betaFeedback

WorksPhysics and Astronomy

Anomalous transport of small polarons arises from transient lattice relaxation or immovable boundaries

Abstract

Elucidating transport mechanisms and predicting transport coefficients is crucial for advancing material innovation, design, and application. Yet, state-of-the-art calculations are restricted to exact simulations of small lattices with severe finite-size effects or approximate simulations that assume the nature of transport. We leverage recent algorithmic advances to perform exact simulations of the celebrated Holstein model that systematically quantify and eliminate finite-size effects to gain insights into small polaron formation and the nature and timescales of its transport. We perform the first systematic comparison of the performance of two distinct approaches to predict charge carrier mobility: equilibrium-based Green-Kubo relations and nonequilibrium relaxation methods. Our investigation uncovers when and why disparities arise between these ubiquitously used techniques, revealing that the equilibrium-based method is highly sensitive to system topology whereas the nonequilibrium approach requires bigger system sizes to reveal its diffusive region. Contrary to assumptions made in standard perturbative calculations, our results demonstrate that small polarons exhibit anomalous transport and that it manifests transiently, due to nonequilibrium lattice relaxation, or permanently, as a signature of immovable boundaries. These findings have consequences for applications including the utilization of organic polymers in organic electronics and transition metal oxides in photocatalysis.

Materials design and their subsequent application have driven scientific and technological innovation for decades. Understanding how a material responds to external factors, such as temperature gradients or mechanical stress, is crucial for effective design principles [1, 2]. Transport coefficients encode how materials respond to these external stimuli and offer a figure of merit in the search for novel materials: diffusion coefficients of biomolecules for drug delivery [3], Seebeck coefficients for thermoelectric generators [4], proton conductivity in membrane channels for flow batteries [5], and electron mobility in organic electronics and biocompatible device design [6, 7]. Great strides in spectroscopy have recently enabled transport measurements of photogenerated energy carriers with unprecedented detail [8] and advances in electronic structure theory have transformed our ability to parameterize physically transparent models [9]. Yet, while progress on the quantitative dependence of transport coefficients on the underlying chemical structure is promising [10, 11], the challenge is complicated by the difficulty of accurately predicting the quantum dynamics of carriers, which can exhibit either typical or anomalous transport. Indeed, questions of when, how, and over what timescales one should expect anomalous transport to emerge engender intense interest [12]. Hence, it is critical to interrogate transport in fundamental models and elucidate their mechanistic dependence on a system’s microscopic details [13].

Anomalous transport is an important feature across many important processes ranging from biomolecular flow in cells [14], to nanoparticle coordination [15], gas diffusion through porous materials [16], and even penetrant transport in glassy polymers [17]. The ubiquity of this phenomenon has ignited an intense theoretical focus on uncovering the sources and timescales of anomalous transport [18, 19, 20]. Theory and experiment indicate that it requires at least one of the following conditions: heterogeneous environments [21], mixed motion with both diffusion and trapping or binding [22], or the presence of memory effects [23]. In contrast to standard perturbative approaches that impose typical transport at the level of their rate expressions [24, 25], here we will leverage numerically exact simulations—of unprecedented reach in system size and time—to establish that polaron transport can be transiently or permanently anomalous and investigate the source of these anomalous behaviors to determine which of the aforementioned conditions are at play in polaron-mediated solid-state charge transport.

Theoretical approaches for calculating transport coefficients fall into two major categories: equilibrium and nonequilibrium. Equilibrium approaches invoke linear response theory-derived Green-Kubo relations to encode the nonequilibrium response of the system through the correlation of its equilibrium fluctuations [26], while nonequilibrium methods prepare systems in perturbed states that mirror the desired relaxation process. Regardless of the approach, computing transport coefficients is challenging. A persistent difficulty is the fact that simulations rely on finite systems, which introduce unwanted contributions that only vanish in the macroscopic limit [27, 28, 29] and misidentifying the influence of the system size has been shown to lead to misdiagnosis of the nature of the transport [30]. Despite the need to account for finite-size effects, differences in their severity between equilibrium and nonequilibrium simulations (in a given system and parameter regime) remain unknown. Yet, the two protocols consider qualitatively different densities in real space (e.g., of matter, charge, or energy), and the idea that the fluxes that they engender experience the system’s finite extent equivalently is dubious.

In the following, we compare the equilibrium and nonequilibrium frameworks for computing transport coefficients. Because these simulations cannot be performed on arbitrarily large systems due to their prohibitive cost, we interrogate a particular example of this problem: the electronic conductivity of the dispersive Holstein model [25]—a minimal model for conductivity in materials ranging from organic crystals and disordered polymers [31], to covalent organic frameworks [32], and nanomaterials [33, 34]. For example, ionizable functional groups in a polymer, such as unsaturated carbon atoms, are mapped onto a lattice of coupled sites that host mobile electronic excitations (excitons or charges). These excitations couple to, and therefore shift the equilibrium of, a site’s local vibrational (bosonic) environment, which spans a broad frequency range characteristic of condensed phase systems [35]. This model describes the stabilization of carriers by lattice distortions, resulting in polaron formation [25, 36, 37].

While the quantum dynamics of the dispersive Holstein model admit exact numerical solution [38, 39], only relatively short simulation times have been reached. Leveraging algorithmic advances that have become available only in recent years [40, 41], we have achieved unprecedented simulations in their combination of simulation times (∼25\sim25 ps) and system sizes (N≤40N \le40). These data allow us to capture the physics of polaron formation, determine the timescale of its phenomenology, and disentangle its influence on the transport it exhibits. We show that although both equilibrium and nonequilibrium approaches are expected to yield the same results, these are sensitive to system size in different ways. Further, in certain parameter regimes and model topologies, the nonequilibrium method exhibits diffusion while the equilibrium approach fails to converge at all.

What is the justification for the equivalence of equilibrium and nonequilibrium approaches? Since the 1950s, researchers have relied on linear response theory, and especially the Green-Kubo relations, to compute transport coefficients spanning thermal conductivities, diffusion constants, magnetic susceptibilities, and viscosity. These relations, built upon Onsager’s regression hypothesis, connect the nonequilibrium response of the system to an applied field and the system’s equilibrium fluctuations. In our equilibrium protocol, we apply linear response to determine the charge carrier mobility,

μ=12e0kBT∫−∞∞ds CJJ(s).(1)\mu= \frac{1}{2e_0 k_{\mathrm{B}}T}\int_{-\infty}^{\infty} ds\, C_{JJ}(s). \tag*{(1)}

Here, e0e_0 is the electron charge, kBk_{\mathrm{B}} is the Boltzmann constant, TT is the temperature, and CJJ(t)C_{JJ}(t) is the quantum current-autocorrelation function at thermal equilibrium.

In contrast, the widely adopted nonequilibrium protocol for determining charge mobility employs the Einstein relation, μ=e0D/kBT\mu= e_0D/k_{\mathrm{B}}T, to connect μ\mu to the diffusion constant DD of a charge (see the Methods section for details). At early times the mean square displacement (MSD) can display complex dynamics. Once the motion transitions from ballistic to diffusive transport, the MSD exhibits linear growth in time. In an infinite system, diffusion continues indefinitely, inspiring the definition

D=lim⁡t→∞12d MSD(t)dt.(2)D = \lim_{t\to\infty}\frac{1}{2}\frac{d\,MSD(t)}{dt}. \tag*{(2)}

We begin our analysis by performing a series of equilibrium and nonequilibrium simulations for the dispersive Holstein model parameterized to embody typical organic semiconductors [42]. We have employed the numerically exact hierarchical equations of motion (HEOM) [38] to obtain the dynamics. In the equilibrium protocol, one must evaluate the integral of CJJ(t)C_{JJ}(t) over infinite limits (Eq. (1)) [43]. At long times, this correlation function approaches zero since lim⁡t→∞CJJ(t)=⟨J^⟩2\lim_{t\to\infty} C_{JJ}(t) = \langle\hat{J}\rangle^2 and ⟨J^⟩=0\langle\hat{J}\rangle= 0. We define the timescale at which this occurs as the equilibrium time, teqt_{\mathrm{eq}}. Consequently, we compute the integral as a function of the limits, anticipating that the curve plateaus at the value 2e0kBTμ2e_0k_\mathrm{B}T\mu.

Yet, Figure 1a demonstrates that, for an open chain, the integral in Eq. (1) does not converge with tt. Examining the correlation function (Figure S1) reveals that CJJC_{JJ} approaches zero from below, converging faster for longer chains. This reveals the presence of a finite-size effect. Although our simulations reach unprecedented system sizes and simulation times, the parameters required to converge this integral remain unaffordable. The failure to return a meaningful value for μ\mu suggests an anomalous transport beyond the scope of linear response.

Figure 1. HEOM dynamics performed with dt=0.25dt = 0.25 fs, η=323\eta= 323 cm−1^{-1}, γ=41\gamma= 41 cm−1^{-1} and vij=50v_{ij} = 50 cm−1^{-1} a: Equilibrium simulation: the integral of CJJC_{JJ} against its integration limit for different linear chain lengths. The integral (Eq. (1)) does not converge over reasonable timescales. Inset: Longer simulation for 20-site lattice. b: Nonequilibrium simulation: time derivative of the MSD against time for different linear chain lengths. The curve for a system size of 20 sites shows a distinct diffusive region, which disappears for shorter chains. For all these studies, we initialize the charge at the middle of the linear chain. Inset: Time derivative of the MSD for 20-site lattice with linear time scale.

Figure S1. HEOM calculations were performed with dt=0.25dt = 0.25 fs, η=323\eta= 323 cm−1^{-1}, γ=41\gamma= 41 cm−1^{-1} and vij=50v_{ij} = 50 cm−1^{-1} to compute the current autocorrelation function for different system size NN. Plot displays how the real part of current autocorrelation approaches towards zeros as we increase NN.

What kind of transport does the nonequilibrium treatment reveal for this same system? Figure 1b reveals that the protocol exhibits a well-behaved diffusion region for an open chain of N≥20N \ge20 sites. Thus, the nonequilibrium method converges with increasing system size without any hint of persistent anomalous transport. This striking difference—that one method can circumvent finite-size effects fatal to the other—proves the two approaches interact with the system size in qualitatively different ways and motivates a deeper analysis.

Unlike the equilibrium case, the nonequilibrium protocol is sensitive to the initial state of the system. Hence, we place the charge on a single site of the chain and test the sensitivity of the results to the extent of local lattice equilibration around the charge prior to its migration. We test two extremes for the initial state preparation:

  1. With equilibration (w/ eq.)—the charge is allowed to completely equilibrate with its local vibrational environment on a particular site. This corresponds to a fully localized polaron state. The charge migration starts after local equilibration is complete.

  1. Without equilibration (w/o eq.)—the charge is placed on a lattice site and released before its local vibrational environment has reacted. Polaron formation and charge migration begin simultaneously.

These two different initializations produce different early-time derivatives of the MSD, evident in the difference between the full and dashed lines before ∼550\sim550 fs in Figure 2. If the system has a well-defined diffusion constant, these two curves must exhibit the same (flat) slope at later times. Therefore we can divide the dynamics into four distinct nonequilibrium regimes.

Figure 2. Nonequilibrium HEOM results for a 20-site linear chain. These calculations were performed with dt=0.25dt = 0.25 fs, η=323\eta= 323 cm−1^{-1}, γ=41\gamma= 41 cm−1^{-1}, and vij=50v_{ij} = 50 cm−1^{-1}. a: dMSD/dtdMSD/dt when the charge is initialized at the center of the linear chain. We can observe an approximately flat region ∼600\sim600–16001600 fs. b: Corresponding plot for α\alpha, which shows the initial ballistic region, followed by a subdiffusion towards a steady value of the α\alpha before the finite-size effect. The dotted line is the same simulation but for N=40N=40 c: dMSD/dtdMSD/dt when the charge is initialized next to one wall. In this case, the derivative never plateaus. d: Corresponding plot for α\alpha showing a similar initial ballistic region, followed by a superdiffusive region instead.

The first, ballistic regime quickly transitions to a period of semi-local polaron equilibration that ends when the two traces meet at τP∼550\tau_P \sim550 fs. We qualitatively identify the third regime as diffusive motion, beginning immediately after τP\tau_P. In an infinite system, this plateau would continue indefinitely as described by Eq. (2). However, here, the system enters the fourth regime beginning ∼1600\sim1600 fs and the curve decreases to zero. This time, which we call τR\tau_R, corresponds to when the charge reaches the boundary of the system, reflects back, and interferes with itself. Thus the decrease in dMSD/dtdMSD/dt toward zero signals the emergence of the finite-size effect for this method (see Figure S2). It follows that the success of the nonequilibrium simulation relies on the separation in timescales between the reflection time τR\tau_R and the polaron relaxation time τP\tau_P.

Figure S2. Top Left: Cartoon showing how α\alpha changes from ballistic to subdiffusive, and then exhibits a plateau region. The corresponding maximum value α\alpha is chosen as αlongtime\alpha_{\mathrm{longtime}} and the time is defined as τR\tau_R, after this α\alpha falls due to reflection. Relation between length scale and timescale for observing long time steady α\alpha value. Bottom: How αlongtime\alpha_{\mathrm{longtime}} converges towards 1 with increasing system length (Right) and reflection time (Left). Top Right: Plot showing length scale and time scale aa directly correlated in both w/ eq. and w/o eq. initializations.

The semi-local equilibration of the polaron determining τP\tau_P can be understood as the decreasing difference in on-site charge density for the w/ eq. and w/o eq. cases as a function of both time elapsed and distance from the initial site (see Figure S3) 2. In other words, both nonequilibrium simulations converge to the same charge distribution after an initial period of polaron relaxation. We quantify this via the scaling exponent of the MSD∝∣t∣αMSD \propto|t|^\alpha against (log) time. As Figure 2b shows, α=2\alpha= 2 for ballistic expansion at early times, but quickly decreases to α=1\alpha= 1, signaling diffusion. The region of the curve between these two limits holds essential information about the nature of τP\tau_P. Specifically, the relatively larger value of α\alpha for the w/o eq. case during the first ∼100\sim100 fs reports on the initially lower drag compared to the w/ eq. case, where a fully formed polaron presents an activation barrier to charge transfer. Further, careful inspection of α\alpha reveals that the w/o eq. simulation does not actually reach the diffusive regime before the onset of finite-size effects. If our argument is correct, we should expect to see convergence to α=1\alpha= 1 from below if we perform the (extremely costly) N=40N = 40 simulation to increase the time of unadulterated dynamics. Indeed, the dotted line in Figure 2b confirms our prediction. Moreover, even the w/ eq. simulation visits the region α<1\alpha< 1, which offers hints at the relative contribution from polaron formation on different sites. Physically, the formation of polarons on sites increasingly far from the original site corresponds to an increasing drag experienced by the charge carriers as they migrate along the chain, which dynamically manifests as ‘subdiffusion’ or α<1\alpha< 1.

Figure S3. Nonequilibrium HEOM results for a 20-site open chain. These calculations were performed with dt=0.25dt = 0.25 fs, η=323\eta= 323 cm−1^{-1}, γ=41\gamma= 41 cm−1^{-1}, and vij=50v_{ij} = 50 cm−1^{-1}. The dynamic charge density as a function of lattice site for both types of nonequilibrium simulation for both initial conditions. The difference between the top two panels confirms the initial polaron relaxation effect decreases with time.

Confirmation of τR\tau_R as the reflection time is simpler, as we can numerically ascertain the time at which the charge first reaches the final site. At some critical system size, this causes τP>τR\tau_P > \tau_R and the disappearance of the diffusive region. At this point, the finite-size effect becomes fatal for the nonequilibrium protocol, as is evident from the results obtained when reducing the lattice from N=16N = 16 to N=8N = 8 in Figure 1b. The dependence on system size implies that, even on a long chain, if one were to initiate the charge on a site closer to a boundary, reflection would onset sooner. However, we find the situation becomes murky when considering an extreme case like that

depicted in Figure 2c. Here, initiating the charge immediately next to the wall leads to a qualitatively different behavior of the MSDMSD, where dMSD/dtdMSD/dt increases even after τP\tau_P, preventing the identification of a diffusive region before τR\tau_R altogether. This suggests the presence of a persistent asymmetric bias or ‘wall effect’ in the dynamics, which is unrelated to the chain length. Physically, after an initial rise, the carrier density on each site is found to monotonically approach the equilibrium value from above (see Figure S4), meaning that a persistent directional bias emerges which precludes the onset of normal diffusion. We can further confirm this by plotting the value of α\alpha in Figure 2d, which for both cases reaches a stable value of ∼1.05\sim1.05 before τR\tau_R, i.e. a steady-state with superdiffusive character. As a result, this simulation never exhibits typical transport, even when NN is increased many times.

Figure S4. Nonequilibrium HEOM results for a 20-site open chain with dt=0.25dt = 0.25 fs, η=323\eta= 323 cm−1^{-1}, γ=41\gamma= 41 cm−1^{-1}, and vij=50v_{ij} = 50 cm−1^{-1}. Initializing the charge next to the wall in an open chain configuration, the population of each plotted site approaches the equilibrium value (solid grey line) from above.

Given that the proximity of the charge to a wall can prevent the nonequilibrium simulations from capturing diffusive behavior, one may ask if a similar effect could cause the lack of convergence in the equilibrium approach. After all, the initial condition of the system in a linear response method is the canonical equilibrium density, ∝e−βH^\propto e^{-\beta\hat{H}}, which places charge on every site. Therefore, regardless of the chain length, the equilibrium calculation always contains a component with knowledge of the wall, suggesting that the failure to determine the transport coefficient is unique to the open chain topology. Recent simulations have noted a similar lack of convergence in such topologies [45]. Hence, to test our hypothesis we turn to equilibrium simulations under periodic boundary conditions.

Figure 3 presents the results of equilibrium simulations for a periodic chain with all parameters being equal to those of the linear case analyzed above. As the chain length increases from N=4N = 4, we obtain rapid convergence to the N=15N = 15 value of 0.337 Å2^2 fs−1^{-1} as early as N=8N = 8. In all cases, the integral in Eq. (1) converges fully by 700 fs, in stark contrast to the linear case of Figure 1a, which failed to converge even after 25 ps. This confirms the validity of our hypothesis and demonstrates that the failure of the equilibrium simulation is dependent on the linear chain topology. This wall effect has a profound implication: since knowledge of the wall is independent of chain length, fully eliminating the finite-size effect in the equilibrium dynamics of open chains requires an infinitely large system. This has deep implications for the use of methods that employ tensor network decompositions, which perform most efficiently in open-chain topologies [46].

Figure 3. Equilibrium HEOM results for the periodic topology. The dynamics were calculated with dt=0.25dt = 0.25 fs, vij=50v_{ij} = 50 cm−1^{-1}, η=323\eta= 323 cm−1^{-1}, and γ=41\gamma= 41 cm−1^{-1}. a: Integral of CJJC_{JJ} against its integration limit for different periodic chain lengths. The integral converges both with teqt_{\mathrm{eq}} and system size. b: Convergence plot of the integral against ring size (final point of the left panel). The grey dashed line in both panels shows the final, converged value as being 0.337 Å2^2 fs−1^{-1}.

Throughout this analysis, we employed a parameter regime in which the nonequilibrium simulation clearly exhibits a diffusive region. However, timescale competition between τP\tau_P and τR\tau_R can cause a breakdown in the nonequilibrium protocol where it is impossible to identify a plateau in dMSD/dtdMSD/dt, as in the small-NN traces of Figure 1b. Understanding when the separation of these timescales allows the diffusion region to emerge and, therefore, yields the transport coefficient determines the minimum NN required to eliminate the finite-size effect. We thus turn to a thorough interrogation of the physically relevant parameter space. Figure 4 shows the nonequilibrium dynamics for a periodic Holstein chain as a function of two parameters: increasing charge-phonon coupling in the form of the reorganization energy (from bottom to top) and increasing phonon-environment decorrelation speed (from left to right). The results recapitulate the four nonequilibrium regimes: ballistic transport, semi-local polaron relaxation, the putative diffusion regime for which α→1\alpha\to1 at long times (when it appears), and the reflection time that emerges as a consequence of finite system sizes. Inspection of Figure 4 reveals that:

Figure 4. Nonequilibrium HEOM results for a 20-site periodic chain. These calculations were performed with dt=0.25dt = 0.25 fs and wij=50w_{ij} = 50 cm−1^{-1}. From top to bottom: Decreasing reorganization energy (i.e., charge-phonon interaction strength) which manifests as an earlier onset of reflection timescale τR\tau_R and decrease of the polaron relaxation effect’s magnitude (but not τP\tau_P itself). From left to right: Increasing the bath speed which reduces the initial polaron effect’s signature (both relaxation time τP\tau_P and magnitude)

  • While a homogeneous periodic lattice eliminates the issue of choosing the initialization site from the nonequilibrium perspective, one still observes the downturn in dMSD/dtdMSD/dt, which arises from the charge interfering with itself when it reaches the diametrically opposite point in the ring. Thus, periodic lattices also exhibit a competition of timescales that manifests as a finite-size effect.

  • As the phonon environment speed increases (going across a row), the duration of the semi-local polaron relaxation shortens, leading to a faster onset of diffusive motion and a larger diffusion constant.

  • Diminishing the reorganization energy (going down a column) reduces the semi-local polaron relaxation. Additionally, the reflection regime onsets earlier, consistent with the barrier to hopping lowering. While less evident, the onset of the diffusion region (τP\tau_P) remains unchanged.

  • As the reorganization energy becomes smaller than the intersite coupling (bottom row), the diffusive region disappears, as the polaron relaxation and reflection timescales overlap. Capturing the diffusion region in this parameter regime requires significantly larger lattices.

These results demonstrate that the severity of finite-size effects depends strongly on the parameter regime. To observe diffusion when γ=0.5v\gamma= 0.5v and η=0.6v\eta= 0.6v (Figure 4g) one needs a larger system where τR>τP\tau_R > \tau_P. This begs the question: does an equivalent size requirement emerge in the equilibrium approach? That is, do weakly coupled systems require bigger rings for equilibrium calculations to converge, and how does this value of NN compare to the nonequilibrium case? We find that the equilibrium approach, for this choice of parameters, converges by N=20N = 20 (see Figure S5), whereas the nonequilibrium simulations for N=20N = 20 (Figure 4g) fail to reveal a plateau region. What is more, even in the far easier parameter regime of Figure 2 (Figure 4b), the nonequilibrium simulation requires around N=40N = 40 to reach truly diffusive behavior, and therefore the correct value of dMSD/dtdMSD/dt. This competition of timescales in nonequilibrium simulations and,

Figure S5. Equilibrium simulation converged with time and system size: Equilibrium HEOM results for the periodic topology. The dynamics were calculated with dt=0.25dt = 0.25 fs, vij=50v_{ij} = 50 cm−1^{-1}, η=30\eta= 30 cm−1^{-1}, and γ=25\gamma= 25 cm−1^{-1}. Left: Integral of CJJC_{JJ} against its integration limit for different periodic chain lengths. Inset: Zoom of the region indicated by the black box ∼800\sim800 fs, showing the N>6N > 6 curves in detail. Right: Convergence plot of the integral against ring size (final points of the left panel). The grey dashed line in both panels shows the final, converged value as being 1.862 Å fs−1^{-1}.

in particular, the system size-dependent reflection time, means that nonequilibrium simulations require large system sizes to reveal the nature of transport and a converged transport coefficient. In contrast, the equilibrium approach, in this periodic topology where it functions correctly, is considerably cheaper to converge.

Our results thus enable us to determine the source of anomalous transport in small polaron-forming systems: semi-local nonequilibrium lattice relaxation, over a broad range of timescales, that generates spatial heterogeneity. Prior to complete lattice equilibration, which occurs long after τR\tau_R, each site encounters varying charges. This transient variability results in different extents of polaron formation across the lattice, akin to the heterogeneity observed in biomolecules within the cytoplasm or particles in crowded fluids. This leads to anomalous transport. Only close to equilibrium, when the charge discrepancy between neighboring sites diminishes considerably, does heterogeneity become vanishingly small. For this reason, equilibrium simulations easily capture the transport coefficient (indicating normal transport) when implemented in a periodic system. Nevertheless, in most parameter regimes, one can afford sufficiently large system sizes, yielding a nonequilibrium value for α\alpha which is close to one. Such α\alpha values enable one to adequately determine the transport coefficient in the nonequilibrium case.

We anticipate that our work, especially when combined with advances in electronic structure theory [9], will offer a unifying framework to understand and control small polarons across experiments. Indeed, cutting-edge measurements in systems spanning perovskites [47], conjugated polymers [48], nanomaterials [49], and transition metal oxide photocatalysts [50] bolster the idea that polaron dynamics stands to guide materials design for optimized transport properties.

Methods

Hamiltonian

The Holstein Hamiltonian has been widely used to capture the physics of charge transport phenomena in condensed phase systems. While the original Holstein model assumes local coupling to a single optical phonon mode [25, 36, 37], we study the dispersive Holstein model, which is more appropriate for organic crystals and disordered polymers [42],

H^=∑iNϵia^i†a^i+∑⟨ij⟩Nvija^i†a^j+∑iN[H^B,i+V^B,ia^i†a^i],(3)\hat{H} = \sum_{i}^{N} \epsilon_i \hat{a}_i^\dagger\hat{a}_i + \sum_{\langle ij\rangle}^{N} v_{ij}\hat{a}_i^\dagger\hat{a}_j + \sum_i^N \left[\hat{H}_{B,i}+\hat{V}_{B,i}\hat{a}_i^\dagger\hat{a}_i\right], \tag*{(3)}
H^B,i=12∑α[P^i,α2+ωiα2X^i,α2],(4)\hat{H}_{B,i} = \frac{1}{2}\sum_\alpha\left[\hat{P}_{i,\alpha}^2+\omega_{i\alpha}^2\hat{X}_{i,\alpha}^2\right], \tag*{(4)}
V^B,i=∑αci,αX^i,α.(5)\hat{V}_{B,i} = \sum_\alpha c_{i,\alpha}\hat{X}_{i,\alpha}. \tag*{(5)}

Here a^i\hat{a}_i (a^i†\hat{a}_i^\dagger) is the annihilation (creation) operator for an electron/hole on lattice site ii and X^i,α\hat{X}_{i,\alpha} (P^i,α\hat{P}_{i,\alpha}) is the dimensionless position (momentum) operator for α\alpha-th phonon mode connected with ii-th lattice site. The total number of sites is NN and each site is connected to a bath of phonon modes. The first term of Eq. M.1a sums over the energy contributions of occupied sites and the second term allows charges to hop from one site to its nearest neighbors (denoted by the angular brackets). The third term (explicitly given in Eq. M.1b) accounts for the free vibrational Hamiltonian, and the final term contains the on-site electron-phonon coupling, which is linear in the local phonon bath coordinates (see Eq. M.1c). This coupling is responsible for the fluctuation of the on-site energies and ultimately results in polaron formation, where the lattice distorts around a charge offering additional stabilization to that particular (localized) configuration.

The charge-phonon interaction through which the ii-th site is connected with its phonon modes is described by the spectral density,

Ji(ω)=π2∑αci,α2ωiαδ(ω−ωiα).(6)J_i(\omega)=\frac{\pi}{2}\sum_\alpha\frac{c_{i,\alpha}^2}{\omega_{i\alpha}}\delta(\omega-\omega_{i\alpha}). \tag*{(6)}

We assume that all Ji(ω)J_i(\omega) are equivalent for all sites and follow a Debye form commonly used to capture the dissipation in the condensed phase [35]

J(ω)=ηγωω2+γ2,(7)J(\omega)=\frac{\eta\gamma\omega}{\omega^2+\gamma^2}, \tag*{(7)}

where η/2\eta/2 is the reorganization energy and 1/γ1/\gamma is the timescale at which the phonon environment decorrelates. Aligning with previous studies [43], we set ϵi\epsilon_i to 0. The lattice distance between two consecutive sites is chosen to be a=5a=5 Å without loss of generality. We also choose all the nearest neighbor coupling to be the same, i.e., vij=vv_{ij}=v.

Current Autocorrelation Function

The current autocorrelation function CJJC_{JJ} is defined as

CJJ=1ZTr⁡[e−βH^J^(t)J^(0)],(8)C_{JJ}=\frac{1}{Z}\operatorname{Tr}\left[e^{-\beta\hat{H}}\hat{J}(t)\hat{J}(0)\right], \tag*{(8)}

where H^\hat{H} is the full Hamiltonian of the system, Z=Tr⁡[e−βH^]Z=\operatorname{Tr}[e^{-\beta\hat{H}}] is the partition function of the total system, J^(t)\hat{J}(t) is the current operator at time tt, and β=1/kBT\beta=1/k_{\mathrm{B}}T. The current operator is defined as [36]

J^=−iae0ℏ∑⟨mn⟩vmn(a^m†a^n−a^n†a^m),(9)\hat{J}=-\frac{iae_0}{\hbar}\sum_{\langle mn\rangle}v_{mn}\left(\hat{a}_m^\dagger\hat{a}_n-\hat{a}_n^\dagger\hat{a}_m\right), \tag*{(9)}

where i=−1i=\sqrt{-1} and ℏ\hbar is Planck’s constant. In all results shown, we have set e0=1e_0=1.

Mean Squared Displacement

We initialize the charge at a particular site of the lattice and observe the population at different sites as a function of time. One can compute its mean-squared displacement,

MSD(t)=∑idi2Pi(t),(10)MSD(t)=\sum_i d_i^2P_i(t), \tag*{(10)}

where did_i represents the distance between the iith site and the initial lattice site, and Pi(t)=Tr⁡[ρ^(t)a^i†a^i]P_i(t)=\operatorname{Tr}[\hat{\rho}(t)\hat{a}_i^\dagger\hat{a}_i] is the charge population at the iith site at time tt.

Simulation Details

We have employed the hierarchical equations of motion (HEOM) to obtain the numerically exact dynamics of the dispersive Holstein system at finite temperatures [38]. HEOM maps the entire open quantum system into auxiliary reduced density matrices (ADMs) and solves coupled differential equations for all the ADMs at each time step. We converged the HEOM calculation with respect to the internal parameters of the method: hierarchical depth L=26L=26 and Matsubara parameter K=2K=2. However, these calculations are prohibitively expensive both in memory and simulation time. Therefore, to render them affordable, we have employed dynamical filtering and an advanced nn-particle approximation that allows us to retain the numerical exactness of our result while minimizing the cost [40, 41]. To ensure consistency with previous works, we have performed all the simulations in the dilute electron/hole limit where there is only one charge in the system.

To compute the nonequilibrium relaxation simulation, we initialize the charge at the iith site by making the iith diagonal element of the reduced density matrix (RDM) equal to unity. To implement the w/ eq. strategy, we initiate the charge, set the nearest neighbor hopping term vij=0v_{ij}=0, and propagate all ADMs using HEOM until the density stops changing. This allows for local polaron formation on the ith site, yielding ρ^(t→∞)=a^i†a^iρ^B,ishifted∏k≠iρ^B,k\hat{\rho}(t \to\infty) = \hat{a}_i^\dagger\hat{a}_i \hat{\rho}_{B,i}^{\mathrm{shifted}} \prod_{k \ne i} \hat{\rho}_{B,k}, where ρ^B,ishifted=e−β(H^B,i+V^B,i)/Tr⁡B,i[e−β(H^B,i+V^B,i)]\hat{\rho}_{B,i}^{\mathrm{shifted}} = e^{-\beta(\hat{H}_{B,i}+\hat{V}_{B,i})}/\operatorname{Tr}_{B,i}[e^{-\beta(\hat{H}_{B,i}+\hat{V}_{B,i})}] and ρ^B,k=e−βH^B,k/Tr⁡B,k[e−βH^B,k]\hat{\rho}_{B,k} = e^{-\beta\hat{H}_{B,k}}/\operatorname{Tr}_{B,k}[e^{-\beta\hat{H}_{B,k}}]. After this initial equilibration, we allow the charge to hop by making vijv_{ij} nonzero and trace the charge density at each site with time. In contrast, the w/o eq. initial condition is simply ρ^(0)=a^i†a^i∏kρ^B,k\hat{\rho}(0) = \hat{a}_i^\dagger\hat{a}_i \prod_k \hat{\rho}_{B,k}.

We follow the protocol in Ref. [43] to compute equilibrium correlation functions using HEOM. This calculation is split into two parts. First, we propagate the reduced density matrix of the electronic degrees of freedom and all ADMs for timescales that are sufficiently long to achieve equilibrium. This calculation yields the correlated equilibrium density, ρ^eq=e−βH^/Z\hat{\rho}_{eq} = e^{-\beta\hat{H}}/Z. Then all the ADMs associated with this correlated state are multiplied by the current operator J^\hat{J} to make a modified initial condition ρ~(0)=J^ρ^eq\tilde{\rho}(0) = \hat{J}\hat{\rho}_{eq}, which we then propagate forward in time. The correlation function is calculated as CJJ=Tr⁡[J^ρ~(t)]C_{JJ} = \operatorname{Tr}[\hat{J}\tilde{\rho}(t)]. This expression is equivalent to that in Eq. (M.4).

The dynamic filtering parameters are δ=10−7\delta= 10^{-7} for nonequilibrium simulations (both w/ eq. and w/o eq. cases) and initial state preparation for equilibrium simulation and δ=10−10\delta= 10^{-10} (in atomic units) for the post-equilibration propagation in the calculation of equilibrium correlation functions.

Acknowledgments

A.M.C. acknowledges the start-up funds from the University of Colorado Boulder for partial support of this research. Acknowledgment is made to the donors of the American Chemical Society Petroleum Research Fund for partial support of this research (No. PRF 66836-DNI6). We thank Prof. Qiang Shi for sharing his HEOM code with us. This work utilized the Alpine high-performance computing resource at the University of Colorado Boulder. Alpine is jointly funded by the University of Colorado Boulder, the University of Colorado Anschutz, Colorado State University, and the National Science Foundation (award 2201538).

References

Supplementary Information

References

  1. [1]X. Yu, J. Zhou, H. Liang, Z. Jiang, and L. Wu, Progress in Materials Science 94, 114 (2018).
  2. [2]M. Park, J. Ryu, W. Wang, and J. Cho, Nature Reviews Materials 2, 1 (2016).
  3. [3]D. M. Mitrea, M. Mittasch, B. F. Gomes, I. A. Klein, and M. A. Murcko, Nature Reviews Drug Discovery 21, 841 (2022).
  4. [4]P. S. Mahapatra, K. Sarkar, H. Krishnamurthy, S. Mukerjee, and A. Ghosh, Nano Letters 17, 6822 (2017).
  5. [5]L. Xin, D. Zhang, K. Qu, Y. Lu, Y. Wang, K. Huang, Z. Wang, W. Jin, and Z. Xu, Advanced Functional Materials 31, 2104629 (2021).
  6. [6]C. Hermosa, J. Vicente Álvarez, M.-R. Azani, C. J. Gómez-García, M. Fritz, J. M. Soler, J. Gómez-Herrero, C. Gómez-Navarro, and F. Zamora, Nature Communications 4, 1709 (2013).
  7. [7]C. D. Bostick, S. Mukhopadhyay, I. Pecht, M. Sheves, D. Cahen, and D. Lederman, Reports on Progress in Physics 81, 026601 (2018).
  8. [8]N. S. Ginsberg and W. A. Tisdale, Annual Review of Physical Chemistry 71, 1 (2020).
  9. [9]F. Giustino, Reviews of Modern Physics 89, 015003 (2017).
  10. [10]M. A. C. Stuart, W. T. Huck, J. Genzer, M. Müller, C. Ober, M. Stamm, G. B. Sukhorukov, I. Szleifer, V. V. Tsukruk, M. Urban, et al., Nature Materials 9, 101 (2010).
  11. [11]M. Kadic, G. W. Milton, M. van Hecke, and M. Wegener, Nature Reviews Physics 1, 198 (2019).
  12. [12]R. Klages, G. Radons, and I. M. Sokolov, Anomalous Transport (Wiley Online Library, 2008).
  13. [13]H. Oberhofer, K. Reuter, and J. Blumberger, Chemical Reviews 117, 10319 (2017).
  14. [14]F. Höfling and T. Franosch, Reports on Progress in Physics 76, 046602 (2013).
  15. [15]S. Srivastava, L. A. Archer, and S. Narayanan, Physical Review Letters 110, 148302 (2013).
  16. [16]J. S. Kim and P. K. Kang, Advances in Water Resources 135, 103467 (2020).
  17. [17]N. Peppas and J. Sinclair, Colloid and Polymer Science 261, 404 (1983).
  18. [18]J. Szymanski and M. Weiss, Physical Review Letters 103, 038102 (2009).
  19. [19]F. Thiel, F. Flegel, and I. M. Sokolov, Physical Review Letters 111, 010601 (2013).
  20. [20]S. Gopalakrishnan and R. Vasseur, Physical Review Letters 122, 127202 (2019).
  21. [21]Y. Liu, X. Zheng, D. Guan, X. Jiang, and G. Hu, ACS Nano 16, 16054 (2022).
  22. [22]B. Berkowitz and H. Scher, Physical Review Letters 79, 4038 (1997).
  23. [23]S. C. Weber, A. J. Spakowitz, and J. A. Theriot, Physical Review Letters 104, 238102 (2010).
  24. [24]R. A. Marcus, Angewandte Chemie International Edition in English 32, 1111 (1993).
  25. [25]T. Holstein, Annals of Physics 8, 325 (1959).
  26. [26]R. Kubo, Journal of the Physical Society of Japan 12, 570 (1957).
  27. [27]R. Steinigeweg, J. Gemmer, and W. Brenig, Physical Review Letters 112, 120601 (2014).
  28. [28]Y. Zhang, G. Guo, K. Refson, and Y. Zhao, Journal of Physics: Condensed Matter 16, 9127 (2004).
  29. [29]F. Grasselli, The Journal of Chemical Physics 156, 134705 (2022).
  30. [30]M. Žnidarič, A. Scardicchio, and V. K. Varma, Physical Review Letters 117, 040601 (2016).
  31. [31]F. Ortmann, F. Bechstedt, and K. Hannewald, New Journal of Physics 12, 023011 (2010).
  32. [32]R. Ghosh and F. Paesani, The Journal of Physical Chemistry Letters 12, 9442 (2021).
  33. [33]H. Mousavi and H. Rezania, Modern Physics Letters B 24, 2947 (2010).
  34. [34]R. P. Cline, J. K. Utterback, S. E. Strong, G. Dukovic, and J. D. Eaves, The Journal of Physical Chemistry Letters 9, 3532 (2018).
  35. [35]U. Weiss, Quantum Dissipative Systems (World Scientific, 2012).
  36. [36]G. D. Mahan, Many-Particle Physics (Springer Science & Business Media, 2000).
  37. [37]A. S. Alexandrov, Polarons in Advanced Materials, Vol. 103 (Springer Science & Business Media, 2008).DOI
  38. [38]Y. Tanimura and R. Kubo, Journal of the Physical Society of Japan 58, 101 (1989).
  39. [39]D. Suess, A. Eisfeld, and W. Strunz, Physical Review Letters 113, 150403 (2014).
  40. [40]Q. Shi, L. Chen, G. Nan, R.-X. Xu, and Y. Yan, The Journal of Chemical Physics 130 (2009).
  41. [41]K. Song, S. Bai, and Q. Shi, The Journal of Chemical Physics 143, 064109 (2015).
  42. [42]T. Nematiaram and A. Troisi, The Journal of Chemical Physics 152 (2020).
  43. [43]L. Song and Q. Shi, The Journal of Chemical Physics 142, 174103 (2015).
  44. [44]D. Golež, J. Bonča, L. Vidmar, and S. A. Trugman, Physical Review Letters 109, 236402 (2012).
  45. [45]B. Bertini, F. Heidrich-Meisner, C. Karrasch, T. Prosen, R. Steinigeweg, and M. Žnidarič, Reviews of Modern Physics 93, 025003 (2021).
  46. [46]A. Strathearn, P. Kirton, D. Kilda, J. Keeling, and B. W. Lovett, Nature Communications 9, 3322 (2018).
  47. [47]M. Delor, H. L. Weaver, Q. Yu, and N. S. Ginsberg, Nature Materials 19, 56 (2020).
  48. [48]S. B. Penwell, L. D. Ginsberg, R. Noriega, and N. S. Ginsberg, Nature Materials 16, 1136 (2017).
  49. [49]G. M. Akselrod, F. Prins, L. V. Poulikakos, E. M. Lee, M. C. Weidman, A. J. Mork, A. P. Willard, V. Bulovic, and W. A. Tisdale, Nano Letters 14, 3556 (2014).
  50. [50]L. M. Carneiro, S. K. Cushing, C. Liu, Y. Su, P. Yang, A. P. Alivisatos, and S. R. Leone, Nature Materials 16, 819 (2017).

Paper details

vlnc:W_xge432je
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents