Electron transport in disordered insulating lattice under nonlinear electric field Preprint: prb Kunal Mozumdar, Herbert F. Fotso, Jong E. Han Abstract Transport in disordered systems often occurs via the variable range hopping (VRH) in the dilute carrier density limit, where electrons hop between randomly distributed localized levels. We study the nonequilibrium transport by a uniform DC electric field on a one-dimensional insulating tight-binding chain with the on-site disorder, using a finite-lattice calculation and the coherent potential approximation. We develop a theory of electric-field-assisted variable range hopping as a mechanism for nonlinear transport in a disordered chain. Our finite-lattice calculations of the electron propagation distance and the electron mobility determine the range of the variable range hopping as Δ < W ≲ 2 ​ Δ \Delta0}v_{g}(k)\frac{dk}{\dot{k}}=\frac{4t}{E}, (13) with the group velocity v_{g}(k)=2t\sin(k) and the semi-classical equation of motion \hbar\dot{k}=eEa . This behavior corroborates with the discussion Mozumdar et al. 2025 that, at high fields, the agreement of the CPA and the finite-lattice spectral functions indicates the tendency of delocalization. In Fig. 2 (b), we vary the disorder strength W at a fixed E=0.03 where Bloch oscillation effect is weak. We observe that the width of C(x) reaches a maximum at W=0.4\sim 0.45 . This non-monotonic behavior is a result of an interplay of the number of accessible localized levels near the Fermi energy with increasing disorder Mozumdar et al. 2025 , and the localization of the electronic wave functions at very high disorder. Figure 3: (a) Conductivity vs electric field ( \Gamma=0.0005 , N=501 , t=1 , T=0.01 ). The average conductivity slowly increases with the electric field. In the low electric field region the conductivity strongly varies with the disorder strengths [ W\in\{0.2,0.3,0.35,0.4,0.45,0.5,0.6,0.7,0.8,1.0,1.2\} : increasing in order indicated by the arrow in 3 (b)] with a rapid rise in the VRH regime 0.30.5 ) the mobility typically follows roughly the Fermi golden-rule ( \sim W^{-2} ) shown by the black dashed line. (c) CPA calculation of conductivity. While the overall behavior is consistent with the finite-lattice calculation, the high-disorder limit strongly overestimates the conductivity. (d) CPA calculation of mobility. Despite the similarity of the overall lineshape, the CPA overestimates the mobility and fails to produce the correct trends in the high- W limit. The electron propagation under the electric field in Fig. 2 (a-b) can be summarized by computing the drift distance L_{\rm drift} defined as L_{\rm drift}=\frac{\int_{0}^{\infty}C(x)xdx}{\int_{0}^{\infty}C(x)dx}. (14) We make a note here that, the drift distance is the travel distance of wave-propagation under the effect of an external field during the charge carrier’s lifetime. This is conceptually distinguished from the optimal hopping distance as proposed by Mott’s formalism (discussed in later sections). Drift distance incidentally could be a cumulative result of several hops over large distances in the lattice. As seen in Fig. 2 (c), the drift distance increases rapidly at small electric fields, and reaches a maximum as E increases. Initially, L_{\rm drift}(E) starts with a zero slope in the low disorder limit (blue, W=0.2 ) since localized states lie above the Fermi energy. In the range 0.020.5 ), a significant fraction of sites are filled and the VRH mechanism gives way to strongly disordered transport with a much-reduced variation of L_{\rm drift} by E , as discussed earlier. The crossover of the VRH transport to the Bloch oscillation behavior is most visible at the maximal VRH disorder at W=0.4-0.45 , with the dashed line denoting the behavior of x_{B}\propto 1/E . Fig. 2 (d) shows the variation of L_{\rm drift} with the disorder strength W . Within the VRH window (dashed vertical lines) of \Delta1/3 for W>W^{*} since the disorder values fall outside of the VRH Regime. (b) Log of Conductivity is plotted against the electric field for different disorder strengths [ W\in\{0.2,0.3,0.35,0.4,0.45,0.5,0.6,0.7,0.8\} : increasing in the order indicated by the arrow] with a fit (dashed line) of Eq. ( 20 ). The fit exponent is computed as \nu . (c) \text{Log}(\sigma) plotted against E^{-1/3} which depicts a straight line for all disorder values. We scale all the curves to match the fit of W=0.4 to shows a general agreement between the scaling law and the numerical result. We contrast the finite-lattice calculations with the CPA method. The results for conductivity in the CPA calculation are shown in Fig. 3 (c). For weaker disorder strengths ( W<2\Delta=0.6 ), the behavior is similar to the lattice calculations where the conductivity rises slowly at some value with respect to the electric field. For stronger disorder ( W>2\Delta ) the CPA-conductivity is strongly overestimated compared to the finite-lattice calculation, especially in the low-field limit. The insulator-to-metal transition occurs at W=0.6 for our chosen values of \Delta and \Gamma with the CPA conductivity decreasing with the field in the high disorder limit, in contrast to (a). As shown in Fig. 3 (d), the mobility undergoes a sharp transition at W=2\Delta . This is the disorder strength at which the CPA spectral function band edge crosses the Fermi level. In contrast, the mobility from the finite-lattice calculations shows a gradual rise before capping at the critical disorder, where the Lifshitz tail smears into the Fermi sea at zero field limit. This contrast between both calculations indicates the difference in capturing the physics of the model. While the CPA method is generally successful in capturing the spectral properties at very high fields, it is unable to capture the transport physics of Anderson localization at lower electric fields hence it does not show any signatures of VRH which typically arise from the transport in localized levels. III.2 Electric-Field Scaling Law of VRH In this section, we discuss an extension of Mott’s VRH behavior of Eq. ( 1 ) into the strong-field limit. The discussions on the strong-field scaling have been quite controversial. Differences in the E -field strength versus the temperature and the statistical criteria for hops under a field have led to competing predictions for the scaling exponent Mott 1970 ; Shklovskii 1973 ; Fogler and Kelley 2005 ; Apsley and Hughes 1974 . Here, we present our own interpretation of the strong- E -field scaling and provide microscopic and unbiased lattice calculations to support our relation. The electric field accelerates the electrons which can be excited to non-local sites in a chain. We modify Mott’s model by incorporating the field-driven excitations in the Miller-Abraham’s hopping probability {\cal W} , Eq. ( 2 ), as \mathcal{W}=\mathcal{W}_{0}\exp\left[-\frac{2x}{\xi}-\frac{\epsilon}{eEx}\right], (16) with the hopping distance x between two localized states of level-spacing \epsilon , the electric-field E , and the localization length \xi . The first term in the exponent captures the spatial overlap between two localized levels and the second term is modified to incorporate electric field excitations in place of the thermal excitation k_{\text{B}}T , as demonstrated in a dissipative field-driven lattice model Han 2013 . At low temperatures, the electric field is the major source of excitations in the system and the scaling behavior may be dictated by the nonlinear electric field. Here, we imply that an electron gains the energy eEx by traveling the distance x in the direction of the field, and after making the hop, the excess energy relaxes into thermal energy. In the steady-state limit, repeated inelastic scattering establishes the effective temperature as a function of the most probable hopping distance x , which needs to be determined as below. As in the Drude model Ashcroft and Mermin 1976 , we assume that energetic electrons relax their excess energy and thermalize before the next hop, often by emitting phonons. In our lattice calculations, the dissipation is provided by the hot-electron exchange with the fermion baths. We minimize the function in the exponent of the Miller-Abraham’s relation, Eq. ( 16 ) f(x,\epsilon)=\frac{2x}{\xi}+\frac{\epsilon}{eEx} (17) for some optimum values of x and \epsilon which maximize the overall hopping probability. Following Mott’s statistical argument, if there is at least one probable hop in some spatial distance and energy window \epsilon , we can write in one-dimension \epsilon=\frac{1}{2g(E_{F})x} (18) where g(E_{F}) is the density of states at the Fermi level E_{F} , and for simplicity we set g(E_{F})=1/(2Wa_{0}) with the unit-cell normalization with the lattice constant a_{0} . The exponent becomes f(x)=\frac{2x}{\xi}+\frac{Wa_{0}}{eEx^{2}}. (19) We can then optimize Eq. ( 17 ) with respect to x following similar steps to Mott’s approach Mott 1968 , and obtain the optimal x^{*}=(Wa_{0}\xi/eE)^{1/3} , from which we obtain \sigma(E)\sim\exp\left[-3\left(\frac{Wa_{0}}{eE\xi^{2}}\right)^{1/3}\right], (20) with the exponent \nu=1/3 . We may define the effective temperature T_{\rm eff} at the most probable x^{*} as f(x^{*})=eEx^{*}/T_{\rm eff} and obtain Marianer and Shklovskii 1992 T_{\rm eff}(E)=\frac{1}{3}eE\xi. (21) In the strong-field limit E\to\infty , the electron motions are accelerated along the potential slope and the thermal factor can be ignored when the second term in f(x) becomes small roughly when Wa_{0}/(eEx^{2})\sim 1 or x\sim\sqrt{Wa_{0}/eE}\ll\xi , beyond which the thermal factor becomes irrelevant. This threshold x determines the most likely hopping distance and the Miller-Abraham’s formula takes the first term of the wave-function overlap Mott 1970 ; Shklovskii 1973 , and we approximate the conductivity as \sigma(E)\sim e^{-2x/\xi}=\exp\left[-2\sqrt{\frac{Wa_{0}}{eE\xi^{2}}}\right], (22) with the exponent \nu=1/2 , in agreement with Ref. Shklovskii 1973 in the one-dimensional limit. As in previous calculations, we perform straightforward numerical calculations on a finite-size electronic lattice with the electrostatic potential created by the field, by using the Keldysh Green’s function method. We then confirm the above prediction by performing a nonlinear curve-fit on our numerical simulation of the conductivity shown earlier in Fig. 3 (a) to this scaling law in Eqs. ( 20 ) and ( 22 ). We compute the scaling exponent \nu by treating \sigma_{\infty} , E_{0} and \nu as free parameters. The exponent \nu is shown in Fig. 4 (a) for different disorder strength. The estimated exponent is close to the theoretical value of 1/3 predicted in Eq. ( 20 ) which is marked by the dashed line in Fig. 4 (a). However, at strong disorders ( W>W^{*} ) where stronger electric fields are required for significant variations of conductivity, the numerically obtained exponent \nu\approx 1/2 becomes more consistent with Eq. ( 22 ). The nonlinear fit is depicted in Fig 4 (b) which shows \log({\sigma}) against the electric field, with Eq. ( 20 ) shown as dashed lines. In the low-field limit, [ E<0.01 , this range is omitted in Fig. 4 (b)], the dissipation effects dominate over the VRH transport and the conductivity is constant with electric field. While in the high-field limit ( E>0.4 ) the fit overestimates the numerical conductivity as the system exhibits Bloch oscillations. Therefore, we note that Eq. ( 20 ) is a good fit at intermediate electric fields 0.13\lesssim E\lesssim 0.4 where the VRH transport is observed. In Fig. 4 (c), \log({\sigma}) vs E^{-1/3} shows a linear behavior which captures the electric field scaling law in Eq. ( 20 ) reasonably well in the intermediate field regime. Figure 5: (a) Conductivity vs electric field ( \Gamma=0.0005 , N=501 , t=1 , W=0.4 ) in the VRH regime for different Temperatures. In the low-field limit, the constant part of the conductivity follows Mott’s T^{-1/2} -scaling of the VRH. Conductivity crosses over to a temperature-independent limit at high fields. The threshold field at which the conductance starts to deviate from the zero-field limit is evaluated using an empirical fit method detailed in the text, shown as the black dashed line for T=0.02 . The threshold field progresses to higher values, as T increases. (b) Threshold field E_{\text{th}} vs temperature T . We emphasize that the electric-field scaling is studied by performing quantum mechanical calculations on an electronic model without any reference to Mott’s statistical arguments. The derived scaling law applies to the regime in which the nonequilibrium excitations dominate the thermal effects of the bath, in contrast to statistical models Fogler and Kelley 2005 ; Conchuir et al. 2016 and semi-classical approaches Malinin et al. 2004 ; Apsley and Hughes 1974 . The E^{-1/3} scaling with electric field has been observed in some experiments van Ancum et al. 1995 ; Govor et al. 2002 while the exponent \nu=1/2 also has been reported Rodin and Fogler 2009 ; Tremblay et al. 1989 . In the literature, different theoretical assumptions have led to different power-law scaling behaviors Conchuir et al. 2016 ; Yu et al. 2004 ; Tremblay et al. 1989 ; Rahman and Sanyal 2010 ; Park et al. 2017 ; Ladieu et al. 2000 , and our calculations directly based on quantum mechanical equations of motion may offer some insights into the nonlinear VRH transport in low dimensional disordered systems. We conclude this section by presenting experimentally relevant temperature dependence of the VRH in the high-field regime. A log-log plot of the conductivity vs the electric field, shown in Fig. 5(a), displays two transport regime separated by the crossover behavior. The low-field behavior of the constant conductivity is governed by the linear conductivity, which is consistent with Mott’s T^{-1/2} -scaling. The high-field limit, however, is nearly independent of the temperature. This is again consistent with the view that a high E destroys the localization. Constrained by the two limits, the temperature dependence of the localization-delocalization crossover is of a strong experimental interest. The debate also relates to resistive switching, the case of more distinct insulator-to-metal transitions driven by an electric field, where the relation between the threshold field and the temperature may point to the nature of the transition as classical vs quantum Han et al. 2023 . To evaluate the threshold electric field E_{\rm th} as a function of temperature, we use an empirical fit motivated by the activation mechanism similar to the discussions in Appendix Eq. (A6). \sigma(E,T)=\sigma_{1}(T)+\sigma_{2}\exp[-(E_{c}/E)] (23) and we compute the E_{\rm th} from the fit parameter \sigma_{1} , \sigma_{2} and E_{c} as E_{\rm th}=\frac{E_{c}}{\ln[\sigma_{2}/\sigma_{1}]}. (24) The E_{\rm th} vs. T , as shown in Fig. 5 (b), is an increasing relation. Such relations often suggest that a stronger electric field is required to overcome the thermal fluctuations, which means that the thermal effects counteract the nonequilibrium bias and it has been taken for evidence of non-thermal transitions Han et al. 2023 ; Yuen et al. 2009 ; Rahman and Sanyal 2010 from many-body effects. Our scenario based on the disorder model suggests something less dramatic. The \sigma_{1}(T) -term strongly varies with temperature over a few orders of magnitude due to the VRH mechanism, while the high- E limit of the conductivity \sigma(E,T) has much weaker T -dependence since the noneqiulibrium-driven effective temperature would overwhelm the bath temperature. Then, by writing the conductivity as \sigma_{1}(T)+\sigma^{\prime}(E) where \sigma^{\prime}(E) is T -independent and an increasing function of E , the crossover would happen when \sigma_{1}(T)\approx\sigma^{\prime}(E_{\rm th}) , thus E_{\rm th} is an increasing function of T . IV Conclusion We investigated the phenomenology of the nonlinear electron transport in a disordered insulator lattice in the steady-state limit under of large electric field. The Keldysh Green’s function method was used to solve the steady-state nonequilibrium rigorously with the dissipation modeled by local fermion thermostats. We set up the lattice in the bulk limit with a large enough self-averaging disorder region. In the experimentally relevant energy scale with the electric field of 10-10^{4} kV/cm, we showed that the variable range hopping (VRH) is the dominant transport mechanism when the disorder parametrized by W begins to overcome the gap \Delta in the range \Delta\lesssim W\lesssim 2\Delta . In the VRH regime, the electron drift distance, electric conductivity, and mobility showed strong dependence on E and W . In a narrow window of E , the hopping range varied more than over an order of magnitude as a function of E and W with a maximum that divides the regime of the VRH and the Bloch oscillation. Conductivity and mobility demonstrated disorder-assisted transport with a huge enhancement over the VRH regime. A comprehensive comparison with the coherent-potential approximation (CPA) highlighted the role of non-local disorder physics in the finite-lattice model. The nonequilibrium extension of Mott’s VRH argument is confirmed for an electric-field scaling of the conductivity \sigma(E) through an electronic model. Over electric fields outside the linear response and the Bloch oscillation regime, we verified the scaling law \sigma(E)\sim\exp[-(E_{0}/E)^{\nu}] with \nu=1/3 and \nu=1/2 in the intermediate and strong disorder regime, respectively, in one-dimension. We envision that experimental verification of the nonlinear transport is one of the most pressing issues in nonequilibrium-disorder physics. While the VRH physics has been well documented in the linear regime, its extension to the strong field regime awaits further studies. The transport mechanism in transition-metal oxides, for instance, is often considered as mediated by transport through localized in-gap states, and the nonlinear IV relation before the switching may be examined for the nonlinear transport scaling Zeng et al. 2006 ; van Ancum et al. 1995 ; Yu et al. 2004 ; Tremblay et al. 1989 ; Rahman and Sanyal 2010 . Finally, we addressed the VRH mechanism for the unconventional temperature dependence of the conductance threshold which can be experimentally examined in quasi-one-dimensional systems Rahman and Sanyal 2010 ; Yuen et al. 2009 . Acknowledgements. We thank J. P. Bird, G. Sambandamurthy, M. Randle, H. Zeng, and Z. Zhang for their helpful discussions. In particular we greatly benefited from insightful discussions with B. I. Shklovskii. We acknowledge computational support from the CCR at University at Buffalo. KM and HFF acknowledge the support from the Department of Energy, Office of Science, Basic Energy Sciences, under grant number DE-SC0024139. Appendix A Nonlinear Transport in Clean Limit Adopting the similar approach in the previous section, we can write the local electron density n_{\rm loc} in the electric field in terms of the same-time lesser Green’s function, after the change of variable p+ES\to p in the expression defined in Appendix A of Paper I, as \displaystyle n_{\rm loc}(E) \displaystyle= \displaystyle\frac{i\Gamma}{\pi}\int\frac{dp}{2\pi}\int_{-\infty}^{\infty}ds\int_{-\infty}^{-|s|/2}dS\frac{e^{2\Gamma S}}{s+i\eta}\times (25) \displaystyle\times \displaystyle\exp\left[is\epsilon(p)+\frac{iE^{2}s^{3}}{24m}\right] \displaystyle= \displaystyle\frac{i}{4\pi^{2}}\int dp\int_{-\infty}^{\infty}ds\frac{e^{-\Gamma|s|}}{s+i\eta}\exp\left[is\epsilon(p)+\frac{iE^{2}s^{3}}{24m}\right]. The Gaussian integral over p and an integral over S results in the expression \frac{ie^{i\pi/4}m^{1/2}}{(2\pi)^{3/2}}\int_{-\infty}^{\infty}ds\frac{e^{-\Gamma|s|}}{(s+i\eta)^{3/2}}\exp\left[i\Delta s+\frac{iE^{2}s^{3}}{24m}\right]. Rescaling the variable s\to(\sqrt{8m\Delta}/E)s , we obtain ie^{i\pi/4}\left(\frac{\sqrt{m}E}{(2\pi)^{3}\sqrt{8\Delta}}\right)^{1/2}\int_{-\infty}^{\infty}ds\frac{e^{-\gamma|s|}}{(s+i\eta)^{3/2}}e^{i\lambda(s+s^{3}/3)}, with \lambda=\sqrt{8m\Delta^{3}}/E and \gamma=\Gamma\sqrt{8m\Delta}/E . We are interested in the limit of \lambda\gg 1 and \gamma\ll 1 . By using the steepest descent technique for the \lambda\gg 1 limit, the integral simplifies to \displaystyle n_{\rm loc}(E) \displaystyle= \displaystyle\left(\frac{\sqrt{m}E}{(2\pi)^{3}\sqrt{8\Delta}}\right)^{1/2}\sqrt{\frac{\pi}{\lambda}}e^{-\frac{2}{3}\lambda} (26) \displaystyle= \displaystyle\frac{E}{8\pi\Delta}\exp\left[-\frac{2\sqrt{8m\Delta^{3}}}{3E}\right]. For the electric current in high field, however, we have to be cautious with the approximation of using the quadratic dispersion relation instead of that from a lattice model. Unlike the calculation of the occupation number above, the mechanical momentum p+eEt increases beyond the first Brillouin zone in solids, and the continuum approximation will deviate from the lattice results, which will be discussed below. The electric current density J is calculated as \displaystyle J(E) \displaystyle= \displaystyle\frac{iE}{8\pi^{2}m\Gamma}\int dp\int_{-\infty}^{\infty}ds\frac{(1+\Gamma|s|)e^{-\Gamma|s|}}{s+i\eta} (27) \displaystyle\times\exp\left[is\epsilon(p)+\frac{iE^{2}s^{3}}{24m}\right] \displaystyle= \displaystyle\frac{e^{3\pi i/4}}{8\pi m}\left(\frac{\sqrt{2m}E^{3}}{2\pi\sqrt{\Delta}\Gamma^{2}}\right)^{1/2} \displaystyle\times\int_{-\infty}^{\infty}ds\frac{1+\gamma|s|}{(s+i\eta)^{3/2}}e^{-\gamma|s|+i\lambda(s+s^{3}/3)}. The integral can be approximated by the steepest descent method up to the leading orders of \gamma as \displaystyle\int_{-\infty}^{\infty}ds\frac{1-\frac{1}{2}\gamma^{2}s^{2}}{(s+i\eta)^{3/2}}e^{i\lambda(s+s^{3}/3)} (28) \displaystyle\approx \displaystyle e^{-3i\pi/4}\left(1+\frac{\gamma^{2}}{2}\right)\sqrt{\frac{\pi}{\lambda}}e^{-\frac{2}{3}\lambda}, and J(E)\approx\frac{E^{2}}{8\pi m\Delta\Gamma}\left(1+\frac{4m\Delta\Gamma^{2}}{E^{2}}\right)\exp\left[-\frac{2\sqrt{8m\Delta^{3}}}{3E}\right]. (29) The first term in the parenthesis corresponds to the Drude current J_{D}=n_{\rm loc}(E)E/(m\Gamma) in the continuum model with n_{\rm loc}(E) from Eq. ( 26 ). However, in the nonlinear regime, a lattice model at E\gg\Gamma develops the Bloch oscillation with the period T=\pi/(eEa) (we use the unit e=a=1 ). With a scattering time much exceeding this time scale, the Bloch oscillation averages out the current. In the regime of our interest [see Fig. 3(a)], E/\Gamma\gtrsim 20 after nonlinear threshold fields, and the Bloch oscillation eliminates the first contribution in Eq. ( 29 ). Therefore, the conductance in the lattice model can be approximated with the next leading order of \Gamma as \sigma(E)\approx\frac{\Gamma}{2\pi E}\exp\left[-\frac{2\sqrt{8m\Delta^{3}}}{3E}\right]. (30) The result can be expressed as \sigma(E)=4(\Delta\Gamma/E^{2})n_{\rm loc}(E) . In the metallic lattice Han 2013 , the conductivity in the strong Bloch oscillation limit (E\gg\Gamma) is rigorously given as \sigma_{\rm metallic}(E)=(8/\pi)(t\Gamma/E^{2})n_{\rm metallic}(E) in the half-filled limit n_{\rm metallic}(E)=1/2 , following a form similar to the above. While Eq. ( 30 ) may be improved by considering the tight-binding dispersion relation for comparison with the numerical results, Fig. 3 (a) shows that Eq. ( 30 ) matches the numerical nonlinear conductivity reasonably well after the nonlinear threshold. The leading linear dependence in \Gamma is consistent with numerical results. By equating Eq. (A12) of Mozumdar et al. 2025 to Eq. ( 30 ) we may estimate the nonlinear threshold field at which the linear behavior crosses over to the nonlinear regime from the relation \Gamma/(2\Delta)\sim\lambda e^{-(2/3)\lambda} . *