Magnetic Gaps related to Spin Glass Order in Fermionic Systems

Abstract

We provide evidence for spin glass related magnetic gaps in the fermionic density of states below the freezing temperature. Model calculations are presented and proposed to be relevant for explaining resistivity measurements which observe a crossover from variable–range– to activated behavior. The magnetic field dependence of a hardgap and the low temperature decay of the density of states are given. In models with fermion transport a new metal-insulator transition is predicted to occur due to the spin–glass gap, anteceding the spin glass to quantum paramagnet transition at smaller spin density. Important fluctuation effects due to finite range frustrated interactions are estimated and discussed.

Magnetic hard gaps were observed in a large number of systems such as In–doped CdMnTe [1, 2], amorphous GeCr–films [3], boron– and arsenic–doped silicon [4, 5, 6, 7], in amorphous $Si_{1-x}Mn_{x}$–films [8], and irradiated polymers [9, 10] by manifestation of a thermal crossover from variable range hopping– or Mott–behavior to activated $exp(E_{H}/T)$–behavior in the low T resistivity. Reduction of the activation energy associated with the gap under an increased external magnetic field was held responsible for both the generally observed feature of a negative magnetoresistance and the magnetic origin of the gap. In all experiments the crossover in resistivity behavior was seen at temperatures close to the corresponding spin glass freezing temperature. An existing theory involving localized magnetic polarons [11, 12], which explained the gradual hardening of Coulomb gaps [13, 14], was considered to be a possible and appropriate explanation of some of the experiments [1, 8]. For other experiments [3, 5, 9, 10] spin glass order was suggested to be at the origin of the observed activated behavior. The random position of the implanted ions in these materials leads to a broad distribution of exchange integrals which favors random freezing of the magnetic moments.
Fermionic spin glasses are marked by the fascinating combination of a complex energy landscape with phenomena well known from strongly correlated systems like quantum phase transitions, non Fermi liquid behavior, phase separation, and interaction induced metal–insulator transitions. The paramagnet to spin glass quantum phase transition occuring in these systems was studied in [15, 16, 17], while strong correlation effects deep inside the spin glass phase were discarded. Recently the properties of highly correlated metals and interaction–induced insulators have attracted much interest [18].
The goal of this Letter is to report theoretical results, obtained by replica–methods and TAP–equations, for comparable phenomena in several different fermionic spin glasses, which develop either a hard gap or a pseudogap in the fermionic density of states at zero temperature, depending in size and shape on the spin glass order parameter(s). In any case the manifestation of spin glass order in this fermionic property represents a surprisingly large effect.
The magnetic field induced reduction of the gap, implying a negative magnetoresistance, and its stability against fermion hopping transport are demonstrated. Since the system stays insulating due to spin glass order it may be characterized as a SG (spin glass) based insulator. The existence of a SG based metal insulator transition is conjectured from the combination of our new calculations with results obtained in the vicinity of the quantum paramagnet to spin glass transition in metallic systems [19]. The low energy behavior of the Green function in the SG based insulating phase is derived and an estimation for the location of the abovementioned transition is given. We discuss implications of our results for the interpretation of the experiments [1, 2, 3, 4, 5, 6, 7, 8, 9, 10].
The fermionic Ising spin glass $ISG_{f}$ is defined by the Hamiltonian ${\cal{H}}=-\frac{1}{2}\sum J_{ij}\sigma_{i}\sigma_{j}-\mu\sum n_{i}$ with spins $\sigma=n_{\uparrow}-n_{\downarrow}$, the fermion number operator $n=n_{\uparrow}+n_{\downarrow}$, chemical potential $\mu$ , and Gaussian–distributed $J_{ij}$ with variance $J^{2}$. It is the simplest model that takes charge–spin couplings and corresponding fluctuations into account and, by virtue of its exact solvability in infinite space dimensions and for $T\rightarrow 0$ provides solid theoretical background to be compared with related experimental situations. Due to the possibility of double and zero occupancy of sites the freezing temperature is slightly lower than the one of its standard counterpart (SK–model) with one spin at each site. Thermally activated hopping of the localized fermions is effectively allowed for by working in the grand canonical ensemble with statistically fluctuating particle numbers at each site. While spin correlation functions of the model are static, correlators with an odd number of equal time fermion operators display the rich quantum–dynamical structure of an interacting fermion system. The fermionic path integral technique provides the tool for the calculation of quantities such as density of states, (dc or ac) conductivities et cetera for all kinds of spin glass models with charge degrees of freedom. We consider important the question, whether such a purely magnetic model, whose Hamiltonian consists just of a frustrated spin interaction, can have specific and strong effects on charge correlations and fluctuations, and whether a continuous distribution of exchange energies can cause a rigid gap of finite width.
The density of states $\rho_{\sigma}(\epsilon)=-\frac{1}{\pi}ImG_{\sigma}^{R}(\epsilon)$, calculated by means of the replica–method from the fermion Green’s function of the $ISG_{f}$ , reads

$$\rho_{\sigma}(\epsilon)=\frac{ch(\beta\mu)+ch(\beta(\epsilon+\mu))}{\sqrt{2\pi T\chi}\hskip 2.84544ptJ}$$ (1)

$$\hskip-14.22636pt\int_{-\infty}^{\infty}\frac{dz}{\sqrt{2\pi}}\frac{exp\left[-\frac{z^{2}}{2}-\beta(\epsilon+\mu+\sigma\tilde{H}(z))^{2})/(2J^{2}\chi)\right]}{ch(\beta\mu)+exp(\frac{1}{2}\beta J^{2}\chi)ch((\beta\tilde{H}(z))}\ ,$$

where $\tilde{H}(z)=H+J\sqrt{q}z$,$H$, $q$, $\chi$ denote effective field, external magnetic field, spin glass order parameter, and linear susceptibility. Effects from replica symmetry breaking (RSB), involving $\infty$–many order parameters $q_{i}$ and variables $z_{i}$, and from finite–range interactions $J_{ij}$ are not yet included here, but are considered below. The existence of a hardgap of size $2E_{g}(H)$ is nevertheless an unexpected and surprisingly large effect of spin glass order; it is best appreciated in our result for the density of states result at $T=0$, which is given by

$$\rho_{\sigma}(E)|_{{}_{T=0}}=\frac{{}^{\Theta(|E|-E_{g}(H))}}{{}_{\sqrt{2\pi}J}}e^{{}^{-\frac{1}{2J^{2}}\left[E-E_{g}(H)sgn(E)+\sigma H\right]^{2}}},$$ (2)

with $E:=\epsilon+\mu$ and the field–dependent gap energy $E_{g}(H)=J\sqrt{2/\pi}exp(-H^{2}/(2J^{2}))$. Eq.(2) is valid within the regime $|\mu|<E_{g}/2$, which corresponds to a half filled system at $T=0$ with filling factor $\nu=1+(1-q-T\chi)tanh(\beta\mu)$. In terms of $E_{g}(H)$ the following low temperature expansions (for linear susceptibility and order parameter), required for the exact evaluation of the DoS formula (1) at low T, were obtained as $q=1-E_{g}(H)T+O(T^{2}),\chi=E_{g}(H)+\frac{1}{2}E^{2}_{g}(H)(1-H^{2})T+O(T^{2})$. The last equation allows one to express the zero field gapwidth as $E_{g}=a(\mu)\chi T_{c}^{2}$ completely in terms of experimentally accessible quantities, where $a$ increases with the chemical potential from $2.1833$ at $\mu=0$ to $2.4075$ at $\mu=J/\sqrt{2\pi}$. These results show that i) the density of states is zero at $T=0$ in the finite interval given by $|\epsilon+\mu|<E_{g}(H)$ and ii) the gapwidth in a magnetic field is independent of spin orientation (in contrast to the local–limit Hubbard gap in an external field) and shrinks as the field is increased. This will be the source of the negative magnetoresistance in the extended Ising spin glass model with charge transport as discussed below. The physical consequences of these spin glass related properties are hence in agreement with the experimental observations of crossover behavior in the low T resistivities mentioned above. For single fermion energies $E\equiv\epsilon+\mu$ smaller than the gap energy $E_{g}(H)$ and $|\mu|<\frac{1}{2}E_{g}(H)$ the density of states decays to zero exponentially as given by

$$\hskip-14.22636pt\rho_{\sigma}(E)=\frac{1}{{}_{2\hskip 2.84544ptJ\hskip 2.84544ptcos(\frac{\pi}{2}\frac{E}{E_{g}(H)})}}\hskip 2.84544pte^{-\frac{1}{4}\frac{E^{2}_{g}(H)-E^{2}}{J^{2}}(1-\frac{H^{2}}{J^{2}})-\frac{1}{2}\frac{H^{2}}{J^{2}}}$$

$$(\beta E_{g}(H))^{-\frac{1}{2}}\hskip 2.84544pt\left[ch(\beta\mu)+ch(\beta E)\right]\hskip 2.84544pte^{-\frac{1}{2}\beta\frac{E^{2}_{g}(H)+E^{2}}{E_{g}(H)}}.$$ (3)

Figure 1: Single particle density of states (DoS) for the fermionic Ising spin glass ($ISG_{f}$) versus energy and magnetic field in units of J.

The infinite range insulating spin glass model allows to evaluate effects of replica symmetry breaking (RSB). Let us discuss the ($T=0,H=0$) result for the fermion Green’s function

$$G^{R}(E)=\sum_{\lambda=\pm 1}\int_{0}^{\infty}d\tilde{H}\;P(\tilde{H})\;\frac{1}{{}_{E+i0-\lambda\left[\tilde{H}+\bar{\chi}\right]}}\ \ .$$ (4)

The weight $P(\tilde{H})$ depends on the set of Parisi order parameters and simplifies to $\frac{1}{\sqrt{2\pi q}}e^{-\frac{\tilde{H}^{2}}{2q}}$ under replica symmetry reproducing thus Eq.(2) together with $\overline{\chi}\equiv\chi$. Hence the one particle excitations reveal a gap if the nonequilibrium susceptibility $\bar{\chi}$, which enters as a self energy and represents the effect of the Onsager reaction field, is finite. The influence of the randomly distributed local fields on the other hand is contained in the weight $P(\tilde{H})$. For the infinite–range model, the RSB–solution shows that $\bar{\chi}$ vanishes like T for $T\rightarrow 0$. Our explicit calculations show that in this case the hardgap turns into a pseudogap (details will be published elsewhere). Finite range interactions however lead to corrections that render $\bar{\chi}$ finite [21], which indicates that fluctuations will tend to reinstall the gap. Whereas a controlled theory for self energy corrections to Eq.(4) does not yet exist, an estimation of these small fluctuation contributions to the density of states is given below.
The gap also turns out to be robust against an additional fermion hopping term until the bandwidth exceeds a critical value. The gap and related properties of this extended itinerant spin glass model, whose magnetic phase diagram and –transitions have previously been analyzed [15, 17], are also discussed below.
Coulomb interaction effects are of diverse nature: the long range part will still tend to depress the (remaining) DoS near the Fermi level and thus stay particularly relevant when either spin glass order is weak or absent, or (in the case of fully developed spin glass order) when the Fermi energy lies close to the gap edges. The Hubbard coupling U leads to a shift of the chemical potential ($\mu\rightarrow\mu-\frac{U}{2}\nu$), and also to a shift $H\rightarrow H^{\prime}\equiv H+\frac{U}{2}m(H)$ of the applied magnetic field.
Corrections associated with a finite interaction range are estimated by considering TAP–equations for the $ISG_{f}$. Similarly to the spin case an Onsager reaction term $-\sum_{j=1}^{z}J_{ij}^{2}\sigma_{i}\chi_{jj}m_{i}$ (coordination number z) must be subtracted from the local field on the level of frozen in magnetizations, but to account for spin fluctuations a self energy $\frac{1}{2}\sum_{j=1}^{z}J_{ij}^{2}\sigma_{i}\chi_{jj}\sigma_{i}$ must be added again. Thus the polarization $\sigma_{i}J_{ij}\chi_{jj}$ of surrounding spins lowers the energy of magnetic sites by $E_{g}=[\sum_{j=1}^{z}\chi_{jj}(J_{ij})^{2}]_{av}$. As this gap energy originates in the formation of polarized spin clusters it differs from the Hubbard gap width U caused by onsite repulsion. Now we argue that for a finite interaction range the local susceptibility behaves like $\overline{\chi}=const.+O(T)$ instead of the $\tilde{\chi}\sim T$ found by solving the TAP–equations for the infinite–range SK–model [20]. The latter is valid for a system stuck in a fixed free energy valley with infinitely high barriers and is intimately related to a pseudogap $P(h_{loc})\sim h$ in the distribution of local fields, which is absent in finite dimensional systems [21]. Thus fluctuation effects in systems with finite–range frustrated interactions must be expected (despite the not yet available finite–range spin glass theory) to favor the hardgap as predicted by the replica symmetric solution eq.(2) of the infinite range model. Hence we conjecture $\rho(\epsilon)\sim[\Theta(\epsilon-\chi\sum_{j=1}^{z}(J_{ij})^{2})]_{av}\sim\epsilon^{\frac{z}{2}}$ and therefore along the lines of [13] $\sigma_{DC}\sim\exp(-B/T^{\frac{z+2}{z+8}})$ for bonds with a Gaussian distribution. A sufficiently large (magnetic) coordination number z or a long–ranged RKKY–interaction leads to an activated behavior of the conductivity, otherwise nonuniversal temperature exponents may occur. A bimodal $\pm J$ – distribution on the other hand is expected to result in a true hardgap.
Multi–valley correlations of statistical DoS - fluctuations as given by $C_{\rho}^{(k)}\equiv\overline{\rho_{a_{1}}(E_{1})\rho_{a_{2}}(E_{2})...\rho_{a_{k}}(E_{k})}$, where the $a_{i}$ denote distinct replicas, do not factorize within the spin glass phase. For $|E_{k}|=E<E_{g}(H)$ we obtain $C_{\rho}^{(k)}(E)=(\rho(E))^{k}\frac{B(\frac{k}{2}d_{-}^{E},\frac{k}{2}d_{+}^{E})}{(B(\frac{1}{2}d_{-}^{E},\frac{1}{2}d_{+}^{E}))^{k}}$ with $d_{\pm}^{E}\equiv\frac{E_{g}(H)\pm E}{E_{g}(H)}$ and Beta–function B. These multi–valley correlations also vanish exponentially within the gap as $T\rightarrow 0$ and, weaker than the averaged DoS though, they increase as either gap edge is approached at fixed low T.
A comparison of the free energy of ordered and disordered phase predicts a first order transition to the paramagnetic phase at $\mu=0.9$, but when analyzing the stability of the magnetic solution according to the scheme of de Almeida and Thouless [21] we found two negative eigenvalues of the Hessian matrix instead of the expected one negative value indicating just the instability towards RSB. As described in [21] for spin glass problems, one has to choose the stable solution with the lowest free energy, hence the system undergoes a first order transition from half filling to the completely filled paramagnetic state at $\mu=\frac{1}{2}E_{g}$.
The electronic kinetic energy is properly embedded in an itinerant $ISG_{f}$–model by adding the random hopping Hamiltonian $H_{t}=\sum_{i,j;\sigma}t_{ij}c_{i,\sigma}^{\dagger}c_{j,\sigma}.$ The real–symmetric $t_{ij}$ are Gaussian–distributed and chosen to describe a semi–circular band of width $E_{0}$. The 4–fermion term which results from $t_{ij}$–averages is decoupled by means of a quaternionic matrix field $\underline{\underline{R}}$ in a way familiar from Anderson localization theory. The saddle point value $r_{\sigma}(\epsilon_{l})$ of this field obeys the selfconsistency equation $r_{\sigma}(\epsilon_{l})=\frac{iE_{0}^{2}}{4}G_{\sigma}(\epsilon_{l})=-\frac{iE_{0}^{2}}{4n}\sum_{a}\int_{0}^{\beta}d\tau e^{i\epsilon_{l}\tau}<T_{\tau}c_{a,\sigma}(\tau)c_{a,\sigma}^{\dagger}(0)>$, which becomes exact for infinite space dimensions. The expectation value employs the quantum–dynamical action ${\cal A}=\int_{0}^{\beta}\int_{0}^{\beta}d\tau d\tau^{\prime}(\sum_{a,\sigma}c_{a,\sigma}^{\dagger}(\tau)[\delta(\tau-\tau^{\prime})(\partial_{\tau}+\mu)+ir_{\sigma}(\tau,\tau^{\prime})]c_{a,\sigma}(\tau)+\sum_{a,b}S^{a}(\tau)S^{b}(\tau^{\prime})Q^{ba}(\tau^{\prime},\tau))$ containing the disorder averaged spin interaction. In contrast to the quaternionic charge fields $\underline{\underline{R}}$, $\underline{\underline{Q}}(\tau^{\prime},\tau)$ denotes the quantum spin glass order parameter field; quantum dynamics necessitate the solution of dynamical selfconsistency equations for the spin autocorrelation function $<Q^{aa}(r,\tau^{\prime},\tau)>$ at saddle point level. Apart from the different interaction term in the Lagrangian, the selfconsistency equation for $r_{l,\sigma}$ resembles the LISA equation [18] for a model with nonrandom nearest neighbor hopping on a Bethe lattice with infinite coordination number, commonly used to study strong correlations in the Hubbard model [18]. Quantum dynamics of the $\underline{\underline{Q}}$–fields can be important close to the $T=0$ paramagnet - SG– transition; yet the influence of the fermion kinetic energy on the DoS is described sufficiently well by a static, replica symmetric approximation for the spin autocorrelation function. For $H=0$ and half filling the Green function follows from

$$r_{l}=\frac{1}{4}E_{0}^{2}<(\epsilon_{l}+r_{l})/((\epsilon_{l}+r_{l})^{2}+\tilde{H}^{2})>_{\Phi}$$ (5)

with $<f(\tilde{H})>_{\Phi}=\int_{z}^{G}f(\tilde{H})\int_{y}^{G}\frac{e^{\Phi(\tilde{H}(z,y))}}{\int_{\tilde{y}}^{G}e^{\Phi(\tilde{H}(z,\tilde{y}))}}$. The regularized determinant $\phi(\tilde{H})=Tr\log[1+\frac{\tilde{H}^{2}}{(\epsilon_{l}+r_{l})^{2}}]$ results from integrating out the decoupled fermion fields, and $\tilde{H}(z,y)=J(\sqrt{q}z+\sqrt{T\chi}y)$. In order to proof the stability of the hard gap against fermionic hopping we present a solution of the above equation valid in the low energy regime. The leading contribution in a double expansion in $E_{0}$ and frequency $\epsilon_{l}$ is $r_{l}=\epsilon_{l}\frac{E_{0}^{2}}{4}<\frac{1}{\tilde{H}^{2}}>_{\Phi{ISG_{f}}}$ giving just back the $ISG_{f}$ - result. Now we note that the ansatz $r_{l}=\alpha\ \epsilon_{l}$ i)satisfies eq.(5) to all orders in $E_{0}^{2}$ and to $O(\epsilon_{l})$ for $\alpha=\frac{E_{0}^{2}<J^{2}/\tilde{H}^{2}>_{\Phi_{\alpha}}}{4J^{2}-E_{0}^{2}<J^{2}/\tilde{H}^{2}>_{\Phi_{\alpha}}}$ and ii) corresponds to a real $G^{R}(\epsilon)=-\frac{4\alpha\epsilon}{E_{0}^{2}}$ and hence to a vanishing DoS. Solving a set of coupled self consistency equations one obtains $q_{\alpha}=1+O(T)$, $\chi_{\alpha}=\frac{\sqrt{2/\pi}}{1+\alpha}+O(T)$, and finally $<\frac{1}{\tilde{H}^{2}}>_{\Phi_{\alpha}}=<\frac{1}{\tilde{H}^{2}}>_{\Phi_{ISG_{f}}}=0.568659$. The last result implies that for $E_{0c}=2.65219J$ the low energy approximation breaks down since $\alpha$ diverges. On the other hand it is known that for $E_{0}=32J/(3\pi)$ the system undergoes a paramagnet to spin glass quantum phase transition and has a semielliptic DoS at zero temperature [19], hence we conclude that the divergence of $\alpha$ at $E_{0c}$ indicates the neighborhood of a metal insulator transition exclusively caused by random interactions. A mobility edge induced by either randomness in fermion hopping or in the interaction can interfere with the SG–based MIT. A transition between gapped and non–gapped insulating behavior appears possible too and requires further analysis. Several different lower critical dimensions such as the one for Ising spin glass order at $T_{c}>0$ and the one for Anderson localization will also play a role in this context. Anderson localization of states neighboring a gap had been studied in Ref.([22])
Using these results for the interpretation of experiments one should distinguish two classes of systems. In $GeCr$ [3], irradiated polymers [9, 10], $SiB$ [5], ion implanted [7] and Czochralski grown [6] $SiAs$ localized impurity spins are the main source of magnetic behavior. The interaction between them is due to direct wave function overlap, hence the spatial randomness in their positions translates itself into a broad distribution of exchange energies favoring spin glass freezing in the ground state. We believe that this experimental situation is well described by the $ISG_{f}$ and that the observed activated conductivity is explained by the spin freezing induced hard gap in the DoS. However, in Indium–doped $CdMnTe$ [1] and annealed $SiMn$ [8] films the carrier spins interact with Manganese spins, may form magnetic polarons and thus create a finite activation energy for hopping transport. In [1, 8] this mechanism is held responsible for the observed low T behavior of the conductivity. Since $CdMnTe$ is a prototype insulating spin glass a SG–origin of the activation energy can only be ruled out by checking whether the thermal crossover of the conductivity (in In–doped material) appears already above the freezing temperature. If this were not the case an explanation of these experiments could be found in a spin glass of randomly interacting magnetic polarons. Other mechanisms discussed in the literature as cause for the observed activated conduction are a hardening of the Coulomb gap by formation of electronic polarons [11] and nearest neighbor hopping. The former can be ruled out because of the observed negative magnetoresistance, the latter because of Efros–Sklovskii variable–range hopping above the crossover temperature. The possibility of a dimensional crossover at low T needs to be considered, but it would lead to a $(T_{0}/T)^{1/3}$ behavior and not to the appearance of a hardgap.
We acknowledge support by the SFB410 of the DFG. We thank G. Landwehr and C. Rigaux for valuable hints, and a referee for bringing ref.(22) to our attention.