Datenanalyse und Stochastische Modellierung 6. Active Movement
Reminder: Exponential Decay of Autocorrelations and Brownian Motion
AR(1) \[ x_{t+1} = a x_{t} + \xi_t \]
with autocorrelation function \[ C(\Delta) = e^{-\frac{\Delta}{\tau}} \]
and
Random walk \[ y_{t+1} = y_{t} + \xi_t \]
with MSD \[ \langle y_t^2 \rangle = 2Dt \]
Correlated 'Random' Walks
Look at the cummulative process with correlated increments \[ y_t = \sum_{n=1}^{t} x_n \]
In this case, the MSD for long times scales linearly, but for short times it scales quadratically \[ \langle y^2(t) \rangle = 2D [t-\tau (1-e^{-t/\tau})] \]
The central limit theorem is still valid for long times
A two Dimensionsional model
In this course, we mostly concentrate on one dimension - generalizing to two dimensions is ofter straight forward. In this case, models and applications are usually in two or three dimensions.
For a constant velocity with random angle \[ \dot{\vec{r}} = v \left(\begin{array}{c}\cos\phi\\ \sin\phi\end{array}\right) + \sqrt{2D_T}\vec{\xi}, \;\;\;\;\;\; \dot\phi = \sqrt{2D_R}\xi_\phi \;\; [+w] \]
If w is different from 0, the particle is chiral
The solution (no chirality case) is \[ \langle r_x(t) \rangle = \frac{v}{D_R}[1-\exp(-D_Rt)] \]
The MSD goes to a linear scaling for long times, for short times it is steeper, for very short times again linear \[ \langle r^2(t) \rangle = 4D_Tt + 2\frac{v^2}{D_R^2} (D_Rt + e^{-D_Rt} -1) \]
e.g. Particles with characteristic time for sticking together build clusters
...as do self-propelled particles with a force towards higher density
Zhang et al. Active phase separation by turning towards regions of higher density. Nat. Phys. 17, 961–967 (2021).
The Joseph-Effect
ARFIMA(0,d,0) model with d=J-1/2 \[ x_t=\sum_{k=1}^\infty (-1)^{k+1}\frac{\prod_{a=0}^{k-1}(J-1/2-a)}{k!}x_{t-k}+\xi_t \]
Anomalous diffusive scaling in the cummulative process \[ y_t=\sum_{n=1}^t x_n \;\;\;\;\;\;\;\; \langle y_t^2 \rangle \propto t^{2J} \]
Continuous time version of this process: fractional Brownian motion
Power law decay of increment autocorrelations \[ \langle x(t+\Delta)x(t) \rangle = \sigma^2 \frac{\Gamma(2-2J)}{\Gamma(3/2-J)\Gamma(J-1/2)} \frac{\Gamma(\Delta-1/2+J)}{\Gamma(\Delta+3/2-J)} \]
Diverging autocorrelation time
Look at the autocorrelation function \[ \langle x(t+\Delta)x(t) \rangle = \sigma^2 \frac{\Gamma(2-2J)}{\Gamma(3/2-J)\Gamma(J-1/2)} \frac{\Gamma(\Delta-1/2+J)}{\Gamma(\Delta+3/2-J)} \stackrel{\Delta\rightarrow\infty}{\sim} \frac{\Gamma(\Delta)\Delta^{J-1/2}}{\Gamma(\Delta)\Delta^{3/2-J}}=\Delta^{2J-2} \] with \[ \Gamma(z)=\int_0^\infty t^{z-1}e^{-t} \mathrm{d}t \stackrel{z\in\mathbb{N}}{=} (z-1)! \;\;\mbox{with}\;\; \Gamma(\Delta+a) \stackrel{\Delta\rightarrow\infty}{\sim} \Gamma(\Delta)\Delta^{a} \]
The autocorrelation time is the integral \[ \tau=\int_0^\infty \langle x(t+\Delta)x(t) \rangle \mathrm{d}\Delta \]
It can be found to yield long range anipersistence or long memory \[ \tau = 0 \;\;\mbox{for}\;\; 1/2>J>0 \;\;\;\;\;\; \tau \rightarrow\infty \;\;\mbox{for}\;\; 1>J>1/2 \]
Non-Independent Increments and the Central Limit Theorem
Long range anipersistence or long memory imply non-independence of the increments at all times, so the central limit theorem is not valid, not even in the long time limit
Gaussian increments lead to Gaussian processes
The MSD scales with an anomalous exponent
However, the system is ergodic (time averages equal ensemble averages)
Fractional Brownian Motion as a model for diffusion in crowded environments
e.g. cytoplasms of living cells or artificial solutions; up to 400 mg/ml of macromolecules
In the previous chapters we concentrated on simple forms auf autocorrelations (random walks with white-noise-increments and exponential decay of autocorrelations)
In this chapter we discussed correlated correlated random walks and showed their relevance in soft matter research
While increments of correlated random walks still obey the central limit theorem, power-law decay of autocorrelations violates with some exponent 2J-2 the premise of independent increments
Fractional Brownian Motion (with Fractional Gaussion noise increments (ARFIMA(0,d,0))) is a long-range-correlated random walk - it is also ergodic