Part 7 Mean Field Theory

Part VII: More is the Same
Issues Abstract phases of matter phase diagram water basic physics in phase transitions Gibbs denition; Ehrenfest classication Magnetic Phase Diagram a qualitative change in Behavior from Gibbs to singularities Mean Field Theory more is the same for Ising model Mean Field Theory is Only Partially Right Calculate results simplied phase diagram Graph of Order Parameter behavior in neighborhood of critical point go after magnetization go after energy go after correlations critical opalescence basic equation correlation function susceptability
Perimeter Institute Statistical Physics Lecture Notes part 7. Mean Field Theory Version 1,7 9/11/09 Leo Kadanoff 1
Mean eld theory of uids from van der Waals to Weiss van der Waals rst effect van der Waals second effect Many Different Phase Transitions After van der Waals Landau Mean Field Theory order parameter generalized generalized mean eld scheme recent example why minimize very M to vary F h=0, no space variation jump like square root of Tc-T Summary order parameter and free energy were basic Look ahead: a worry
Perimeter Institute Statistical Physics Lecture Notes part 7. Mean Field Theory Version 1,7 9/11/09 Leo Kadanoff
More is the Same Phase Transitions and Mean Field Theories

Leo P. Kadanoff email:leop@UChicago.edu Abstract This talk summarizes concepts derived from the study of phase transitions mostly within condensed matter physics. In its original form, the talk was aimed equally at condensed matter physicists and philosophers of science. The latter group are particularly interested in the logical structure of science. This talk bears some traces of its history. The key technical ideas go under the names of ``singularity'', ``order parameter'', ``mean eld theory'', ``variational method'', and ``correlation length''. The key ideas here go under the names of mean eld theory, phase transitions, universality, variational method, and scaling.
Issues
Matter exists in different phases, different states of matter with qualitatively different properties: These phases are interesting in modern physics and provocative to modern philosophy. For example, no phase transition can ever occur in a nite system. Thus, in some sense phase transitions are not products of the nite world but of the human imagination.
Phase Diagram for Water
liquid-gas phase transition line
Phase Transitions Describe Basic Physics

Ideas derived from phase transitions and other intellectual products of condensed matter physics are crucial for many branches of modern science and for modern theoretical physics. The basic ideas include: a. Phase transitions always require an innite number of degrees of freedom. This innity may come either from something that happens over an innite period of time, over an innite amount of space, or via the development of some sort of innite complexity within the system. b. The structure of space and time are important determinants of what we see happen. The dimensions of space matter as do whether the system in question repeats itself innitely often. The topology of the surrounding space and of created structure are quite important. c. Condensed matter systems provide a good area for study because they provide an observable platform for amazing diverse and rich phenomena, well beyond the untutored imagination of scientists.
Gibbs: A phase transition is a singularity in thermodynamic behavior.

This occurs only in an innite system Ehrenfest: First order= discontinuous jump in thermodynamic quantities. Second order has continuous thermodynamic quantities, but innity in derivative of thermodynamic quantities. ....... P. Ehrenfest
Magnetic Phase Diagram
A Phase Transition is a change from one behavior to another

A rst order phase transition involves a discontinuous jump in some statistical variable. The discontinuous property is called the order parameter. Each phase transition has its own order parameter. The possible order parameters range over a tremendous variety of physical properties. These properties include the density of a liquid-gas transition, the magnetization in a ferromagnet, the size of a connected cluster in a percolation transition, and a condensate wave function in a superuid or superconductor. A continuous transition occurs when the discontinuity in the jump approaches zero. This section is about the development of mean eld theory as a basis for a partial understanding of phase transition phenomena.
http://blogs.trb.com/news/local/ longisland/politics/blog/2008/04/
http://azahar.les.wordpress.com/2008/12/ 9
theory starts from: phase transitions are singularities in free energy
J. Willard Gibbs:
gets to: phase transitions are a property of innite systems proof: .... consider Ising model for example
problem dened by free energy dened by
H/(kT ) = K
nn
r s + h
r
F/(kT ) = ln
{r =1}
exp H{r }/(kT )
H is a smooth function of K and h. Since a nite sum of exponentials of smooth functions is a positive smooth function, it follows that the free energy is smooth too.
Mean Field Theory: more is the same

one spin
Ising model, spin, simplied = 1 atom one spin in a magnetic eld H = B = kTh statistical average: < >= tanh(h )
many spins
spin in a magnetic eld, dimension d H / kT = K r s + h r
nn r
focus on one spin, at r: that spin feels h and K s nn to rs
11
more is the same

one spin
statistical average: <>= tanh h
many spins
focus on one spin statistical average: Heff / (kT ) = r [hr + K < s >]
s
heff = [h + Kz < >] < >= tanh(heff )
z=number of nn
or, if there is space variation, heff = hr +K s nn to r <s>

Mean Field Theory is Only Partially Right

Mean eld theory says that spin moves in the average eld produced by all other spins. But actual value is often larger in magnitude than mean value and uctuates in sign. Net result is error, with unknown sign. The same ideas can be applied to lots of problems. (In particle physics mean eld theory often goes with the words one loop approximation.) As we shall discuss in detail, mean eld theory gives an interesting and instructive theory of phase transitions, but one which is only partially right. Near the critical point, for lower dimensional systems, including three dimensions, uctuations dominate the system behavior and mean eld theory gives the wrong answer, badly wrong. Very near rst order phase transitions, uctuations also count, but in a less obvious manner. However in high dimensions, usually above four, mean eld theory gives a good picture of phase transitions. It also has features which point the way toward the right theory. It is also simple to use
Calculate results of mean eld theory

Numerical method: I calculate <> at high temperatures, small K, by using Newtons method starting from <>=h. I then increase K step by step at xed h, and nd <> in each step by using the last step as the starting point for Newtons method. I then calculate - F using the fact that h(- F) = <> and the known value of the free energy at high temperatures. One of the useful results that emerges from this numerical calculation is the value of the coupling K which produces the rst splitting in the two h=0 magnetization curves. This bifurcation occurs at Kz=1. Consequently we identify this value of K as the critical one. Since K is a physical coupling strength divided by temperature, we can write Kx in terms of the temperature and the critical temperature as Kz=Tc/T.
simplied phase diagram for ferromagnet

The basic variables dening the state of the system are the magnetic eld and the temperature. We describe what is happening by looking at the magnetization. The magnitude of the magnetization measures the extent to which the spins in the system are lined up with each other. Its sign describes the direction of the alignment. If the temperature is sufciently low, the system has a non-zero magnetization even at zero magnetic eld. At these lower temperatures, the zero-eld magnetization has two possible values, for the two possible directions in which the spins may align themselves. The heavy line is the locus of points at which this spontaneous magnetization is nonzero. As one crosses this line, there is a discontinuous jump in the magnetization, which maintains its magnitude but reverses its direction. This jump is a rst order phase transition. Typically, this jump decreases in size as the temperature gets higher until, at some critical point, the jump goes continuously to zero. This point is then the position of a continuous phase transition.
order parameter in mean eld transition

In legend H should be h
1
0.5
magnetization
H=0 H=0.2 H=0.5 H=0H=-0.2 H=-0.5
-0.5
-1 0 0.5 1
T c /T
1.5
First order phase transition critical point

Behavior in neighborhood of critical Point

The mean eld equation is <r>= tanh heff with heff = hr +K s nn to r <s>. We assume that we are near the critical point both the eld and the magnetization are small. Then we can expand the equation for <r> in a power series in heff <r>= tanh heff heff -heff 3/3 We also assume slow variation in space and that the temperature is close to criticality. To lowest order heff Kz <r> <r>. The lowest order result is used in the cubic term and the rest is evaluated exactly as <r>=hr+ <r>[1-t] +K s nn to r [<s> - <r>] - <r> 3/3 Here, t is the temperature deviation from criticality t=1-Tc/T. Finally replace the remaining K by its critical value, 1/z and expand to second order in the difference s-r to obtain 0=hr -t <r> + a2 2 <r>/z - <r>3/3 This is the near-critical mean eld theory in the Weiss model of a ferromagnet. It is the continuum version. On the lattice a2 2 <r>
------>
s nn to r [<s> - <r>]
17
Conclusions from Weiss Mean Field Theory For example, go after magnetization when h=0, no variation in space
0=t <> + <>3/z t is proportional to (T-Tc) if t is positive there is but one real solution <>=0. if t is negative we have a possible solution <>=0, and also the solutions <> = (-tz)1/2 One should choose the solution which actually minimizes the free energy. It turns out that this solution is the one which has the same sign as h. Therefore as h passes through zero for T less than Tc there is a jump in the magnetization proportional to the square root of -t.
Conclusions from Weiss Mean Field Theory For example, go after <H> when h=0, no variation in space H / kT = K + h
r s r nn r
We assume neighboring spins are uncorrelated -<H>/(kT) = Kz<> <> + other effects near Tc <>2 = 0 above Tc <>2 = - tz below Hence specic heat, d <H>/dT has a jump at Tc. It looks like
dE/dT
Tc
19
Critical Opalescence and Fluctuations

In the early 20th century experimenters observed what is called critical opalescence. As the critical point is approached, a uid which is otherwise clear and transparent is seen to become milky and reects light. This phenomenon was explained by Smoluchowski (1908) and Einstein (1910) Einstein particularly showed that the scattering came from density uctuations, and noted that these uctuations diverged at the critical point. Soon thereafter, in 1916 and 1914, Ornstein and Zernike put together a theory which show that the uctuations came from large regions of correlated uctuations and derived a qualitatively correct mean eld theory treatment of the phenomenon. This development would prove to be particularly important because spatial and correlational structure would turn out to be the key to understanding phase transitions.
Weiss Mean Field Theory for Correlations

For example, go after <rs>, when h=0, no variation in space, T>Tc 0=hr -t <r> + s nn to r [<s> - <r>]/z - <r> 3/3 neglect cubic term, differentiate equation with respect to hu 0=(d hr /d hu )-t (d<r>/d hu ) + s nn to r [d<s>/d hu - d<r>/d hu]/z d<r>/d hu is correlation function g(r-u)= <[r - <r>] [u - <u>] >, which then obeys 0=r,u -t g(r-u) + s nn to r [g(s-u) - g(r-u)]/z dene G(q)=exp[-iq.(r-u)] g(r-u)
r
so that
0=1 +[-t + 2i=1,2...d (cos qi - 1)/z ] G(q)

Fourier transform of correlation function obeys

0=1 +{-t +2 i=1,2...d [cos(aqi) - 1]/z } G(q) We can now solve for G(q) = 1/{t+2i=1,2...d [1 - cos (aqi)]/z} and invert the Fourier transform to get g(r-s)=dq exp[iq.(r-s) ]/(2)d 1/{t+2i=1,2...d [1 - cos (aqi)]/z} Here the integrals over each component of q extend from -/a to /a. To get a result for large spatial separation expand to second order in q and extend the integrations to cover all q g(r-s)=dq {exp[iq.(r-s) ]/(2)d } 1/[t+a2q2/z] One can perform the integral to nd that g(r-s)~ exp[-| r-s |/]/ | r-s |d-2 when the exponent is of order one, and d >2. The correlation length, , is given by =a/(zt)1/2
Correlation function looks like

The result is then of the form g(r-s)~ exp[-| r-s |/]/ | r-s |d-2 The correlation length, , is given by =a/(zt)1/2 This result is important. It says that the correlations in space extend over longer distances as one gets closer to criticality. So the fact that the phase transition will only occur in an innite system partially shows up in mean eld theory. The correlation extending to innity is a crucial fact in phase transitions. However, mean eld theory allows a phase transition in a periodic but nite system. Also, mean eld theory gets an answer which is somewhat incorrect in the neighborhood of the critical point. Near the critical point ~a/(t) but =1 in d=2 and 0.64 in three dimensions.
Conclusions from Weiss Mean Field Theory:

The magnetic susceptibility The quantity <>/ h, calculated at xed T, is called the magnetic susceptibility. It is given by the Fourier transform of the spin correlation function as <>/ h=G(0). Thus. we know its value in mean eld theory, which is <>/ h ~ 1/t. So the innite correlation length has as its direct consequence the innity in this important thermodynamic derivative.
24
Mean Field Theory of Fluids

I did not describe the development of mean eld theory in anything like historical order. Van der Waals derived the rst mean eld theory in the 1870s. He was interested in understanding the data of Andrews (1869) which showed a phase transition line separating a liquid and a vapor phase that then ends in a critical point in which the two become identically the same. Pierre Curie noticed that the behavior of a ferromagnet has a close analogy to that of a uid in 1895. Later in 1907, Pierre Weiss worked this remark into a theory like the one mentioned here. Fluids no doubt came rst because they were more familiar and because experiments existed.
T. Andrews, ``On the continuity of the gaseous and liquid states of matter,'' Phil. Trans. Roy. Soc. 159, 575-590(1869). Reprinted in: T. Andrews,,The Scientic Papers, Macmillan, London (1889). J. D. van der Waals, thesis Leiden, 1873. P. Curie, Ann. Chem. Phys. 5, 289 (1895). P. Weiss, J. Phys. 6, 661 (1907).
25 Perimeter Institute Statistical Physics Lecture Notes part 7. Mean Field Theory Version 1,7 9/11/09 Leo Kadanoff
van der Waals derivation

The starting point was the perfect gas law P = N T perfect gas law, which of course shows no phase transition. Van der Waals then introduced two corrections to describe what he had inferred about uids and the atoms or molecules which formed them. First, he argued that the molecules could not approach each other too closely because of an inferred short-ranged repulsive interaction among the molecules. He probably based his understanding of this repulsion upon the fact that it is very hard to compress liquids like water. This repulsive effect should reduce the volume available to the molecules by an amount proportional to the number of molecules in the system. Thus, in the perfect gas law should be replaced by the available or effective volume, -Nb, where b would be the excluded volume around each molecule.
26
van der waals second effect

The second effect is more subtle. The pressure, P is a force per unit area produced by the molecules hitting the walls of the container. However, van der Waals inferred that there was an attractive interaction pulling each molecule towards its neighbors. This attraction is the fundamental reason why a drop of liquid can hold together and form an almost spherical shape. Again we see that familiar facts can be translated into theoretical understanding. As the molecules move toward the walls they are pulled back by the molecules they have left behind them, and their velocity is reduced. Because of this reduced velocity, their impacts impart less momentum to the walls The equation of state contains the pressure as measured at the wall, P. This pressure is the one produced by molecular motion inside the liquid, NT/(-Nb), minus the correction term coming from the interaction between the molecules near the walls. That correction term is proportional to the density of molecules squared. In symbols the correction is a(N/)2 where a is proportional to the strength of the interaction between molecules. Van der Waals' corrected expression for the pressure is thus P= NkT/(-Nb)-a(N/)2 Here, a and b are parameters that are different for different uids.
27
Many Different Phase Transitions:

liquid -gas paramagnetic to ferromagnetic (un)mixing of solids and liquids superconducting, ..... van der Waals: (1873) Different simple liquids-gas transitions have very similar thermodynamic properties. Derives mean eld theory of liquid. Curie-Weiss (1907) mean eld theory of magnets. But, each different phase transition calls for its own theory.
After van der Waals Many Different Mean eld theories

The theory of phase transitions involving the unmixing of uids was developed by van der Waals himself, while such unmixing in solids was described by W. L. Bragg and E. J. Williams. Literally dozens of such theories were dened, culminating in the theory of superconductivity of Bardeen, Cooper and Schrieffer. These theories are all different in that they have different physical quantities playing the roles we have given to the magnetic eld, or T-Tc, or serving as the order parameters. The order parameter is the quantity that can undergo a discontinuous jump in the rst order transition, and describes the symmetry group of the physical situation. The magnetization is the order parameter of the ferromagnet, the density is the order parameter in the liquidgas transition. Much effort and ingenuity has gone into the discovery and description of the order parameter in other phase transitions. In the antiferromagnetic transition the order parameter is a magnetization that points in opposite directions upon alternating lattice sites. In ferroelectrics, it is the electric eld within the material. The superuid and superconducting transition have as their order parameter the quantum wave function for a macroscopically occupied state. Liquid crystals have order parameters reecting possible different kinds of orientation of the molecules within a liquid. The description of these different manifestations of the phase transition concept reect more than a century of work in condensed matter physics, physical chemistry, etc.
Order Parameter, generalized

Landau (~1937)suggested that phase transitions were manifestations of a broken symmetry, and used the order parameter to measure the extent of breaking of the symmetry. in ferromagnet, parameter = magnetization in uid, parameter = density
L.D. Landau
30
Generalized Mean Field Schemes I

Many different mean-eld schemes developed:. Each one has an order parameter, an average of a microscopic quantity. Landau generalized this by assuming an expansion of the free energy in an order parameter, note: no cubic term. This free energy applies to symmetry of Ising model
F= dr[a-hM+tM2+cM4+(M)2
expansion assumes a small order parameter (works near critical point) and small uctuations (works far away?!) h is magnetic eld t is proportional to
(T-Tc)
minimize F in M: result General Solution M(h, (T-Tc)) singularity as t,h both go through zero! singularity as h goes through zero for T< Tc
32
Why minimize?
Thermodynamics says that the free energy is extremized by the variation in any macroscopic parameter, e.g. any extensive variable. This is part of a general idea that the free energy is a probability, which arises from how the partition function is used. In statistical mechanics only the most probable things happen. This applies to all macroscopic phenomena, Now M(r) is not quite macroscopic, but Landaus idea was that at the long wavelength part of it it was macroscopic enough so that one could neglect its uctuations. This point of view has turned out to be wildly successful.
33
Vary M(r) to vary F

F= dr[a-hM+tM2+cM4+(M)2
integrate by parts in gradient term F=
dr M(r) [-h+2tM(r)+4cM(r)3- 2 M(r)]
to minimize F, coefcient of M(r) must vanish 0= h+2tM(r)+4cM(r)3-2 M(r) Hence we have an equation for M! That is a general equation for the order parameter in mean eld theory with the Z2 symmetry, i.e. symmetry under sign change of M.
34
For example, go after magnetization when h=0, no variation in space

0=- h+2tM(r)+4cM(r)3-2 M(r) 0=2tM+4cM3 t is proportional to (T-Tc) if t is positive there is but one real solution M=0. if t is negative we have a possible solution M=0, and also the solutions
M = t
2c
One should choose the solution which actually minimizes the free energy. It turns out that this solution is the one which has the same sign as h. Therefore as h passes through zero for T less than Tc there is a jump in the magnetization proportional to the square root of t.
Crucial part of the solution: for T< Tc , jump in order parameter goes as M ~ (Tc T ) with =1/2
Generalized Mean Field Schemes II
magnetization
This square root ( =1/2) appears to be a Universal result. Mean Field Theory predicts all near-critical behavior
0.5
H=0 H=0.2 H=0.5 H=0H=-0.2 H=-0.5
-0.5
order parameter, jump, and free energy were crucial concepts
-1 0 0.5 1
T c /T
1.5
order parameter in mean eld transition
36
order parameter and free energy were crucial concepts

free energy could be expressed in terms on any descriptors of systems behavior. It is a minimized by the correct value of any one of them, We have thus come loose from the particular thermodynamic variables handed to us by our forefathers, order parameter could be anything which might jump in the transition. other variables could be anything at all. In the meantime Schwinger was working on electromagnetic elds for World War II radar. He use variational methods and effective elds (lumped variables) to build electromagnetic circuits.
A worry?
Mean eld theory gives M ~ (Tc T ) and = 1/2 This power is, however, wrong. Experiments are closer to 1880-1960: No one worries much about discrepancies
M (Tc T )1/3 in 3-D
order parameter: density versus Temperature in liquid gas phase transition. After E. A. Guggenheim J. Chem. Phys. 13 253 (1945)
38
vapor
liquid
39
STOP Here

Part 7 Mean Field Theory

Uploaded by

Copyright:

Available Formats

Part 7 Mean Field Theory

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Part 7 Mean Field Theory

Uploaded by

Copyright:

Available Formats

Part VII: More is the Same

More is the Same Phase Transitions and Mean Field Theories

Phase Diagram for Water

liquid-gas phase transition line

Phase Transitions Describe Basic Physics

Gibbs: A phase transition is a singularity in thermodynamic behavior.

Magnetic Phase Diagram

A Phase Transition is a change from one behavior to another

theory starts from: phase transitions are singularities in free energy

problem dened by free energy dened by

exp H{r }/(kT )

Mean Field Theory: more is the same

focus on one spin, at r: that spin feels h and K s nn to rs

more is the same

heff = [h + Kz < >] < >= tanh(heff )

or, if there is space variation, heff = hr +K s nn to r <s>

Mean Field Theory is Only Partially Right

Calculate results of mean eld theory

simplied phase diagram for ferromagnet

order parameter in mean eld transition

H=0 H=0.2 H=0.5 H=0H=-0.2 H=-0.5

First order phase transition critical point

Behavior in neighborhood of critical Point

Critical Opalescence and Fluctuations

Weiss Mean Field Theory for Correlations

0=1 +[-t + 2i=1,2...d (cos qi - 1)/z ] G(q)

Fourier transform of correlation function obeys

Correlation function looks like

Conclusions from Weiss Mean Field Theory:

Mean Field Theory of Fluids

van der Waals derivation

van der waals second effect

Many Different Phase Transitions:

After van der Waals Many Different Mean eld theories

Order Parameter, generalized

Generalized Mean Field Schemes I

Vary M(r) to vary F

dr M(r) [-h+2tM(r)+4cM(r)3- 2 M(r)]

For example, go after magnetization when h=0, no variation in space

Generalized Mean Field Schemes II

order parameter, jump, and free energy were crucial concepts

order parameter in mean eld transition

order parameter and free energy were crucial concepts

You might also like