1037 words
5 minutes
Statistical Mechanics: The Directed Polymer Problem

This one is from a statistical mechanics problem set (the original PDF is linked at the end of the post). I like it as an example because the “polymer” turns out to be nothing more than a random walk, and its typical size comes out of a partition function trick that is worth knowing.

A directed polymer is made of atoms i=0,1,2,…,Ni = 0, 1, 2, \dots, N sitting at positions (xi,yi)∈Z2(x_i, y_i) \in \mathbb{Z}^2 on a square lattice. The atom at the origin is fixed at x0=y0=0x_0 = y_0 = 0, and the other atoms are chained together such that

  1. xi−xi−1=1x_i - x_{i-1} = 1
  2. ∣yi−yi−1∣=1|y_i - y_{i-1}| = 1

The polymer is therefore oriented along the xx-direction, and it cannot intersect itself.

Setup#

With these two rules we can draw the polymer as a set of points on Z2\mathbb{Z}^2 joined by straight segments.

  • Origin. O=(0,0)O = (0, 0).
  • Steps in xx. Each atom sits exactly one unit to the right of the previous one, xi−xi−1=1x_i - x_{i-1} = 1.
  • Steps in yy. Each step goes either one unit up or one unit down, yi−yi−1=±1y_i - y_{i-1} = \pm 1.
x = 0x = 1x = 2x = 3paths1331O
All 23 = 8 paths up to x = 3. The numbers on the right count how many paths end at each height (1, 3, 3, 1, the binomial coefficients). One path is highlighted.

So at every step to the right, the polymer has two options in yy (up or down), and for a polymer of length NN this gives

(2 choices after origin)×(2 choices after node 1)×⋯=2N total choices(2 \text{ choices after origin}) \times (2 \text{ choices after node 1}) \times \dots = 2^N \text{ total choices}

Counting all microstates#

The first question is how many microstates the polymer has in total. Put a particle on every node. The first one is fixed at the origin, and every following particle has 2 choices (up or down) relative to the one before it,

Particle1 (Origin)23…N+1
Choices122…2
Node (x)012…N

so the total number of microstates W\mathcal{W} is

W=1⏟Fixed×2×2×⋯×2⏟N=2N\mathcal{W} = \underbrace{1}_{\text{Fixed}} \times \underbrace{2 \times 2 \times \dots \times 2}_{N} = 2^N

Microstates with a given deflection#

Next, how many microstates W(y)\mathcal{W}(y) end at the height yN=yy_N = y?

Let σi=yi−yi−1∈{+1,−1}\sigma_i = y_i - y_{i-1} \in \{+1, -1\} be the sign of step ii, and let n+n_+ and n−n_- count the steps up and down. The total deflection is then

yN=∑i=1Nσi=n+−n−=yy_N = \sum_{i=1}^N \sigma_i = n_+ - n_- = y

and the steps add up to NN,

n++n−=N  ⟹  n−=N−n+n_+ + n_- = N \implies n_- = N - n_+

Putting the second relation into the first, we can solve for the number of up steps,

y=n+−(N−n+)=2n+−Ny = n_+ - (N - n_+) = 2n_+ - N  ⟹  n+=y+N2\implies n_+ = \frac{y + N}{2}

So fixing the end height yy is the same as fixing n+n_+, and the number of ways to choose which n+n_+ of the NN steps go up is a binomial coefficient,

W(y)=(Nn+)=(Ny+N2)\mathcal{W}(y) = \binom{N}{n_+} = \binom{N}{\frac{y+N}{2}}

The probability of ending at height yy is then P(y)=W(y)W=12N(Nn+)P(y) = \frac{\mathcal{W}(y)}{\mathcal{W}} = \frac{1}{2^N} \binom{N}{n_+}.


Typical deflection#

Finally, how far does the end of the chain typically wander from the axis, i.e. what is ⟨yN2⟩\langle y_N^2 \rangle? One could do the sum over binomial coefficients by brute force, but there is a cleaner route.

The partition function trick#

We define a “partition function” Z(β)Z(\beta) for the ensemble (the name is borrowed, β\beta here is just a bookkeeping parameter),

Z(β)=∑yeβyW(y)Z(\beta) = \sum_y e^{\beta y} \mathcal{W}(y)

At β=0\beta = 0 it simply counts all states,

Z(0)=∑yW(y)=2NZ(0) = \sum_y \mathcal{W}(y) = 2^N

and each derivative with respect to β\beta brings down one power of yy. Two derivatives at β=0\beta = 0 therefore give the mean square deflection,

⟨yN2⟩=∑yy2W(y)∑yW(y)=1Z(0)∂2Z∂β2∣β=0\langle y_N^2 \rangle = \frac{\sum_y y^2 \mathcal{W}(y)}{\sum_y \mathcal{W}(y)} = \left. \frac{1}{Z(0)} \frac{\partial^2 Z}{\partial \beta^2} \right|_{\beta=0}

Calculating Z(β)Z(\beta)#

We write y=2n+−Ny = 2n_+ - N and turn the sum over yy into a sum over n+n_+,

Z(β)=∑n+=0Neβ(2n+−N)(Nn+)=∑n+=0N(eβ)n+(e−β)N−n+(Nn+)Z(\beta) = \sum_{n_+=0}^N e^{\beta(2n_+ - N)} \binom{N}{n_+} = \sum_{n_+=0}^N (e^\beta)^{n_+} (e^{-\beta})^{N-n_+} \binom{N}{n_+}

This is exactly the binomial expansion (a+b)N=∑(Nk)akbN−k(a+b)^N = \sum \binom{N}{k} a^k b^{N-k} with a=eβa=e^\beta and b=e−βb=e^{-\beta}, so

Z(β)=(eβ+e−β)N=(2cosh⁡β)N=2Ncosh⁡NβZ(\beta) = (e^\beta + e^{-\beta})^N = (2 \cosh \beta)^N = 2^N \cosh^N \beta

Finding the deflection#

The first derivative is

Z′(β)=2N⋅Ncosh⁡N−1β⋅sinh⁡βZ'(\beta) = 2^N \cdot N \cosh^{N-1} \beta \cdot \sinh \beta

and the second derivative is

Z′′(β)=2NN[(N−1)cosh⁡N−2βsinh⁡2β+cosh⁡N−1βcosh⁡β]Z''(\beta) = 2^N N \left[ (N-1) \cosh^{N-2} \beta \sinh^2 \beta + \cosh^{N-1} \beta \cosh \beta \right]Z′′(0)=2NN[(N−1)(1)(0)+(1)(1)]=2N⋅NZ''(0) = 2^N N [ (N-1)(1)(0) + (1)(1) ] = 2^N \cdot N

where at β=0\beta = 0 we used sinh⁡0=0\sinh 0 = 0 and cosh⁡0=1\cosh 0 = 1. Dividing by Z(0)Z(0),

⟨yN2⟩=Z′′(0)Z(0)=2N⋅N2N=N\langle y_N^2 \rangle = \frac{Z''(0)}{Z(0)} = \frac{2^N \cdot N}{2^N} = N

This is what one should expect. The directed polymer is a random walk of NN steps of size ±1\pm 1, and for a random walk the variance (the typical deflection squared) grows linearly with the number of steps NN. So the end of the chain typically sits about N\sqrt{N} away from the xx-axis.


TIP

Full Reference: You can download the original problem set with full diagrams here: Download Directed Polymer PDF

Statistical Mechanics: The Directed Polymer Problem
https://rohankulkarni.me/posts/notes/directed-polymer/
Author
Rohan Kulkarni
Published at
2020-12-23
License
CC BY-NC-SA 4.0
Share Post
Comments