correlations in the SU(4) &\& SU(6) Anderson impurity

In this section, we discuss the behavior of charge and spin susceptibilities defined in Eqs. (37) and (41). In particular, we focus on the derivatives χ¯C/ϵd\partial\overline{\chi}_{C}/\partial\epsilon_{d} and χ¯S/ϵd\partial\overline{\chi}_{S}/\partial\epsilon_{d}, which can also be expressed in terms of the three-body correlation functions:

χ¯Cϵd=\displaystyle\frac{\partial\overline{\chi}_{C}}{\partial\epsilon_{d}}\,= χσσσ[3]+ 3χ~σσσ[3]+ 2χ~σσσ′′[3],\displaystyle\ \chi_{\sigma\sigma\sigma}^{[3]}\,+\,3\,\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\,+\,2\,\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]}\;, (55)
χ¯Sϵd=\displaystyle\frac{\partial\overline{\chi}_{S}}{\partial\epsilon_{d}}\,= χσσσ[3]+N3N1χ~σσσ[3]2N1χ~σσσ′′[3].\displaystyle\ \chi_{\sigma\sigma\sigma}^{[3]}\,+\,\frac{N-3}{N-1}\,\,\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\,-\,\frac{2}{N-1}\,\,\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]}\,. (56)

The NRG results reveal the fact that these derivatives in the left-hand side are suppressed in the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2 for large UU. This implies that the linear combinations of the three-body correlations in left-hand side of Eqs. (55) and (56) approach zero, reducing the number of independent components of the three-body correlation functions χσ1σ2σ3[3]\chi_{\sigma_{1}\sigma_{2}\sigma_{3}}^{[3]}, as demonstrated below.

Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 2: Inverse energy scale 1/T1/T^{*}, charge susceptibility χ¯C\overline{\chi}_{C} and spin χ¯S\overline{\chi}_{S} susceptibilities are plotted vs ξd\xi_{d} (=ϵd+U/2=\epsilon_{d}+U/2) for N=4N=4 and 66. Here, 4T4T^{*} (=1/χσσ=1/\chi_{\sigma\sigma}), and TKT|ξd=0T_{K}\equiv T^{*}\big{|}_{\xi_{d}=0}. Interaction strengths are chosen such that, for N=4N=4, U/(πΔ)=2/3()U/(\pi\Delta)=2/3(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), and 6(×)6(\times), at which TK/(πΔ)=0.20T_{K}/(\pi\Delta)=0.20, 0.180.18, 0.130.13, 0.0920.092, 0.0630.063, 0.0410.041, and 0.0260.026, respectively. For N=6N=6, U/(πΔ)=2/5()U/(\pi\Delta)=2/5(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), and 6(×)6(\times), at which TK/(πΔ)=0.22T_{K}/(\pi\Delta)=0.22, 0.190.19, 0.150.15, 0.130.13, 0.100.10, 0.0850.085, and 0.0680.068, respectively.
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 3: (4T)2χ¯C/ϵd(4T^{*})^{2}\partial\overline{\chi}_{C}/\partial\epsilon_{d}, (4T)2χ¯S/ϵd(4T^{*})^{2}\partial\overline{\chi}_{S}/\partial\epsilon_{d}, and (4T)2χB[3](4T^{*})^{2}\chi_{B}^{[3]} are plotted vs ξd\xi_{d} for N=4N=4 and 66. Interaction strengths are chosen such that, for N=4N=4, U/(πΔ)=2/3()U/(\pi\Delta)=2/3(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times). For N=6N=6, U/(πΔ)=2/5()U/(\pi\Delta)=2/5(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times).

VI.1 Charge and spin susceptibilities: χ¯C\overline{\chi}_{C} & χ¯S\overline{\chi}_{S}

One of the most fundamental quantities that play a central role in the low-energy physics of quantum impurities is the characteristic energy scale T1/(4χσσ)T^{*}\equiv 1/(4\chi_{\sigma\sigma}), defined in Eq. (34) as an inverse of the diagonal susceptibility Nχσσ=χ¯C+(N1)χ¯SN\chi_{\sigma\sigma}=\overline{\chi}_{C}+(N-1)\overline{\chi}_{S}. We will use, in the following discussions, the Kondo temperature TKT|ξd=0T_{K}\equiv T^{*}\big{|}_{\xi_{d}=0}, defined as the value of TT^{*} at the electron-hole symmetric point ξd=0\xi_{d}=0.

The NRG results for 1/T1/T^{*} in the SU(4) and SU(6) cases are plotted vs ξd\xi_{d} in Figs.  2(a) and 2(b), respectively, by multiplying them by TKT_{K}. We see that 1/T1/T^{*} has N1N-1 local maxima for strong interactions, at integer-filling points, i.e., ξd0,±U,,±(N2)U/2\xi_{d}\simeq 0,\pm U,\ldots,\pm(N-2)U/2, reflecting the oscillatory behavior of the wave function renormalization factor zz (=ρdσ/χσσ=\rho_{d\sigma}/\chi_{\sigma\sigma}) described in Appendix E. At |ξd|(N1)U/2|\xi_{d}|\gg(N-1)U/2, the energy scale TT^{*} approaches the non-interacting value, TK/T|ξd|(N1)U/2Δ2/ξd2T_{K}/T^{*}\xrightarrow{\,|\xi_{d}|\gg(N-1)U/2\,}\Delta^{2}/\xi_{d}^{2}, as the electron filling of the impurity levels approaches Nd0N_{d}\simeq 0 or NN.

The charge susceptibilities for N=4N=4 and 66 are plotted in Figs. 2(c) and 2(d), using TT^{*} as a normalization factor, i.e., 4Tχ¯C=1K~4T^{\ast}\overline{\chi}_{C}=1-\widetilde{K} from Eq. (37). Therefore, the normalized value 4Tχ¯C4T^{\ast}\overline{\chi}_{C} is determined by the rescaled Wilson ratio K~\widetilde{K}, described in Appendix E. As UU increases, 4Tχ¯C4T^{\ast}\overline{\chi}_{C} decreases in a wide region of the impurity level |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, where the impurity levels are partially filled 1NdN11\lesssim N_{d}\lesssim N-1. In this filling range, the charge susceptibility is significantly suppressed by the Coulomb repulsion, and it vanishes 4Tχ¯CUπΔ04T^{\ast}\overline{\chi}_{C}\xrightarrow{\,U\gg\pi\Delta\,}0 in the strong interaction limit. Outside this region, i.e., at |ξd|(N1)U/2|\xi_{d}|\gg(N-1)U/2, the charge susceptibility approaches the noninteracting value 4Tχ¯C|ξd|14T^{*}\overline{\chi}_{C}\xrightarrow{\,|\xi_{d}|\to\infty\,}1 as the filling of the impurity levels approaches Nd0N_{d}\simeq 0 or NN.

Figures 2(e) and 2(f) show the spin susceptibilities for N=4N=4 and 66, which are normalized with the same scaling factor, i.e., 4Tχ¯S=1+K~/(N1)4T^{*}\overline{\chi}_{S}=1+\widetilde{K}/(N-1) [see Eq. (41)]. As the interaction UU increases, 4Tχ¯S4T^{*}\overline{\chi}_{S} increases from the non-interacting value 11. In the strong interaction limit, it approaches the value 4Tχ¯SN/(N1)4T^{*}\overline{\chi}_{S}\to N/(N-1), i.e., 4/34/3 for N=4N=4 and 6/56/5 for N=6N=6, and exhibits a wide plateau structure in the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2. At |ξd|(N1)U/2|\xi_{d}|\gg(N-1)U/2, where the occupation number approaches Nd0N_{d}\simeq 0 or NN, the spin susceptibility also approaches the noninteracting value 4Tχ¯S|ξd|14T^{*}\overline{\chi}_{S}\xrightarrow{\,|\xi_{d}|\to\infty\,}1.

VI.2 Derivative of χ¯C\overline{\chi}_{C} and χ¯S\overline{\chi}_{S} with respect to ϵd\epsilon_{d}

The three-body correlation functions can be obtained from the derivatives of χσ1σ2\chi_{\sigma_{1}\sigma_{2}} with respect to ϵd\epsilon_{d} and bb, using Eqs. (42)–(44). In particular, χσ1σ2/ϵd\partial\chi_{\sigma_{1}\sigma_{2}}/\partial\epsilon_{d} can be rewritten in terms of the derivatives of the charge and spin susceptibilities, as

χσσϵd=\displaystyle\frac{\partial\chi_{\sigma\sigma}}{\partial\epsilon_{d}}\,= 1Nχ¯Cϵd+N1Nχ¯Sϵd,\displaystyle\ \frac{1}{N}\frac{\partial\overline{\chi}_{C}}{\partial\epsilon_{d}}\,+\,\frac{N-1}{N}\frac{\partial\overline{\chi}_{S}}{\partial\epsilon_{d}}\,, (57)
(N1)χσσϵd=\displaystyle(N-1)\frac{\partial\chi_{\sigma\sigma^{\prime}}}{\partial\epsilon_{d}}\,= N1Nχ¯CϵdN1Nχ¯Sϵd,\displaystyle\ \frac{N-1}{N}\frac{\partial\overline{\chi}_{C}}{\partial\epsilon_{d}}\,-\,\frac{N-1}{N}\frac{\partial\overline{\chi}_{S}}{\partial\epsilon_{d}}\,, (58)

for σσ\sigma\neq\sigma^{\prime}. Figures 3(a) and 3(b) show the NRG results for χ¯C/ϵd\partial\overline{\chi}_{C}/\partial\epsilon_{d} for N=4N=4 and 66, respectively. Similarly, the derivatives of the spin susceptibility χ¯S/ϵd\partial\overline{\chi}_{S}/\partial\epsilon_{d} for N=4N=4 and 66 are plotted in Figs. 3 (c) and 3(d). Note that in these figures, the derivatives have been multiplied by a factor of (4T)2(4T^{*})^{2} to make them dimensionless. These derivatives are significantly suppressed in the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2 for large UU. More specifically, |χ¯C/ϵd||χ¯S/ϵd|1/(4T)2|\partial\overline{\chi}_{C}/\partial\epsilon_{d}|\ll|\partial\overline{\chi}_{S}/\partial\epsilon_{d}|\ll 1/(4T^{*})^{2}: the derivative of spin susceptibility becomes much smaller than 1/(4T)21/(4T^{*})^{2} while it is still larger than the derivative of charge susceptibility. Therefore, both χσσ/ϵd\partial\chi_{\sigma\sigma}/\partial\epsilon_{d} and χσσ/ϵd\partial\chi_{\sigma\sigma^{\prime}}/\partial\epsilon_{d} are suppressed in a wide range of electron fillings 1NdN11\lesssim N_{d}\lesssim N-1 for large UU.

We next consider χB[3]\chi_{B}^{[3]}, defined in Eq. (45) as a derivative of a linear combination of the two different diagonal susceptibilities with respect to the magnetic fields bb. Figures 3(e) and 3(f) show (4T)2χB[3](4T^{*})^{2}\chi_{B}^{[3]} for N=4N=4 and 66, respectively. We see that (4T)2χB[3](4T^{*})^{2}\chi_{B}^{[3]} exhibits a staircase structure with a flat plateau emerging around the integer filling points ξd=0\xi_{d}=0, ±U\pm U, ±2U\pm 2U, \ldots, (N2)U/2(N-2)U/2 for large UU. The magnitude |χB[3]|\bigl{|}\chi_{B}^{[3]}\bigr{|} becomes much larger than the derivative of the charge and spin susceptibilities, |χ¯C/ϵd||\partial\overline{\chi}_{C}/\partial\epsilon_{d}| and |χ¯S/ϵd||\partial\overline{\chi}_{S}/\partial\epsilon_{d}|. Therefore, χB[3]\chi_{B}^{[3]} dominates the terms in the right-hand side of Eqs. (42)–(44) in the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, and the three independent components of three-body correlations approach one another in such a way that

χσσσ[3]χ~σσσ[3]χ~σσσ′′[3]N1NχB[3].\displaystyle\chi_{\sigma\sigma\sigma}^{[3]}\,\simeq\,-\,\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\,\simeq\,\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]}\ \simeq\ -\frac{N-1}{N}\,\chi_{B}^{[3]}\,. (59)

This means that the three-body correlations are described by a single parameter χB[3]\chi_{B}^{[3]} in a wide filling range 1NdN11\lesssim N_{d}\lesssim N-1 for large UU.

Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 4: Dimensionless three-body corrections ΘI()\Theta_{\mathrm{I}}(\blacksquare) , Θ~II()-\widetilde{\Theta}_{\mathrm{II}}(\blacklozenge), and Θ~III()\widetilde{\Theta}_{\mathrm{III}}(\bullet) are plotted vs ξd\xi_{d}, for interaction strengths U/(πΔ)=2/3U/(\pi\Delta)=2/3 (a), 2/52/5 (b), 22 (c,d), and 66 (e,f).

VI.3 Three-body correlations for N=4N=4 and 66

We have calculated the three-body correlation functions χσσσ[3]\chi_{\sigma\sigma\sigma}^{[3]}, χ~σσσ[3]\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]} and χ~σσσ′′[3]\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]} for σσσ′′σ\sigma\neq\sigma^{\prime}\neq\sigma^{\prime\prime}\neq\sigma, using Eqs. (42)–(44). Figure 4 shows the dimensionless three-body correlations ΘI\Theta_{\mathrm{I}}, Θ~II\widetilde{\Theta}_{\mathrm{II}} and Θ~III\widetilde{\Theta}_{\mathrm{III}}, defined in Eqs.  (48) and (49). The left and right panels describe the NRG results in the SU(4) and SU(6) cases, respectively, and three different interaction strengths (from weak to strong) are chosen for the top, middle, and bottom panels.

All components of the three-body correlation functions vanish, ΘI=Θ~II=Θ~III=0\Theta_{\mathrm{I}}=\widetilde{\Theta}_{\mathrm{II}}=\widetilde{\Theta}_{\mathrm{III}}=0, at the electron-hole symmetric point ξd=0\xi_{d}=0, and evolve as ξd\xi_{d} deviates from this point. Among the three independent components, the intra-level component ΘI\Theta_{\mathrm{I}} has the largest magnitude, and exhibits plateau structures for large UU at integer filling points Nd1N_{d}\simeq 1 , 22, \ldots, N1N-1. The other components, Θ~II\widetilde{\Theta}_{\mathrm{II}} and Θ~III\widetilde{\Theta}_{\mathrm{III}}, involve inter-level correlations and evolve as the Coulomb interaction UU increases. In particular, the correlation between the three different levels Θ~III\widetilde{\Theta}_{\mathrm{III}} becomes the weakest. In the limit of |ξd||\xi_{d}|\to\infty, the diagonal component ΘI\Theta_{\mathrm{I}} approaches the noninteracting value while the other two vanish:

ΘI|ξd|2,Θ~II|ξd| 0,Θ~III|ξd| 0.\displaystyle\Theta_{\mathrm{I}}\,\xrightarrow{|\xi_{d}|\to\infty\,}\,-2,\quad\widetilde{\Theta}_{\mathrm{II}}\,\xrightarrow{|\xi_{d}|\to\infty\,}\,0,\quad\widetilde{\Theta}_{\mathrm{III}}\,\xrightarrow{|\xi_{d}|\to\infty\,}\,0. (60)

Figures 4(e) and 4(f) clearly demonstrate the relation Eq. (59), which holds at |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2 for large UU:

ΘIΘ~IIΘ~III.\displaystyle\Theta_{\mathrm{I}}\,\simeq\,-\widetilde{\Theta}_{\mathrm{II}}\,\simeq\,\widetilde{\Theta}_{\mathrm{III}}\,. (61)

It means that all the three-body components are determined by a single parameter χB[3]\chi_{B}^{[3]} in the strong-coupling region for large UU, as mentioned. The dimensionless three-body correlation functions clearly exhibit the plateau structure around the integer filling points ξd=0\xi_{d}=0, ±U\pm U, ±2U\pm 2U, \ldots, ±(N2)U/2\pm(N-2)U/2, where the SU(NN) Kondo effect occurs. These structures reflect the behavior of χB[3]\chi_{B}^{[3]}, described in Figs. 3(e) and 3(f), and evolve as the interaction strength UU increases. We see, in Figs.  4(e) and 4(f), that the plateaus appear much clearer for N=4N=4 than N=6N=6, for the same interaction strength U/(πΔ)=6.0U/(\pi\Delta)=6.0.

We will examine, in the subsequent sections, how these three-body correlation functions affect the next-leading order terms of the transport coefficients in the low-energy Fermi liquid regime away from half filling.

VII Nonlinear current noise of
SU(4) &\& SU(6) quantum dots

In this section, we discuss the nonlinear terms of the steady current JJ and the current noise SnoiseQDS_{\mathrm{noise}}^{\mathrm{QD}}, specifically, the order (eV)3(eV)^{3} term for symmetric junctions, i.e., ΓL=ΓR\Gamma_{L}=\Gamma_{R} and μL=μR=eV/2\mu_{L}=-\mu_{R}={eV}/{2}. To provide a comprehensive view of the low-bias behavior of these terms, we begin with a brief review on the previous results for the coefficient CVC_{V} of the nonlinear conductance [29], extending slightly the interaction range up to U/(πΔ)=6.0U/(\pi\Delta)=6.0. We will then discuss the results for CSC_{S}, i.e., the order |eV|3|eV|^{3} term of nonlinear noise.

VII.1 CVC_{V}: order (eV)2(eV)^{2} term of dJ/dVdJ/dV

The leading-order term of the conductance, at T=0T=0, is determined by the transmission probability sin2δ\sin^{2}\delta, i.e., the first term in the right-hand side of Eq. (51). It exhibits the well-known Kondo plateaus, which develop in the strong-coupling region, as shown in Figs. 15(b) and 16(b) in Appendix E.

The next-leading order term CVC_{V} of the nonlinear conductance can be decomposed into the two-body part WVW_{V}, defined in Table 2, and the three-body part ΘV\Theta_{V}, as

CV=π264(WV+ΘV),ΘVΘI+3Θ~II.\displaystyle C_{V}\,=\,\frac{\pi^{2}}{64}\,\bigl{(}W_{V}+\Theta_{V}\bigr{)},\qquad\Theta_{V}\,\equiv\,\Theta_{\mathrm{I}}+3\,\widetilde{\Theta}_{\mathrm{II}}. (62)

These coefficients CVC_{V}, WVW_{V} and ΘV\Theta_{V} are plotted vs ξd\xi_{d} in Figs. 5(a)–5(f) for N=4N=4 and 66.

The two-body part WVW_{V} dominates the next-leading order term near the electron-hole symmetric point ξd=0\xi_{d}=0, where δ=π/2\delta=\pi/2 and the three-body part ΘV\Theta_{V} vanishes:

WVξd=0 1+5K~2N1,ΘVξd=00.\displaystyle W_{V}\,\xrightarrow{\,\xi_{d}=0}\,1+\frac{5\widetilde{K}^{2}}{N-1},\qquad\Theta_{V}\xrightarrow{\,\xi_{d}=0\,}0. (63)

Therefore, in the strong interaction limit, the peak at ξd=0\xi_{d}=0 reaches (64/π2)CVξd=0&U8/3(64/\pi^{2})C_{V}\xrightarrow{\,\xi_{d}=0\,\&U\to\infty}8/3 and 22 for N=4N=4 and 66, respectively. Note that the rescaled Wilson ratio approaches K~1\widetilde{K}\simeq 1 in a wide region of |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2 for large UU. Outside this region, the two-body and three-body parts of CVC_{V} approach to the noninteracting values,

WV|ξd|1,ΘV|ξd|2,\displaystyle W_{V}\,\xrightarrow{\,|\xi_{d}|\to\infty}\,-1,\qquad\Theta_{V}\,\xrightarrow{\,|\xi_{d}|\to\infty}\,-2, (64)

since cos2δ|ξd|1\cos 2\delta\xrightarrow{\,|\xi_{d}|\to\infty}1, K~|ξd|0\widetilde{K}\xrightarrow{\,|\xi_{d}|\to\infty}0, and the three-body contributions approach the values given in Eq. (60).

Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 5: ξd\xi_{d} dependence of CV=(π2/64)(WV+ΘV)C_{V}=(\pi^{2}/64)(W_{V}+\Theta_{V}), two-body part WVW_{V}, and three-body part ΘVΘI+3Θ~II\Theta_{V}\equiv\Theta_{\mathrm{I}}+3\,\widetilde{\Theta}_{\mathrm{II}}. Left panels: N=4N=4, for U/(πΔ)=2/3()U/(\pi\Delta)=2/3(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times). Right panels: N=6N=6, for U/(πΔ)=2/5()U/(\pi\Delta)=2/5(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times).

The SU(NN) Kondo effect occurs in the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2 at the integer filling points. The plateau structure evolves as UU increases, especially in the three-body part ΘV\Theta_{V}, as seen in Figs. 5(e) and 5(f). In the strong-coupling region, it can be expressed in the form ΘV2ΘI\Theta_{V}\simeq-2\Theta_{\mathrm{I}} due to Eq. (61). The two-body part WVW_{V}, shown in Figs. 5(c) and 5(d) for U/(πΔ)6.0U/(\pi\Delta)\leq 6.0, does not exhibit a clear plateau other than the one appearing near half filling. For N=4N=4, this is because the factor cos2δ\cos 2\delta for WVW_{V} vanishes at δ=π/4\delta=\pi/4 and 3π/43\pi/4, i.e., at the quarter and three-quarters filling points. As a sum of WVW_{V} and ΘV\Theta_{V}, the coefficient CVC_{V} exhibits a wide and rather flat structure in the region of |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2 for large UU, seen in Figs. 5(a) and 5(b). There also emerge some weak local maxima in this flat structure at the integer filling points ξd=0\xi_{d}=0, ±U\pm U, ±2U\pm 2U, \ldots, ±(N2)U/2\pm(N-2)U/2. In particular, the peak is most pronounced at ξd±(N2)U/2\xi_{d}\simeq\pm(N-2)U/2, where Nd1N_{d}\simeq 1 or N1N-1.

Refer to caption
Refer to caption
Figure 6: Linear part, sin2δ(1sin2δ)\sin^{2}\delta\,(1-\sin^{2}\delta), of the noise SnoiseQDS_{\mathrm{noise}}^{\mathrm{QD}}. Interaction strengths are chosen for (a) N=4N=4 to be U/(πΔ)=2/3()U/(\pi\Delta)=2/3(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times). For (b) N=6N=6, U/(πΔ)=2/5()U/(\pi\Delta)=2/5(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times).

VII.2 Order |eV||eV| and Order |eV|3|eV|^{3} terms of SnoiseQDS_{\mathrm{noise}}^{\mathrm{QD}}

We next consider the current noise SnoiseQDS_{\mathrm{noise}}^{\mathrm{QD}}, the low-energy asymptotic form of which is given in Eq. (52) and Table 2. The leading-order term, sin2δ(1sin2δ)=(1cos4δ)/8\sin^{2}\delta\,(1-\sin^{2}\delta)=(1-\cos 4\delta)/8, corresponds to the linear-response noise, the NRG results for which are shown as a function of ξd\xi_{d} in Figs. 6(a) and 6(b), for N=4N=4 and 66, respectively. The linear noise is maximized at the points where the phase shift reaches δ=π/4\delta=\pi/4 and 3π/43\pi/4. It occurs at the integer filling points Nd=1N_{d}=1 and 33 for N=4N=4, at which the SU(4) Kondo effect makes the peaks wide and flat. In contrast, for N=6N=6, the peak emerges at the half-integer filling points Nd=3/2N_{d}=3/2 and 9/29/2. More generally, the peak of the linear noise forms a flat plateau structure for N0N\equiv 0 (mod 44), whereas the peak becomes round for N2N\equiv 2 (mod 44) due to the fluctuations occurring between two adjacent integer filling states. For large UU, the quarter and three-quarters fillings occur near |ξd|/UN/4|\xi_{d}|/U\simeq N/4, at which the ground state is highly correlated for multilevel systems of N4N\geq 4. In contrast, these fillings occur at ξd±U/2\xi_{d}\simeq\pm U/2 for SU(2) quantum dots, where the electron correlation becomes less important due to the valence fluctuations (see Appendix F).

The coefficient CSC_{S} for the order |eV|3|eV|^{3} term of current noise SnoiseQDS_{\mathrm{noise}}^{\mathrm{QD}} can also be decomposed into the two-body WSW_{S} and three-body ΘS\Theta_{S} parts, as shown in Table 2:

CS=π2192(WS+ΘS),ΘS(ΘI+3Θ~II)cos2δ.\displaystyle\!\!C_{S}=\frac{\pi^{2}}{192}\bigl{(}W_{S}+\Theta_{S}\bigr{)},\quad\Theta_{S}\equiv-\bigl{(}\Theta_{\mathrm{I}}+3\widetilde{\Theta}_{\mathrm{II}}\bigr{)}\cos 2\delta. (65)

The behavior of three-body part ΘS\Theta_{S} can be deduced from the one for dJ/dVdJ/dV, i.e., ΘV=ΘI+3Θ~II\Theta_{V}=\Theta_{\mathrm{I}}+3\widetilde{\Theta}_{\mathrm{II}} shown in Figs. 5(e) and 5(f), by multiplying them by a factor of “cos2δ-\cos 2\delta” which induces the modulations. One of the most distinctive features of CSC_{S}, compared to the other coefficients CC’s listed in Table 2, is that it depends upon the higher harmonics cos4δ\cos 4\delta and sin4δ\sin 4\delta with respect to the phase shift, which enter through not only through WSW_{S} but ΘS\Theta_{S}: note that ΘI\Theta_{\mathrm{I}} and Θ~II\widetilde{\Theta}_{\mathrm{II}} defined in Eq. (48) are proportional to the factor sin2δ\sin 2\delta. As ξd\xi_{d} varies, these higher harmonics evolve continuously in the range 04δ4π0\leq 4\delta\leq 4\pi, simultaneously with the electron filling 0NdN0\leq N_{d}\leq N. Figures 7(a)–7(f) show results for CSC_{S}, WSW_{S}, and ΘS\Theta_{S} for N=4N=4 and 66.

Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 7: ξd\xi_{d} dependence of CS=(π2/192)(WS+ΘS)C_{S}=(\pi^{2}/192)(W_{S}+\Theta_{S}), two-body part WSW_{S}, and three-body part ΘS(ΘI+3Θ~II)cos2δ\Theta_{S}\equiv-\bigl{(}\Theta_{\mathrm{I}}+3\,\widetilde{\Theta}_{\mathrm{II}}\bigr{)}\cos 2\delta. Left panels: N=4N=4, for U/(πΔ)=2/3()U/(\pi\Delta)=2/3(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times). Right panels: N=6N=6, for U/(πΔ)=2/5()U/(\pi\Delta)=2/5(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times).

Each curve for CSC_{S}, shown in Figs. 7(a) and 7(b), exhibits two valleys situated at the valence fluctuation regions ξd±(N1)U/2\xi_{d}\simeq\pm(N-1)U/2 for large UU. These valleys rise all around as UU increases. Notably, for N=4N=4, the bottom value of CSC_{S} turns positive for large interactions U/(πΔ)5U/(\pi\Delta)\gtrsim 5, while for N=6N=6, it remains negative even for the largest interaction U/(πΔ)=6U/(\pi\Delta)=6 examined in this study.

Near the electron-hole symmetric point ξd=0\xi_{d}=0, where δ=π/2\delta=\pi/2 and Nd=N/2N_{d}=N/2, the two-body part WSW_{S} dominates the nonlinear noise coefficient CSC_{S} since the three-body correlations disappear around this point:

WSξd=0 1+9K~2N1,ΘSξd=00.\displaystyle W_{S}\,\xrightarrow{\,\xi_{d}=0\,}\,1+\frac{9\widetilde{K}^{2}}{N-1}\,,\qquad\Theta_{S}\xrightarrow{\,\xi_{d}=0\,}0. (66)

In particular, in the limit of UU\to\infty, the rescaled Wilson ratio approaches the saturation value K~1\widetilde{K}\to 1. Hence, the coefficient for the order |eV|3|eV|^{3} term reaches (192/π2)CSξd=0&U4(192/\pi^{2})C_{S}\xrightarrow{\xi_{d}=0\,\&\,U\to\infty}4 for N=4N=4, and it reaches 14/514/5 for N=6N=6: the height of this ridge decreases as NN increases.

In contrast, in the opposite limit |ξd||\xi_{d}|\to\infty, both the two-body and the three-body parts approach the noninteracting values

WS|ξd| 1,ΘS|ξd| 2,\displaystyle W_{S}\,\xrightarrow{\,|\xi_{d}|\to\infty}\,1,\qquad\Theta_{S}\,\xrightarrow{\,|\xi_{d}|\to\infty}\,2, (67)

since cos2δ1\cos 2\delta\to 1, K~0\widetilde{K}\to 0, and ΘI2\Theta_{\mathrm{I}}\to-2 in this limit. Therefore, WSW_{S} and ΘS\Theta_{S} contribute comparably to CSC_{S} in the region |ξd|(N1)U/2|\xi_{d}|\gtrsim(N-1)U/2, where the impurity levels are either almost empty Nd0N_{d}\simeq 0 or fully occupied NdNN_{d}\simeq N. In particular, as seen in Figs. 7(c) and 7(d), the two-body part approaches the saturation value WS1W_{S}\to 1 already at |ξd|(N1)U/2|\xi_{d}|\simeq(N-1)U/2 for large interactions U/(πΔ)3U/(\pi\Delta)\gtrsim 3.

In the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2 for large UU, the behavior of two-body part WSW_{S} is determined by the higher-harmonic cos4δ\cos 4\delta term, as

WS12(3+5N1)+12(13N11)cos4δ,\displaystyle W_{S}\,\simeq\,\frac{1}{2}\left(3+\frac{5}{N-1}\right)\,+\,\frac{1}{2}\left(\frac{13}{N-1}-1\right)\cos 4\delta\,, (68)

since the Wilson ratio is locked at K~1.0\widetilde{K}\simeq 1.0 in this region. As seen in Figs. 7(c) and 7(d), the two-body part WSW_{S} has local minima at quarter and three-quarters filling points, which occur at δ=π/4\delta=\pi/4 and 3π/43\pi/4, or |ξd|/UN/4|\xi_{d}|/U\simeq N/4 for large UU. At these local minima, WSW_{S} take a positive value for N4N\geq 4, as it can be deduced from Eq. (68). This is in contrast to the SU(2) case where the local minima of WSW_{S} take a negative value, as shown in Appendix F. In the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, the three-body part takes the following form, for large UU,

ΘSχσσσ[3]2πχσσ2sin4δ,\displaystyle\Theta_{S}\,\simeq\,\frac{\chi_{\sigma\sigma\sigma}^{[3]}}{2\pi\,\chi_{\sigma\sigma}^{2}}\,\sin 4\delta\,, (69)

due to the property described in Eq. (61). Equation (69) clearly shows that the three-body part vanishes, ΘS=0\Theta_{S}=0, at quarter δ=π/4\delta=\pi/4 and three-quarters δ=3π/4\delta=3\pi/4 fillings. Therefore, at these filling points, ΘS\Theta_{S} does not exhibit the plateau structures for N=4N=4 in Fig. 7(e), despite the fact that ΘV\Theta_{V} clearly exhibits the plateau structures as seen in Fig. 5(e). In contrast, for N=6N=6, ΘS\Theta_{S} shows the clear plateau structures in Fig. 7(f) at the fillings of Nd=2N_{d}=2 and 44. The three-body parts ΘS\Theta_{S} for both N=4N=4 and 66 cases also have pronounced local minima near the valence fluctuation regions ξd±(N1)U/2\xi_{d}\simeq\pm(N-1)U/2, which cause the valley structure appearing in CSC_{S}. A similar valley structure also emerges in CSC_{S} for N=2N=2, as shown in Appendix F. However, it stems from the two-body correlations WSW_{S}, instead of ΘS\Theta_{S}, in the SU(22) case.

VIII Thermoelectric transport of
SU(4) &\& SU(6) quantum dots

We next consider the order T2T^{2} term of the linear conductance g=dJ/dV|eV=0g=dJ/dV\big{|}_{eV=0} and the order T3T^{3} term of thermal conductance κQD\kappa_{\mathrm{QD}} of SU(NN) quantum dots.

VIII.1 CTC_{T}: order T2T^{2} term of dJ/dVdJ/dV

The coefficient CTC_{T} for the order T2T^{2} conductance, defined in Table 2, also consists of two-body parts WTW_{T} and three-body part ΘT\Theta_{T}:

CT=π248(WT+ΘT),ΘTΘI+Θ~II.\displaystyle C_{T}\,=\,\frac{\pi^{2}}{48}\,\bigl{(}W_{T}+\Theta_{T}\bigr{)}\,,\qquad\Theta_{T}\,\equiv\,\Theta_{\mathrm{I}}+\widetilde{\Theta}_{\mathrm{II}}\,. (70)

In particular, the three-body part ΘT\Theta_{T} is solely determined by the derivatives of the charge and spin susceptibilities, given in Eqs. (55) and (56), and does not depend on χB[3]\chi_{B}^{[3]}:

ΘT=(4T)2N[χ¯Cϵd+(N1)χ¯Sϵd]sin2δ2π.\displaystyle\Theta_{T}\,=\,\frac{(4T^{*})^{2}}{N}\left[\,\frac{\partial\overline{\chi}_{C}}{\partial\epsilon_{d}}\,+(N-1)\frac{\partial\overline{\chi}_{S}}{\partial\epsilon_{d}}\,\,\right]\frac{\sin 2\delta}{2\pi}\,. (71)

This is a quite distinct characteristics of CTC_{T} from the next-leading order terms of the other transport coefficients. Figures 8(a)–8(f) show the NRG results for CTC_{T}, WTW_{T}, and ΘT\Theta_{T}.

We see in Figs. 8(e) and 8(f) that the three-body contribution almost vanishes, ΘT0.0\Theta_{T}\simeq 0.0, in the wide strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, in which the occupation number varies with ξd\xi_{d}, in the range of 1NdN11\lesssim N_{d}\lesssim N-1. This is because the magnitudes of the derivatives χ¯C/ϵd\partial\overline{\chi}_{C}/\partial\epsilon_{d} and χ¯S/ϵd\partial\overline{\chi}_{S}/\partial\epsilon_{d}, appearing in the right-hand side of Eq. (71), are significantly suppressed by the Coulomb repulsion in this region, as demonstrated in Figs. 3(a)–3(d).

Therefore, the two-body part WTW_{T} dominates CTC_{T} in the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, and it takes the following form for large UU,

WT[ 1+2N1](2sin2δ1),ΘT0,\displaystyle W_{T}\simeq\left[\,1+\frac{2}{N-1}\,\right]\,(2\sin^{2}\delta-1),\qquad\Theta_{T}\simeq 0, (72)

as the rescaled Wilson ratio reaches the saturation value K~1\widetilde{K}\to 1. Hence, the plateau structure of CTC_{T} is determined by the sin2δ\sin^{2}\delta term of WTW_{T} in Eq. (72). In particular, the plateau around the half filling point |ξd|U/2|\xi_{d}|\lesssim U/2 reaches the height of (48/π2)CTU5/3(48/\pi^{2})C_{T}\xrightarrow{U\to\infty\,}5/3 and 7/57/5 for N=4N=4 and 66, respectively, since δπ/2\delta\simeq\pi/2 in this region. The order T2T^{2} conductance, CTC_{T}, vanishes at |ξd|/UN/4|\xi_{d}|/U\simeq N/4, more specifically at the quarter and the three-quarter fillings where the phase shift reaches δ=π/4\delta=\pi/4 or 3π/43\pi/4. The zero points of CTC_{T} emerge at integer fillings in the case of N0N\equiv 0 (mod 44) at which the SU(NN) Kondo effect is occurring, whereas for N2N\equiv 2 (mod 44) the zeros emerge at half-integer filings in between the two adjacent Kondo states. This explains the reason why the SU(4) Kondo state at quarter filling exhibits universal T/TT/T^{*}-scaling behavior, which shows the (T/T)4(T/T^{*})^{4} dependence at low temperatures instead of the (T/T)2(T/T^{*})^{2} dependence [68].

In the valence fluctuation and empty (or fully-occupied) orbital regimes, which spread over the regions of |ξd|(N1)U/2|\xi_{d}|\gtrsim(N-1)U/2, the three-body part ΘT\Theta_{T} becomes comparable to the two-body part WTW_{T}. Both of these parts approach the noninteracting values in the limit of |ξd||\xi_{d}|\to\infty:

WT|ξd|1,ΘT|ξd|2.\displaystyle W_{T}\,\xrightarrow{\,|\xi_{d}|\to\infty}\,-1,\qquad\Theta_{T}\,\xrightarrow{\,|\xi_{d}|\to\infty}\,-2. (73)
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 8: ξd\xi_{d} dependence of CT=(π2/48)(WT+ΘT)C_{T}=(\pi^{2}/48)(W_{T}+\Theta_{T}), two-body part WTW_{T}, and three-body part ΘTΘI+3Θ~II\Theta_{T}\equiv\Theta_{\mathrm{I}}+3\,\widetilde{\Theta}_{\mathrm{II}}. Left panels: N=4N=4, for U/(πΔ)=2/3()U/(\pi\Delta)=2/3(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times). Right panels: N=6N=6, for U/(πΔ)=2/5()U/(\pi\Delta)=2/5(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times).
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 9: ξd\xi_{d} dependence of CκQD=(7π2/80)(WκQD+ΘκQD)C_{\kappa}^{\mathrm{QD}}=(7\pi^{2}/80)(W_{\kappa}^{\mathrm{QD}}+\Theta_{\kappa}^{\mathrm{QD}}), two-body part WκQDW_{\kappa}^{\mathrm{QD}}, and three-body part ΘκQD=ΘI+521Θ~II\Theta_{\kappa}^{\mathrm{QD}}=\Theta_{\mathrm{I}}+\frac{5}{21}\widetilde{\Theta}_{\mathrm{II}}. Left panels: N=4N=4, for U/(πΔ)=2/3()U/(\pi\Delta)=2/3(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times). Right panels: N=6N=6, for U/(πΔ)=2/5()U/(\pi\Delta)=2/5(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times).

VIII.2 CκQDC_{\kappa}^{\mathrm{QD}}: order T3T^{3} term of κQD\kappa_{\mathrm{QD}}

The coefficient CκQDC_{\kappa}^{\mathrm{QD}} for the order T3T^{3} term of the thermal conductance κQD\kappa_{\mathrm{QD}} can also be decomposed into two parts, WκQDW_{\kappa}^{\mathrm{QD}} and ΘκQD\Theta_{\kappa}^{\mathrm{QD}}, as shown in Table 2:

CκQD=7π280(WκQD+ΘκQD),ΘκQDΘI+521Θ~II.\displaystyle C_{\kappa}^{\mathrm{QD}}\,=\,\frac{7\pi^{2}}{80}\,\bigl{(}W_{\kappa}^{\mathrm{QD}}+\Theta_{\kappa}^{\mathrm{QD}}\bigr{)},\qquad\Theta_{\kappa}^{\mathrm{QD}}\,\equiv\,\Theta_{\mathrm{I}}+\frac{5}{21}\widetilde{\Theta}_{\mathrm{II}}. (74)

In contrast to ΘT\Theta_{T} of the conductance given in Eq. (71), the three-body part ΘκQD\Theta_{\kappa}^{\mathrm{QD}} of the thermal conductance depends on χB[3]\chi_{B}^{[3]} as well as χ¯C/ϵd\partial\overline{\chi}_{C}/\partial\epsilon_{d} and χ¯S/ϵd\partial\overline{\chi}_{S}/\partial\epsilon_{d}. The contribution of χB[3]\chi_{B}^{[3]} enters through χσσσ[3]\chi_{\sigma\sigma\sigma}^{[3]} and χ~σσσ[3]\widetilde{\chi}_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}, described in Eqs. (42) and (43). This component χB[3]\chi_{B}^{[3]} yields the plateau structure in ΘκQD\Theta_{\kappa}^{\mathrm{QD}}, which reflects the staircase behavior seen in Figs. 3(e) and 3(f). Figure 9 shows the NRG results for CκQDC_{\kappa}^{\mathrm{QD}}, WκQDW_{\kappa}^{\mathrm{QD}}, and ΘκQD\Theta_{\kappa}^{\mathrm{QD}} in the SU(4) and SU(6) cases.

Near half filling in the region of |ξd|U/2|\xi_{d}|\lesssim U/2, the two-body part WκQDW_{\kappa}^{\mathrm{QD}} dominates CκQDC_{\kappa}^{\mathrm{QD}} as the three-body part ΘκQD\Theta_{\kappa}^{\mathrm{QD}} disappears near half filling ξd=0\xi_{d}=0, i.e., δ=π/2\delta=\pi/2:

WκQDξd=0 1+6K~27(N1),ΘκQDξd=00.\displaystyle W_{\kappa}^{\mathrm{QD}}\xrightarrow{\,\xi_{d}=0\,}\,1+\frac{6\widetilde{K}^{2}}{7(N-1)}\,,\qquad\Theta_{\kappa}^{\mathrm{QD}}\xrightarrow{\,\xi_{d}=0\,}0\,. (75)

In particular, in the strong interaction limit UU\to\infty, the coefficient CκQDC_{\kappa}^{\mathrm{QD}} for N=4N=4 and 66 approach the values of [80/(7π2)]CκQDξd=0&U9/7[80/(7\pi^{2})]C_{\kappa}^{\mathrm{QD}}\xrightarrow{\,\xi_{d}=0\,\&\,U\to\infty\,}9/7 and 41/3541/35, respectively.

The rescaled Wilson ratio takes the values very close to the saturation value K~1\widetilde{K}\to 1 for large UU in the strong-coupling region of |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, as described in Appendix E. Similarly, in this region, the three-body correlations show the property described in Eq. (61), and thus WκQDW_{\kappa}^{\mathrm{QD}} and ΘκQD\Theta_{\kappa}^{\mathrm{QD}} take the form

WκQD\displaystyle W_{\kappa}^{\mathrm{QD}}\,\simeq 121[10+(11+18N1)(2sin2δ1)],\displaystyle\ \frac{1}{21}\left[10+\left(11+\frac{18}{N-1}\right)\left(2\sin^{2}\delta-1\right)\right],
ΘκQD\displaystyle\Theta_{\kappa}^{\mathrm{QD}}\,\simeq 1621ΘI.\displaystyle\ \frac{16}{21}\,\Theta_{\mathrm{I}}\,. (76)

Therefore, the plateau structure of WκQDW_{\kappa}^{\mathrm{QD}} is determined by the sin2δ\sin^{2}\delta term appearing in the right-hand side, while the structure of ΘκQD\Theta_{\kappa}^{\mathrm{QD}} reflects the behavior of ΘI\Theta_{\mathrm{I}} shown in Fig. 4(e) and 4(f). The coefficient CκQDC_{\kappa}^{\mathrm{QD}} changes sign in the strong-coupling region at two points of ξd\xi_{d}, which are incommensurate with the occupation number NdN_{d} in contrast to CTC_{T} that changes sign at 1/41/4 and 3/43/4 fillings.

In the opposite limit |ξd||\xi_{d}|\to\infty, both the two-body and three-body parts approach the noninteracting values,

WκQD|ξd|121,ΘκQD|ξd|2.\displaystyle W_{\kappa}^{\mathrm{QD}}\xrightarrow{\,|\xi_{d}|\to\infty\,}\,-\frac{1}{21},\qquad\Theta_{\kappa}^{\mathrm{QD}}\xrightarrow{\,|\xi_{d}|\to\infty\,}\,-2. (77)
Refer to caption
Refer to caption
Figure 10: Order T2T^{2} term CLQD=CκQDCTC_{L}^{\mathrm{QD}}=C_{\kappa}^{\mathrm{QD}}-C_{T} of Lorenz number LQDL_{\mathrm{QD}} is plotted vs ξd\xi_{d}. Left panel: N=4N=4, for U/(πΔ)=2/3()U/(\pi\Delta)=2/3(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times). Right panel: N=6N=6, for U/(πΔ)=2/5()U/(\pi\Delta)=2/5(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times).

VIII.3 CLQDC_{L}^{\mathrm{QD}}: order T2T^{2} term of LQDL_{\mathrm{QD}}

The Lorenz number LQDκQD/(gT)L_{\mathrm{QD}}\equiv\kappa_{\mathrm{QD}}/(g\,T) for quantum dots is defined as the ratio of the thermal conductance κQD/T\kappa_{\mathrm{QD}}/T to the electrical conductance gg. It takes the universal Wiedemann-Franz value at zero temperature: LQDT0π2/(3e2)L_{\mathrm{QD}}\xrightarrow{T\to 0}\pi^{2}/(3e^{2}). However, as the temperature rises, it deviates from the universal value, showing the T2T^{2} dependence described in Tables 1 and 2. The coefficient for the order T2T^{2} term is given by CLQD=CκQDCTC_{L}^{\mathrm{QD}}=C_{\kappa}^{\mathrm{QD}}-C_{T}, as a difference between the next-leading order terms of κQD/T\kappa_{\mathrm{QD}}/T and gg.

Figure 10(a) and 10(b) show the NRG results for CLQDC_{L}^{\mathrm{QD}} in the SU(4) and SU(6) cases, respectively. Near half filling |ξd|U/2|\xi_{d}|\lesssim U/2, the coefficient CLQDC_{L}^{\mathrm{QD}} is determined by the two-body part WLQDW_{L}^{\mathrm{QD}} as the three-body part ΘLQD16ΘI\Theta_{L}^{\mathrm{QD}}\equiv 16\,\Theta_{\mathrm{I}} vanishes at ξd=0\xi_{d}=0, as seen in Fig. 4(e) and 4(f):

CLQDξd=0π230(2+K~2N1)>0.\displaystyle C_{L}^{\mathrm{QD}}\xrightarrow{\,\xi_{d}=0\,}\,\frac{\pi^{2}}{30}\left(2+\frac{\widetilde{K}^{2}}{N-1}\right)\ >0\,. (78)

It takes a positive value and reaches CLQDξd=0&U=0,767C_{L}^{\mathrm{QD}}\xrightarrow{\xi_{d}=0\,\&\,U=\infty}0,767\cdots for N=4N=4, and 0.7230.723\cdots for N=6N=6, in the limit of UU\to\infty where K~1\widetilde{K}\to 1.

In the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, the coefficient CLQDC_{L}^{\mathrm{QD}} exhibits the plateau structures for large UU, reflecting the corresponding structures of CTC_{T} and CκQDC_{\kappa}^{\mathrm{QD}}. The coefficient CLQDC_{L}^{\mathrm{QD}} changes sign at the points where the order T2T^{2} terms of κQD/T\kappa_{\mathrm{QD}}/T and gg coincide, i.e., CκQD=CTC_{\kappa}^{\mathrm{QD}}=C_{T}.

In the other regions at |ξd|(N1)U/2|\xi_{d}|\gtrsim(N-1)U/2, the impurity levels approach the empty state Nd0N_{d}\simeq 0 or fully occupied NdNN_{d}\simeq N state. In particular, in the limit of |ξd||\xi_{d}|\to\infty, both the two-body and three-body parts approach the noninteracting values, WLQD|ξd|4W_{L}^{\mathrm{QD}}\xrightarrow{|\xi_{d}|\to\infty}4 and 16ΘI|ξd|3216\Theta_{\mathrm{I}}\xrightarrow{|\xi_{d}|\to\infty}-32, and thus

CLQD|ξd|7π260=1.151<0.\displaystyle C_{L}^{\mathrm{QD}}\xrightarrow{\,|\xi_{d}|\to\infty\,}\,-\frac{7\pi^{2}}{60}\ =\ -1.151\cdots\ \ <0\,. (79)

IX Fermi liquid description for
SU(NN) symmetric magnetic alloys

We have discussed, in the previous sections, low-energy transport properties of SU(NN) quantum dots, by extending the Fermi liquid description to the next-leading order terms which contribute to the transport at finite temperatures or at finite bias voltages. Our formulation is applicable to a wide class of Kondo systems other than quantum dots, particularly to dilute magnetic alloys (MA) composed of 3d3d, 4f4f, or 5f5f electrons [2]. In this and the next sections, we apply this formulation to dilute magnetic alloys away from half filling, taking into account exactly the order ω2\omega^{2} and T2T^{2} energy shifts of quasiparticles that enter through the real part of the self-energy Σσr(ω)\Sigma_{\sigma}^{r}(\omega) given in Appendix B.

The thermoelectric transport coefficients of magnetic alloys in the linear-response regime can be derived from the function n,σMA\mathcal{L}_{n,\sigma}^{\mathrm{MA}} for n=0n=0, 11, and 22, defined by [43]

n,σMA=𝑑ωωnπΔAσ(ω)(f(ω)ω).\displaystyle\mathcal{L}_{n,\sigma}^{\mathrm{MA}}=\int_{-\infty}^{\infty}d\omega\,\frac{\omega^{n}}{\pi\Delta A_{\sigma}(\omega)}\left(-\frac{\partial f(\omega)}{\partial\omega}\right)\,. (80)

Here, the inverse spectral function 1/Aσ(ω)1/A_{\sigma}(\omega) in the integrand represents the relaxation time of conduction electrons, which depends on TT as well as ω\omega. The low-temperature expansion of n,σMA\mathcal{L}_{n,\sigma}^{\mathrm{MA}} can be deduced from the exact low-energy asymptotic of 1/Aσ(ω)1/A_{\sigma}(\omega), given in Appendix B. We have presented the expansion formulas for the standard N=2N=2 Anderson impurity model in a previous paper [21]. In this work, we extend the formulation to multi-level impurities in a general form without assuming the SU(NN) symmetry. Details of the derivation are given in Appendix G.

In the following, we consider the behavior of the next-leading order terms of electrical conductivity σMA\sigma_{\mathrm{MA}} and thermal conductivity κMA\kappa_{\mathrm{MA}} of magnetic alloys in the SU(NN) symmetric case, where ϵdσϵd\epsilon_{d\sigma}\equiv\epsilon_{d} for all σ\sigma, and UσσUU_{\sigma\sigma^{\prime}}\equiv U for all σ\sigma and σ\sigma^{\prime}. In this case, the formulas given in Appendix G are simplified, and as a result, the electrical resistivity ϱMA=1/σMA\varrho_{\mathrm{MA}}=1/\sigma_{\mathrm{MA}}, the thermal resistivity 1/κMA1/\kappa_{\mathrm{MA}}, and the Lorenz number LMA=κMA/(σMAT)L_{\mathrm{MA}}=\kappa_{\mathrm{MA}}/(\sigma_{\mathrm{MA}}\,T) can be expressed in the form,

ϱMA=1σMAunit[sin2δCϱMA(πTT)2+],\displaystyle\varrho_{\mathrm{MA}}\,=\,\frac{1}{\sigma_{\mathrm{MA}}^{\mathrm{unit}}}\,\left[\,\sin^{2}\delta\,-\,C_{\varrho}^{\mathrm{MA}}\left(\frac{\pi T}{T^{*}}\right)^{2}+\cdots\,\right]\,, (81)
1κMA=3e2π2σMAunit1T[sin2δCκMA(πTT)2+],\displaystyle\!\!\frac{1}{\kappa_{\mathrm{MA}}}\,=\,\frac{3\,e^{2}}{\pi^{2}\,\sigma_{\mathrm{MA}}^{\mathrm{unit}}}\frac{1}{T}\left[\,\sin^{2}\delta-C_{\kappa}^{\mathrm{MA}}\,\left(\frac{\pi T}{T^{*}}\right)^{2}+\cdots\,\right], (82)
LMA=π23e2[ 1CLMAsin2δ(πTT)2+].\displaystyle\!\!L_{\mathrm{MA}}\,=\,\frac{\pi^{2}}{3\,e^{2}}\,\left[\,1\,-\,\frac{C_{L}^{\mathrm{MA}}}{\sin^{2}\delta}\,\left(\frac{\pi T}{T^{*}}\right)^{2}\ +\ \cdots\,\right]\,. (83)

Here, σMAunit\sigma_{\mathrm{MA}}^{\mathrm{unit}} is the unitary-limit value of electrical conductivity. The explicit expressions of the dimensionless coefficients CϱMAC_{\varrho}^{\mathrm{MA}}, CκMAC_{\kappa}^{\mathrm{MA}}, and CLMAC_{L}^{\mathrm{MA}} are listed in Table 2. These coefficients CC’s for magnetic alloys can also be decomposed into the two-body WW part and the three-body Θ\Theta part, as those for quantum dots. Note that the following relations hold between the coefficients for magnetic alloys and quantum dots:

CϱMA=\displaystyle C_{\varrho}^{\mathrm{MA}}\,= π212cos2δ+CT,\displaystyle\ \,\frac{\pi^{2}}{12}\,\cos^{2}\delta+C_{T}\,, (84)
CκMA=\displaystyle C_{\kappa}^{\mathrm{MA}}\,= 11π260cos2δ+CκQD,\displaystyle\ \,\frac{11\pi^{2}}{60}\,\cos^{2}\delta+C_{\kappa}^{\mathrm{QD}}\,, (85)
CLMA=\displaystyle C_{L}^{\mathrm{MA}}\,= π210cos2δCLQD.\displaystyle\ -\frac{\pi^{2}}{10}\,\cos^{2}\delta-C_{L}^{\mathrm{QD}}\,. (86)

In particular, the cos2δ\cos^{2}\delta term appearing in the right-hand side vanishes at half filling, i.e., δ=π/2\delta=\pi/2. In this case the coefficients for magnetic alloys in the left-hand side coincide with their quantum-dot counterparts in the right-hand side, except for the signs of CLMAC_{L}^{\mathrm{MA}} and CLQDC_{L}^{\mathrm{QD}}. The behavior of these coefficients for magnetic alloys also reflects the properties of low-lying energy states and significantly depends on the occupation number NdN_{d} and the interaction strength UU.

X Thermoelectric transport of
SU(4) &\& SU(6) magnetic alloys

We consider here the next-leading order terms of the electrical resistivity ϱMA\varrho_{\mathrm{MA}} and thermal resistivity 1/κMA1/\kappa_{\mathrm{MA}} of SU(NN) symmetric magnetic alloys for N=4N=4 and 66. For comparison, we also provide the NRG results for these transport coefficients in the SU(2) case in Appendix F, and the analytic expressions of CC’s for noninteracting magnetic alloys in Appendix H.

X.1 CϱMAC_{\varrho}^{\mathrm{MA}}: order T2T^{2} term of ϱMA\varrho_{\mathrm{MA}}

The coefficient CϱMAC_{\varrho}^{\mathrm{MA}} for the order T2T^{2} resistivity, is defined in Table 2. It consists of two-body part WϱMAW_{\varrho}^{\mathrm{MA}} and three-body part ΘϱMA\Theta_{\varrho}^{\mathrm{MA}}:

CϱMA=π248(WϱMA+ΘϱMA),ΘϱMAΘI+Θ~II.\displaystyle\!\!\!C_{\varrho}^{\mathrm{MA}}=\frac{\pi^{2}}{48}\,\left(W_{\varrho}^{\mathrm{MA}}+\Theta_{\varrho}^{\mathrm{MA}}\right),\qquad\Theta_{\varrho}^{\mathrm{MA}}\equiv\Theta_{\mathrm{I}}+\widetilde{\Theta}_{\mathrm{II}}. (87)

Note that ΘϱMAΘT\Theta_{\varrho}^{\mathrm{MA}}\equiv\Theta_{T}, i.e., the three-body part for the T2T^{2} resistivity of magnetic alloys is identical to the one for the T2T^{2} conductance of quantum dots. Therefore, ΘϱMA\Theta_{\varrho}^{\mathrm{MA}} does not depend on χB[3]\chi_{B}^{[3]} and is determined by the derivative of the charge and spin susceptibilities through Eq. (71). Figures 11(a)–11(d) show the NRG results for CϱMAC_{\varrho}^{\mathrm{MA}}, WϱMAW_{\varrho}^{\mathrm{MA}}, and ΘϱMA\Theta_{\varrho}^{\mathrm{MA}} for both the SU(4) and SU(6) symmetric cases.

The coefficient CϱMAC_{\varrho}^{\mathrm{MA}} is positive and is less sensitive to the impurity level position ξd\xi_{d} as compared to the quantum-dot counterpart CTC_{T} shown in Fig. 8. This difference is caused by the first term, (π2/12)cos2δ(\pi^{2}/12)\cos^{2}\delta, appearing in the right-hand side of Eq. (84). In particular, in the noninteracting case, it becomes a constant, (48/π2)CϱMAU=01(48/\pi^{2})C_{\varrho}^{\mathrm{MA}}\xrightarrow{U=0}1, independent of the level position, as all effects due to ϵd\epsilon_{d} are absorbed into the characteristic energy TT^{*} for U=0U=0 (see Appendix H). Therefore, it is the strong electron correlation that makes CϱMAC_{\varrho}^{\mathrm{MA}}, shown in Figs. 11(a) and 11(b), deviates from the constant value.

The three-body part almost vanishes, ΘϱMA0\Theta_{\varrho}^{\mathrm{MA}}\simeq 0, for large UU in the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, as mentioned for ΘT\Theta_{T} in Eq. (71). This is caused by the fact that the magnitudes of the derivatives, χ¯C/ϵd\partial\overline{\chi}_{C}/\partial\epsilon_{d} and χ¯S/ϵd\partial\overline{\chi}_{S}/\partial\epsilon_{d}, are significantly suppressed by the Coulomb repulsion in the wide range of electron filling 1NdN11\lesssim N_{d}\lesssim N-1, as demonstrated in Figs. 3(a)–3(d). Therefore, in this region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, the coefficient CϱMAC_{\varrho}^{\mathrm{MA}} is determined solely by the two-body part WϱMAW_{\varrho}^{\mathrm{MA}}, which takes the form

CϱMAπ248[2+(12N1)(12sin2δ)]> 0,\displaystyle C_{\varrho}^{\mathrm{MA}}\,\simeq\,\frac{\pi^{2}}{48}\left[2+\left(1-\frac{2}{N-1}\right)\left(1-2\sin^{2}\delta\right)\right]\,>\,0, (88)

as the rescaled Wilson ratio is saturated to K~1\widetilde{K}\simeq 1. Hence, the plateaus emerging around the integer filling points δ/π=1\delta/\pi=1, 22, \ldots, N1N-1, reflect the structures of the Kondo ridge occurring for the transmission probability sin2δ\sin^{2}\delta, seen in Figs. 15(b) and 16(b) in Appendix E. Among these plateaus, the one at half filling, where δ=π/2\delta=\pi/2, takes the smallest value: (48/π2)CϱMAξd=0&U5/3(48/\pi^{2})C_{\varrho}^{\mathrm{MA}}\xrightarrow{\,\xi_{d}=0\,\&\,U\to\infty}5/3 and 7/57/5 for N=4N=4 and 66, respectively. As ξd\xi_{d} moves away from half filling, the coefficient CϱMAC_{\varrho}^{\mathrm{MA}} increases in the region of |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2. Equation (88) also indicates that the plateau becomes highest at the electron fillings of Nd1N_{d}\simeq 1 and N1N-1: (48/π2)CϱMAU,δ=π/N2(48/\pi^{2})C_{\varrho}^{\mathrm{MA}}\xrightarrow{U\to\infty,\,\delta=\pi/N}2 and 23/1023/10 for N=4N=4 and 66, respectively. The NRG results for CϱMAC_{\varrho}^{\mathrm{MA}}, shown in Figs. 11(a) and 11(b), clearly demonstrate these behaviors, which are quite different form the behaviors of CTC_{T} of quantum dots. Equation (88) also shows that, in the SU(2) symmetric case where N=2N=2, the coefficient CϱMAC_{\varrho}^{\mathrm{MA}} takes the maximum value at half filling, as demonstrated also in Appendix F.

In contrast, at |ξd|(N1)U/2|\xi_{d}|\gtrsim(N-1)U/2, the occupation number approaches the empty Nd0N_{d}\simeq 0 or the fully occupied NdNN_{d}\simeq N states as the impurity level moves further away from the Fermi level. In this region, both the two-body WϱMAW_{\varrho}^{\mathrm{MA}} and three-body ΘϱMA\Theta_{\varrho}^{\mathrm{MA}} parts give comparable contributions to CϱMAC_{\varrho}^{\mathrm{MA}}, and approach the noninteracting values:

WϱMA|ξd| 3,ΘϱMA|ξd|2,\displaystyle W_{\varrho}^{\mathrm{MA}}\,\xrightarrow{\,|\xi_{d}|\to\infty}\,3,\qquad\Theta_{\varrho}^{\mathrm{MA}}\,\xrightarrow{\,|\xi_{d}|\to\infty}\,-2, (89)

and thus (48/π2)CϱMA|ξd|1(48/\pi^{2})C_{\varrho}^{\mathrm{MA}}\,\xrightarrow{\,|\xi_{d}|\to\infty}1.

Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 11: ξd\xi_{d} dependence of CϱMA=(π2/48)(WϱMA+ΘϱMA)C_{\varrho}^{\mathrm{MA}}=(\pi^{2}/48)(W_{\varrho}^{\mathrm{MA}}+\Theta_{\varrho}^{\mathrm{MA}}), two-body part WϱMAW_{\varrho}^{\mathrm{MA}}, and three-body part ΘϱMA=ΘI+Θ~II\Theta_{\varrho}^{\mathrm{MA}}=\Theta_{\mathrm{I}}+\widetilde{\Theta}_{\mathrm{II}}. Left panels: N=4N=4, for U/(πΔ)=2/3()U/(\pi\Delta)=2/3(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times). Right panels: N=6N=6, for U/(πΔ)=2/5()U/(\pi\Delta)=2/5(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times).

X.2 CκMAC_{\kappa}^{\mathrm{MA}}: order T3T^{3} term of κMA\kappa_{\mathrm{MA}}

We next consider the order T3T^{3} term of thermal conductivity κMA\kappa_{\mathrm{MA}} of the SU(NN) symmetric magnetic alloys. The coefficient CκMAC_{\kappa}^{\mathrm{MA}}, defined in Tables 1 and 2, consists of two-body WκMAW_{\kappa}^{\mathrm{MA}} and three-body ΘκMA\Theta_{\kappa}^{\mathrm{MA}} parts:

CκMA=7π280(WκMA+ΘκMA),ΘκMAΘI+521Θ~II.\displaystyle C_{\kappa}^{\mathrm{MA}}\,=\,\frac{7\pi^{2}}{80}\,\bigl{(}W_{\kappa}^{\mathrm{MA}}+\Theta_{\kappa}^{\mathrm{MA}}\bigr{)},\qquad\Theta_{\mathrm{\kappa}}^{\mathrm{MA}}\,\equiv\,\Theta_{\mathrm{I}}+\frac{5}{21}\widetilde{\Theta}_{\mathrm{II}}. (90)

Note that ΘκMAΘκQD\Theta_{\mathrm{\kappa}}^{\mathrm{MA}}\equiv\Theta_{\mathrm{\kappa}}^{\mathrm{QD}}, i.e., the three-body part for CκMAC_{\kappa}^{\mathrm{MA}} of magnetic alloys is identical to the one for the T3T^{3} thermal conductance of quantum dots. The NRG results for these coefficients CκMAC_{\kappa}^{\mathrm{MA}}, WκMAW_{\kappa}^{\mathrm{MA}}, and ΘκMA\Theta_{\kappa}^{\mathrm{MA}} are shown in Figs. 12(a)–12(d) for N=4N=4 and 66.

The coefficient CκMAC_{\kappa}^{\mathrm{MA}} for magnetic alloys is less sensitive to ξd\xi_{d} as compared to CκQDC_{\kappa}^{\mathrm{QD}} for for quantum dots. This difference is caused by the contribution of the first term, (11π2/60)cos2δ(11\pi^{2}/60)\cos^{2}\delta, in the right-hand side of Eq. (85). The coefficient CκMAC_{\kappa}^{\mathrm{MA}} is positive and has a broad peak at ξd=0\xi_{d}=0, the height of which is determined by the two-body contribution WκMAW_{\kappa}^{\mathrm{MA}} as three-body part ΘκMA\Theta_{\kappa}^{\mathrm{MA}} vanishes in the electron-hole symmetric case:

WκMAξd=0 1+6K~27(N1),ΘκMAξd=00.\displaystyle W_{\kappa}^{\mathrm{MA}}\,\xrightarrow{\,\xi_{d}=0}\,1+\frac{6\widetilde{K}^{2}}{7(N-1)}\,,\qquad\Theta_{\kappa}^{\mathrm{MA}}\xrightarrow{\,\xi_{d}=0\,}0\,. (91)

In the limit of UU\to\infty, it reaches the maximum possible value, [80/(7π2)]CκMAξd=0&U9/7[80/(7\pi^{2})]C_{\kappa}^{\mathrm{MA}}\xrightarrow{\,\xi_{d}=0\,\&\,U\to\infty}9/7 and 41/3541/35 for N=4N=4 and 66, respectively.

Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 12: ξd\xi_{d} dependence of CκMA=(7π2/80)(WκMA+ΘκMA)C_{\kappa}^{\mathrm{MA}}=(7\pi^{2}/80)(W_{\kappa}^{\mathrm{MA}}+\Theta_{\kappa}^{\mathrm{MA}}), two-body part WκMAW_{\kappa}^{\mathrm{MA}}, and three-body part ΘκMA=ΘI+521Θ~II\Theta_{\kappa}^{\mathrm{MA}}=\Theta_{\mathrm{I}}+\frac{5}{21}\widetilde{\Theta}_{\mathrm{II}}. Left panels: N=4N=4, for U/(πΔ)=2/3()U/(\pi\Delta)=2/3(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times). Right panels: N=6N=6, for U/(πΔ)=2/5()U/(\pi\Delta)=2/5(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times).

In the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, the coefficient CκMAC_{\kappa}^{\mathrm{MA}} takes the following form for large UU,

CκMAπ2240[32+(1118N1)cos2δ+ 16ΘI].\displaystyle C_{\kappa}^{\mathrm{MA}}\,\simeq\,\frac{\pi^{2}}{240}\left[32+\left(11-\frac{18}{N-1}\right)\,\cos 2\delta\,+\,16\,\Theta_{\mathrm{I}}\right]. (92)

This is because, in this region, the rescaled Wilson ratio is almost saturated K~1\widetilde{K}\simeq 1 (see Appendix E) and the three-body part is parameterized by a single component, ΘκMA1621ΘI\Theta_{\kappa}^{\mathrm{MA}}\simeq\frac{16}{21}\,\Theta_{\mathrm{I}}, due to the property described in Eq. (61). The plateau structures of CκMAC_{\kappa}^{\mathrm{MA}}, appearing in Figs. 12(a) and 12(b) around the integer filling points Nd=1N_{d}=1, 22, \ldots, N1N-1, are determined by both the cos2δ\cos 2\delta term in WκMAW_{\kappa}^{\mathrm{MA}} and the plateaus occurring in ΘκMA\Theta_{\kappa}^{\mathrm{MA}} seen in Figs. 12(c) and 12(d).

In contrast, at |ξd|(N1)U/2|\xi_{d}|\gtrsim(N-1)U/2, the electron filling approaches the empty Nd0N_{d}\simeq 0 or the fully occupied NdNN_{d}\simeq N states as the impurity level goes very far away from the half filled point. Therefore, the order T3T^{3} thermal conductivity for magnetic alloys also approaches the noninteracting value in this limit:

WκMA|ξd|4321,ΘκMA|ξd|2,\displaystyle W_{\kappa}^{\mathrm{MA}}\,\xrightarrow{\,|\xi_{d}|\to\infty}\,\frac{43}{21},\qquad\Theta_{\kappa}^{\mathrm{MA}}\,\xrightarrow{\,|\xi_{d}|\to\infty}\,-2, (93)

and thus [80/(7π2)]CκMA|ξd|1/21[80/(7\pi^{2})]\,C_{\kappa}^{\mathrm{MA}}\,\xrightarrow{\,|\xi_{d}|\to\infty}1/21.

X.3 CLMAC_{L}^{\mathrm{MA}}: order T2T^{2} term of LMAL_{\mathrm{MA}}

The Lorenz number LMAκMA/(σMAT)L_{\mathrm{MA}}\equiv\kappa_{\mathrm{MA}}/(\sigma_{\mathrm{MA}}\,T) for magnetic alloys is defined as the ratio of the thermal conductivity κMA/T\kappa_{\mathrm{MA}}/T to electrical conductivity σMA\sigma_{\mathrm{MA}}. It takes the universal Wiedemann-Franz value at zero temperature: LMAT0π2/(3e2)L_{\mathrm{MA}}\xrightarrow{T\to 0}\pi^{2}/(3e^{2}). However, it deviates from this value as the temperature increases, showing the T2T^{2} dependence as described in Eq. (83). The precise expansion formula for the coefficient CLMAC_{L}^{\mathrm{MA}} for the order T2T^{2} term is shown in Table 2. It is given by the difference, CLMA=CϱMACκMAC_{L}^{\mathrm{MA}}=C_{\varrho}^{\mathrm{MA}}-C_{\kappa}^{\mathrm{MA}}, between the order T2T^{2} term of the resistivities, 1/σMA1/\sigma_{\mathrm{MA}} and T/κMAT/\kappa_{\mathrm{MA}}, defined in Eqs. (81) and (83). The coefficient CLMAC_{L}^{\mathrm{MA}} for magnetic alloys and the quantum-dot counterpart CLQMC_{L}^{\mathrm{QM}} are related to each other through Eq. (86): these two coefficients tend to have opposite signs. In Appendix H, we have also provide an analytic formula for CLMAC_{L}^{\mathrm{MA}} in the noninteraction case, for comparison.

The NRG results for CLMAC_{L}^{\mathrm{MA}} are shown in Figs. 13(a) and 13(b) for N=4N=4 and 66, respectively. Near half filling |ξd|U/2|\xi_{d}|\lesssim U/2, the coefficient CLMAC_{L}^{\mathrm{MA}} is determined by the two-body part WLMAW_{L}^{\mathrm{MA}} as the three-body part, given by 16ΘI-16\Theta_{\mathrm{I}}, vanishes at ξd=0\xi_{d}=0:

CLMAξd=0π230(2+K~2N1)<0.\displaystyle C_{L}^{\mathrm{MA}}\,\xrightarrow{\,\xi_{d}=0}\,-\frac{\pi^{2}}{30}\left(2+\frac{\widetilde{K}^{2}}{N-1}\right)\ <0\,. (94)

This coefficient attains its greatest possible negative value in the limit of UU\to\infty where K~1\widetilde{K}\to 1: CLMAξd=0&U0,767C_{L}^{\mathrm{MA}}\xrightarrow{\xi_{d}=0\,\&\,U\to\infty}-0,767\cdots for N=4N=4, and 0.723-0.723\cdots for N=6N=6.

In the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, the CLMAC_{L}^{\mathrm{MA}} also exhibits the plateau structures around the integer filling points ξd/U=0\xi_{d}/U=0, ±1\pm 1, \ldots, ±(N2)/2\pm(N-2)/2 for large UU, reflecting the structures that appear for both CϱMAC_{\varrho}^{\mathrm{MA}} and CκMAC_{\kappa}^{\mathrm{MA}}. In particular, the plateaus at NdN2±1N_{d}\simeq\frac{N}{2}\pm 1 fillings, seen in Figs. 13(a) and 13(b) for U/(πΔ)3U/(\pi\Delta)\gtrsim 3, take a negative value for both N=4N=4 and N=6N=6, whereas the other plateaus become positive for N=6N=6 as the electrical resistivity dominates, i.e., CϱMA>CκMAC_{\varrho}^{\mathrm{MA}}>C_{\kappa}^{\mathrm{MA}}.

As the impurity level goes far away from the Fermi level in the region |ξd|(N1)U/2|\xi_{d}|\gtrsim(N-1)U/2, the occupation number approaches Nd0N_{d}\simeq 0 or NdNN_{d}\simeq N. In the limit of |ξd||\xi_{d}|\to\infty, the two-body and three-body parts take the noninteracting values, WLMA|ξd|28W_{L}^{\mathrm{MA}}\xrightarrow{|\xi_{d}|\to\infty}-28 and 16ΘI|ξd|32-16\Theta_{\mathrm{I}}\xrightarrow{|\xi_{d}|\to\infty}32, and the coefficient CLMAC_{L}^{\mathrm{MA}} converges to the positive value,

CLMA|ξd|π260= 0.164>0.\displaystyle C_{L}^{\mathrm{MA}}\,\xrightarrow{\,|\xi_{d}|\to\infty}\,\frac{\pi^{2}}{60}\ =\ 0.164\cdots\ >0\,. (95)
Refer to caption
Refer to caption
Figure 13: Order T2T^{2} term CLMA=CϱMACκMAC_{L}^{\mathrm{MA}}=C_{\varrho}^{\mathrm{MA}}-C_{\kappa}^{\mathrm{MA}} of Lorenz number LMAL_{\mathrm{MA}} is plotted vs ξd\xi_{d}. Left panel: N=4N=4, for U/(πΔ)=2/3()U/(\pi\Delta)=2/3(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times). Right panel: N=6N=6, for U/(πΔ)=2/5()U/(\pi\Delta)=2/5(\star), 1()1(\blacktriangledown), 2()2(\blacklozenge), 3()3(\blacksquare), 4()4(\blacktriangle), 5()5(\bullet), 6(×)6(\times).

XI Summary

We have presented a comprehensive Fermi liquid description for nonlinear current and thermoelectric transport through quantum dots and magnetic alloys, which is asymptotically exact at low energies up to the next-leading order terms. Our formulation is based on the multilevel Anderson model and is applicable to arbitrary impurity-level structures ϵdσ\epsilon_{d\sigma} for σ=1\sigma=1, 22, \ldots, NN, including the spin degrees of freedom. The coefficients for the next-leading order terms have been shown to be expressed in terms of a set of the correlation functions defined with respect to the equilibrium ground state: the phase shift δσ\delta_{\sigma}, the static susceptibilities χσ1,σ2\chi_{\sigma_{1},\sigma_{2}}, and the three-body correlations χσ1,σ2,σ3[3]\chi_{\sigma_{1},\sigma_{2},\sigma_{3}}^{[3]} which emerge when the system does not have both the electron-hole and time-reversal symmetries.

Extending Yamada-Yosida’s field-theoretical approach, we have obtained the formulas for the differential conductance dI/dVdI/dV, current noise SnoiseQDS_{\mathrm{noise}}^{\mathrm{QD}} and thermal conductance κQD\kappa_{\mathrm{QD}} of quantum dots, and also the electrical resistivity ϱMA\varrho_{\mathrm{MA}} and thermal conductivity κMA\kappa_{\mathrm{MA}} of dilute magnetic alloys. In the SU(NN) symmetric case, these transport coefficients take simplified forms, as listed in Tables 1 and 2, and the three-body correlations can be deduced from the derivatives of the susceptibilities: χ¯C/ϵd\partial\overline{\chi}_{C}/\partial\epsilon_{d}, χ¯S/ϵd\partial\overline{\chi}_{S}/\partial\epsilon_{d}, and χB[3]\chi_{B}^{[3]} through Eqs.  (42)–(45).

We have also calculated the correlation functions for the SU(4) and SU(6) cases, using the NRG approach over the whole region of impurity-electron fillings NdN_{d}, which includes the Kondo and the valence-fluctuation regimes. In the SU(NN) case, the three-body correlations have three linearly independent components that approach each other closely as the Coulomb interaction UU increases: χσσσ[3](N1)χσσσ[3](N1)(N2)2χσσσ′′[3]\chi_{\sigma\sigma\sigma}^{[3]}\simeq-(N-1)\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\simeq\frac{(N-1)(N-2)}{2}\chi_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]} for σσσ′′σ\sigma\neq\sigma^{\prime}\neq\sigma^{\prime\prime}\neq\sigma, in a wide filling range 1NdN11\lesssim N_{d}\lesssim N-1. This is caused by the suppression of the derivatives of the charge and spin susceptibilities, occurring at large UU: |χ¯C/ϵd|(T)2|\partial\overline{\chi}_{C}/\partial\epsilon_{d}|\ll(T^{*})^{-2} and |χ¯S/ϵd|(T)2|\partial\overline{\chi}_{S}/\partial\epsilon_{d}|\ll(T^{*})^{-2}, with T1/(4χσσ)T^{*}\equiv 1/(4\chi_{\sigma\sigma}) the characteristic energy scale of the SU(NN) Fermi liquid. This property of three-body correlations is also related to a similar property of linear susceptibilities, χσσ(N1)χσσ\chi_{\sigma\sigma}\simeq-(N-1)\chi_{\sigma\sigma^{\prime}}, which reflects the suppression of charge fluctuations, χ¯C0\overline{\chi}_{C}\simeq 0, occurring at large UU.

The coefficients CC’s for the next-leading order terms can be decomposed into the two-body part WW’s and the three-body part Θ\Theta’s, as listed in Table 2. The NRG results show that the three-body part ΘV\Theta_{V} of the coefficient CVC_{V} for the order (eV)2(eV)^{2} term of dI/dVdI/dV exhibits Kondo plateau structures at integer filling points Nd=1N_{d}=1, 22, \ldots, N1N-1 for large UU. These plateaus of ΘV\Theta_{V} complement the two body part WVW_{V}, which decreases away from half filling, to form a wide ridge structure in CVC_{V} that spreads over the region of 1NdN11\lesssim N_{d}\lesssim N-1.

The linear-response term of current noise SnoiseQDS_{\mathrm{noise}}^{\mathrm{QD}} is maximized at quarter filling Nd/N=1/4N_{d}/N=1/4 and three-quarters 3/43/4 filling, and the peaks exhibit flat structures for N=4N=4, while they are round for N=6N=6. This difference is caused by the fact that, at these fillings, the SU(NN) Kondo effects occur for N0N\equiv 0 (mod 44), while the intermediate valence fluctuations occur for N2N\equiv 2 (mod 44). The coefficient CSC_{S} for the order |eV|3|eV|^{3} nonlinear term of SnoiseQDS_{\mathrm{noise}}^{\mathrm{QD}} has a peak at half filling, which evolves into a plateau as UU increases. As the impurity level ξd\xi_{d} deviates from the half filling point, CSC_{S} decreases rapidly for N=4N=4, whereas it varies more modestly for N=6N=6. This is mainly due to the higher-harmonic “sin4δ\sin 4\delta” dependence of the three-body part ΘS\Theta_{S}, which vanishes at the quarter and three-quarters filling points. As |ξd||\xi_{d}| increases further, ΘS\Theta_{S} has a pronounced negative minimum in the valence fluctuation regions at ξd±(N1)U/2\xi_{d}\simeq\pm(N-1)U/2 for both N=4N=4 and 66, which yield the valley structures appearing in CSC_{S}. This is in marked contrast to the SU(2) case, where the valley structure emerging in CSC_{S} is caused by the two-body contributions WSW_{S}, instead of ΘS\Theta_{S}.

We have also studied the coefficient CTC_{T} for the order T2T^{2} term of the linear conductance gdJ/dV|eV=0g\equiv\left.dJ/dV\right|_{eV=0} and the coefficient CκQDC_{\kappa}^{\mathrm{QD}} for the order T3T^{3} term of thermal conductance κQD\kappa_{\mathrm{QD}}, for SU(NN) quantum dots. The three-body part ΘT\Theta_{T} of CTC_{T} is determined solely by the derivatives of charge and spin susceptibilities, i.e., χ¯C/ϵd\partial\overline{\chi}_{C}/\partial\epsilon_{d} and χ¯S/ϵd\partial\overline{\chi}_{S}/\partial\epsilon_{d}, and is independent of χB[3]\chi_{B}^{[3]}. Therefore, ΘT\Theta_{T} is significantly suppressed and almost vanishes for strong interactions UU in a wide range of the electron filling 1NdN11\lesssim N_{d}\lesssim N-1. In contrast, the three-body part ΘκQD\Theta_{\kappa}^{\mathrm{QD}} for thermal conductance involves χB[3]\chi_{B}^{[3]} and becomes comparable to the two-body part WκQDW_{\kappa}^{\mathrm{QD}}, except for the plateau region near half filling. The Lorenz number LQDκQD/(gT)L_{\mathrm{QD}}\equiv\kappa_{\mathrm{QD}}/(g\,T) for quantum dots takes the universal Wiedemann-Franz value π2/(3e2)\pi^{2}/(3e^{2}) at T=0T=0 and shows a T2T^{2} dependence at low temperatures. The coefficient for the order T2T^{2} term of LQDL_{\mathrm{QD}} is given by CLQD=CκQDCTC_{L}^{\mathrm{QD}}=C_{\kappa}^{\mathrm{QD}}-C_{T}, i.e., the difference between the next-leading order terms of κQD\kappa_{\mathrm{QD}} and gg. This coefficient CLQDC_{L}^{\mathrm{QD}} also exhibits the Kondo plateau structures at integer filling points Nd=1N_{d}=1, 22, \ldots, N1N-1.

The three-body correlations also play a significant role in the low-energy transport of magnetic alloys. We have investigated the behaviors of the coefficient CϱMAC_{\varrho}^{\mathrm{MA}} for the order T2T^{2} resistivity ϱMA\varrho_{\mathrm{MA}} and CκMAC_{\kappa}^{\mathrm{MA}} for the order T3T^{3} thermal conductivity κMA\kappa_{\mathrm{MA}}. In the SU(NN) symmetric case, the three-body parts of these coefficients become identical to the related ones for quantum dots, i.e., ΘϱMAΘT\Theta_{\varrho}^{\mathrm{MA}}\equiv\Theta_{T} and ΘκMAΘκQD\Theta_{\kappa}^{\mathrm{MA}}\equiv\Theta_{\kappa}^{\mathrm{QD}}. Therefore, the difference between the coefficients for MA and that for the QD counterparts arises from the two-body parts, more specifically, from the additional cos2δ\cos^{2}\delta terms appearing in the right-hand side of Eqs. (84) and (85). The additional cos2δ\cos^{2}\delta terms make the coefficients CϱMAC_{\varrho}^{\mathrm{MA}} and CκMAC_{\kappa}^{\mathrm{MA}} positive definite and less sensitive to the impurity level position ξd\xi_{d}, compared to the QD counterparts CTC_{T} and CκQDC_{\kappa}^{\mathrm{QD}}. The coefficient CϱMAC_{\varrho}^{\mathrm{MA}} takes the maximum value at the electron fillings of Nd1N_{d}\simeq 1 and N1N-1 for N4N\geq 4, while in the SU(2) case CϱMAC_{\varrho}^{\mathrm{MA}} has a single peak at half filling. The coefficient CLMA=CϱMACκMAC_{L}^{\mathrm{MA}}=C_{\varrho}^{\mathrm{MA}}-C_{\kappa}^{\mathrm{MA}} for the order T2T^{2} term of the Lorenz number LMAκMA/(σMAT)L_{\mathrm{MA}}\equiv\kappa_{\mathrm{MA}}/(\sigma_{\mathrm{MA}}\,T) for magnetic alloys also becomes less sensitive to the electron filling NdN_{d} compared to CLQDC_{L}^{\mathrm{QD}} for quantum dots.

The three-body correlation functions can be determined experimentally by measuring the coefficients CC’s for the next leading order terms. These experimental values can then be used to infer the behaviors of other unmeasured transport coefficients.

Acknowledgements.
This work was supported by JSPS KAKENHI Grants No. JP18K03495, No. JP18J10205, No. JP21K03415, and No. 23K03284 and JST CREST Grant No. JPMJCR1876. K. M.  was supported by JST Establishment of University Fellowships towards the Creation of Science Technology Innovation Grant No. JPMJFS2138. Y. T.  was supported by Sasakawa Scientific Research Grant from the Japan Science Society No. 2021-2009, and by the Shigemasa and Shigeaki Nakazawa Fellowship of Graduate School of Science, Osaka City University.

Appendix A Fermi liquid parameters

A.1 Linear and nonlinear static susceptibilities

The ground-state properties and the leading Fermi liquid corrections due to the low-lying excitations can be described in terms of the occupation number and the linear susceptibilities of the impurity level, derived from the free energy Ω(1/β)log[Treβ]\Omega\equiv-(1/\beta)\,\log\,\bigl{[}\,\mathrm{Tr}\,e^{-\beta\mathcal{H}}\,\bigr{]}:

ndσ=\displaystyle\bigl{\langle}n_{d\sigma}\bigr{\rangle}\,= Ωϵdσ,\displaystyle\ \frac{\partial\Omega}{\partial\epsilon_{d\sigma}}\,, (96)
χσσ\displaystyle\chi_{\sigma\sigma^{\prime}}\,\equiv 2Ωϵdσϵdσ=0β𝑑τδndσ(τ)δndσ.\displaystyle\ -\frac{\partial^{2}\Omega}{\partial\epsilon_{d\sigma}\partial\epsilon_{d\sigma^{\prime}}}\,=\,\int_{0}^{\beta}\!d\tau\,\bigl{\langle}\delta n_{d\sigma}(\tau)\,\delta n_{d\sigma^{\prime}}\bigr{\rangle}\,. (97)

Here, δndσndσndσ\delta n_{d\sigma}\equiv n_{d\sigma}-\langle n_{d\sigma}\rangle, and thermal-equilibrium averages are defined as O=Tr[eβ𝒪]/Treβ\langle O\rangle=\mathrm{Tr}\,\bigl{[}\,e^{-\beta\mathcal{H}}\,\mathcal{O}\,\bigr{]}/\mathrm{Tr}\,e^{-\beta\mathcal{H}}, with β=1/T\beta=1/T the inverse temperature.

In addition to the linear susceptibilities, the nonlinear susceptibilities χσ1σ2σ3[3]\chi_{\sigma_{1}\sigma_{2}\sigma_{3}}^{[3]} play an essential role in the next-leading order terms of the transport coefficients when the system does not have both the electron-hole and time-reversal symmetries:

χσ1σ2σ3[3]3Ωϵdσ1ϵdσ2ϵdσ3=χσ1σ2ϵdσ3\displaystyle\chi_{\sigma_{1}\sigma_{2}\sigma_{3}}^{[3]}\,\equiv\,-\frac{\partial^{3}\Omega}{\partial\epsilon_{d\sigma_{1}}\partial\epsilon_{d\sigma_{2}}\partial\epsilon_{d\sigma_{3}}}\,=\,\frac{\partial\chi_{\sigma_{1}\sigma_{2}}}{\partial\epsilon_{d\sigma_{3}}}
=\displaystyle= 0β𝑑τ10β𝑑τ2Tτδndσ1(τ1)δndσ2(τ2)δndσ3.\displaystyle-\int_{0}^{\beta}\!d\tau_{1}\!\int_{0}^{\beta}\!d\tau_{2}\,\bigl{\langle}T_{\tau}\delta n_{d\sigma_{1}}(\tau_{1})\,\delta n_{d\sigma_{2}}(\tau_{2})\,\delta n_{d\sigma_{3}}\bigr{\rangle}\,. (98)

Here, TτT_{\tau} is the imaginary-time ordering operator. This correlation function has the permutation symmetry: χσ1σ2σ3[3]=χσ2σ1σ3[3]=χσ3σ2σ1[3]=χσ1σ3σ2[3]=.\chi_{\sigma_{1}\sigma_{2}\sigma_{3}}^{[3]}=\chi_{\sigma_{2}\sigma_{1}\sigma_{3}}^{[3]}=\chi_{\sigma_{3}\sigma_{2}\sigma_{1}}^{[3]}=\chi_{\sigma_{1}\sigma_{3}\sigma_{2}}^{[3]}=\cdots. Specifically, in our formulation we are using the ground state values for ndσ\langle n_{d\sigma}\rangle, χσσ\chi_{\sigma\sigma^{\prime}} and χσ1σ2σ3[3]\chi_{\sigma_{1}\sigma_{2}\sigma_{3}}^{[3]}, determined at T=0T=0.

The occupation number can be related to the phase shift δσ\delta_{\sigma} through the Friedel sum rule: ndσT0δσ/π\langle n_{d\sigma}\rangle\xrightarrow{\,T\to 0\,}\delta_{\sigma}/\pi. The phase shift corresponds to the argument of the Green’s function, given by Gσr(0)=|Gσr(0)|eiδσG_{\sigma}^{r}(0)=-\left|G_{\sigma}^{r}(0)\right|e^{i\delta_{\sigma}}, at ω=T=eV=0\omega=T=eV=0 [8], and determines the value of the spectral function ρdσ(ω)\rho_{d\sigma}(\omega) at ω=0\omega=0:

ρdσ(ω)\displaystyle\rho_{d\sigma}(\omega)\,\equiv Aσ(ω)|T=eV=0,\displaystyle\ A_{\sigma}(\omega)\Big{|}_{T=eV=0}\,, (99)
ρdσ\displaystyle\rho_{d\sigma}\,\equiv ρdσ(0)=sin2δσπΔ,\displaystyle\ \rho_{d\sigma}(0)\,=\,\frac{\sin^{2}\delta_{\sigma}}{\pi\Delta}\,, (100)

where Aσ(ω)A_{\sigma}(\omega) is the nonequilibrium spectral function defined in Eq. (9). The derivative of ρdσ(ω)\rho_{d\sigma}(\omega) also contributes to the next-leading order terms and can be expressed in terms of the diagonal susceptibility χσσ\chi_{\sigma\sigma}, using Eq. (106),

ρdσρdσ(ω)ω|ω=0=ρdσϵdσ=χσσΔsin2δσ.\displaystyle\rho_{d\sigma}^{\prime}\equiv\left.\frac{\partial\rho_{d\sigma}(\omega)}{\partial\omega}\right|_{\omega=0}=\,-\frac{\partial\rho_{d\sigma}}{\partial\epsilon_{d\sigma}}\ =\ \frac{\chi_{\sigma\sigma}}{\Delta}\,\sin 2\delta_{\sigma}. (101)

One of the most typical Fermi liquid corrections due to many-body scatterings arises in the TT-linear specific heat 𝒞impheat\mathcal{C}_{\mathrm{imp}}^{\mathrm{heat}} of impurity electrons:

𝒞impheat=γimpT,γimpπ23σχσσ.\displaystyle\mathcal{C}_{\mathrm{imp}}^{\mathrm{heat}}\,=\,\gamma_{\mathrm{imp}}\,T\,,\qquad\gamma_{\mathrm{imp}}\,\equiv\,\frac{\pi^{2}}{3}\sum_{\sigma}\chi_{\sigma\sigma}\,. (102)

The coefficient γimp\gamma_{\mathrm{imp}} can be expressed in terms of the diagonal components of the linear susceptibility χσσ\chi_{\sigma\sigma} using the Ward identities [6, 8, 9], i.e., Eqs. (106) and (107), which follow from a relationship between the derivative of the self-energy with respect to ω\omega and the derivative with respect to ϵdσ\epsilon_{d\sigma}.

A.2 Ward identities

The local Fermi liquid state of quantum impurity systems can be microscopically described using the retarded Green’s function, defined in Eq. (8), which can also be written in the form,

Gσr(ω)=\displaystyle G_{\sigma}^{r}(\omega)\,= 1ωϵdσ+iΔΣσr(ω).\displaystyle\ \frac{1}{\omega-\epsilon_{d\sigma}+i\Delta-\Sigma_{\sigma}^{r}(\omega)}\,. (103)

The information about the low-lying energy states can be extracted from the equilibrium self-energy Σeq,σr(ω)Σσr(ω)|T=eV=0\Sigma_{\mathrm{eq},\sigma}^{r}(\omega)\equiv\left.\Sigma_{\sigma}^{r}(\omega)\right|_{T=eV=0}, by expanding it, step by step, around the Fermi energy ω=0\omega=0. The expansion up to linear terms in ω\omega describes the renormalized resonance state of the form,

Gσr(ω)\displaystyle G_{\sigma}^{r}(\omega)\,\simeq zσωϵ~dσ+iΔ~σ.\displaystyle\ \frac{z_{\sigma}}{\omega-\widetilde{\epsilon}_{d\sigma}+i\widetilde{\Delta}_{\sigma}}\,. (104)

Here, the renormalized parameters are defined by

ϵ~dσ\displaystyle\widetilde{\epsilon}_{d\sigma}\,\equiv zσ[ϵdσ+Σeq,σr(0)]=Δ~σcotδσ,\displaystyle\ z_{\sigma}\left[\epsilon_{d\sigma}+\Sigma_{\mathrm{eq},\sigma}^{r}(0)\right]\,=\,\widetilde{\Delta}_{\sigma}\cot\delta_{\sigma}\,,
Δ~σ\displaystyle\widetilde{\Delta}_{\sigma}\,\equiv zσΔ,1zσ 1Σeq,σr(ω)ω|ω=0.\displaystyle\ z_{\sigma}\Delta,\qquad\frac{1}{z_{\sigma}}\,\equiv\ 1-\left.\frac{\partial\Sigma_{\mathrm{eq},\sigma}^{r}(\omega)}{\partial\omega}\right|_{\omega=0}.\rule{0.0pt}{17.07182pt} (105)

The wavefunction renormalization factor zσz_{\sigma} can be related to the derivative of Σeq,σr(0)\Sigma_{\mathrm{eq},\sigma}^{r}(0) with respect to the impurity level ϵdσ\epsilon_{d\sigma^{\prime}}, using the Ward identity [6, 8, 9]:

1zσ=χ~σσ,χ~σσδσσ+Σeq,σr(0)ϵdσ.\displaystyle\frac{1}{z_{\sigma}}\,=\,\widetilde{\chi}_{\sigma\sigma}\,,\qquad\quad\widetilde{\chi}_{\sigma\sigma^{\prime}}\,\equiv\,\delta_{\sigma\sigma^{\prime}}+\frac{\partial\Sigma_{\mathrm{eq},\sigma}^{r}(0)}{\partial\epsilon_{d\sigma^{\prime}}}\,. (106)

The coefficient χ~σσ\widetilde{\chi}_{\sigma\sigma^{\prime}} determines the extent to which the susceptibility is enhanced at T=0T=0:

χσσ=ndσϵdσT0ρdσχ~σσ.\displaystyle\chi_{\sigma\sigma^{\prime}}\,=\,-\frac{\partial\bigl{\langle}n_{d\sigma}\bigr{\rangle}}{\partial\epsilon_{d\sigma^{\prime}}}\ \xrightarrow{\,T\to 0\,}\ \rho_{d\sigma}\widetilde{\chi}_{\sigma\sigma^{\prime}}\,. (107)

Recently studies have been clarified that the order ω2\omega^{2} real part of the self-energy can be expressed in terms of the the diagonal component of the three-body correlation function, χσσσ[3]\chi_{\sigma\sigma\sigma}^{[3]}, as [18, 19, 20, 21]

2ω2ReΣeq,σr(ω)|ω0=2Σeq,σr(0)ϵdσ2=χ~σσϵdσ.\displaystyle\!\!\!\left.\frac{\partial^{2}}{\partial\omega^{2}}\mathrm{Re}\,\Sigma_{\mathrm{eq},\sigma}^{r}(\omega)\right|_{\omega\to 0}\,=\,\frac{\partial^{2}\Sigma_{\mathrm{eq},\sigma}^{r}(0)}{\partial\epsilon_{d\sigma}^{2}}\,\ \ =\ \frac{\partial\widetilde{\chi}_{\sigma\sigma}}{\partial\epsilon_{d\sigma}}\,. (108)

Physically, this coefficient determines the energy shifts of quasiparticles of order ω2\omega^{2}, T2T^{2} and (eV)2(eV)^{2}, which affect the low-energy transport of the next-leading order.

Appendix B Low-energy asymptotic form of spectral function

The low-energy asymptotic form of the retarded self-energy Σσr(ω)\Sigma_{\sigma}^{r}(\omega) for multilevel Anderson impurity model has been derived up to terms of order ω2\omega^{2}, T2T^{2}, and (eV)2(eV)^{2} in the previous works [27, 29]. For symmetric tunnel junctions with ΓL=ΓR\Gamma_{L}=\Gamma_{R} (=Δ/2=\Delta/2) and μL=μR\mu_{L}=-\mu_{R} (=eV/2=eV/2), it takes the form,

ImΣσr(ω)=\displaystyle\mathrm{Im}\,\Sigma_{\sigma}^{r}(\omega)\,= π21ρdσσ(σ)χσσ2[ω2+34(eV)2+(πT)2]+,\displaystyle\ -\,\frac{\pi}{2}\,\frac{1}{\rho_{d\sigma}}\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}}^{2}\,\left[\,\omega^{2}+\frac{3}{4}\,(eV)^{2}+(\pi T)^{2}\,\right]\ +\,\cdots\,, (109)
ϵdσ+ReΣσr(ω)=\displaystyle\epsilon_{d\sigma}+\mathrm{Re}\,\Sigma_{\sigma}^{r}(\omega)\,= Δcotδσ+(1χ~σσ)ω+12χ~σσϵdσω2+161ρdσσ(σ)χσσσ[3][34(eV)2+(πT)2]+.\displaystyle\ \ \Delta\,\cot\delta_{\sigma}\,+\bigl{(}1-\widetilde{\chi}_{\sigma\sigma}\bigr{)}\,\omega\,+\frac{1}{2}\,\frac{\partial\widetilde{\chi}_{\sigma\sigma}}{\partial\epsilon_{d\sigma}}\,\omega^{2}\,+\frac{1}{6}\,\frac{1}{\rho_{d\sigma}}\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\left[\frac{3}{4}\,(eV)^{2}+\left(\pi T\right)^{2}\right]\ +\,\cdots\,. (110)

Substituting these expansion results into Eq. (103), we obtain the asymptotic form spectral function,

πΔAσ(ω)=\displaystyle\pi\Delta\,A_{\sigma}(\omega)\,= sin2δσ+π23(32cos2δσσ(σ)χσσ2sin2δσ2πσ(σ)χσσσ[3])[34(eV)2+(πT)2]\displaystyle\ \sin^{2}\delta_{\sigma}+\frac{\pi^{2}}{3}\left(\frac{3}{2}\cos 2\delta_{\sigma}\,\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}}^{2}-\frac{\sin 2\delta_{\sigma}}{2\pi}\,\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\,\right)\left[\,\frac{3}{4}\left(eV\right)^{2}+\left(\pi T\right)^{2}\,\right]
+πsin2δσχσσω+π2[cos2δσ(χσσ2+12σ(σ)χσσ2)sin2δσ2πχσσσ[3]]ω2+.\displaystyle\ +\pi\sin 2\delta_{\sigma}\,\chi_{\sigma\sigma}\,\omega+\pi^{2}\left[\,\cos 2\delta_{\sigma}\left(\chi_{\sigma\sigma}^{2}+\frac{1}{2}\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}}^{2}\right)-\frac{\sin 2\delta_{\sigma}}{2\pi}\,\chi_{\sigma\sigma\sigma}^{[3]}\,\right]\,\omega^{2}\ +\ \cdots. (111)

Correspondingly, the inverse of the spectral function that determines the thermoelectric transport of magnetic alloys takes the following form, up to terms order ω2\omega^{2} and T2T^{2} at eV=0eV=0,

1πΔAσ(ω)\displaystyle\frac{1}{\pi\Delta A_{\sigma}(\omega)}\,\simeq 1πΔρdσ[116Δρdσ( 3πcos2δσσ(σ)χσσ2sin2δσσ(σ)χσσσ[3])(πT)2sin2δσχσσΔρdσω\displaystyle\ \frac{1}{\pi\Delta\rho_{d\sigma}}\Biggl{[}1-\frac{1}{6\Delta\rho_{d\sigma}}\left(\,3\pi\cos 2\delta_{\sigma}\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}}^{2}-\sin 2\delta_{\sigma}\,\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\,\right)\left(\pi T\right)^{2}-\frac{\sin 2\delta_{\sigma}\,\chi_{\sigma\sigma}}{\Delta\rho_{d\sigma}}\,\omega
+πΔρdσ{(cos2δσ+2)χσσ212cos2δσσ(σ)χσσ2+sin2δσ2πχσσσ[3]}ω2]+.\displaystyle\qquad\qquad+\frac{\pi}{\Delta\rho_{d\sigma}}\left\{\,\left(\cos 2\delta_{\sigma}+2\right)\,\chi_{\sigma\sigma}^{2}-\frac{1}{2}\cos 2\delta_{\sigma}\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}}^{2}+\frac{\sin 2\delta_{\sigma}}{2\pi}\,\chi_{\sigma\sigma\sigma}^{[3]}\,\right\}\,\omega^{2}\Biggr{]}+\cdots\;. (112)

Appendix C Properties of χσ1σ2σ3[3]\chi_{\sigma_{1}\sigma_{2}\sigma_{3}}^{[3]} in SU(NN) case

We briefly describe here some relations between the three-body correlation functions and the derivative of the linear susceptibilities with respect to the center of mass coordinate of the impurity levels, ϵd(1/N)σϵdσ\epsilon_{d}\equiv(1/N)\sum_{\sigma}\epsilon_{d\sigma}.

The derivative of the diagonal susceptibility χσσ\chi_{\sigma\sigma} can be written as

χσσϵd\displaystyle\frac{\partial\chi_{\sigma\sigma}}{\partial\epsilon_{d}} =χσσϵdσ+σ(σ)χσσϵdσ\displaystyle=\ \frac{\partial\chi_{\sigma\sigma}}{\partial\epsilon_{d\sigma}}+\sum_{\sigma^{\prime}(\neq\sigma)}\frac{\partial\chi_{\sigma\sigma}}{\partial\epsilon_{d\sigma^{\prime}}}
SU(N)χσσσ[3]+(N1)χσσσ[3],\displaystyle\xrightarrow{\,\mathrm{SU}(N)\,}\,\chi_{\sigma\sigma\sigma}^{[3]}+(N-1)\,\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\,, (113)

where σσ\sigma\neq\sigma^{\prime}. Note that χσσσ[3]=χσσσ[3]\chi_{\sigma\sigma\sigma^{\prime}}^{[3]}=\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]} in the SU(NN) symmetric case. Similarly, the derivative of the off-diagonal susceptibility χσσ\chi_{\sigma\sigma^{\prime}} for σσ\sigma\neq\sigma^{\prime} takes the form

χσσϵd\displaystyle\frac{\partial\chi_{\sigma\sigma^{\prime}}}{\partial\epsilon_{d}} =χσσϵdσ+χσσϵdσ+σ′′(σσ)χσσϵdσ′′\displaystyle=\ \frac{\partial\chi_{\sigma\sigma^{\prime}}}{\partial\epsilon_{d\sigma}}+\frac{\partial\chi_{\sigma\sigma^{\prime}}}{\partial\epsilon_{d\sigma^{\prime}}}+\sum_{\sigma^{\prime\prime}(\neq\sigma\atop\neq\sigma^{\prime})}\frac{\partial\chi_{\sigma\sigma^{\prime}}}{\partial\epsilon_{d\sigma^{\prime\prime}}}
SU(N) 2χσσσ[3]+(N2)χσσσ′′[3],\displaystyle\xrightarrow{\,\mathrm{SU}(N)\,}\,2\,\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}+(N-2)\,\chi_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]}\;, (114)

for σσσ′′σ\sigma\neq\sigma^{\prime}\neq\sigma^{\prime\prime}\neq\sigma. In the SU(2) symmetric case, Eqs. (113) and (114) provide enough information to determine the two independent components χσσσ[3]\chi_{\sigma\sigma\sigma}^{[3]} and χσσσ[3]\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]} from the two differential coefficients χσσ/ϵd\partial\chi_{\sigma\sigma}/\partial\epsilon_{d} and χσσ/ϵd\partial\chi_{\sigma\sigma^{\prime}}/\partial\epsilon_{d}. However, for N3N\geq 3, there are three independent three-body components, i.e., χσσσ[3]\chi_{\sigma\sigma\sigma}^{[3]}, χσσσ[3]\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}, and χσσσ′′[3]\chi_{\sigma\sigma^{\prime}\sigma^{\prime\prime}}^{[3]}, so that we need additional information to determine all these components.

In order to obtain another independent relation, we consider the derivative of the susceptibilities with respect to the magnetic field bb, which induces the level splitting in impurity levels ϵdσ\epsilon_{d\sigma} in a such way that ϵd,m,=ϵdb\epsilon_{d,m,\uparrow}=\epsilon_{d}-b and ϵd,m,=ϵd+b\epsilon_{d,m,\downarrow}=\epsilon_{d}+b, with σ=(m,s)\sigma=(m,s) for m=1,2,,N/2m=1,2,\ldots,N/2 and s=,s=\uparrow,\downarrow:

χm1s1,m2s2b=m3=1N/2s3=,ϵd,m3,s3bχm1s1,m2s2ϵd,m3,s3\displaystyle\!\!\!\frac{\partial\chi_{m_{1}s_{1},m_{2}s_{2}}}{\partial b}\,=\,\sum_{m_{3}=1}^{N/2}\sum_{s_{3}=\uparrow,\downarrow}\frac{\partial\epsilon_{d,m_{3},s_{3}}}{\partial b}\,\frac{\partial\chi_{m_{1}s_{1},m_{2}s_{2}}}{\partial\epsilon_{d,m_{3},s_{3}}}
=m3=1N/2χm1s1,m2s2,m3[3]+m3=1N/2χm1s1,m2s2,m3[3].\displaystyle\!\!\!\!=\,-\sum_{m_{3}=1}^{N/2}\,\chi_{m_{1}s_{1},m_{2}s_{2},m_{3}\uparrow}^{[3]}\,+\sum_{m_{3}=1}^{N/2}\chi_{m_{1}s_{1},m_{2}s_{2},m_{3}\downarrow}^{[3]}. (115)

From this derivative, we obtain the following relation, taking m1=m2(m)m_{1}=m_{2}\,(\equiv m),

b(χm,mχm,m2)\displaystyle\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\frac{\partial}{\partial b}\left(\frac{\chi_{m\uparrow,m\uparrow}-\chi_{m\downarrow,m\downarrow}}{2}\right)
=\displaystyle= m=1N/212(χm,m,m[3]χm,m,m[3]\displaystyle\ \ -\sum_{m^{\prime}=1}^{N/2}\frac{1}{2}\,\biggl{(}\chi_{m\uparrow,m\uparrow,m^{\prime}\uparrow}^{[3]}-\chi_{m\uparrow,m\uparrow,m^{\prime}\downarrow}^{[3]}
+χm,m,m[3]χm,m,m[3])\displaystyle\qquad\qquad\quad+\chi_{m\downarrow,m\downarrow,m^{\prime}\downarrow}^{[3]}-\chi_{m\downarrow,m\downarrow,m^{\prime}\uparrow}^{[3]}\biggr{)}
b0\displaystyle\xrightarrow{\,b\to 0\,} m3=1N/2[χm,m,m3[3]χm,m,m3[3]]\displaystyle\,-\sum_{m_{3}=1}^{N/2}\left[\,\chi_{m\uparrow,m\uparrow,m_{3}\uparrow}^{[3]}\,-\,\chi_{m\uparrow,m\uparrow,m_{3}\downarrow}^{[3]}\,\right]
=\displaystyle= (χm,m,m[3]χm,m,m[3]).\displaystyle\ -\left(\chi_{m\uparrow,m\uparrow,m\uparrow}^{[3]}\,-\,\chi_{m\uparrow,m\uparrow,m\downarrow}^{[3]}\right). (116)

Here, we set the magnetic field to be zero, b=0b=0, in the last two lines. This relation can also be rewritten into the following form at b=0b=0, using the original label σ=(m,s)\sigma=(m,s),

χB[3]b(χm,mχm,m2)b=0=χσσσ[3]+χσσσ[3],\displaystyle\chi_{B}^{[3]}\,\equiv\,\frac{\partial}{\partial b}\left(\frac{\chi_{m\uparrow,m\uparrow}-\chi_{m\downarrow,m\downarrow}}{2}\right)_{b=0}=\,-\chi_{\sigma\sigma\sigma}^{[3]}+\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}, (117)

for σσ\sigma^{\prime}\neq\sigma.

Refer to caption
Figure 14: Comparison of (\bullet) NRG and (—) exact results of sin2δ\sin^{2}\delta for U=0U=0. The NRG calculation was performed for N=6N=6, taking Λ=20\Lambda=20 and keeping Ntrunc=40000N_{\mathrm{trunc}}=40000 low-energy states.

Appendix D NRG procedures

We have carried out NRG calculations, dividing the NN conduction channels into N/2N/2 pairs and using the SU(2) spin and U(1) charge symmetries for each of the pairs, i.e., k=1N2{SU(2)U(1)}k\prod_{k=1}^{\frac{N}{2}}\left\{\mbox{SU(2)}\otimes\mbox{U(1)}\right\}_{k} symmetries. The discretization parameter Λ\Lambda and the number of retained low-lying excited states NtruncN_{\mathrm{trunc}} are chosen to be (Λ,Ntrunc)=(6,10000)(\Lambda,N_{\mathrm{trunc}})=(6,10000) for N=4N=4. Note that the SU(4) symmetry is preserved in our iteration scheme since the truncation of higher energy states has been carried out after adding all new states from the N/2N/2 pairs.

For N=6N=6, we have also exploited the method of Stadler et al. [85]. In this case, the truncation procedure is carried out at each of the steps (k=k=1, 22, \ldots, N/2N/2) after adding the states constructed with one of the channel pairs, by using Oliveira’s 𝒵\mathcal{Z}-trick [86], choosing different 𝒵\mathcal{Z} values for each of the N/2N/2 steps: 𝒵k=1/2+k/N\mathcal{Z}_{k}=1/2+k/N for the kk-th pair. We have carried out calculations for N=6N=6, taking rather large values for the NRG parameters, such that (Λ,Ntrunc)=(20,40000)(\Lambda,N_{\mathrm{trunc}})=(20,40000) for small interactions U/(πΔ)=2/5U/(\pi\Delta)=2/5 and 11, and (20,30000)(20,30000) for large interactions U/(πΔ)=2U/(\pi\Delta)=2, 33, 44, 55, and 66. This method significantly reduces the computational cost for obtaining low-lying energy states and enables us to calculate the three-body correlation functions for 66, although it does not faithfully preserve the SU(66) symmetry. We have checked whether this method reproduces the noninteracting results. Figure 14 compares the NRG result for sin2δ\sin^{2}\delta with the exact one for U=0U=0; the results show reasonable agreement. It indicates that this truncation procedure works effectively for deducing the SU(6) FL parameters.

In order to calculate χB[3]\chi_{B}^{[3]} defined in Eq. (45), we have also introduced a small external potential ϵsp,k\epsilon_{\mathrm{sp},k}, which depends on the channel index k=k=1, 22, \ldots, N/2N/2 and shifts the impurity level from ϵd\epsilon_{d}. Specifically, for N=4N=4, it is applied in a way equivalent to the local Zeeman field: ϵsp,1=b\epsilon_{\mathrm{sp},1}=-b and ϵsp,2=b\epsilon_{\mathrm{sp},2}=b. For N=6N=6, we have extended the potential such that ϵsp,1=b\epsilon_{\mathrm{sp},1}=-b, ϵsp,2=0\epsilon_{\mathrm{sp},2}=0, and ϵsp,3=b\epsilon_{\mathrm{sp},3}=b, and have deduced χB[3]\chi_{B}^{[3]} from the derivatives of the channel susceptibilities with respect to bb.

Appendix E Two-body FL parameters for
SU(4) & SU(6) symmetric cases

We provide a quick overview of the behavior of the renormalized parameters in the SU(4) and SU(6) cases, which can be derived from the phase shift and the linear susceptibilities [87, 88, 89, 90]. Our discussion here is based on the NRG results, plotted in Figs. 15 and 16 as functions of the impurity level position ξd\xi_{d}, for several different interaction strengths: from weak to strong interactions up to U/(πΔ)=6.0U/(\pi\Delta)=6.0.

The SU(NN) Kondo effect occurs for strong interactions when the impurity levels are filled by an integer number of electrons, i.e., at the fillings of Nd=1N_{d}=1, 22, \ldots, N1N-1. It takes place at ξd0\xi_{d}\simeq 0, ±U\pm U, …, ±N22U\pm\frac{N-2}{2}U, and gives an interesting variety in the low-energy properties. As NN increases, a greater interaction strength is required to clearly observe the Kondo behavior. This is because the quantum fluctuations caused by the Coulomb interaction are suppressed for large NN. In particular, the mean-field theory becomes exact in the limit NN\to\infty that is taken keeping the scaled interaction U(N1)UU^{*}\equiv(N-1)U constant [90].

Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 15: Fermi liquid parameters for the SU(4) symmetric Anderson model are plotted vs ξd\xi_{d}: (a) NdN_{d}, (b) sin2δ\sin^{2}\delta, (c) renormalization factor zz, (d) K~=(N1)(R1)\widetilde{K}=(N-1)(R-1). (e) renormalized level ϵ~d\widetilde{\epsilon}_{d}, and (f) enlarged view of ϵ~d\widetilde{\epsilon}_{d}, Interaction strengths are chosen to be U/(πΔ)=2/3(),1(),2(),3(),4(),5(),6(×)U/(\pi\Delta)=2/3(\star),1(\blacktriangledown),2(\blacklozenge),3(\blacksquare),4(\blacktriangle),5(\bullet),6(\times). The dashed line in (a) represents NdN_{d} in the atomic limit Δ0\Delta\to 0.
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 16: Fermi liquid parameters for the SU(6) symmetric Anderson model are plotted vs ξd\xi_{d}: (a) NdN_{d}, (b) sin2δ\sin^{2}\delta, (c) renormalization factor zz, (d) K~=(N1)(R1)\widetilde{K}=(N-1)(R-1). (e) renormalized level ϵ~d\widetilde{\epsilon}_{d}, and (f) enlarged view of ϵ~d\widetilde{\epsilon}_{d}, Interaction strengths are chosen to be U/(πΔ)=2/5(),1(),2(),3(),4(),5(),6(×)U/(\pi\Delta)=2/5(\star),1(\blacktriangledown),2(\blacklozenge),3(\blacksquare),4(\blacktriangle),5(\bullet),6(\times). The dashed line in (a) represents NdN_{d} in the atomic limit Δ0\Delta\to 0.

Figures 15(a) and 16(a) show the occupation number NdN_{d} for N=4N=4 and 66, respectively. As UU increases, the Coulomb staircase structure emerges: NdN_{d} varies steeply at ξd±U/2\xi_{d}\simeq\pm U/2, ±3U/2\pm 3U/2, \ldots, ±(N1)U/2\pm(N-1)U/2. Correspondingly, as UU increases, the transmission probability sin2δ\sin^{2}\delta, shown in Figs. 15(b) and 16(b), exhibits a plateau structure that develops around ξd0\xi_{d}\simeq 0, ±U\pm U, \ldots, ±(N2)U/2\pm(N-2)U/2. In particular, the plateau at the half-filling point ξd0\xi_{d}\simeq 0 reaches the unitary limit value sin2δ=1.0\sin^{2}\delta=1.0.

The renormalization factor zz, shown in Figs. 15(c) and 16(c) for N=4N=4 and 66, exhibits a broad valley structure at |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, where 1NdN11\lesssim N_{d}\lesssim N-1. The valley becomes deeper as UU increases, and the local minima emerge for U/(πΔ)3.0U/(\pi\Delta)\gtrsim 3.0 at the integer-filling points, reflecting the occurrence of the SU(NN) Kondo effects. The renormalization factor also has local maxima at intermediate valence states in between two adjacent local minima. Note that zz is significantly suppressed by the strong electron correlations even at these local maxima. Figures 15(d) and 16(d) show the rescaled Wilson ratio K~(N1)(R1)\widetilde{K}\equiv(N-1)(R-1). For large interactions, K~\widetilde{K} exhibits a wide flat structure at |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, the height of which approaches the saturation value K~1.0\widetilde{K}\simeq 1.0, especially for U/(πΔ)3.0U/(\pi\Delta)\gtrsim 3.0, reflecting the suppression of charge fluctuations in this region of ξd\xi_{d}.

Figures 15(e) and 16(e) show the renormalized resonance-level position ϵ~dzΔcotδ\widetilde{\epsilon}_{d}\equiv z\Delta\cot\delta for N=4N=4 and 66, respectively, as a function of ξd\xi_{d}. For strong interactions, the renormalized level is almost locked at the Fermi level, ϵ~d0.0\widetilde{\epsilon}_{d}\simeq 0.0, in the strong-coupling region |ξd|(N1)U/2|\xi_{d}|\lesssim(N-1)U/2, which corresponds to the filling range of 1NdN11\lesssim N_{d}\lesssim N-1. Figures 15(f) and 16(f) show an enlarged view of ϵ~d\widetilde{\epsilon}_{d} in the vicinity of the Fermi level. We see that ϵ~d\widetilde{\epsilon}_{d} exhibits a fine structure which reflects the staircase behavior of the phase shift δ\delta and the oscillatory behavior of the renormalization factor zz. Outside the strong-coupling region, ϵ~d\widetilde{\epsilon}_{d} approaches the bare value ϵ~dϵd\widetilde{\epsilon}_{d}\simeq\epsilon_{d} at ξd(N1)U/2\xi_{d}\gg(N-1)U/2, or the Hartree-Fock value ϵ~dϵd+(N1)U\widetilde{\epsilon}_{d}\simeq\epsilon_{d}+(N-1)U at ξd(N1)U/2\xi_{d}\ll-(N-1)U/2: these asymptotic forms of ϵ~d\widetilde{\epsilon}_{d} are shown as the dashed lines in Figs. 15(e) and 16(e).

Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 17: Current noise for SU(2) symmetric quantum dots vs ξd\xi_{d}: (a) sin2δ\sin^{2}\delta, (b) linear noise sin2δ(1sin2δ)\sin^{2}\delta\,(1-\sin^{2}\delta), (c) (192/π2)CS=WS+ΘS(192/\pi^{2})\,C_{S}=W_{S}+\Theta_{S}, (d) ΘS=(ΘI+3Θ~II)cos2δ\Theta_{S}=-\bigl{(}\Theta_{\mathrm{I}}+3\,\widetilde{\Theta}_{\mathrm{II}}\bigr{)}\cos 2\delta (see Table 2). Interaction strengths are chosen to be U/(πΔ)=1(),2(),3(),4(),5()U/(\pi\Delta)=1(\blacktriangledown),2(\blacklozenge),3(\blacksquare),4(\blacktriangle),5(\bullet).

Appendix F Current noise and thermoelectric transport for SU(2) symmetric case

We have discussed the behavior of the current noise and thermoelectric transport of the SU(4) and SU(6) Anderson model in Secs. VII, VIII, and X. For comparison, here we briefly describe the corresponding results in the SU(2) case, specifically, the current noise and the thermoelectric transport coefficients are plotted in Figs. 17 and 18, respectively.

We can see in Fig. 17(b) that the linear noise is suppressed in the strong-coupling region |ξd|U/2|\xi_{d}|\lesssim U/2 for large UU, where the transmission probability exhibits a wide Kondo plateau in Fig. 17(a). However, the linear noise increases and has the peaks, the width of which is of the order of Δ\Delta at valence fluctuation regions near the quarter and three-quarters filling points where the phase reaches δ=π/4\delta=\pi/4 or 3π/43\pi/4. At these filling points, the coefficient CSC_{S} for the order |eV|3|eV|^{3} nonlinear current noise has the negative minima, as seen in Fig. 17(c). We see in Fig. 17(d) that the three-body part ΘS\Theta_{S} of CSC_{S}, is also suppressed over a wide region |ξd|U/2|\xi_{d}|\lesssim U/2. Therefore, in this strong-coupling region, the nonlinear noise CSC_{S} is determined solely by the two-body part WSW_{S} in the SU(2) case:

WS= 4(R1)2+[ 1+5(R1)2]cos4δ,\displaystyle W_{S}\,=\,4(R-1)^{2}+\Bigl{[}\,1+5(R-1)^{2}\Bigr{]}\cos 4\delta\,, (118)

where RR is the Wilson ratio. In particular, the dip structures of CSC_{S} at the quarter and three-quarters fillings reflect the mimina of cos4δ\cos 4\delta in WSW_{S}. In contrast, at |ξd|U/2|\xi_{d}|\gtrsim U/2, the three-body part ΘS\Theta_{S} becomes comparable to WSW_{S}, and contributes to CSC_{S}.

Figure 18 shows the results for the next-leading order terms of the thermoelectric transport coefficients for the SU(2) quantum dots (a) CTC_{T} and (b) CκQDC_{\kappa}^{\mathrm{QD}}, and for SU(2) magnetic alloys (c) CϱMAC_{\varrho}^{\mathrm{MA}} and (d) CκMAC_{\kappa}^{\mathrm{MA}}. All these coefficients exhibit the plateau structures due to the Kondo effect in the strong-coupling region |ξd|U/2|\xi_{d}|\lesssim U/2 for large UU, where the phase shit is locked at δπ/2\delta\simeq\pi/2. In this region, the three-body contributions, ΘI\Theta_{\mathrm{I}} and Θ~II\widetilde{\Theta}_{\mathrm{II}}, almost vanish (see Ref. 28 for more details), and the height of these plateaus approach the saturation values for UU\to\infty, i.e., (48/π2)CT3(48/\pi^{2})C_{T}\to 3, (48/π2)CϱMA3(48/\pi^{2})C_{\varrho}^{\mathrm{MA}}\to 3, [80/(7π2)]CκQD13/7[80/(7\pi^{2})]C_{\kappa}^{\mathrm{QD}}\to 13/7, and [80/(7π2)]CκMA13/7[80/(7\pi^{2})]C_{\kappa}^{\mathrm{MA}}\to 13/7. The coefficients CTC_{T} and CϱMAC_{\varrho}^{\mathrm{MA}} for charge transport show a similar behavior in the strong-coupling region. Furthermore, in this region, the coefficients for thermal conductivities, CκQDC_{\kappa}^{\mathrm{QD}} and CκMAC_{\kappa}^{\mathrm{MA}}, also show a similar behavior. In contrast, at |ξd|U/2|\xi_{d}|\gtrsim U/2, the coefficients CTC_{T} and CκQDC_{\kappa}^{\mathrm{QD}} for quantum dots change sign and become negative as the occupation number approaches Nd0.0N_{d}\simeq 0.0 or 2.02.0, whereas the coefficients CϱMAC_{\varrho}^{\mathrm{MA}} and CκMAC_{\kappa}^{\mathrm{MA}} for magnetic alloys remain positive definite.

Refer to caption
Refer to caption
Refer to caption
Refer to caption
Figure 18: Next-leading order terms of transport coefficients for an SU(2) symmetric Anderson impurity: (a) (48/π2)CT(48/\pi^{2})C_{T}, (b) [80/(7π2)]CκQD[80/(7\pi^{2})]C_{\kappa}^{\mathrm{QD}}, (c) (48/π2)CϱMA(48/\pi^{2})C_{\varrho}^{\mathrm{MA}}, and (b) [80/(7π2)]CκMA[80/(7\pi^{2})]C_{\kappa}^{\mathrm{MA}}, defined in Table 2. Interaction strengths are chosen to be U/(πΔ)=1(),2(),3(),4(),5()U/(\pi\Delta)=1(\blacktriangledown),2(\blacklozenge),3(\blacksquare),4(\blacktriangle),5(\bullet).

Appendix G Three-body Fermi liquid corrections to thermoelectric transport of magnetic alloys

In this appendix, we describe the low-energy asymptotic form of the electrical resistivity ϱMA=1/σMA\varrho_{\mathrm{MA}}=1/\sigma_{\mathrm{MA}}, the thermopower 𝒮MA\mathcal{S}_{\mathrm{MA}}, and the thermal conductivity κMA\kappa_{\mathrm{MA}} of magnetic alloys:

σMA=\displaystyle\sigma_{\mathrm{MA}}\ = σMAunit1Nσ0,σMA,\displaystyle\ \sigma_{\mathrm{MA}}^{\mathrm{unit}}\,\frac{1}{N}\,\sum_{\sigma}\mathcal{L}_{0,\sigma}^{\mathrm{MA}}\,, (119)
𝒮MA=\displaystyle\mathcal{S}_{\mathrm{MA}}\ = 1|e|Tσ1,σMAσ0,σMA,\displaystyle\ \frac{-1}{|e|T}\frac{\sum_{\sigma}\mathcal{L}_{1,\sigma}^{\mathrm{MA}}}{\sum_{\sigma}\mathcal{L}_{0,\sigma}^{\mathrm{MA}}}\,, (120)
κMA=\displaystyle\kappa_{\mathrm{MA}}\ = σMAunite2T1N[σ2,σMA(σ1,σMA)2σ0,σMA].\displaystyle\ \frac{\sigma_{\mathrm{MA}}^{\mathrm{unit}}}{e^{2}T}\,\frac{1}{N}\,\left[\,\sum_{\sigma}\mathcal{L}_{2,\sigma}^{\mathrm{MA}}-\frac{\left(\sum_{\sigma}\mathcal{L}_{1,\sigma}^{\mathrm{MA}}\right)^{2}}{\sum_{\sigma}\mathcal{L}_{0,\sigma}^{\mathrm{MA}}}\,\right]. (121)

For magnetic alloys, the response functions n,σMA\mathcal{L}_{n,\sigma}^{\mathrm{MA}} for n=0,1n=0,1 and 22 are given by

n,σMA=𝑑ωωnπΔAσ(ω)(f(ω)ω).\displaystyle\mathcal{L}_{n,\sigma}^{\mathrm{MA}}=\int_{-\infty}^{\infty}d\omega\,\frac{\omega^{n}}{\pi\Delta A_{\sigma}(\omega)}\left(-\frac{\partial f(\omega)}{\partial\omega}\right). (122)

Equation (119) defines the electrical conductivity relative to its unitary-limit value σMAunit\sigma_{\mathrm{MA}}^{\mathrm{unit}}. Correspondingly, the prefactor for κMA\kappa_{\mathrm{MA}} is defined in such a way that the TT-linear thermal conductivity should take the form,

limT0κMAT=π23e2σMAunit.\displaystyle\lim_{T\to 0}\frac{\kappa_{\mathrm{MA}}}{T}=\frac{\pi^{2}}{3\,e^{2}}\,\sigma_{\mathrm{MA}}^{\mathrm{unit}}\,. (123)

The asymptotic form of n,σMA\mathcal{L}_{n,\sigma}^{\mathrm{MA}} can be calculated, using the low-energy expansion of the inverse spectral function 1/Aσ(ω)1/A_{\sigma}(\omega) given in Eq. (112):

0,σMA=\displaystyle\mathcal{L}_{0,\sigma}^{\mathrm{MA}}= 1sin2δσ[ 1+a0,σMAsin2δσ(πT)2]+,\displaystyle\ \frac{1}{\sin^{2}\delta_{\sigma}}\left[\,1+\frac{a_{0,\sigma}^{\mathrm{MA}}}{\sin^{2}\delta_{\sigma}}\left(\pi T\right)^{2}\,\right]\ +\ \cdots\,, (124)
1,σMA=\displaystyle\mathcal{L}_{1,\sigma}^{\mathrm{MA}}\,= 13ρdσπΔρdσ2(πT)2+,\displaystyle\ -\frac{1}{3}\frac{\rho_{d\sigma}^{\prime}}{\pi\Delta\rho_{d\sigma}^{2}}\,(\pi T)^{2}\ +\ \cdots\,, (125)
2,σMA=\displaystyle\mathcal{L}_{2,\sigma}^{\mathrm{MA}}= (πT)23sin2δσ[ 1+a2,σMAsin2δσ(πT)2]+.\displaystyle\ \frac{\left(\pi T\right)^{2}}{3\sin^{2}\delta_{\sigma}}\left[\,1+\frac{a_{2,\sigma}^{\mathrm{MA}}}{\sin^{2}\delta_{\sigma}}\left(\pi T\right)^{2}\,\right]\ +\ \cdots\,. (126)

Here, the coefficients a0,σMAa_{0,\sigma}^{\mathrm{MA}} and a2,σMAa_{2,\sigma}^{\mathrm{MA}} are given by

a0,σMA=\displaystyle\!\!\!a_{0,\sigma}^{\mathrm{MA}}= π23[(cos2δσ+2)χσσ22cos2δσσ(σ)χσσ2\displaystyle\,\frac{\pi^{2}}{3}\Biggl{[}\left(\cos 2\delta_{\sigma}+2\right)\chi_{\sigma\sigma}^{2}-2\cos 2\delta_{\sigma}\,\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}}^{2}
+sin2δσ2π(χσσσ[3]+σ(σ)χσσσ[3])],\displaystyle\ \ +\frac{\sin 2\delta_{\sigma}}{2\pi}\,\left(\chi_{\sigma\sigma\sigma}^{[3]}+\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\right)\Biggr{]}, (127)
a2,σMA=\displaystyle\!\!\!a_{2,\sigma}^{\mathrm{MA}}= 7π25[(cos2δσ+2)χσσ267cos2δσσ(σ)χσσ2\displaystyle\,\frac{7\pi^{2}}{5}\Biggl{[}\left(\cos 2\delta_{\sigma}+2\right)\chi_{\sigma\sigma}^{2}-\frac{6}{7}\cos 2\delta_{\sigma}\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}}^{2}
+sin2δσ2π(χσσσ[3]+521σ(σ)χσσσ[3])].\displaystyle\ \ +\frac{\sin 2\delta_{\sigma}}{2\pi}\left(\chi_{\sigma\sigma\sigma}^{[3]}+\frac{5}{21}\sum_{\sigma^{\prime}(\neq\sigma)}\chi_{\sigma\sigma^{\prime}\sigma^{\prime}}^{[3]}\right)\Biggr{]}. (128)

We obtain the low-temperature expressions of ϱMA=1/σMA\varrho_{\mathrm{MA}}=1/\sigma_{\mathrm{MA}}, 𝒮MA\mathcal{S}_{\mathrm{MA}} and 1/κMA1/\kappa_{\mathrm{MA}}, substituting Eqs. (124)–(126) into Eqs. (119)–(121):

ϱMA=\displaystyle\varrho_{\mathrm{MA}}= 1σMAunit[(sin2δ¯)HMcϱMA(πT)2+],\displaystyle\,\frac{1}{\sigma_{\mathrm{MA}}^{\mathrm{unit}}}\,\left[\,\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}}-c_{\varrho}^{\mathrm{MA}}\,\left(\pi T\right)^{2}+\cdots\right], (129)
𝒮MA=\displaystyle\mathcal{S}_{\mathrm{MA}}= π23σρdσσρdσT|e|+,\displaystyle\,\frac{\pi^{2}}{3}\,\frac{\sum_{\sigma}\rho_{d\sigma}^{\prime}}{\sum_{\sigma}\rho_{d\sigma}}\,\frac{T}{|e|}\,+\,\cdots, (130)
1κMA=\displaystyle\frac{1}{\kappa_{\mathrm{MA}}}\,= 3e2π2σMAunit1T[(sin2δ¯)HMcκMA(πT)2+].\displaystyle\,\frac{3\,e^{2}}{\pi^{2}\,\sigma_{\mathrm{MA}}^{\mathrm{unit}}}\,\frac{1}{T}\,\Bigl{[}\,\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}}\,-\,c_{\kappa}^{\mathrm{MA}}\,\left(\pi T\right)^{2}\,+\,\cdots\,\Bigr{]}. (131)

Here, (sin2δ¯)HM\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}} is the harmonic mean (HM) of sin2δσ\sin^{2}\delta_{\sigma}, defined by

(sin2δ¯)HM11Nσ1sin2δσ.\displaystyle\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}}\,\equiv\,\frac{1}{\frac{1}{N}\,\sum_{\sigma}\,\frac{1}{\sin^{2}\delta_{\sigma}}}\,. (132)

The coefficients cϱMAc_{\varrho}^{\mathrm{MA}} and cκMAc_{\kappa}^{\mathrm{MA}} of the next-leading order terms are given by

cϱMA=\displaystyle c_{\varrho}^{\mathrm{MA}}\,= {(sin2δ¯)HM}21Nσa0,σMAsin4δσ,\displaystyle\,\left\{\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}}\right\}^{2}\frac{1}{N}\,\sum_{\sigma}\,\frac{a_{0,\sigma}^{\mathrm{MA}}}{\sin^{4}\delta_{\sigma}}\,, (133)
cκMA=\displaystyle c_{\kappa}^{\mathrm{MA}}\,= {(sin2δ¯)HM}2[1Nσa2,σMAsin4δσ\displaystyle\,\left\{\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}}\right\}^{2}\left[\frac{1}{N}\,\sum_{\sigma}\,\frac{a_{2,\sigma}^{\mathrm{MA}}}{\sin^{4}\delta_{\sigma}}\,\right.
π23(sin2δ¯)HM{1Nσsin2δσsin4δσχσσ}2].\displaystyle\left.\,-\,\frac{\pi^{2}}{3}\,\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}}\left\{\frac{1}{N}\,\sum_{\sigma}\,\frac{\sin 2\delta_{\sigma}}{\sin^{4}\delta_{\sigma}}\,\chi_{\sigma\sigma}\right\}^{2}\,\right]. (134)

Correspondingly, the electrical conductivity and the thermal conductivity take the following form,

σMA=\displaystyle\sigma_{\mathrm{MA}}\,= σMAunit(sin2δ¯)HM[ 1+cϱMA(sin2δ¯)HM(πT)2+],\displaystyle\,\frac{\sigma_{\mathrm{MA}}^{\mathrm{unit}}}{\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}}}\,\left[\,1\,+\frac{c_{\varrho}^{\mathrm{MA}}}{\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}}}\,\left(\pi T\right)^{2}\,+\,\cdots\,\right], (135)
κMA=\displaystyle\mathcal{\kappa}_{\mathrm{MA}}= π2σMAunit3e2T(sin2δ¯)HM\displaystyle\,\frac{\pi^{2}\,\sigma_{\mathrm{MA}}^{\mathrm{unit}}}{3\,e^{2}}\,\frac{T}{\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}}}
×[ 1+cκMA(sin2δ¯)HM(πT)2+].\displaystyle\qquad\times\left[\,1\,+\,\frac{c_{\kappa}^{\mathrm{MA}}}{\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}}}\,\left(\pi T\right)^{2}\,+\,\cdots\,\right]. (136)

Furthermore, the Lorenz number LMAκMA/(σMAT)L_{\mathrm{MA}}\equiv\kappa_{\mathrm{MA}}/(\sigma_{\mathrm{MA}}T) can also be deduced up to terms of order T2T^{2}, as

LMA=\displaystyle L_{\mathrm{MA}}\,\,= π23e2[1cLMA(sin2δ¯)HM(πT)2+],\displaystyle\ \frac{\pi^{2}}{3\,e^{2}}\,\left[1\,-\,\frac{c_{L}^{\mathrm{MA}}}{\bigl{(}\overline{\sin^{2}\delta}\bigr{)}_{\mathrm{HM}}}\,\left(\pi T\right)^{2}\ +\ \cdots\,\right], (137)
cLMA=\displaystyle c_{L}^{\mathrm{MA}}\,= cϱMAcκMA.\displaystyle\ c_{\varrho}^{\mathrm{MA}}\,-\,c_{\kappa}^{\mathrm{MA}}\,. (138)

In the limit of zero temperature, the Lorenz number takes a constant value LMAT0π23e2L_{\mathrm{MA}}\xrightarrow{\,T\to 0\,}\frac{\pi^{2}}{3\,e^{2}}, and the Wiedemann-Franz law holds. However, it deviates as temperature increases, showing the T2T^{2} dependence.

Appendix H Thermoelectric transport coefficients for noninteracting magnetic alloys

We provide here the noninteracting results for the thermoelectric transport coefficients of magnetic alloys. For U=0U=0, the response functions n,σMA\mathcal{L}_{n,\sigma}^{\mathrm{MA}} for n=0,1,2n=0,1,2 can be calculated, substituting the explicit form of the spectral function πAσ(ω)U=0Δ/[(ωϵdσ)2+Δ2]\pi A_{\sigma}(\omega)\xrightarrow{\,U=0\,}\Delta/[(\omega-\epsilon_{d\sigma})^{2}+\Delta^{2}] into Eq. (122):

0,σMAU=0\displaystyle\mathcal{L}_{0,\sigma}^{\mathrm{MA}}\,\xrightarrow{\,U=0\,} [1+(ϵdσΔ)2]+13(πTΔ)2,\displaystyle\ \left[1+\left(\frac{\epsilon_{d\sigma}}{\Delta}\right)^{2}\right]\,+\,\frac{1}{3}\left(\frac{\pi T}{\Delta}\right)^{2}, (139)
1,σMAU=0\displaystyle\mathcal{L}_{1,\sigma}^{\mathrm{MA}}\,\xrightarrow{\,U=0\,} 2ϵdσ3(πTΔ)2,\displaystyle\ \frac{-2\,\epsilon_{d\sigma}}{3}\left(\frac{\pi T}{\Delta}\right)^{2}, (140)
2,σMAU=0\displaystyle\mathcal{L}_{2,\sigma}^{\mathrm{MA}}\,\xrightarrow{\,U=0\,} Δ23[1+(ϵdσΔ)2](πTΔ)2+7Δ215(πTΔ)4.\displaystyle\ \frac{\Delta^{2}}{3}\left[1+\left(\frac{\epsilon_{d\sigma}}{\Delta}\right)^{2}\right]\left(\frac{\pi T}{\Delta}\right)^{2}+\frac{7\Delta^{2}}{15}\left(\frac{\pi T}{\Delta}\right)^{4}. (141)

The transport coefficients, defined in Eqs. (119)–(121), can be deduced from these analytic expressions for n,σMA\mathcal{L}_{n,\sigma}^{\mathrm{MA}}. In particular, in the SU(NN) symmetric case where ϵdσϵd\epsilon_{d\sigma}\equiv\epsilon_{d}, the electrical resistivity ϱMA(0)\varrho_{\mathrm{MA}}^{(0)} for U=0U=0 is given by

ϱMA(0)=\displaystyle\varrho_{\mathrm{MA}}^{(0)}\,= 1σMAunit[sin2δ0CϱMA(0)(πTT0)2],\displaystyle\ \frac{1}{\sigma_{\mathrm{MA}}^{\mathrm{unit}}}\,\left[\,\sin^{2}\delta_{0}\,-\,C_{\varrho}^{\mathrm{MA}(0)}\,\left(\frac{\pi T}{T^{*}_{0}}\right)^{2}\,\right], (142)
sin2δ0=\displaystyle\sin^{2}\delta_{0}\,= 11+(ϵd/Δ)2,CϱMA(0)=π248.\displaystyle\ \frac{1}{1+(\epsilon_{d}/\Delta)^{2}}\,,\qquad C_{\varrho}^{\mathrm{MA}(0)}\,=\,\frac{\pi^{2}}{48}\,. (143)

Here, T0πΔ/(4sin2δ0)T_{0}^{*}\equiv\pi\Delta/(4\sin^{2}\delta_{0}) is the characteristic energy scale in the noninteracting case. The coefficient CϱMA(0)C_{\varrho}^{\mathrm{MA}(0)} for the T2T^{2}-resistivity is given by a constant π2/48\pi^{2}/48 (=0.205=0.205\cdots) as the ϵd\epsilon_{d} dependence is absorbed into T0T_{0}^{*}.

The thermopower 𝒮MA0\mathcal{S}_{\mathrm{MA}}^{0} and thermal resistivity 1/κMA(0)1/\kappa_{\mathrm{MA}}^{(0)} for U=0U=0 are given by

𝒮MA(0)=1|e|2π23ϵdϵd2+Δ2T,\displaystyle\mathcal{S}_{\mathrm{MA}}^{(0)}\,=\,\frac{1}{|e|}\,\frac{2\pi^{2}}{3}\,\frac{\epsilon_{d}}{\epsilon_{d}^{2}+\Delta^{2}}\,T\;, (144)
1κMA(0)=3e2π2σMAunit1T[sin2δ0CκMA(0)(πTT0)2],\displaystyle\frac{1}{\kappa_{\mathrm{MA}}^{(0)}}\,=\,\frac{3\,e^{2}}{\pi^{2}\,\sigma_{\mathrm{MA}}^{\mathrm{unit}}}\,\frac{1}{T}\,\left[\,\sin^{2}\delta_{0}\,-\,C_{\kappa}^{\mathrm{MA}(0)}\,\left(\frac{\pi T}{T_{0}^{*}}\right)^{2}\,\right], (145)
CκMA(0)=7π2801+20sin2δ021.\displaystyle C_{\kappa}^{\mathrm{MA}(0)}\,=\,\frac{7\pi^{2}}{80}\,\frac{1+20\sin^{2}\delta_{0}}{21}\,. (146)

Furthermore, the dimensionless coefficient for the order T2T^{2} term of the Lorenz number LMAL_{\mathrm{MA}} is given by CLMA(0)CϱMA(0)CκMA(0)C_{L}^{\mathrm{MA}(0)}\equiv C_{\varrho}^{\mathrm{MA}(0)}-C_{\kappa}^{\mathrm{MA}(0)} and it takes the form,

CLMA(0)=π260( 15sin2δ0).\displaystyle C_{L}^{\mathrm{MA}(0)}\,=\,\frac{\pi^{2}}{60}\,\left(\,1-5\sin^{2}\delta_{0}\,\right)\,. (147)

These results for the next-leading order terms in the noninterating case are plotted as functions of ϵd\epsilon_{d} in Fig.  19. The coefficient CκMA(0)C_{\kappa}^{\mathrm{MA}(0)} takes a Lorentzian form with an offset value of CκMA(0)|ϵd|π2/240C_{\kappa}^{\mathrm{MA}(0)}\xrightarrow{\,|\epsilon_{d}|\to\infty\,}\pi^{2}/240 (=0.041=0.041\cdots). Correspondingly, CLMA(0)C_{L}^{\mathrm{MA}(0)} has a dip at ϵd=0\epsilon_{d}=0, and vanishes at the points ϵd=±2Δ\epsilon_{d}=\pm 2\Delta where CκMA(0)C_{\kappa}^{\mathrm{MA}(0)} and CϱMA(0)C_{\varrho}^{\mathrm{MA(0)}} give equal contributions to the Lonrez number LMAL_{\mathrm{MA}}. As the impurity level moves further away from the Fermi level |ϵd|Δ|\epsilon_{d}|\gg\Delta, it approaches the value of CLMA(0)|ϵd|π2/60C_{L}^{\mathrm{MA}(0)}\xrightarrow{\,|\epsilon_{d}|\to\infty\,}\pi^{2}/60 (=0.164=0.164\cdots).

Refer to caption
Figure 19: Coefficients CϱMA(0)C_{\varrho}^{\mathrm{MA}(0)}, CκMA(0)C_{\kappa}^{\mathrm{MA}(0)}, and CLMA(0)C_{L}^{\mathrm{MA}(0)} for noninteracting U=0U=0 magnetic alloys plotted vs ϵd\epsilon_{d}.

References