Researchers analyze non-Markovian barrier-crossing dynamics in simulations and analytical theory; three scaling regimes for mean first-passage time.

Reaction dynamics in non-Markovian systems with non-monotonically decaying memory

Whereas the effect of monotonically decaying memory kernels on reaction dynamics has been studied extensively, the effect of oscillatory memory, observed for widely different systems, is much less explored. We analyze non-Markovian barrier-crossing dynamics in the presence of a single exponentially decaying oscillatory friction memory kernel by simulations and analytical theory. We identify three scaling regimes: two in which the mean first-passage time τ MFP \tau_{\mathrm{MFP}} scales quartically and linearly with the memory kernel decay time τ o \tau_{\mathrm{o}} , respectively, and a third in which τ MFP \tau_{\mathrm{MFP}} scales quartically with the oscillation period τ φ \tau_{\varphi} of the memory kernel. Our results are relevant to a broad class of many-body dynamical systems such as dihedral isomerization processes and chemical reactions in solvents.

Physical, chemical, and biological many-body systems are commonly described in terms of low-dimensional reaction coordinates [ 45 , 62 , 18 , 58 , 61 ] . A prominent example is protein folding, which can be monitored using suitably defined scalar reaction coordinates. [ 6 , 21 , 55 , 56 ] . Solving the equations of motion of nonlinear many-body systems analytically is generally impossible, while direct simulations are computationally expensive [ 6 ] . Having an analytic framework that relates the many-body dynamics of a complex system to the reduced or “coarse-grained” dynamics of a low-dimensional collective variable is therefore essential for modelling and gaining a deeper understanding of the dynamics of many-body systems [ 66 , 17 , 5 ] . Projecting the many-body dynamics onto reaction coordinates gives rise to an effective free-energy profile [ 12 , 23 ] . The free energy profile defines states of the system, so that transitions between states can be described as free energy barrier-crossings. According to Arrhenius’ law, the barrier-crossing or mean first-passage times scale exponentially with the potential barrier height U 0 U_{0} [ 4 ] , as

where the prefactor τ ∗ \tau_{} has been predicted by transition state theory [ 24 , 53 , 63 , 46 ] , Kramers theory [ 45 ] , reactive flux theory [ 37 , 2 , 9 , 18 ] , Grote-Hynes theory [ 27 , 29 ] and by simulations [ 10 , 33 , 11 , 43 ] . As has recently been shown [ 52 ] , τ ∗ \tau_{} can be related to the integral over the squared two-point correlation function of the reaction coordinate. This formulation accounts for memory and inertial effects and serves as the theoretical framework used in this work for comparison with our simulation results. A systematic framework for coarse-graining is provided by the Generalized Langevin Equation (GLE) [ 66 , 54 , 51 , 65 , 50 ] , which reduces the microscopic degrees of freedom of the many-body system to only one or a few relevant degrees of freedom through projection. Starting from the Liouville equation, the projection formalism results in a non-Markovian equation of motion that features a time-dependent friction memory function Γ ⁡ ( t ) \Gamma(t) and an orthogonal force F R ( t ) F_{R}(t) , that is typically modelled as a random force.

A wide range of systems have been successfully modelled using the GLE, such as isomerizing molecules [ 18 , 8 , 22 ] , absorption spectra of fluids [ 14 , 3 ] , protein folding [ 1 , 15 , 6 , 55 , 56 , 21 ] , pair reactions of molecules [ 13 , 19 , 22 ] , cell mobility [ 49 , 41 , 40 ] or meteorological data [ 39 ] . The GLE approach has also been used to analyse non-equilibrium systems outside physics such as financial time series data [ 39 , 28 , 30 ] . Whereas free-energy profiles can often be determined comparatively straightforwardly, extracting friction memory kernels is considerably more challenging. Recently, there has been much progress in the development of numerical tools that can directly extract friction memory kernels Γ ⁡ ( t ) \Gamma(t) from time series data generated from experiments or large-scale simulations [ 60 , 32 , 19 , 31 , 16 , 11 , 42 , 59 , 26 , 34 , 44 , 26 , 6 , 25 ] . The specific functional form of Γ ⁡ ( t ) \Gamma(t) depends on the system and can often be fitted to a sum of monotonically decaying exponential functions. However, in many instances, one also finds oscillating components [ 6 , 13 , 22 , 44 , 48 ] .

For systems with monotonically decaying friction kernels the effect of single-exponential memory on the barrier-crossing kinetics depends on the ratio of the memory decay time τ \tau to the diffusive timescale τ D \tau_{\text{D}} of the system. The barrier-crossing kinetics are accelerated for intermediate τ ≈ τ D

10 [ 27 , 59 ] , and slowed down for long τ > τ D \tau>\tau_{\text{D}} , showing a quadratic increase of the reaction time with τ \tau for large τ \tau [ 35 ] , demonstrating that a simple timescale separation between environment and the barrier-crossing process does not work for long memory times [ 20 ] . While the influence of exponentially decaying memory on barrier-crossing kinetics is well understood from numerical studies for single-exponential [ 27 , 59 , 35 ] , bi-exponential [ 36 ] , tri-exponential [ 47 ] kernels, as well as analytic work [ 27 , 29 , 52 ] , comparatively little is known about the effect of oscillatory memory kernels on barrier-crossing kinetics. Apart from the general Grote–Hynes framework [ 27 ] , dedicated studies of oscillatory memory in barrier-crossing dynamics remain scarce.

In this paper, we systematically investigate how oscillations in the friction memory kernel affect mean barrier-crossing kinetics by studying the GLE in a bistable free-energy potential using Markovian embedding simulations [ 57 ] and analytic theory [ 52 ] . We find new scaling regimes of the reaction kinetics with respect to the timescales of the memory kernel: the mean first-passage times scale quartically with the memory decay time τ o \tau_{\text{o}} and the oscillation period τ φ \tau_{\varphi} of the friction kernel for intermediate values of τ o \tau_{\text{o}} and τ φ \tau_{\varphi} , and scale linearly with τ o \tau_{\text{o}} for long values of τ o \tau_{\text{o}} . Further, we show that the mean first-passage time scales as τ MFP ∼ 1

m for small mass m m and long memory decay time τ o \tau_{\text{o}} and finite τ φ \tau_{\varphi} , in stark contrast to the behaviour for non-oscillating memory kernels. Our analytical approach agrees well with our simulation results.

Non-Markovian model and Markovian-embedding simulation method

We consider the one-dimensional Generalized Langevin equation (GLE) for the reaction coordinate x ⁡ ( t ) x(t)

with x x -independent mass m m , time dependent friction kernel Γ ⁡ ( t ) \Gamma(t) , deterministic force arising from the free energy U ⁡ [ x ⁡ ( t ) ] U[x(t)] and the random force F R ( t ) F_{R}(t) which has zero mean ⟨ F R ( t ) ⟩ = 0 \langle F_{R}(t)\rangle=0 and second moment ⟨ F R ( t ) F R ( t ′ ) ⟩ = k B T Γ ( | t − t ′ | ) \langle F_{R}(t)F_{R}(t^{\prime})\rangle=k_{B}T\Gamma(|t-t^{\prime}|) , thereby relating the random force to the dissipative friction kernel [ 66 ] . We consider an oscillatory-exponential memory kernel

with friction coefficient γ = ∫ 0 ∞ Γ osc ( t ) 𝑑 t \gamma=\int^{\infty}{0}\Gamma{\mathrm{osc}}(t)dt , exponential decay time τ o \tau_{\text{o}} and oscillation period 2 π τ φ 2\pi\ \tau_{\varphi} . The GLE eq.( 2 ) for Γ ( t ) = Γ osc ( t ) \Gamma(t)=\Gamma_{\mathrm{osc}}(t) can be mapped exactly onto a system of coupled Langevin equations

with v ⁡ ( t ) v(t) being the velocity of the reaction coordinate x x , and y ⁡ ( t ) , w ⁡ ( t ) , m y , γ y(t),w(t),m_{y},\gamma being position, velocity, mass and friction of an auxiliary variable that is harmonically coupled to the reaction coordinate x x with coupling constant k y k_{y} and is subject to a Gaussian random force F y ( t ) F_{y}(t) with zero mean and second moment ⟨ F y ( t ) F y ( t ′ ) ⟩ = 2 γ k B T δ ( t − t ′ ) \langle F_{y}(t)F_{y}(t^{\prime})\rangle=2\gamma k_{B}T\delta(t-t^{\prime}) , see SI Sec. 1A for details of the Markovian embedding. The timescales of the kernel in eq.( 3 ) are related to the parameters of the Langevin eqs.( 4 ) by

As reference timescales, we choose

The diffusion time τ D \tau_{\text{D}} is the timescale for a particle to diffuse over the length L L in a heat bath with thermal energy k B T k_{B}T and total friction γ \gamma in the absence of a free energy profile. The inertial time τ m \tau_{\text{m}} marks the crossover between ballistic and diffusive particle motion [ 66 ] . We consider a symmetric double-well potential

with barrier height U 0 U_{0} and well separation 2 L 2L , see Figure 1 a. We quantify the barrier-crossing time by the mean first-passage time τ MFP \tau_{\text{MFP}} (MFPT). The τ MFP \tau_{\text{MFP}} is obtained from the mean of all first-passage events between x = − L x=-L and x = L x=L in a long continuous simulation trajectory, see Figure 1 b; numerical details are provided in SI Sec. 1B–C. This definition of τ MFP \tau_{\text{MFP}} corresponds to the reaction or escape time [ 64 ] , see SI Sec. 2 for details. We simulate the reaction kinetics by integrating eqs.( 4 ) and analyse the barrier-crossing kinetics as a function of the four characteristic timescales { τ D , τ m , τ o , τ φ } {\tau_{\text{D}},\tau_{\text{m}},\tau_{\text{o}},\tau_{\varphi}} of the GLE.

In Figure 1 c–h, we show oscillatory memory kernels for different memory times { τ o

τ D , τ φ

\tau_{\text{D}}} and the corresponding trajectories for τ m

τ D = 0.1 \tau_{\text{m}}

k B T = 3 U_{0}

k_{B}T=3 . We observe frequent barrier-crossing events in Figure 1 f and g, but in Figure 1 h, when both the exponential decay time τ o \tau_{\text{o}} and the oscillation period τ φ \tau_{\varphi} are set equal to the diffusion time, the trajectory exhibits extended phases of rapid state re-crossings [ 64 ] and long phases without transitions, typical for GLE systems in the regime of memory-induced slowdown [ 35 ] . These observations suggest that for GLE systems with oscillatory memory, the slowdown of the reaction kinetics is strongest when both τ o \tau_{\text{o}} and τ φ \tau_{\varphi} are long compared to τ D \tau_{\text{D}} .

Limit of monotonically decaying exponential memory

We start with a discussion of the oscillatory-exponential memory in terms of the parameters of the Markovian embedding eqs.( 4 ), because in this formulation the monotonically decaying exponential memory limit is straightforwardly obtained by considering m y → 0 m_{y}\to 0 . In this limit, the oscillatory-exponential kernel eq.( 3 ) converges to a purely exponentially decaying function

with the asymptotic decay time given by

see Appendix A for details. In Figure 2 a, we show τ MFP

\tau_{\text{D}} as a function of τ a

\tau_{\text{D}} for various ratios m y

m . For small m y

m , we recover the previously studied single-exponential kernel scenario [ 35 ] : for small decay times τ a ≪ τ D \tau_{\text{a}}\ll\tau_{\text{D}} , τ MFP \tau_{\text{MFP}} converges to the Markovian limit; for intermediate τ a

τ D ≈ 0.1 \tau_{\text{a}}

\tau_{\text{D}}\approx 0.1 , the barrier-crossing dynamics is accelerated, as reflected by a decrease in τ MFP \tau_{\text{MFP}} , as shown in the inset of Figure 2 a. For long τ a ≫ τ D \tau_{\text{a}}\gg\tau_{\text{D}} , the barrier-crossing kinetics slows down, as reflected by a quadratic increase of τ MFP \tau_{\text{MFP}} with τ a \tau_{\text{a}} . Increasing the ratio m y

m turns the quadratic into a quartic scaling, see Figure 2 a. We also show τ MFP \tau_{\text{MFP}} for m y

m=0 , which corresponds to a purely exponential kernel, exhibiting perfect agreement with the results for m y

m=0.001 . In Figure 2 b, we s