Statistical Mechanics for Applications

Midterm, due Monday, October 28 at 11:00am

Numerical problems can be done in any language of your choosing.

Do not turn in your code.

Figures should be legible with appropriate labels.

========================

1. In class, we dealt with two- and three-dimensional hard spheres. Here,

we consider putting one-dimensional disks along a line. In fact, as they are
only one-dimensional disks, it will be simpler to think of them as one-
dimensional bars. Each bar is of length 2(σ + ), where  is some small
additional length so that each bar is a bit longer than 2σ . You could say
that the bar sticks out at both ends by some amount .
Two bars are to placed uniformly on the one-dimensional lattice of points
x/σ = 1, 2, 3, 4, 5. Of course, for a legal (= valid) arrangement, the bars can
not overlap. (BTW, there are no other obstacles. For example, there is no
wall at x=0 or x = 6.)
Slightly changing Sethna's wording in his Problem 1.1 so as to t this
problem, The probability of nding a particular legal sequence of bars is
larger by a constant factor due to discarding the illegal sequences. This
factor is just one over the probability of a given arrangement of the bars
being legal, Z = ∑ α Pα summed over legal sequences α.

(a) Determine the probability Pα of any one arrangement of the bars, legal
or illegal.

(b) Determine ...the probability of a given arrangement of the bars being

legal, Z = ∑α Pα summed over legal sequences α.

(c) Use the results of parts (a) and (b) to nd the probability of a legal
sequence.

1
2. Using Direct Sampling, write a program that places N disks of diam-
eter2σ (or equivalently, bars of length 2σ ) on a line of length L, namely
0 ≤ x ≤ L. The line is terminated at both ends by rigid boundaries. So,
a legal conguration of bars can not overlap the boundaries. Nor, of course,
can a legal conguration contain any overlapping bars.
Take L=2 and σ = 0.05.

(a) For at least several values of N, plot the acceptance ratio as a func-
tion of N. The acceptance ratio is the number of legal congurations
divided by the number of attempts. Choose values of N that are as
large as appear feasible.

(b) Also plot the logarithm of the acceptance ratio as a function of N.

This may be a better plot to use in answering the next question.

(c) If for N =6 it takes your old computer 1 second to generate 104 legal
congurations, estimate,from your data, the time that it would take to
generate 104 legal congurations for N = 15 and N = 19.

(d) (i) Write the acceptance ratio in terms of the phase space volume of
accepted congurations and the phase space volume of all attempts.
(ii) What is the relationship between the acceptance ratio and the
partition function for the bars?

Seeing how unearthly long it takes to place bars densely on the interval,
we are prepared to consider another method which, remarkably, has an ac-
ceptance rate of 100%! (Which, unfortunately, only works for some one-
dimensional systems.) Let's rst reiterate the variables. Let L be the length
of the interval bounded at x = 0 and x = L by rigid walls. Let σ be the radius
of the N disks. Here are the instructions for writing this code. Do not place
the disks directly onto the interval of length L. Take a shortened interval of
length L − 2N σ and place N points on this interval drawn from a uniform
distribution. Let the positions of these points be called yi for i = 1, ..., N .
Now sort the N points into order from the smallest value to the largest.
Relabel these points so the y1 is the smallest value (i.e. the furthest left), y2
the next largest, etc. up to the largest value yN (i.e. the one closest to the
right-hand boundary). Now, let's put these points onto the actual 0≤x≤L
interval, by setting their actual xi locations according to xi = yi + (2 i − 1)σ
for i = 1, ..., N .
The run time for this program should be independent of the size of the
bars. For the questions below, L=2 and σ = 0.05.

2
(e) Plot the histogram of the probability distribution function for N = 15.
(Make sure that you plot a PDF, not counts or probability or something
not the PDF.)

(f ) You will nd that the histogram in part (e) is not uniform across x. In
a paragraph or so, explain why you think that while the locations were
chosen uniformly, the distribution of disks on the line is not uniform.

(g) Plot a second histogram but now the PDF for N = 20.

(h) How many dierent congurations are there for N = 20?

(i) The area of the histogram under a PDF should integrate to unity. Show
that this is the case for the histogram for N = 20.

3
3. From your homeworks, you know how to estimate the area of the unit
circle by sampling. Here, we'll assume that we already know that the area
of the unit circle is π. And, we'd like to nd an analytical expression for
the area in terms of the number NC of points that fall within the circle and
number NS outside of the circle.
Stated with more detail... Consider that you know that the area of the
unit circle is π. N
Find an analytic expression for the probability that of
attempts chosen uniformly within the square of side 2, NC of them will fall
within the inscribed unit circle of area π , where NC + NS = N , and NS is the
number of points outside of the circle.
Hint: Find an analytical expression for the probability that N coins,
where pH is the probability of heads and pT is the probability of tails, will
result in NH heads, where NH + NT = N , and NT is the number of tails.

4
4. Recall the homework problembased on the journal article of Kellerof
predicting the heads or tails outcome of a coin toss. You found that as the
value of the rotation rate increased, the initial conditions would need to be
specied to an increasing accuracy in order to predict the outcome. We
can imagine that at some value of the rotation rate, it would be practically
impossible to render the initial conditions to the desired accuracy for a de-
terministic outcome. And, so, the outcome would then only be predictable
as a probability of heads or tails. Consequently and surprisingly, then, a
deterministic systemdue to sensitive dependence on initial conditionsmay
have only a probabilistic outcome.
As another example, consider a gas of hard spheres. (Let's think of
the low density limit as is appropriate for a model of an ideal gas.) It
was mentioned in class that due to the collisions with the convex shape
of hard spheres, the trajectory of the spheres deviates increasingly from
the trajectory that would be followed if all positions and speeds could be
computed with innite accuracy. Here, as with the coin example above,
we have a deterministic system that realistically has only a probabilistic
outcome of positions and velocities.
Let's try to model the amplication of error that takes place through
multiple collisions. We'll assume that the size of the error after each collision
isr/n, where r is a characteristic error size which gets spread over n collisions.
To say it otherwise, r/n is the error per collision. After M collisions, the
M
error has been amplied by (1 + r/n) . For our calculations, let's assume
that the time between collisions is some constant interval.

(a) Fit the curve (for n = 10) of error versus number of collisions to an
αt
exponential e , where t is time and α is the tting parameters. Plot
both curves.

(b) (i) Write the exponent αt in terms of r, n and M. (ii) What is α in

terms of these variables? (Strictly speaking, the exponent is convergent
to (1 + r/n)M only in the limit n → ∞.)

(c) You recall that (Nk ) ≡ N !/ [(N − k)!k!] generates the binomial coe-
N
cients. What expression using ( ) ≡ N !/ [(N − k)!k!] generates the
k
same curve(s) as you found in part (a)? Hint: Write out the product
(1 + r/n)M for, say, M = 0, 1, 2, 3. (You might nd it more revealing
to replace r/n with x.) These expressions are, of course, polynomials
in r/n (or, more simply, in x). The coecients of the polynomials are
the binomial coecients.

5
(d) Plot this expression that you found in (c) on the same gure as used
above, which now will have three curves (using n = 10): (i) a curve
based on (1 + r/n)M , (ii) a curve t to an exponential, and (iii) a
N
curve based on ( ).
k