One-dimensional diffusion on a finite region

Markus Deserno
Department of Chemistry and Biochemistry, UCLA, USA
(Dated: December 10, 2004)
We study the problem of simple diffusion of one particle on a finite line with reflecting boundaries.
Probability density and variance are expressed as an expansion in eigenmodes of the Fokker-Planck
operator for a particle which starts to diffuse in the middle of the line. We find the well known
short-time behavior σ 2 (t) = 2Dt, but the full solution also yields precise asymptotics for long times.


I. THE DIFFUSION EQUATION


We want to solve the diffusion equation
∂ ∂2
P (x, t) = D 2 P (x, t), (1)  
∂t ∂x
 
increasing
where P (x, t) is the probability density of finding a parti-   time

cle at position x at time t and D is the diffusion constant.
The factorization ansatz

P (x, t) = ϕ(x) e−λt (2)
leads to the eigenvalue equation 
· 2
d
¸
λ           
+ ω 2
ϕ(x) = 0 with ω2 = . (3) 

dx2 D
We want to find a solution of the diffusion problem on FIG. 1: Time evolution of the probability density P (x, t) from
the line [−L/2; L/2]. The eigenfunctions of the harmonic Eqn. (10) for the time steps Dt/L2 = 0.0001, 0.0002, 0.0005,
oscillator type equation (3) are sine and cosine functions. 0.001, . . ., 1.0.
In the present case the eigenfunctions can additionally be
classified by the symmetry of the mode:
II. START IN THE MIDDLE
ϕo,n (x) = sin(ωo,n x) , ϕe,n (x) = cos(ωe,n x) (4)
The eigenvalues ωo,n and ωe,n follow from the boundary A. Probability density
conditions. We will assume reflecting boundaries and
hence set the probability current at ±L/2 to zero [1]: We now want to find the particular solution for the
∂ symmetric initial condition P (x, 0) = δ(x), i. e., a par-
!
0 = −D P (x, t) ⇒ ϕ0 (±L/2) = 0. (5) ticle starting to diffuse in the middle of the line. Since
∂x δ(x) is an even “function”, all Ao,n must vanish. Fur-
This determines the eigenvalues thermore, on the interval [−L/2; L/2] the delta function
can be represented as
2π(n + 1/2) 2πn
ωo,n = , ωe,n = . (6)
L L ∞
X 2πnx
From there we get λn = Dωn2
for odd and even modes. L δ(x) = 1 + 2 cos . (9)
n=1
L
We now can write down the spatial part of the solution:
X∞ ·
2πnx 2π(n + 1/2)x
¸ The right hand side is actually periodic with period L,
ϕ(x) = Ae,n cos + Ao,n sin . but this does not matter since we are only interested
L L
n=0 in its values within [−L/2; L/2]. Comparing coefficients
(7) with Eqn. (7) shows that Ae,n = 2/L for n > 0. The full
The normalization condition time-dependent solution of the diffusion problem is thus
Z L/2 · ¸
! 2πnx 2π(n + 1/2)x ∞
1 = dx Ae,n cos + Ao,n sin 1 2 X 2πnx n 4π 2 n2 Dt o
−L/2 L L P (x, t) = + cos exp − . (10)
L L n=1 L L2
= Ae,0 L (8)
fixes Ae,0 = 1/L, but leaves all other amplitudes unspec- An illustration of the time evolution of this probability
ified. density is given in Fig. 1.
 2

< *'
To get the asymptotic behavior at small t is a little
' bit more tricky: A naive expansion of the exponential
A' ( exp{−4π 2 n2 Dt/L2 } for small t is barred by the fact that
n2 will always become large during the course of perform-
ing the sum [3]. Instead, we will use the Euler-Maclaurin
6 summation formula [2], which is a controlled way for re-
9: ; 
< > placing a sum by an integral. It reads
67 8 5
+-,
. .
 . ,  / n n
 
!)#%*&' 1£
X Z
¤
 "!$#%
& '&( f (k) = dk f (k) − f (0) + f (n)


  
 0 2


  
 k=1
1£ 0 ¤ 1 £ 000 ¤
+ f (n) − f 0 (0) − f (n) − f 000 (0) ± · · · (13)

<  12 720
 *< 
' *< ( *< @ *< 3? *< > *< =
0  1324
Using this, we can rewrite the sum entering the expres-
sion of P (x, t) as
FIG. 2: Variance σ 2 (t) from Eqn. (11) and some of its asymp-
totics and approximations.
∞ n 4π 2 n2 Dt o
X 2πnx
cos exp −
L L2
B. Variance n=1
Z ∞ n 4π 2 n2 Dt o
2πnx 1
R πn = dn cos exp − −
Using the integral −πn dy y 2 cos(y) = 4πn(−1)n for 0 L L2 2
n ∈ N0 , we can compute the time-dependent variance of exp{−x2 /4Dt} 1
the distribution function P (x, t) from Eqn. (10): = p − , (14)
4 πDt/L2 2
L/2
σ 2 (t) 1
Z
= 2 dx x2 P (x, t)
L 2 L −L/2 from which by inserting into Eqn. (10) we get
∞
1 1 X (−1)n n 4π 2 n2 Dt o
= + 2 exp − . (11) Dt¿L2 1 n x2 o
12 π n=1 n2 L2 P (x, t) = p exp − . (15)
2π(2Dt) 2(2Dt)
This can be viewed as an expansion for large times, but
it is valid for all t. The lowest nontrivial order is This is obviously a Gaussian with variance σ 2 (t) = 2Dt.
σ 2 (t) 1 1 n 4π 2 Dt o A plot of the variance σ 2 (t) and a few of its asymptotics
= − exp − . (12) and approximations is shown in Fig. 2.
L2 12 π 2 L2

[1] H. Risken, “The Fokker-Planck-Equation”, 2nd ed., [3] The only exception
P∞ is t n= 0, in which case the summa-
2 2
Springer, Berlin (1996). tion formula n=1 (−1) /n = −π /12 correctly results
[2] M. Abramowitz and I. A. Stegun (ed.), Handbook of math- in σ 2 (0) = 0.
ematical functions, 9th printing, Dover, New York (1970).