1. Graph Theory and Complex Systems in Statistical Mechanics
Jo Ellis-Monaghan
website:
5/7/2018

2. Getting by with a little (a lot of!) help from my friends….
Patrick Redmond (SMC 2010)
Eva Ellis-Monaghan (Villanova 2010)
Laura Beaudin (SMC 2006)
Patti Bodkin (SMC 2004)
Whitney Sherman (SMC 2004)
Mary Cox (UVM grad)
Robert Schrock (SUNY Stonybrook)
Greta Pangborn (SMC)
Alan Sokal (NYU)
This work is supported by the Vermont Genetics Network through NIH Grant Number 1 P20 RR16462 from the INBRE program of the National Center for Research Resources.
Isaac Newton Institute for Mathematical Sciences
Cambridge University, UK
5/7/2018

3. Applications of the Potts Model
● Liquid-gas transitions
● Foam behaviors
● Magnetism
● Biological Membranes
● Social Behavior
● Separation in binary alloys
● Spin glasses
● Neural Networks
● Flocking birds
● Beating heart cells
These are all complex systems with nearest neighbor interactions.
These microscale interactions determine the macroscale behaviors of the system, in particular phase transitions.
5/7/2018

4. Ernst Ising
Ising, E. Beitrag zur Theorie des Ferromagnetismus. Zeitschrift fr Physik 31 (1925),
72,500 Articles on ‘Potts Model’ found by Google Scholar
5/7/2018

5. The Ising Model Consider a sheet of metal:
It has the property that at low temperatures it is magnetized, but as the temperature increases, the magnetism “melts away”*.
We would like to model this behavior. We make some simplifying assumptions to do so.
The individual atoms have a “spin”, i.e., they act like little bar magnets, and can either point up (a spin of +1), or down (a spin of –1).
Neighboring atoms with the same spins have an interaction energy, which we will assume is constant.
The atoms are arranged in a regular lattice.
5/7/2018
*Mathematicians should NOT attempt this at home…

6. One possible state of the lattice
A choice of ‘spin’ at each lattice point.
Ising Model has a choice of two possible spins at each point
5/7/2018

7. The Kronecker delta function and the Hamiltonian of a state
Kronecker delta-function is defined as:
The Hamiltonian of a system is the sum of the energies on edges with
endpoints having the same spins.
5/7/2018

8. The energy (Hamiltonian) of the state
Endpoints have the same spins, so δ is 1.
Endpoints have different spins, so δ is 0.
of this system is
A state w with the value of δ marked on each edge.
5/7/2018

9. The Potts Model Now let there be q possible states….
Orthogonal vectors, with δ replaced by dot product
Colorings of the points with q colors
States pertinent to the application
Healthy Sick Necrotic
5/7/2018

10. More states--Same Hamiltonian
The Hamiltonian still measures the overall energy of the a state of a system.
The Hamiltonian of a state of a 4X4 lattice with 3 choices of spins (colors) for each element. 1 1 1 1 1 1 1 1 1 1
(note—qn possible states)
5/7/2018 3

11. Probability of a state
The probability of a particular state S occurring depends on the temperature, T
(or other measure of activity level in the application)
--Boltzmann probability distribution--
The numerator is easy. The denominator, called the
Potts Model Partition Function,
is the interesting (hard) piece.
5/7/2018 4

12. Example
The Potts model partition function of a square lattice with two possible spins
Minimum Energy States
5/7/2018 4

13. Probability of a state occurring depends on the temperature
P(all red, T=0.01) = .50 or 50%
P(all red, T=2.29) = 0.19 or 19%
P(all red, T = 100, 000) = = 1/16
Setting J = k for convenience, so
5/7/2018

14. Effect of Temperature
Consider two different states A and B, with H(A) < H(B). The relative probability of the two states is:
At high temperatures (i.e., for kT much larger than the energy difference |D|), the system becomes equally likely to be in either of the states A or B - that is, randomness and entropy "win". On the other hand, if the energy difference is much larger than kT (very likely at low temperatures), the system is far more likely to be in the lower energy state.
5/7/2018

15. Ising Model at different temperatures
Cold Temperature
Hot Temperature
Here H is
and energy is
Here magnetization is average of the spins, and energy per spin is -1/N * sum( a*b) where a, b are spins at endpoint of edge.
Critical Temperature
Images from
5/7/2018

16. Monte Carlo Simulations?
5/7/2018

17. Monte Carlo Simulations
Generate a random number r between 0 and 1.
B (stay old)
B (old)
A (change to new)
5/7/2018

18. Capture effect of temperature
Given r between 0 and 1, and that , with B the current state and A the one we are considering changing to, we have:
High Temp
Low Temp
H(B) < H(A)
B is a lower energy state than A
exp(‘-’/kT) ~1
exp(‘-’/kT) ~ 0
> r, so change states.
< r, so stay in low energy state.
H(B) > H(A)
B is a higher energy state than A
exp(‘+’/kT) ~1
> r, so change to lower energy state.
5/7/2018

19. Foams “Foams are of practical importance in applications as
diverse as brewing, lubrication,
oil recovery, and fire fighting”.
The energy function is modified by the area of a bubble.
Results:
Larger bubbles flow faster.
There is a critical velocity at which the foam starts to flow uncontrollably
5/7/2018 9

20. A personal favorite
Y. Jiang, J. Glazier, Foam Drainage: Extended Large-Q Potts Model Simulation
We study foam drainage using the large-Q Potts model... profiles of draining beer foams, whipped cream, and egg white ...
Olympic Foam:
5/7/2018

21. Life Sciences Applications
This model was developed to see if tumor growth is influenced by the amount and location of a nutrient.
Energy function is modified by the volume of a cell and the amount of nutrients.
Results:
Tumors grow exponentially in the
beginning.
The tumor migrated toward the
nutrient.
5/7/2018 7

22. Sociological Application
The Potts model may be used to “examine some of the individual incentives, and perceptions of difference, that can lead collectively to segregation …”.
(T. C. Schelling won the 2005 Nobel prize in economics for this work)
Variables:
Preferences of individuals
Size of the neighborhoods
Number of individuals
5/7/2018 8

23. What’s a nice graph theorist doing with all this physics?
If two vertices have different spins, they don’t interact, so there might as well not be an edge between them (so delete it).
If two adjacent vertices have the same spin, they interact with their neighbors in exactly the same way, so they might as well be the same vertex (so contract the edge)*.
*with a weight for the interaction energy e
Delete e
Contract e G
G-e
G/e
5/7/2018

24. Tutte Polynomial (The most famous of all graph polynomials)
Let e be an edge of G that is neither a bridge nor a loop. Then,
And if G consists of i bridges and j loops, then
5/7/2018

25. Example The Tutte polynomial of a cycle on 4 vertices… = + = + + = + +
5/7/2018 13

26. Does the order of contraction/deletion matter?
vs.
The Dichromatic Polynomial:
Can show (by induction on the number of edges) that
TheTutte polynomial is independent of the order in which the edges are deleted and contracted.
k(A) = 2
|A| = 2
5/7/2018

27. Universality THEOREM: (various forms—Brylawsky, Welsh, Oxley, etc.)
If f is a function of graphs such that
f(G) = a f(G-e) + b f(G/e) whenever e is not a loop or bridge, and
b) f(GH) = f(G)f(H) where GH is either the disjoint union of G and H
or where G and H share at most one vertex.
Then,
, where
are the number of edges, vertices, and components of G, respectively,
and where
f ( ) = x0 , and f ( ) = y0.
Thus any graph invariant that reduces with a) and b) is an evaluation of the Tutte polynomial.
Proof: By induction on the number of edges.
5/7/2018

28. The q-state Potts Model Partition Function is an evaluation of the Tutte Polynomial
For a loop, note that there are q states, and since both endpoints of a loop necessarily have the same value, the Hamiltonian of each state is 1. Thus,
For a single bridge, there are q states where the spins on the endpoints are equal, giving a Hamiltonian of 1. Then there are q(q-1) states where the spins on the endpoints are different, giving a Hamiltonian of 0. Thus,
If we let , then:
The Potts Model Partition Function is a polynomial in q!!!
(Fortuin and Kasteleyn, 1972)
5/7/2018

29. Example The Tutte polynomial of a 4-cycle:
Compute Potts model partition function from the Universality Theorem result:
Let q = 2 and
5/7/2018

30. A different point of view
Another reason to believe that
What might we be counting here?
Inspiration: A is the set of edges labeled with 1, so these are exactly the number of edges contributing to the Hamiltonian, so give the exponent of v.
However, this then means that all the vertices in a component of A must have the same spin, so q# components ways to assign the spins (sort of…). 1
5/7/2018

31. Thermodynamic Functions
Important thermodynamic functions such as internal energy, specific heat, entropy, and free energy may all be derived from the Potts Model partition function,
U=Internal Energy (sum of the potential and kinetic energy):
C=Specific heat (energy to raise a unit amount of material one degree):
S=Entropy (a measure of the randomness and disorder in a system):
F=Total free energy:
5/7/2018

32. Phase Transitions
For convenience, work with the dimensionless, reduced free energy:
Reduced free energy per unit volume (or in this context, per vertex):
For a fixed graph G, this is clearly an analytic function in both q and T. Any failures of analyticity can only occur in the infinite volume limit.
5/7/2018

33. The infinite volume limit
Let {G} be an increasing sequence of finite graphs (e.g. lattices).
The (limiting) free energy per unit volume is:
Phase transitions (failure of analyticity) arise in the infinite volume limit.
Energy
Energy
Temp
Temp
5/7/2018

34. The antiferromagnetic model at zero temperature
J is negative in the antiferromagnetic model, so minimal energy states have a maximum number of zeros in the summation, i.e. every edge has endpoints with different spins. If we think of the spins as colors, a minimum energy state is then just a proper coloring of the graph.
Now low energy  lots of edges with different spins on endpoints.
At zero temperature, low energy states prevail, i.e. we really need to consider states where the endpoints on every edge are different.
Such a state corresponds to a proper coloring of a graph. A E D C B
5/7/2018

35. Chromatic polynomial
The Chromatic Polynomial counts the ways to vertex color a graph: C(G, n ) = # proper vertex colorings of G in n colors. + = G
G/e
G-e
Do K3 ->
Recursively: Let e be an edge of G . Then,
5/7/2018

36. Example
n(n-1)2 + + = - =
= n(n-1)2 +n(n-1) = n2 (n-1)
Since a contraction-deletion invariant, the chromatic polynomial
is an evaluation of the Tutte polynomial:
5/7/2018

37. The connection Note: Recall: with
These are the same function precisely when
that is, when
which is exactly the zero-temperature model.
5/7/2018

38. Another connection Consider the summands of as , and hence
A summand is 0 except precisely when
in which case it is 1. Thus
simply counts the number of proper colorings of G
with q colors.
5/7/2018

39. Zeros of the chromatic polynomial
In the infinite volume limit, the ground state entropy per vertex of the Potts antiferromagnetic model becomes:
Thus, phase transitions correspond to the accumulation points of roots of the chromatic polynomial in the infinite volume limit.
5/7/2018

40. Locations of Zeros
Mathematicians originally focused on the real zeros of the chromatic polynomial (the quest for a proof of the 4-color theorem…)
Physicists have changed the focus to the locations of complex zeros, because these can approach the real axis in the infinite limit.
Now an emphasis on ‘clearing’ areas of the complex plane of zeros.
5/7/2018

41. Some zeros
(Robert Shrock)
5/7/2018

42. What happens if…. Interaction energy depends on the edge?
Depends on whether the edge is contracted or deleted?
Depends on whether the edge is a bridge or a loop?
What if no longer multiplicative? (!)
There is an external field?
Fortuin and Kasteleyn, Traldi, Zaslavsky, Bollobas and Riordan, Sokal, E-M and Traldi, E-M and Zaslavsky….
5/7/2018

43. Life gets interesting…
No longer necessarily get a well-defined function.
There are necessary and sufficient conditions on the relations among the edge-weights to guarantee this.
These conditions essentially live on three small graphs:
5/7/2018

44. Can lose mulitiplicativity….
Essential characteristics encoded by contraction/deletion
Still need relations to assure well-defined
BUT…
Retain universality properties and that a function is determined by action on ‘smallest’ objects (all discrete matroids not just a single bridge/loop).
5/7/2018

45. Thank you for your attention!
Questions?
5/7/2018