Anisotropy-driven topological quantum phase transition in magnetic impurities

Abstract

A few years ago, a topological quantum phase transition (TQPT) has been found in Anderson and Kondo 2-channel spin-1 impurity models that include a hard-axis anisotropy term $DS_{z}^{2}$ with $D>0$. The most remarkable manifestation of the TQPT is a jump in the spectral density of localized electrons, at the Fermi level, from very high to very low values as $D$ is increased. If the two conduction channels are equivalent, the transition takes place at the critical anisotropy $D_{c}\sim 2.5\;T_{K}$, where $T_{K}$ is the Kondo temperature for $D=0$. This jump might be important to develop a molecular transistor. The jump is due to a corresponding one in the Luttinger integral, which has a topological non-trivial value $\pi/2$ for $D>D_{c}$. Here, we review the main results for the spectral density and highlight the significance of the theory for the interpretation of measurements conducted on magnetic atoms or molecules on metallic surfaces. In these experiments, where $D$ is held constant, the energy scale $T_{K}$ is manipulated by some parameters. The resulting variation gives rise to a differential conductance $dI/dV$, measured by scanning-tunneling spectroscopy, which is consistent with a TQPT at an intermediate value of $T_{K}$. We also show that the theory can be extended to integer spin $S>1$ and two-impurity systems. This is also probably true for half-integer spin and non-equivalent channels in some cases.

• December 2023

1 Introduction

Systems with individual magnetic atoms [1] or molecules [2, 3, 4, 5, 6] on metallic surfaces are being studied extensively in the last years due to their peculiar properties and potential application in spintronics and molecular electronics. A fundamental component within an integrated circuit is the transistor, which can be realized in these magnetic impurity systems through the controlled switching of the electric current by varying some physical parameters [4].

A realistic analysis of magnetic impurity systems requires taking into account their potential multiorbital nature and hybridization with more than one conduction channel. Indeed, a plethora of electronic states, ranging from Landau-Fermi liquid to singular Fermi liquid and non-Fermi liquid, emerges depending on the specific value of the spin of the magnetic impurity and the number of conduction channels [7]. An important, although often overlooked, physical ingredient of multiorbital systems is the single-ion magnetic anisotropy $DS_{z}^{2}$ due to the spin-orbit coupling, enhanced in the low-symmetry arrangements of atoms or molecules on surfaces.

As an example of the relevance of single-ion anisotropy, some years ago, it has been found that the conductance of a system composed of a Ni impurity in a Au chain doped with oxygen has a jump as a function of the anisotropy $D$ of the spin 1 of the Ni atom, suggesting that the system could act as a transistor [8, 9]. The underlying model is the 2-channel spin-1 Anderson model with anisotropy (2CS1AMA) or its integer valent limit, the anisotropic two-channel spin-1 Kondo model (A2CS1KM). Both models are described in Section 2. With such models, it can be seen that the jump is due to a topological quantum phase transition (TQPT) between two phases that differ in the value of the so-called Luttinger integral $I_{L}$, whose zero value had been for decades a hallmark of a Fermi liquid [10]. For low $D/T_{K}$ where $T_{K}$ is the Kondo temperature for $D=0$, the system is in the topologically trivial phase with $I_{L}=0$, characterized by a large spectral density of localized electrons at the Fermi level and large conductance at low temperatures and bias voltage. For large $D/T_{K}$ the system is topologically non-trivial with $I_{L}=\pi/2$, with a pronounced dip in the conductance and spectral density of localized electrons at low energies. This phase has been called “non-Landau ”Fermi liquid, because it cannot be adiabatically connected to a non-interacting system for which $I_{L}=0$ [8, 10].

For degenerated channels and zero magnetic field $B=0$, the transition takes place at the critical anisotropy $D_{c}\sim 2.5\;T_{K}$. However, the TQPT still exists for $B\neq 0$ or inequivalent channels with $B=0$ [11]. For non-equivalent channels and $B\neq 0$, the system is a topologically trivial ordinary Fermi liquid, but a crossover from a dip to a peak can be induced by modifying a parameter like $B$ [11] and has been actually observed, as explained in Section 4.1.

In this article, we review the main results concerning the above mentioned TQPT. While the original model was proposed for a particular system that has not been realized up to date [12], we show that several systems with magnetic atoms or molecules on metallic surfaces can be actually described by the 2CS1AMA, or the simpler A2CS1KM, providing a consistent and unified description of several experiments. Alternatives theories, sometimes inconsistent between them or physically unjustified, have been proposed to interpret the outcome of those experiments. This is probably due to the fact that the theory is rather recent, and that to describe the TQPT, a very accurate technique like the numerical renormalization group (NRG) is required, which is computationally expensive for two or more channels. So far, no other technique has been able to capture the TQPT. We also show that the theory can be extended to a larger spin and two-impurity systems.

The paper is organized as follows. In Sec. 2 we introduce the two models, Kondo and Anderson, for the spin-1 impurity coupled with two conduction channels. In Sec. 3 the NRG predictions for the localized electron spectral density are presented and discussed in the context of the TQPT. In Sec. 4, the differencial conductance ($dI/dV$) measurements for five different systems are discussed and compared with the theoretical calculations: 4.1) FePc on Au(111),  4.2) MnPc on Au(111),  4.3) nickelocene on Cu(100),  4.4) Fe atoms on MoS${}_{2}$/Au(111) and 4.5) Fe porphyrin molecules on Au(111). In Sec. 5 the theory is generalized for $S>1$ and two-impurity systems. Finally, we conclude with a brief summary in Sec. 6.

2 Models

The simplest model to describe the TQPT is the anisotropic two-channel spin-1 Kondo model (A2CS1KM). It takes the form

$$H_{K}=\sum_{k\tau\sigma}\varepsilon_{k\tau}c_{k\tau\sigma}^{\dagger}c_{k\tau\sigma}+\sum_{k\tau\sigma\sigma^{\prime}}\frac{J_{K\tau}}{2}c_{k\tau\sigma}^{\dagger}\vec{\sigma}_{\sigma\sigma^{\prime}}c_{k\tau\sigma^{\prime}}\cdot\vec{S}+DS_{z}^{2},$$ (1)

where $c_{k\tau\sigma}^{\dagger}$ creates a conduction electron with point-group symmetry $\tau$ (channel index), spin $\sigma$ and remaining quantum numbers $k$. The first term describes the substrate conduction bands, the second term is the Kondo exchange interaction between conduction electrons and the localized spin $\vec{S}$ with exchange couplings $J_{K\tau}$, and the last term is the single-ion uniaxial magnetic anisotropy. $\vec{\sigma}$ is the vector of Pauli matrices. The simplest case is when the two channels are equivalent: both are degenerate ($\varepsilon_{k\tau}=\varepsilon_{k}$) and the localized spin is equally coupled to them ($J_{K\tau}=J_{K}$).

When intermediate valence of the magnetic impurity is included, the model is the 2-channel spin-1 Anderson model with anisotropy (2CS1AMA). It can be written in the form used first for Ni compounds with holes in the $xz$ and $yz$ orbitals [9, 13, 14]. Extension to other cases, for example, FePc on Au(111) [11] are straightforward. Neglecting the pair-hopping term [9], which is irrelevant in all cases considered so far as the intra-orbital Coulomb $U$ repulsion is considerably larger that the inter-orbital $U^{\prime}$ one, the Anderson Hamiltonian is

$$H = \sum_{k\tau\sigma}\varepsilon_{k}c_{k\tau\sigma}^{\dagger}c_{k\tau\sigma}+\sum_{k\tau\sigma}\left(v_{\tau}{c}_{k\tau\sigma}^{\dagger}{d}_{\tau\sigma}+\mathrm{H.c.}\right)+$$ (2)

$$+ \sum_{\tau\sigma}\epsilon d_{\tau\sigma}^{\dagger}d_{\tau\sigma}+\sum_{\tau}Un_{\tau\uparrow}n_{\tau\downarrow}+U^{\prime}n_{xz}n_{yz}-J_{H}{\vec{S}}_{xz}\cdot{\vec{S}}_{yz}+DS_{z}^{2},$$

where $d_{\tau\sigma}^{\dagger}$ ($c^{\dagger}_{k\tau\sigma}$) creates a hole with energy $\epsilon$ ($\varepsilon_{k}$) in the $d$ orbital $\tau$ (conduction band $\tau$ with momentum $k$), with $\tau=xz,yz$. $n_{\tau\sigma}=d_{\tau\sigma}^{\dagger}d_{\tau\sigma}$ and $n_{\tau}=\sum_{\sigma}n_{\tau\sigma}$. $v_{\tau}$ is the tunneling or hybridization amplitude between impurity and conduction states (assumed to be independent of $k$), while $J_{H}$ is the strong Hund ferromagnetic exchange responsible for the spin 1 of the impurity.

For small hybridization and when the two-particle configuration dominates, the model (2) reduces to the A2CS1KM [9].

3 Results

For equivalent channels, it has been found that the TQPT in the A2CS1KM takes place for anisotropy $D_{c}\sim 2.5\;T_{K}$, where $T_{K}$ is the Kondo temperature for $D=0$ [9]. The evolution of the spectral density for localized states $\rho(\omega)$ with $D$ for constant $T_{K}$ is shown in Fig. 11 of Ref. [9], where an abrupt change at the Fermi level signals the critical value $D_{c}$

In Fig. 1 we represent $\rho(\omega)$ for different Kondo exchange couplings $J_{K}$, keeping $D=0.0027\;W$ constant, where $W$ is the half-bandwidth of the conduction bands, taken as the energy unit. $r\equiv J_{K}/J_{Kc}$, where $J_{Kc}$ is the critical Kondo coupling for the given $D$ ($J_{K}$ such that $D\simeq 2.5\;T_{K}$). The numerical calculations were performed with the Ljubljana code of the NRG [15, 16]. We assume flat conduction bands extending from $-W$ to $W$ for both symmetries.

Figure 1: (Color online) Spectral density of localized electrons of the A2CS1KM as a function of energy for several values of $r=J_{K}/J_{Kc}$, where $J_{Kc}$ is the value of $J_{K}$ at the TQPT. $\rho_{0}$ is given by the ordinary Friedel sum rule with vanishing Luttinger integral [11]. Vertical dotted lines are at $\omega=\pm D$. $\omega=0$ corresponds to the Fermi level.

For $r$ significantly larger than 1, the spectral density of localized states is similar to that of an ordinary Kondo peak. For $r$ slightly larger than 1, $\rho(\omega)$ has the form of a narrow peak mounted on a broad peak. The latter, at the Fermi level, has around half the magnitude expected for the usual compensated Kondo model and therefore, it is similar to the spectral density expected for the spin-1/2 two-channel Kondo model. However, as long as $r>1$, the system satisfies the ordinary Friedel-Langreth sum rule (the small deviation in the figure is due to numerical errors of the NRG) and $\rho(0)$ has its maximum possible value. However, as soon as $r<1$, $\rho(0)$ jumps to 0. This is due to a jump in the Luttinger integral from 0 to $\pi/2$ at $r=1$ [8, 11], as discussed below. If $r$ is decreased further, the shape of the spectral density tends to be rectangular, with two jumps at $-D$ and $D$, typical of inelastic scattering when only one channel is present [17, 18, 19, 20]. These steps are overbroadened in Fig. 1 due to technical reasons that limit the resolution of the NRG at large energies [21].

A detailed description of what happens with the Friedel-Langreth sum rule and the Luttinger integrals in the more general case, which includes intermediate valence, different channels, and a magnetic field has been explained in detail in the Supplemental Material of Ref. [11]. To avoid including many technical details, we outline the main facts for the simpler case of equivalent conduction channels and zero magnetic field in the two-orbital Anderson model, for which the Luttinger integral $I_{L}$ does not depend on the spin and channel quantum numbers.

Using conservation laws, it can be shown that the spectral function of the localized states for each orbital and spin, at the Fermi level and $T=0,$ is given by

$$\rho_{\tau\sigma}(\omega=0)=\frac{1}{\pi\Delta}\sin^{2}(\delta_{\tau\sigma}),$$ (3)

where $\Delta=\pi v^{2}\rho_{c}$, with $v$ the hybridization and $\rho_{c}$ the density of conduction states assumed independent of energy. The phase shift suffered by the conduction electrons at the Fermi level due to the presence of the impurity is

$$\delta_{\tau\sigma}=\pi\langle n_{\tau\sigma}\rangle-I_{L}.$$ (4)

The Luttinger integral $I_{L}$ (independent of orbital and spin indices in the simplest case) is defined as

$$I_{L}={\rm Im}\int_{-\infty}^{0}d\omega G^{d}_{\tau\sigma}(\omega)\frac{\partial\Sigma^{d}_{\tau\sigma}(\omega)}{\partial\omega},$$ (5)

where $G^{d}_{\tau\sigma}(\omega)$ is the impurity Green function for orbital $\tau$ and spin $\sigma$, and $\Sigma^{d}_{\tau\sigma}(\omega)$ is the corresponding self energy.

For a long time, $I_{L}$ has been assumed to vanish for a Fermi liquid, based on perturbation calculations starting from a non-interacting electronic system [10, 22, 23]. However, rather recently [24, 25] it has been found that this is not always the case for local Fermi liquids. A topological interpretation of $I_{L}$ was provided for extended systems in Ref. [26] and extended to the impurity case in Ref. [8]. Previously a non-trivial value of $I_{L}$ has been found in underscreened one-channel spin-1 models, for which the system is a singular Fermi liquid [27].

An explicit calculation has shown that in the “non-Landau ”phase, for large $D/T_{K}$, $I_{L}=\pi/2$ [11]. In the Kondo limit, $\langle n_{\tau\sigma}\rangle=1/2$ and Eq. (3) gives $\rho_{\tau\sigma}(0)=0$ in this phase, whereas in the ordinary Fermi liquid phase, with $I_{L}=0$, the spectral density at the Fermi level has its maximum possible value $\rho_{\tau\sigma}(0)=1/(\pi\Delta)$. This explains the jump observed in Fig. 1 at the TQPT.

4 Experimental relevance of the topological quantum phase transition

In this section, we present several systems of atoms or molecules on metallic surfaces, and we show that scanning-tunneling spectroscopy experiments performed on them can be explained qualitatively or semiquantitatively by means of the two-channel $S=1$ Kondo or Anderson models with anisotropy. In most of these cases, alternative explanations were proposed in the literature that contradict basic physical principles or are unsatisfactory. As the theory of the topological quantum phase transition is relatively new and it seems to be only captured by NRG calculations, while alternative explanations fail, it is important to show that a consistent explanation using new concepts and models exists. This is our focus, and not to provide an accurate fit of all the experimental curves mentioned below, an objective that lies beyond the scope of this paper.

4.1 FePc on Au(111)

The system of iron phthalocyanine (FePc) on the Au(111) surface has attracted a lot of attention during the last 15 years [11, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38]. According to LDA + U calculations, the basic electronic structure of the molecule is that shown in Fig. 2, taken from Ref. [36]. The partially filled orbitals of Fe are those of symmetry $3z^{2}-r^{2}$ with nearly one electron, and the degenerate so-called $\pi$ orbitals, of symmetry $xz$ and $yz$ with three electrons, resulting in a spin 1. This is in agreement with X-ray magnetic circular dichroism experiments [29].

Figure 2: (Color online) Electronic structure of MnPc and FePc. Reprinted with permission from Ref. [36]. Copyright 2020 American Chemical Society.

The differential conductance $dI/dV$ observed by scanning-tunneling spectroscopy of a single FePc molecule on Au(111) in the on-top position shows a narrow dip mounted on a broad peak [31, 34, 37] (corresponding to the curves at the bottom of Figs. 3 and 4). This is very suggestive of the spectral density of localized electrons near the topological transition, for anisotropy $D$ slightly larger than the critical one [8, 9] (See Fig. 1). However, most experiments were done before the development of the theory of the TQPT and were interpreted in a different fashion, as a two-stage Kondo effect: the $3z^{2}-r^{2}$ orbitals hybridize strongly with the conduction electrons with the same symmetry, giving rise to a first-stage Kondo effect and to the broad peak in $dI/dV$ around 20 meV. At a lower temperature, the Kondo effect due to the $\pi$ orbitals sets in giving rise to a dip of half-width $\sim 0.6$ meV in $dI/dV$ [31, 35]. The fits of the spectrum suggest the following hierarchy of the different orbitals in decreasing order of hopping amplitude to the tip: 3d${}_{3z^{2}-r^{2}}$ , conduction electrons with $3z^{2}-r^{2}$ symmetry, conduction electrons with $\pi$ symmetry and 3d${}_{\pi}$ [35].

However, an experiment that can discern between both scenarios (TQPT or two-stage Kondo effect) has been made. Raising the FePc molecule from the surface [34], the hybridization amplitudes are weakened and, with them, the exchange interactions between localized 3d electrons and conduction electrons. In the two-stage scenario explained above one expects that both features, the broad peak and the Kondo dip narrow since the corresponding Kondo temperatures should decrease. Instead, if the system is a non-Landau Fermi liquid close to the topological transition, decreasing the exchange interactions with respect to $D$ should broaden the dip as the system moves away from the TQPT. This last scenario is what is observed experimentally (see Fig. 3), giving support to the TQPT picture.

Figure 3: Left: experimental differential conductance of FePc on Au(111) as a function of voltage, taken from Fig. 2 (b) of Ref. [34]; and right: theoretical spectral density of 3d $3z^{2}-r^{2}$ electrons as a function of energy, as the molecule is raised from the surface, taken from Supplementary Fig. 3 of Ref. [11]. Parameters are $D=0.005$ $J_{z}=0.44f$, $J_{a}=0.22f$ in units of the band width taken as 1 eV. From top to bottom, the factor $f$ used is 0.1, 0.2, 0.5, 0.6, 0.7 and 0.8.

To construct the adequate model for FePc, one has to take into account the splitting of the $\pi$ orbitals as a consequence of the spin-orbit coupling (SOC) [38], neglected in previous treatments. The states $|\pi\sigma\rangle$ with one hole in the $\pi$ orbitals are (except for an irrelevant phase)

$$|a \uparrow \rangle=\frac{|xz\uparrow\rangle+i|yz\uparrow\rangle}{\sqrt{2}},|a\downarrow\rangle=\frac{|xz\downarrow\rangle-i|yz\downarrow\rangle}{\sqrt{2}},$$

$$|b \uparrow \rangle=\frac{|xz\uparrow\rangle-i|yz\uparrow\rangle}{\sqrt{2}},|b\downarrow\rangle=\frac{|xz\downarrow\rangle+i|yz\downarrow\rangle}{\sqrt{2}}.$$ (6)

The $|b\sigma\rangle$ states lie above the $|a\sigma\rangle$ by an energy of the order of the SOC, estimated in 76 meV for Fe [39]. This also leads to a significant orbital polarization which was observed [29], and to an anisotropy $D\sim 5$ meV.

Therefore, an appropriate model to describe the system is the anisotropic two-channel spin-1 Kondo model (A2CS1KM), one channel with strong exchange coupling $J_{z}$ for the $3z^{2}-r^{2}$ electrons and another channel of the $a$ states $J_{a}$. The effect of raising the molecule is incorporated in our model by reducing both $J_{z}$ and $J_{a}$ by the same factor $f$. In this way, the experiments can be semiquantitatively explained [11], as shown in Fig. 3.

In Ref. [37] the dependence of $G=dI/dV$ with temperature and magnetic field has been measured. The results have been also explained using the A2CS1KM [11]. As the temperature is raised, the dip is reduced and disappears at $\sim 10$ K. The dependence with the magnetic field shown in Fig. 4 is striking: the narrow dip is converted into a narrow peak as the magnetic field is increased. The theoretical results, taken from Ref. [11], reproduce semiquantitatively the experimental data. To take into account the asymmetry of the shape, we assume that the STM tip senses mainly the localized electrons of symmetry $\tau=3z^{2}-r^{2}$ with some admixture of conduction electrons with the same symmetry weighted by the parameter $q$ [40], which we take as $q=0.4$. The conductance given by our model $G_{m}$, represented at the right of Fig. 4 is therefore given by [11]:

$$G_{m}(V)=-\left[(1-q^{2}){\rm Im}G_{\tau\sigma}^{d}(\omega)+2q{\rm Re}G_{\tau\sigma}^{d}(\omega)\right],$$ (7)

where $G_{\tau\sigma}^{d}(\omega)$ is the Green function of localized electrons for symmetry $\tau$ and spin $\sigma$.

A better agreement can probably be obtained by enlarging $J_{z}$, which has the effect of broadening the broad peak, adjusting $J_{a}$ to broaden the dip a little bit, and by including the orbital polarization, which increases the effective coupling with the magnetic field by a factor of 3/2.

Figure 4: Experimental (left) and theoretical (right) differential conductance of FePc on Au(111) as a function of voltage for several values of the magnetic field. The experimental curves are taken from Fig. 2 (b) of Ref. [37], while the theoretical curves are taken from Supplementary Fig. 4 of Ref. [11]. Parameters as in Fig. 3 with $f=1$ and asymmetry parameter $q=0.4$.

4.2 MnPc on Au(111)

Curiously, in spite of having one electron less than FePc, the observed differential conductance in Mn phthalocyanine on Au(111) [41] is qualitatively very similar to that observed in the Fe system. There is a dip of half-width about 0.5 mV, mounted on a broad peak. Under the application of a magnetic field, the dip turns to a peak at $\sim 4$ Tesla, and for a larger magnetic field the peak splits (see Fig. 5). The latter behavior was not observed in the FePc system, but is expected if larger fields were applied in that case.

Figure 5: (Color online) Differential conductance $dI/dV$ of MnPc on Au(111) as a function of magnetic field $B$ and bias voltage $V_{sd}$, taken from Fig. 2(d) of Ref. [41].

It has been suggested that the observed behavior can be explained by a singlet ground state and a triplet excited state with a small excitation energy [41]. However, on one hand, the Hund coupling in 3d transition-metal elements are of the order of 0.7 eV, favoring a total spin 3/2. On the other hand, specific calculations for the singlet-triplet model with a small triplet excitation energy shows a $dI/dV$ that decreases slightly with increasing voltage $V$ for small $V$ (see Fig. 7 of Ref. [42]), in constrast to the experimental observations (see Fig. 5).

The calculated electronic structure for planar MnPc, is different from that in the gas phase (represented in Fig. 2) and corresponds to the intermediate-spin quartet ${}^{4}E_{g}$ $[(xy)^{1}(\pi)^{3}(3z^{2}-r^{2})^{1}]$ [43, 44]. This is in agreement with polarization-dependent N K-edge x-ray absorption spectra for MnPc on Au [45]. Therefore, the difference with the electronic structure for FePc on Au(111) is that, in the Mn system, there is a hole in the 3d${}_{xy}$ which is absent in the Fe system, leading to total spin 3/2 in the Mn case.

A realistic model for MnPc on Au(111) involves therefore three channels [states with symmetry $3z^{2}-r^{2}$, $a$ of Eq. (6) and $xy$] and is almost intractable with NRG. In order to have a qualitative understanding, we have studied a spin 3/2, two-channel Kondo model including anisotropy, assuming $J_{z}\sim 2J_{a}$ and $J_{xy}=0$. From the temperature dependence of the conductance, we see a first-stage Kondo effect in which the contribution of the dominant $3z^{2}-r^{2}$ channel is the usual one for a spin 1/2 system, saturating near $2e^{2}/h$ at low temperatures, without a dip. However, the contribution of the $a$ channel has a dip, so that the total conductance presents a narrow dip mounted on a broad peak. The ground state is a singlet.

The results can be qualitatively understood as follows. At temperatures of the order of the Kondo temperature of the dominant $3z^{2}-r^{2}$ channel (near 20 meV in the Fe system), the spin of that channel is screened and one is left with a spin 1 screened partially by the exchange in the $a$ channel and with anisotropy $D$. This model has been studied and, in presence of any $D>0$ [19, 20], the conductance and the spectral density have a dip, whose width decreases exponentially with $\sqrt{D}$, while the application of a magnetic field leads to a differential conductance of the same form as that shown in Fig. 5 of Ref. [20]. If, in the effective model after screening the spin 1/2 of the dominant channel, one includes the exchange of the third channel $xy$, the model is precisely the A2CS1KM, and one expects a topological quantum phase transition at finite $D$. The parameters should be renormalized, as expected from approximate treatments of similar 3-channel models [35].

4.3 Nickelocene on Cu(100)

The system of isolated double-decker nickelocene (Nc) molecules on Cu(100) substrates have been experimentally studied in detail [13, 46, 47, 48, 49, 50]. Density functional theory (DFT) calculations show that the electronic structure of Ni is basically 3d${}^{8}$, with one hole in each of the nearly degenerate $\pi$ orbitals ($xz$ and $yz$), with some mixing with the 3d${}^{9}$ configuration [48, 13]. Therefore the appropriate model to describe the system is actually the 2CS1AMA proposed for Ni impurities in a Au chain doped with oxygen [8, 9] or its integer-valent limit, the A2CS1KM [9].

Figure 6: (Color online) Differential conductance of several samples of Nc on Cu(111) (a) [(b)] correspond to spectra observed with frequency 2/3 [1/3], approximately. Within each panel, the higher a curve is, the shorter the distance from the molecule to the substrate. Taken from Fig. 5 of Ref. [13]. Copyright 2020 American Physical Society.

The experimentally observed spectra for the differential conductance $dI/dV$ are shown in Fig. 6. Left and right panels correspond to spectra observed with frequency 2/3 (case A) and 1/3 (cases B), respectively. The curves within each panel correspond to different positions of the tip. As the STM tip is approached to the molecule, the hybridization between tip and molecule states increases and a jump from a dip to a peak near $V=0$ takes place, which is consistent with the TQPT.

In the theory that was presented alongside the experimental observations, the ocurrence of a peak or a dip has been tentatively ascribed to a crossover in the spin of the molecule. This transition is noted as shifting from 1/2 in the contact regime (STM tip near the molecule) to 1 in the tunneling regime (STM tip far from the molecule), a deduction based on first-principle calculations [13, 48]. However, on one hand, these calculations miss relevant dynamical correlations and, therefore, they do not properly treat the Kondo effect which tends to screen the spin. On the other hand, the electronic structure does not change much between the two regimes and, as admitted by the authors, the change in the molecular charge is actually insufficient to account for the large change in the spin.

The observed spectra have many similarities with the spectral density of localized electrons in the A2CS1KM (see Fig. 1) but also important differences. For case A, there seems to be a first-order transition as the tip is approached to the molecule, avoiding the transition zone with a very narrow peak or dip, as shown in Fig. 1. For case B, the transition seems to be continuous, without a jump from a dip to a peak at zero voltage, as the hybridization (or exchange) between localized and conduction electrons is increased.

The first-order transition of case A can be understood as follows: as in FePc on Au(111), in which the molecule is raised when the STM approaches it [34], we expect that some variable $\eta$ which determines either the position or the shape of the molecule, modifies the hybridization $v$ of the 2CS1AMA [Eq. (2) assuming $v_{\tau}=v$, the same for both channels] [14], leading to a coupling of $\eta$ with our electronic model. In the absence of this coupling, one expects that the elastic energy is $E_{e}=K\eta^{2}/2$ (shifting the 0 of $\eta$ if necessary). It has been shown that the second derivative of the energy of the electronic model is strong and negative near the TQPT [14]. This means that for a soft spring (small $K$) the second derivative of the total energy is also negative at the TQPT, leading to a first-order transition in a Maxwell construction [14]. This reasoning provides a natural explanation of the observed behavior for case A.

Figure 7: (Color online) Differential conductance as a function of voltage for different values of $\Delta$. Other parameters are $U=3.5$, $U^{\prime}=2.5,$ $J_{H}=0.5$, $\epsilon=-3.0$, $T=0.0005$ [see Eq. (2)].

Case B probably corresponds to a hard spring (large $K$) and the first-order transition does not take place. The reason why an abrupt jump like that in Fig. 1 is not observed is two-fold: i) the magnitude of the jump decreases with the degree of intermediate valence [see Eqs. (3) and (4)] and ii) finite temperature. This is the most relevant parameter. These effects were investigated using the 2CS1AMA described by Eq. (2) [14]. In Fig. 7 we show the evolution of the differential conductance for different values of $\Delta=\pi v^{2}\rho_{c}$ where $\rho_{c}=1/(2W)$ is the density of conduction electrons assumed constant in the range $-W<\omega<W$. The half-band width $W=$ 1 eV is taken as the unit of energy. For small $\Delta$, $dI/dV$ has a dip mounted on a broader peak, as usual. As $\Delta$ increases, the dip narrows, but in contrast to the case of zero temperature, the minimum of the dip increases, and a very sharp dip like that of Fig. 1 is absent. For larger $\Delta$ the dip gradually disappears and the magnitude of $dI/dV$ near zero voltage increases. The overall behavior reproduces semiquantitatively the experimental results for case B shown in the right panel of Fig.6.

4.4 Fe Atoms on MoS${}_{2}$/Au(111)

Trishin et al. [51] studied experimentally a system consisting of an Fe atom on top of a monolayer of MoS${}_{2}$ deposited in turn on a Au(111) surface. As argued below, it is very natural to expect that the system is described by the A2CS1KM. MoS${}_{2}$ on Au(111) forms a Moiré structure, which implies strong local variations of the density of conduction electrons $\rho_{c}$. Therefore, depending on the specific position at which the Fe adatom is located, dramatic variations of the adimensional parameter $J^{\prime}=\rho_{c}J_{K}$ (which determines the Kondo temperature) are expected, and one might expect to observe the TQPT as in Fig. 1.

Figure 8: (Color online) Differential conductance corresponding to the experimental curves (b) to (f) (full lines, each curve is labeled as exp(b),etc.) and the results for our model (dashed lines,shifted upward by the same magnitude as the corresponding experimental curve). Note that (a) is not included and the top figure corresponds to (b). See the main text for the meaning of the $r$, $q$ and $f$ fitting parameters. Figure taken from Ref.  [14]. Copyright 2023 American Physical Society

The validity of the A2CS1KM to describe the system can be justified as follows. DFT calculations of Fe atoms on free-standing MoS${}_{2}$ indicates that the spin state of the atoms is either $S=1$ [52] or $S=2$ [53]. However, for $S=2$, one would expect a second jump in $G(V)$ at larger $|V|$ in the regime of low $J^{\prime}$, which is not observed experimentally [Fig. 3(h) of Ref. [51]]. On the other hand, experiments and DFT calculations indicate that the Fe atoms are located in positions with symmetry corresponding to the point group $C_{3v}$. Therefore, the Fe $3d$ orbitals are split into one $A_{1}$ singlet and two $E$ doublets [54]. Our comparison with the experiment (shown in Fig. 8), indicates that the spin 1 is formed by occupying the two states of an $E$ doublet (the agreement worsens when non-equivalent channels are considered). In this case, it is clear that the spin-orbit coupling originates a hard axis anisotropy $D(S_{z})^{2}$ with $D>0$ [14]. In addition, each of the degenerate orbitals of the doublet hybridizes with conduction states of the same symmetry [8, 9, 14]. This reasoning naturally leads to the 2CS1AMA with degenerate channels used in Refs. [8, 9, 14] and to the A2CS1KM in the integer valence limit.

The differential conductance $G(V)=dI/dV$ has been measured on nearly 40 different Fe positions. Six of them [(a) to (f)] are presented in Ref. [51] and five of them [(b) to (f)] are reproduced in our Fig. 8. In Ref. [51], fits of the different spectra were done using three different approaches: i) perturbation theory in the exchange coupling in an $S=1$ anisotropic one-channel Kondo model for cases (a) to (d), ii) a Frota peak (expected for the simplest Kondo model) for case (f), and iii) a Lorentz peak plus Frota dip (without justification) for case (e). Note that the fit of the peak and the dip requires 3 parameters for each one (determining position, width and intensity) in addition to a linear background. Therefore the resulting good fit is not surprising [51], but it lacks a physical justification.

In contrast, as shown in Fig 8, the A2CS1KM can semiquantitatively explain the data in a unified fashion. A better agreement with the experiment can be obtained by allowing some degree of intermediate valence for cases (e) and (f), as explained below, but we wanted to keep the number of free parameters as minimal as possible.

The numerical calculations were performed with the Ljubljana code of the NRG [15, 16], and were reported previously in Ref. [55]. We assume flat conduction bands extending from $-W$ to $W$ for both symmetries. We take $W=1$ eV and $D=2.7$ meV. The product $J^{\prime}=\rho_{c}J_{K}$ is assumed to vary among the different cases, due to the Moiré modulation. The TQPT is at $J^{\prime}_{c}\sim 0.135$. The different theoretical curves in Fig. 8 correspond to different values of the ratio $r=J^{\prime}/J^{\prime}_{c}$.

The structure at low voltage $V$ of the differential conductance $G(V)=dI/dV$ is determined by the localized and conduction electrons of symmetry $\tau$ included in the model. We assume that the STM tip senses mainly the localized $3d$ states with some admixture of conduction states. Thus, the contribution of the model to $G(V)$ at zero temperature is given by Eq. 7, where the Green functions $G_{\tau\sigma}^{d}(\omega)$ depend only on $r$, and $q$ is a measure of the contribution of the conduction states. In the experiment, there is also a linear background due to the contribution of other states, and $G_{m}(V)$ is affected by a factor $f$ which depends on the distance of the STM tip to the system. Therefore, to fit the experiment, the following expression is used

$$G(V)=fG_{m}(V)+A+BV,$$ (8)

which contains five parameters ($r,f,q,A$ and $B$), but the shape of each curve depends essentially on $r$, while $q$ controls the asymmetry. We have not included in the comparison with experiments the curve (a), which is similar to a rectangular dip formed by two step-like functions, because these steps are overbroadened in our calculations due to the limited resolution of the NRG at large energies [21].

The agreement between theory and experiment for cases (e) and (f) could be considerably improved by introducing intermediate valence effects. This is in fact expected since a larger conduction density of states increases the parameter $\Delta=\pi v^{2}\rho_{c}$ discussed in Section 4.3, which controls the degree of intermediate valence. It is well known that a smaller occupancy of the localized states shifts the Kondo peak to higher energies. Concerning case (e), the dip in the theory is more pronounced than in the experiment. However, as shown in Section 4.3, intermediate valence and finite temperature reduces the magnitude of the dip compared to that predicted by the A2CS1KM.

The comparison with the experiment can also be affected by the assumption of a constant density of conduction states and the effect of other orbitals not considered in our model. In any case, the experimental and theoretical results, including the LDA and NRG ones, strongly suggests that the underlying physics is that of the 2CS1AMA.

4.5 Fe porphyrin molecules on Au(111)

Experiments similar to those for the Nc molecule on Cu(100) (see Section 4.3), in which the STM tip is approached to the molecule, have been carried out for iron porphyrin molecules on Au(111) [56]. The LDA calculations indicate that the Fe spin is 1, with a partial occupancy of two orbitals (of symmetry $3z^{2}-r^{2}$ and $x^{2}-y^{2}$), that belong to different irreducible representations. In other words, the two channels are not equivalent. One expects that the A2CS1KM or the corresponding Anderson model with non-equivalent channels, should be appropriate for the system. In these experiments, the dip narrows as the contact regime is approached, but never turns to a peak, in contrast to Nc on Cu(100). Nevertheless, the results are consistent with our model assuming a weaker hybridization of the localized states with the substrate in comparison to Nc/Cu(100).

Figure 9: (Color online) Differential conductance obtained at different positions (II and III) of a chain of three Fe-porphyrin-based molecules on top of herringbone-reconstructed Au(111) containing a Br atom. The presence of a peak (dip) signals a molecule in a “Kondo” (“spin-flip”) position. The solid curves correspond to phenomenological fits of the data. Reprinted with permission from [57]. Copyright 2023 American Chemical Society.

.

Figure 9: (Color online) Differential conductance obtained at different positions (II and III) of a chain of three Fe-porphyrin-based molecules on top of herringbone-reconstructed Au(111) containing a Br atom. The presence of a peak (dip) signals a molecule in a “Kondo” (“spin-flip”) position. The solid curves correspond to phenomenological fits of the data. Reprinted with permission from [57]. Copyright 2023 American Chemical Society.

More recently similar experiments have been carried out using a Br decorated Au(111) surface [57]. In this case, depending on the particular position of the molecule with respect to the defects, sometimes a broad dip is observed and sometimes a narrow peak (see Fig. 9), consistent with the A2CS1KM for parameters near the TQPT on the non-trivial and trivial topological sectors, respectively.

5 Extension to larger spin and two-impurity systems

For impurities involving transition metal ions, the localized spin $S$ can vary in the interval $0\leq S\leq 5/2$. Clearly, for $S=0$ there is no Kondo effect. The number of channels can also vary reaching up to 5 channels. Since treating more than two channels is computationally too expensive with the NRG, we restrict the present study to two channels, assuming for the moment equivalent channels.

The effects of anisotropy when only one channel is present was studied before [17, 18, 19, 20]. For $S=1$, the transition occurs at $D_{c}=0^{+}$ [19, 20] and the Luttinger integral takes the value $\pi/2$ at this point [27]. For any single-channel $S\geq 1$ impurity, there is a competition between the underscreened Kondo effect [7] and the single-ion anisotropy for $D>0$, giving rise to a complex low energy behavior that strongly depends on the integer or half-integer nature of $S$ and the $D=0$ Kondo temperature $T_{K}$ (see Table 1).

For two channels and $S=1/2$, the anisotropy is irrelevant and one has a non-Fermi liquid behavior corresponding to the spin-1/2 two-channel Kondo model (2CKM) [58]. The $S=3/2$ case has been studied before with NRG [59, 60], and it was found that the two-channel Kondo effect also describes its low-energy physics as in the $S=1/2$ case. On the other hand, for $S=2$ we find a similar topological transition as for $S=1$. Therefore, the low-temperature behavior can be divided into two groups depending if the spin is integer or a half-integer. Table 1 presents a summary of the different cases.

$\#$ ch.	$S$	D	Low energy electronic state
1	1/2	irrelevant	fully compensated KE, ordinary FL
1	integer	$D_{c}=0^{+}$	two-stage KE:
			effective (anisotropic) S=1/2 KE for $D<T_{K};$
			quenched impurity spin for $D>T_{K}$  [17]
1	half-integer	$D_{c}=0^{+}$	two-stage KE: quenched impurity spin for $D<T_{K};$
			effective and complex S=1/2 KE for $D>T_{K}$  [17]
2	1/2	irrelevant	two-channel KE: non-Fermi liquid [7]
2	1	$D_{c}>0$	$D<D_{c}$: fully compensated KE
			$D>D_{c}$: quenched impurity spin, non-Landau FL [8]
2	3/2	$D_{c}=0^{+}$	effective two-channel (anisotropic) KE [59]
2	2	$D_{c}>0$	$D<D_{c}$: fully compensated KE
			$D>D_{c}$: quenched impurity spin, non-Landau FL [8]

Table 1: Low energy states for a magnetic impurity of spin $S$ with $D>0$ and coupled to $\#ch.$ equivalent conduction channels. KE = Kondo effect; FL = Fermi liquid. $T_{K}$ is the Kondo temperature for $D=0$.

Figure 10: (Color online) Entropy as a function of temperature for $J_{K}=0.16$ and different values of $S$ and $D$. $D>0$ ($D<0$) corresponds to $D=0.002$ ($-0.002$). $W$ is taken as the unit of energy.

In Fig. 10 we show the evolution of the entropy with temperature for different spins and anisotropy. For $D=0$ (orange curves), the two equivalent hybridizing channels reduce the ground state spin to a residual value $S^{\prime}=S-1$ and the entropy at zero temperature approaches ln($2S^{\prime}-1$). For negative $D$ (blue curves) the ground state is two-fold degenerate. The most interesting case is for positive $D$. In this case, for half-integer $S$ the low-energy behavior is dominated by the physics of the spin-1/2 two-channel Kondo model and the zero-temperature entropy is $\frac{1}{2}\ln(2)$. For integer $S$, the ground state is non-degenerate.

In Fig. 11 we show the conductance $G(T)$ (related to the spectral density) as a function of temperature for different $S$ and $D$. For $D=0$ (orange curves), $G(T)$ for $T\rightarrow 0$ tends to the unitary value $G_{0}=4e^{2}/h$, characteristic of two orbital- and spin-degenerate conduction channels. The case $D<0$ is complex and is not of interest to the central objective of this article, because in such a case, there is no quantum phase transition. For $D>0$ and half-integer $S$, the low-temperature value is $G_{0}/2$, characteristic of the spin-1/2 two-channel Kondo model. For integer $S$, $G(0)=G_{0}$ (zero) if $D$ is below (above) the critical anisotropy $D_{c}$ of the topological transition. In the figure only the case $D>D_{c}$ is shown.

We have also analyzed what happens for the half-integer case when the couplings of the two channels are different in presence of a small $D>0$. The ground state becomes a singlet and therefore, the entropy goes to zero for $T\rightarrow 0$. The contribution to the conductance of the channel with a larger coupling constant is similar to that of the simplest spin-1/2 1-channel Kondo model, increasing at $T_{K}$ and reaching $2e^{2}/h$ at zero temperature. At a lower characteristic temperature the conductance of the other channel decreases with decreasing temperature, indicating a dip in the corresponding spectral density. This seems to be the case of MnPc on Au(111). See Section 4.2.

Figure 11: (Color online) Conductance as a function of temperature for $J_{K}=0.16$ and different $S$ and $D$. $D>0$ ($D<0$) corresponds to $D=0.002$ ($-0.002$). $W$ is taken as the unit of energy.

Finally, we briefly comment that the two-channel spin-1 Anderson model with single-ion anisotropy [Eq. (2)] can be exactly mapped to a system of two $S=1/2$ impurities coupled through an anisotropic exchange interaction between them. The Hamiltonian is

$$H_{2I} = \sum_{k\alpha\sigma}\varepsilon_{k}c_{k\alpha\sigma}^{\dagger}c_{k\alpha\sigma}+\sum_{k\alpha\sigma}\left(v{c}_{k\alpha\sigma}^{\dagger}{d}_{\alpha\sigma}+\mathrm{H.c.}\right)+$$ (9)

$$+ \sum_{\alpha\sigma}\epsilon^{\rm 2I}d_{\alpha\sigma}^{\dagger}d_{\alpha\sigma}+\sum_{\alpha}U^{\rm 2I}n_{\alpha\uparrow}n_{\alpha\downarrow}+J_{z}s^{z}_{1}s^{z}_{2}+\frac{J_{\perp}}{2}\left(s^{+}_{1}s^{-}_{2}+s^{-}_{1}s^{+}_{2}\right).$$

For simplicity we consider two equivalent channels, and $\alpha=1,2$ refers to both impurities.

Changing the basis of the four impurity states from the individual spin projections $|s^{z}_{1}s^{z}_{2}\rangle$ to the total spin and projection $|SM\rangle$ with $S=0,1$, $M=-1,0,1$, and denoting by $E_{SM}$ the energies of these states in the new basis, one obtains an effective Hund coupling $J_{H}=E_{00}-E_{10}=-J_{\perp}$ and an exchange anisotropy directly related with the single-ion anisotropy of the spin-1 model through the relation $D=E_{11}-E_{10}=\frac{1}{2}(J_{z}-J_{\perp})$. It is interesting to note that the properties of the system are invariant under a rotation of one of the impurity spins (for example $s_{2}$) in $\pi$ around the $z$ axis (or equivalently a change of sign of the states with $s^{z}_{2}=-1/2$). This transformation changes the sign of $J_{\perp}$ and interchanges the states $|00\rangle$ and $|10\rangle$. In particular, the isotropic two-impurity model with $J_{\perp}=J_{z}>0$ is mapped into our A2CS1KM model with $J_{H}=D=J_{z}$, for which one expects a topological phase for low or moderate $T_{K}$.

Therefore, the two-impurity system [Eq.( 9)] will undergo a topological quantum phase transition depending on the exchange anisotropy. In fact, one of the first systems in which the existence of a non-trivial Fermi liquid was detected was the two-impurity system in Ref. [24]. This significantly broadens the range of impurity systems in which the TQPT theory could be useful to interpret scanning-tunneling spectroscopy measurements.

6 Summary and discussion

Since the seminal work of Nozières and Blandin [7], the Kondo effect in situations involving multiorbital magnetic impurities and/or coupled to more than one conduction channels has become a highly active area of research within condensed matter physics. The primary motivation stems from the prediction of exotic electronic states, such as non-Fermi liquids (overscreened Kondo effect) and singular Fermi liquids (underscreened Kondo effect), exhibiting distinctive dynamic and thermal behaviors, that, on one hand, could shed light on unconventional physics in heavy fermion compounds, and, on the other hand, they serve as paradigmatic models for quantum many-body phenomena.

The simplest generalization of the conventional Kondo effect to multiorbital systems, where a spin $S>1/2$ is fully screened by $n=2S$ channels, has been relatively underexplored, perhaps because it was believed that no new physics could be found in this context. However, we have shown in a series of recent publications that the presence of single-ion magnetic anisotropy drives these systems through a topological quantum phase transition, that separates two topologically distinctive local Fermi liquids.

Besides the theoretical relevance of our finding, we have shown that several systems consisting of isolated magnetic atoms or molecules on noble metal surfaces can be described by the 2-channel spin-1 Anderson model with anisotropy or its integer valent limit, the anisotropic two-channel spin-1 Kondo model. Both models exhibits the topological quantum phase transition, as a consequence of which the spectral density at zero temperature of the localized electrons has a jump between high values in the topologically trivial Fermi liquid phase to very low values in the non-trivial ”non-Landau” Fermi liquid phase. Near the transition, the spectral density is characterized by a narrow peak or dip (depending on the phase) mounted on a broad peak.

Several experiments with scanning-tunneling spectroscopy in different systems have in fact identified similar structures in the differential conductance. Five of these systems are listed in Section 4. They were usually interpreted using different, more conventional theories, probably due to the fact of the novelty of the concepts related to the topological quantum phase transition. We expect that this work contributes to disseminate these rather novel ideas to the community of condensed-matter and nanoscience researchers.

We have also shown that the concepts can be extended to larger spin and two-impurity systems.

ORCID iDs

A A Aligia https://orcid.org/0000-0001-9647-3926
L O Manuel https://orcid.org/0000-0001-7793-1709
G G Blesio https://orcid.org/0000-0002-1962-0969

References