Arxiv:1102.0524V1 [Physics.Data-An] 2 Feb 2011

Harmonic Oscillator in Heat Bath: Exact simulation of time-lapse-recorded data, exact analytical benchmark statistics Simon F. Nørrelykke Department of Molecular Biology, Princeton University, Princeton, New Jersey, USA. and Henrik Flyvbjerg Department of Micro- and Nanotechnology, Technical University of Denmark, Kongens Lyngby, Denmark. (Dated: May 29, 2018) The stochastic dynamics of the damped harmonic oscillator in a heat bath is simulated with an algorithm that is exact for time steps of arbitrary size. Exact analytical results are given for correlation functions and power spectra in the form they acquire when computed from experimental time-lapse recordings. Three applications are discussed: (i) Effects of finite sampling-rate and -time, described exactly here, are similar for other stochastic dynamical systems|e.g. motile micro-organisms and their time-lapse recorded trajectories. (ii) The same statistics is satisfied by any experimental system to the extent it is interpreted as a damped harmonic oscillator at finite temperature|such as an AFM cantilever. (iii) Three other models of fundamental interest are limiting cases of the damped harmonic oscillator at finite temperature; it consequently bridges their differences and describes effects of finite sampling rate and sampling time for these models as well. Finally, we give a brief discussion of nondimensionalization. Keywords: Ornstein-Uhlenbeck process, power-spectral analysis, mean-squared displacement, discrete sampling, discretization effects, Langevin, Brownian motion, optical trap, leakage, non-dimensionalization, correlation function I. INTRODUCTION dom motion, e.g. cell migration [8{10, 12{14, 25, 26] and commodity pricing [24]. Consequently, the treatment The damped harmonic oscillator in a heat bath is given here of its experimental statistics for time-lapse the archetypical bounded Brownian dynamical system recorded data should be useful in several ways: with inertia and the simplest possible of this kind. An- (i) All experimental statistics contain effects of finite alytically solvable, it offers insights that are valid also sampling rate, finite sampling time, and finite statis- for more complex systems. Its experimental correlation tics. They do so to various degrees, but the effects are functions and power spectra are given analytically here inherent in measurements. We give exact analytical ex- in the form they take when computed from time-lapse pressions for these effects, for the model treated here. recorded trajectories. Such trajectories are generated Statistics for more complex systems contain the same and analyzed for illustration in an exact Monte Carlo effects, qualitatively, and quantitatively as well, by de- simulation. The results differ significantly from those grees we estimate. derived from the standard continuous-sampling formu- (ii) The statistics we describe below must be found lation. for any experimental system to the extent that system is interpreted as a damped harmonic oscillator at finite In mathematical terms, it is the Ornstein-Uhlenbeck temperature|e.g. an AFM cantilever to be calibrated (OU) process in a Hookean force field we treat, and the by interpretation of its thermal power spectrum. results discussed here extend the ones given in [19, 23] as (iii) Three other models are limiting cases of the well as the free case treated by Gillespie [11]. Three clas- model treated here: sical papers on the OU-process are [7, 29, 30]. Histori- cally, the model was prompted by a question by Smolu- 1. At vanishing mass, Einstein's theory for Brownian chowski regarding how inertia might modify Einstein's motion in a harmonic trap, which, e.g., is a min- theory of Brownian motion [20]. H. A. Lorentz soon af- imal model for the Brownian dynamics of a mi- arXiv:1102.0524v1 [physics.data-an] 2 Feb 2011 ter pointed out that Ornstein's answer to that question crosphere held in an optical trap, with magnetic was insufficient for the case of classical Brownian mo- tweezers [28], or surface-tethered by DNA [1]. tion, i.e., in a fluid of density similar to the Brownian particle [16]: Hydrodynamical effects, entrainment and 2. At vanishing external force, the Ornstein- back-flow, were more important with the time resolu- Uhlenbeck model of free Brownian motion with tion that was available in the first century of Brownian inertia, which, e.g., is a minimal model for the motion. Inertia in classical Brownian motion became persistent random motion seen in trajectories of experimentally relevant only with precision calibration motile cells. of optical tweezers [2, 3] and was directly observed only 3. At vanishing mass and external force, Einstein's in 2005 [17, 27]. original theory for Brownian motion in a fluid at The OU process remained and remains, however, an rest. important model for Brownian motion dominated by inertia, such as massive particles [5, 6, 15] or AFM can- The results for the harmonic oscillator, given below, tilevers [22] in air, and for other kinds of persistent ran- carry over to these three models. The limits are not 2 all obvious, but always enlightening, hence described which, for arbitrary positive ∆t, and with tj = j∆t, below. xj = x(tj), and vj = v(tj), gives us the recursive relation (iv) As the model treated here bridges the differences between the three limiting cases, the material pre- x x ∆x j+1 = e−M∆t j + j ; (7) sented here is well suited for a pedagogical, hands-on vj+1 vj ∆vj computer-based introduction to the four dynamic systems covered here: free/bound diffusion, with/without where inertia, their equilibrium behavior, correlations, and p power spectra, and their transient behavior to equilib- tj +∆t 2D Z 0 ∆xj 0 −M(tj +∆t−t ) 0 rium. = dt e 0 ∆vj η(t ) τ tj (8) and the time-independent matrix exponential can be II. EXACT DISCRETIZED written EINSTEIN-ORNSTEIN-UHLENBECK THEORY −M∆t − ∆t OF BROWNIAN MOTION IN HOOKEAN e = e 2τ [cos(!∆t)I + sin(!∆t)J] (9) FORCE FIELD with The Einstein-Ornstein-Uhlenbeck theory for the 1 1 ! Brownian motion of a damped harmonic oscillator in one 1 0 2!τ ! I ≡ and J ≡ −!2 : (10) dimension is simply Newton's Second Law for the oscil- 0 1 0 −1 lator with a thermal driving force, a.k.a. the Langevin ! 2!τ equation for this system, That the matrix exponential can be written this way, mx¨(t) + γx_(t) + κx(t) = Ftherm(t) (1) can be proven in several ways. We used the algebra of Pauli matrices. Alternatively, one may observe that F (t) = (2k T γ)1=2η(t) : (2) therm B the cosine and sine terms on the RHS are the even and Here x(t) is the coordinate of the oscillator as function odd parts of the LHS. The latter are straight-forwardly, of time t, m its inertial mass, γ its friction coefficient, κ if tediously, computed from the Taylor series for the exponential. Inserting Eq. (9) in Eq. (8) we see that is Hooke's constant, and Ftherm is the thermal force on the oscillator. Equation (2) gives the amplitude of this p Z tj+1 t −t thermal noise explicitly in terms of γ, the Boltzmann 2D − j+1 ∆x = dt e 2τ sin(!(t − t)) η(t) (11) energy k T , and η(t), which is a normalized white-noise j j+1 B !τ tj process, i.e., the time derivative of a Wiener process, η = dW=dt, hence p tj+1 2D Z tj+1−t 0 0 − 2τ hη(t)i = 0 ; hη(t)η(t)i = δ(t − t ) for all t; t : (3) ∆vj = − 2 dt e sin(!(tj+1 − t)) η(t) 2!τ tj Equation (1) can be rewritten as two coupled first-order (12) p differential equations, Z tj+1 t −t 2D − j+1 + dt e 2τ cos(!(t − t)) η(t) p τ j+1 d x(t) x(t) 2D 0 tj = −M + ; (4) dt v(t) v(t) τ η(t) are two correlated random numbers from zero-mean where we have introduced Einstein's relation D = Gaussian distributions. They can be written as a lin- k T/γ and the 2 × 2 matrix ear combination of two independent Gaussian variables: B The four parameters that determine this linear combi- 0 −1 0 −1 nation can be chosen at will, as long as the combination M = = : (5) κ γ !2 1 has the same variance-covariance as the two original cor- m m 0 τ related variables. This is a direct consequence of the Gaussian distribution being completely determined by Here ! = pκ/m is the cyclic frequency of the un- 0 its mean and variance-covariance. Thus, we can write damped oscillator, and τ = m/γ is the characteris- tic time of the exponential decrease with time that the momentum of the particle undergoes in the ab- ∆xj = σxx ξj (13) 2 p 2 4 2 sence of all but friction forces. Below, we shall also ∆vj = σxv/σxx ξj + σvv − σxv/σxx ζj ; (14) need the cyclic frequency of the damped oscillator, p 2 2 p 2 2 ! = κ/m − γ =(4m ) = !0 − 1=(4τ ), which is real where the σs are elements of the variance-covariance ma- for less than critical damping, γ2 < 4mκ. trix (see below), and ξ and ζ are independent random Equation (4) is solved by numbers with Gaussian distribution, unit variance, and p zero mean. This particular choice of linear combina- Z t x(t) 2D 0 −M(t−t0) 0 tion mirrors the structure of Eqs. (11) and (12). Using = dt e 0 ; (6) v(t) τ −∞ η(t ) Eqs. (3), (11), and (12) we calculate that the elements 3 of the variance-covariance matrix are, for ! 6= 0 2 1 2 2 D 2 2 σxx ≡ h(∆xj) i = 2 2 3 4! τ + (15) 0 4! !0τ −1 − ∆t 2 2 τ Positon (nm) e [cos(2!∆t) − 2!τ sin(2!∆t) − 4!0τ ] 6 −2 15 16 17 18 19 20 Time (ms) D σ2 ≡ h(∆v )2i = 4!2τ 2+ (16) vv j 4!2τ 3 4 ∆t − τ 2 2 e [cos(2!∆t) + 2!τ sin(2!∆t) − 4!0τ ] Positon (nm) 2 D − ∆t 2 2 σ ≡ h∆x ∆v i = e τ sin (!∆t) : (17) xv j j !2τ 2 At critical damping (! = 0) this variance-covariance 0 simplifies to 2 2 σxx ≡ h(∆xj) i (18) −2 − ∆t 1 2 0 20 40 60 80 0 500 1000 = 4Dτ 1 − e τ [1 + ∆t/τ + (∆t/τ) ] 2 Time (ms) Count 2 2 FIG.

Load more