1

ERRATA
page 205, equation(4.82) G(r, r ) = page 209, equation (4.92)
1 G ij (r, r ) =

M (r ) 2 L[M ] or G1 (r, r ) = B (r ) M (r )M (r )

L[M ] Mi (r )Mj (r )

M1 =v,M2 =0

page 244, equation following (4.179) D4 page 245, rst equation of the page Z 1 = page 359, equation following (6.81)
T (u = 0) = T = P

D 1 2

=4D

d 1 G (q ) dq 2

q2 =0

page 366, rst equation 0 (x, t) = 0 + Dt

0

dx exp

(x x )2 4Dt

page 512, read: An example of the increase of the H-function is given by Jaynes [60]. ignore: but the argument is incomplete. page 565, equation (9.179) AB (t) = page 577, equation (9.193) t n + equation following (9.196) ( V ) = ( V ) 2 V page 584, rst line of question 1 1. By examining the moments X , P , (P )2 , (X )2 and P X ... 1 [A(t), B (0)] 2 1 gL = 0 m

Solutions for selected exercises and problems

Solutions for chapter 1
Exercise 1.6.2 Internal variable at equilibrium
1. Considering the energy as a function of S and y , one obtains from (1.36) E y At equilibrium (E/S )y = T , and E y
(eq) S

y S

S E

= 1

= T

S y

(eq)

=0
E

because one must have by denition (S/y)E = 0 at equilibrium. Let us also consider a function1 f (S (E, y ), y ) : f y =
E

S y

f S

+
y

f y

=
S

f y

the last equality being valid at equilibrium, and let us use this result in (1.64) 2S y 2
(eq)

=
E

E y

S E

E y

1 2E 2S Ey T y 2

The rst term vanishes at equilibrium, which demonstrates the second equation of (1.63), taking (1.62) into account. 2. The condition of maximum entropy reads dS = 1 1 dE1 + dE2 T1 T2 + P1 P2 dV1 + dV2 T1 T2

Since dE1 = P1 dV1 , the condition dS = 0 is automatic. 3. The condition of minimum energy at xed entropy, volume and particle number, is dE = (T1 T2 )dS1 (P1 P2 )dV1 + (1 2 )dN1 = 0 Because the superuid does not transport entropy, there cannot be entropy transfer between the two compartments: dS1 = 0. Furthermore dV1 = 0 because the volumes are xed. The only condition is that of equality of the chemical potentials: 1 = 2 . Assume that the temperature of compartment 1 is increased by T . As the chemical potential remains unchanged, the condition d1 = 0 added to the Gibbs-Duhem relation (1.41) V S dP d = dT + N N
1 The following equality is a particular case of a general identity which we shall encounter many times. Let f (x ) be a i function of several variables, which are themselves functions of another set of variables: xi (yj ). One easily shows that X f xi f = yj y xi xl=i yj y i
j =k j =k

3 shows that the pressure must increase by P = (S1 /V1 )T .

Problem 1.7.1 Reversible and irreversible free expansion of an ideal gas

1. One starts from the expression of (E/V )T E V =T
T

S V

P =T

P T

where we have used the Maxwell relation (1.27). By assumption, this expression must vanish, which gives another proof of (1.61). taking into account P T

=
V

1 (T ) V

One gets T (T ) = (T ) and by integration (T ) = aT . 2. In the case of an ideal gas S V The expression giving S is
T

=
T

P T

=
V

R V

S (T, V ) = S (T0 , V0 ) +
T0

CV + dT T

dV
V0

R V

that is

T V + R ln T0 V0 In order to express S as a function of the pressure, and not of the volume, one uses S (T, V ) = S (T0 , V0 ) + CV ln V T P0 = V0 T0 P

and, with CP = CV + R S (T, P ) = S (P0 , T0 ) + CP ln 3.

P0 T + R ln T0 P T l/2 V T0 V
l/2

S (T, V ) = S (T0 , V0 ) + R ln A transformation at constant entropy thus corresponds to T l/2 V = const ou

T 1/( 1)V = const

with = (l + 2)/l. If we use the pressure instead of the volume T P (1 )/ = const Tf = Ti Pf Pi

( 1)/

One can compute explicitly the work provided to the gas by integrating the formula for innitesimal work
f

Wif =

P dV
i

for a reversible adiabatic transformation: P V =const, but it is simpler to remark that the increase in work is equal to that in internal energy, as there is no heat exchange Wif = Eif = CV (Tf Ti )

4 the last equality being valid only for an ideal gas, and CV independent of T . It is a good exercise to check that this result coincides with that obtained from calculating the work; one will notice that CV = R/( 1) 4. The work supplied to the gas is determined by the external pressure, which is equal to the nal pressure Pf : Wif = Pf (Vf Vi ). As the transformation is adiabatic and as the gas is ideal, we have once more Wif = Eif = CV (Tf Ti )
Using the ideal gas law, Vf = RTf /Pf , one determines Tf Tf =

Ti

1 + ( 1)

Pf Pi

The nal temperature in the irreversible expansion is higher than that of the reversible expansion: this can be seen by using the concavity of the curve giving Tf as a function of x = Pf /Pi . 5. In the case of the reversible transformation, the entropy variation S vanishes, while in the irreversible case S > 0. Since the nal pressures are identical, one may use in the comparison the result of question 2 S (Tf , Pf ) = S (Ti , Pi ) + Cp ln whence CP ln
Tf

Pi Tf + R ln Ti Pf Tf Ti

Ti

> CP ln

and thus Tf > Tf . The variation of internal energy is minus the work supplied to the external medium W = CV (Ti Tf ). As Tf > Tf , Wirr < Wrev . In order to convince oneself that the external gas does not play any r ole in the entropy balance, one remarks that it can be replaced by a mass M on the piston, with Pf A = M g . The external medium is then entirely mechanical, and does not contribute to the entropy balance. If one puts again the mass on the piston, Pf = Pi , and the expression of the entropy shows that Tf > Ti because one must have S > 0 after completion of both operations. An explicit computation gives = Tf

Ti 2

1 + ( 1)

Pi Pf

1 + ( 1)

Pf Pi

It is easy to check in this expression that Tf > Ti by setting x = Pf /Pi and by remarking that (x +1/x) 2.

6. Since the expansion is adiabatic and does not supply any work to the system, and since the gas is ideal, the internal energy and therefore the temperature remains unchanged. From the results of question 2, dS = R ln[(V + dV )/V ] RdV /V > 0. Since d Q = 0, the transformation cannot be quasi-static: d Q = T dS does not hold. 7. Since the internal energy does not vary (adiabatic wall and no work supply), Tf = Ti , and S = R ln(Vf /Vi ), although no heat was supplied to the gas.

Problem 1.7.3 Equation of state for a solid

1. The knowledge of the equation of state allows us to express the temperature T = E S = f (V )g (S ) = S 1/3 f (V )eS/3R (S + 4R) 3R (1)

The quantity between square brackets staying nite when S 0, one sees in the preceding expression that S and T vanish simultaneously. This behaviour is in agreement with the third law. 2. Expression (1) allows one to display the dominant behaviours of S , and thus of CV , in the two limiting cases

5 1. low temperature: T S 1/3 S T 3 CV T 3 2. high temperature T eS/3R S 3R ln T CV 3R 3. From (1.68), one gets P = E V = f (V )g (S ) = 2Ab(V V0 )eb(V V0 ) g (S )
2

(2)

The physical interpretation of V0 is clearly displayed in (2): V0 is the volume at zero pressure, namely the maximum volume of the solid. For V > V0 , the atoms which form the solid are organized in another phase of matter described by another equation of state. One can point out an inconsistency of the model: it predicts a nite pressure for a vanishing volume! At zero pressure, the dilatation coecient vanishes, since V0 is temperature independent. 4. Let us apply (1.36) to the variables (S, V, T ) S V = T V T S
1 V

f (V )g (S ) f (V )g (S )

There are many possible proofs of relation (1.69), we propose below the proof which seems to us the shortest. Examination of the relevant variables for the calculation of the dilatation coecient lead us to work with the Gibbs potential G(T, P ) = E T S + P V dG = S dT + V dP One may then use the Maxwell relation V T = P S
1 T

Relation (2) provides us with the expression for P in terms of the variables (S, V ). Using a procedure which should be familiar by now, we display explicitly the partial derivatives of P with respect to these variables P S =
T

P S

+
V

P V

V S

(f (V ))2 (g (S ))2 f (V )f (V )g (S )g (S ) = f (V )g (S ) 1 (f (V ))2 (g (S ))2 f (V )f (V )g (S )g (S ) = CV f (V )g (S ) The last line is obtained by noticing that CV = T S T =T
V

T S

1 V

g (S ) g (S )

We invite the reader to work out another demonstration starting from V T T P P V = 1

Problem 1.7.5 Surface tension of a soap lm

1. The T dS equation reads in this case dE = T dS + dx

6 Let us also write the dierential of the free energy dF = S dT + dx and the Maxwell relation which follows from it2 S x = T

2. Using the dierentiation rule for implicit functions and the preceding Maxwell relation, one shows that E T E x 3. At a constant temperature dE = T T dx + dx
x

=
x

E S E = x

S S =T = Cx T x T x E S + = T S x x T T

The second term is identied with the work: d W = dx, and thus the energy which is exchanged in the from of heat is d Q = T dx = a0 T dx > 0 T x

One must supply heat to the lm in order to stretch it. 4. Let us show that (Cx /x)T = 0 Cx x =T
T

S x

T x

= T

2 T 2

=0
x

5. The surface tension, and thus f , are independent of the lm area, and thus of x, whence f x = 0 T =

One recognizes a well-known, and easily observable, property: any uctuation destroys the soap lm. 6. Let us express the T dS equation in terms of the variables (T, x) T dS = Cx dT + a0 T dx When a transformation is done at constant entropy a0 dT dx = T Cx Numerical application: dT = 4.3 102 dx (dx in m).

2 One

can derive directly this relation from (1.27) with the substitutions V x et P .

Solutions for chapter 2

Exercise 2.7.1 Density operator for spin-1/2
1. The second degree equation giving the eigenvalues of D is 2 + a(1 a) |c|2 = 0 and the product of the roots is given by + = a(1 a) |c|2 The positivity condition on D implies + 0 and the condition Tr D = 1, or + + = 1, so that 0 + 1/4, whence (2.105). In the case of a pure state one of the eigenvalues vanishes (the other being equal to one) and a(1 a) = |c|2 . The density matrix of the state | is D= ||2 | |2

and one can indeed check that ||2 | |2 = ( )( ). 2. One writes a = (1 + bz )/2, c = (bx iby )/2, whence (2.107) for D. As Tr i = 0, one deduces from (2.107) Tr i j = 2 ij
2 2 2 and (2.108). One observes that a(1 a) = (1 b2 z )/4 and that |c| = (bx + by )/4. The inequalities (2.105) are equivalent to 0 b 2 1, the pure case corresponding to b 2 = 1.

3. With B parallel to Oz , the Hamiltonian reads 1 H = z 2 and the equation of evolution of the density matrix is 1 1 dD = [H, D] = B (bx y by x ) dt i 2 which is equivalent to dbx = Bby dt dby = Bbx dt dbz =0 dt

that is db = B b dt This is nothing other than the evolution equation of a classical magnetic moment in a constant magnetic eld. 4. The non-zero matrix elements of D are D+;+ = D+;+ = 1 2 D+;+ = D+;+ = 1 2

On taking the partial trace, we obtain the density matrix of the rst spin D(1) D++ = D =
(1) (1)

1 2

D + = D + = 0

(1)

(1)

D(1) then describes a state with zero polarization. One has of course D(1) = D(2) , and D(1) D(2) , which is a diagonal matrix with all elements equal to 1/4, is not equal to D, which would be the case were D a tensor product.

Exercise 2.7.7 Galilean transformation

The occupation probability Pr of a level r is unchanged under a Galilean transformation, otherwise one could detect a translational motion without looking outside the system.This implies D = D and invariance of the entropy. Under a Galilean transformation, the Hamiltonian becomes H = H uP + 1 M u2 2

and the equality of the operator parts of D and D implies H P = (H uP) that is = and u = . One also gets Z = Z exp From (2.66), this gives P = ln Z = M = M u P which is the expected value. Concerning the energy one gets E= ln Z 1 ln Z M 2 = E + M u2 = + 2 2 2 1 M u2 2 = Z exp M 2 2

The total energy is the sum of the rest energy E and the kinetic energy M u2 /2 of the moving mass.

Exercise 2.7.8 Fluctuation-response theorem

1. Let us compute dK/dx dK (x) = ex(A+B ) (A + B )exA ex(A+B ) A exA = ex(A+B ) B exA dx On integrating this equation from x = 0 to x = 1 one gets
1

K (1) K (0) = e

( A +B ) A

1=

dx ex(A+B ) B exA
0

which leads to (2.118) if one multiplies the RHS by exp A. Applying this equation to exp(A + B ) leads to
1

eA+B = eA +
0

dx ex(A+B ) (B ) e(1x)A

One may replace exp(x(A + B )) by exp(xA) with an error on the order of 2 . Using the invariance of the trace under circular permutation leads to (2.119) Tr e(A+B ) Tr eA + Tr (eA B ) By denition d eA() d The substitution d in (2.119) leads to (2.120). Using once more the invariance of the trace under circular permutation allows one to recover (2.57) in a particular case eA(+d) = eA()+dA () = eA() + d

d Tr eA() = Tr d

dA() A() e d

(3)

9 2. From (2.66), or (2.124) Ai = 1 Tr exp Z i k Ak

k

1 Tr Ai exp Z

k k A
k

i ) = Tr (DA

In order to dierentiate Ai with respect to j , one must use (2.120) i exp Tr A j

1

k Ak
k

i Tr A
0 1

dx exp x
k

j exp (1 x) k Ak A

k Ak
k

= from which one deduces Ai j =

Z Tr
0

dxAi Dx Aj D(1x)

1 Tr Ai exp Z j
1

k Ak
k

Ai Aj

=
0

dx Tr (Ai Ai )Dx (Aj Aj )D(1x)

Let us set B =

ai (Ai Ai )
1

ai aj Cij =
ij 0

dx Tr (Dx BD(1x) B) 0

because, in a basis where D is diagonal Tr (Dx BD(1x) B) =

n,m x (1x) Dnn Bnm Dmm Bmn = n,m (1x) x |Bnm |2 0 Dnn Dmm

using the hermiticity of B and the fact that the diagonal elements of D are positive.

Exercise 2.7.9 Phase space volume for N free particles

The condition for the total energy to be E is p2 p2 1 + ...+ N E 2m 2m and thus N (E ) = VN N !hN d3 p1 . . . d3 pN
2 +...+p 2 2mE p1 N

One expects the average energy per particle to be E/N , which suggest to introduce the dimensionless variable u dened as 2mE 2 p= u u2 1 + . . . + uN N N since the average value of u is on the order of 1. This change of variables leads immediately to (2.122), with d3 u1 . . . d3 uN C (N ) =
2 u2 1 +...+uN N

In order to compute C (N ) one calculates in two dierent ways the integral In

+ +

In =

dx1 . . .

dxn e(x1 +...+xN )

10 This integral is the product of n independent Gaussian integrals and its value is n/2 . One can also compute this integral in polar coordinates in a n-dimensional space In = An
0

dr er rn1

where An is the area of the unit sphere in a n-dimensional space. The change of variables v = r2 casts In in the form n 1 n 1 dv ev v 2 1 = An In = An 2 2 2 0 which leads to 2 n/2 An = (n/2)
2 2 1/2 One computes C (N ) in polar coordinates in a 3N -dimensional space (r = (u1 + . . . + uN ) ) N

C (N ) = A3N
0

dr r3N 1 =

3N/2 (N )3N/2 2 3N/2 (N )3N/2 = N N ) + 1) 3N ( 32 ( 32

The nal form of C (N ) is obtained by using Stirling approximation One deduces N (E ) N (E ) = VN N! 4 emE 3 N h2
N 3N/2

3N +1 2

3N 2e

3N/2

(4)

One remarks that the result is well approximated by 1 N! E N

where 1 is the number of levels calculated in (2.22) for a single particle. The density of states (E ) is the derivative of N (E ) with respect to E , (E ) = N (E ), and (E ) is given by N (E )E .

Exercise 2.7.10 Entropy of mixing and osmotic pressure

1. The number of possible congurations available to place molecule B is proportional to the number of A molecules, NA , as each molecule denes a possible localization for B . The entropy is then k ln NA + a function of (T, P ). If one adds NB molecules, this can be done in
B CN NA

(NA )NB NB !

ways, whence the expression of S . This entropy is partly an entropy of mixing (see (2.93)). 2. The terms E and P V do not depend on NA because the energy and volume variations depend only on the interaction of molecule B with its neighbours A, and not on the number of A molecules: E and P V are local terms. If one adds NB molecules of type B , because of the condition NB NA , each B molecule interacts only with its neighbours of type A, and one can add the terms E and P V of each of the B molecules. To the Gibbs potential NA 0 (T, P ) of the solvent one must nally add NB f (T, P ) and the term T S , whence (2.123). In a scale transformation, ln(eNA /NB ) is unchanged, and G G. The chemical potentials follow by dierentiation A B = = G NB kT = 0 (T, P ) NA NA NB G = f (T, P ) + kT ln NB NA

11 One checks that G = A NA + B NB . 3. The chemical potentials of the type A molecules must be the same in both compartments 0 (T, P0 ) = 0 (T, P1 ) By performing a Taylor expansion 0 (T, P0 ) NB kT 0 P T NA NB kT V = 0 (T, P0 ) + (P1 P0 ) NA NA = 0 (T, P0 ) + (P1 P0 ) NB kT V NB kT NA

namely P1 P0

The osmotic pressure is identical to that of an ideal gas with NB molecules in a volume V 3 . However the volume V is typical of that of a liquid, on the order of 103 of that of a gas, and the osmotic pressure can be very large.

3 This remark must be taken as a mnemonic rule: as emphasized in our derivation, the osmotic pressure is of entropic origin .

12

Solutions for chapter 3

Exercise 3.7.5 Solid and liquid vapour pressures
1. The chemical potentials will coincide when the two phases are in mutual equilibrium s = g only if the zeroes of energy have been chosen in an identical way in the two phases. One possible convention is to decide that an atom in the gaseous phase, located at innity with a vanishing kinetic energy, has zero energy. On must then give the solid an internal energy at zero temperature which compensates exactly the ionization energy, so that if the solid is destroyed with all atoms sent to innity, the total energy vanishes. E (T = 0) = N u0 2. The solid being incompressible, the specic heats at constant volume and at constant pressure coincide CP = CV = C = N c; the energy of the solid becomes
T

E (T ) = N u0 + N and its entropy

T

dT c(T )
0

S (T ) = N
0

dT

c(T ) T

This allows one to deduce the expression for the Gibbs potential G = E (T ) T S (T ) + P N vs = N s (T ) T T = N P vs u0 + dT 1 c(T ) T
0

whence (3.149). The equality of the Gibbs potentials s = g leads to an implicit equation for P (T ) = kT /v . In the limit where vs v one gets P (T ) = 2 m h2
3/2

(kT )5/2 exp

s (T ) kT

3. The partition function for the N molecules of the liquid is ZN = 1 N N! l l = N v0 h3 d3 p ep

2

/2m u0

N v0 u0 e 3

from which one derives the liquid Gibbs potential Gl Gl = FN + P V = N P v0 u0 kT ln ev0 3

and l = G/N . Since V = N v0 , N and V are not independent variables, one should not dierentiate with respect to N at xed V . The equality of the chemical potentials g et l and the ideal gas law P v = kT give the relation v0 kT ev0 u0 = ln + + ln kT 3 P 3 v

13 and taking the exponential of this relation eu0 /kT ev0 = kT P v0 /kT e P

Since P v0 /kT = v0 /v , this quantity 1 if the specic volume of the liquid is very small with respect to that of the gas, an excellent approximation in the case of boiling water under standard conditions v0 /v 103 . Within this approximation P kT u0 /kT e ev0

4. The entropies per molecules in the gas and in the liquid phases are deduced from (3.17) or from g l so that = T (g l ) = kT ln On the other hand k dP = dT ev0 1+ u0 kT eu0 /kT = = = g T l T v 5 + k 3 2 P v0 5 = k ln 3 + k 2 P = k ln v v0 1+ u0 kT

P T

which allows one to check that the Clapeyron formula holds true, as must be the case because our starting point was the equality of chemical potentials. In the case (i ) one nds u0 /kT 12.1 and in the case (ii ) u0 /kT 6.4. This dierence by a factor of two illustrates the rough character of the model.

Problem 3.8.4 Models of a boundary surface

A. 1. The length of the interface is obtained from Pythagoras theorem
L

L=

dx
0

1+

dy dx

1/2

1 L+ 2

dx
0

dy dx

(5)

2. On dierentiating (3.158) with respect to x one gets n nx dy An cos = dx n=1 L L whence the expression of the Hamiltonian H = = 2
L 0 2 n,m=1 n,m=1

n L

m nx mx An Am cos cos L L L
L

4L2

nmAn Am
0

dx cos

(n + m)x L

+ cos

(n m)x L

The result (3.159) for H is obtained thanks to

L

dx cos
0

(n + m)x L

+ cos

(n m)x L

= Lnm

This result is nothing other than the decomposition of H into independent normal modes An .

14 3. The probability law for the An s is given by the Boltzmann weight exp[H ]. Since the normal modes An are uncoupled P (An ) exp 2 2 2 n An 4LkT P (An , Am ) exp 2 2 2 n2 A2 n + m Am 4LkT

and average values follow from the equipartition theorem (3.53) A2 n = 2LkT 2 n2

An Am =

2LkT mn 2 n2

This expression allows one to derive y (x)y (x ) y (x)y (x ) = = Using the identity (3.160) 1 2 n n=1

2LkT 2 2LkT 2

1 nx sin sin 2 n L n=1 1 2 n n=1

nx L cos n(x + x) L

cos

n(x x) L

cos

n(x x L

cos

n(x + x L kT x(L x ) L

2 x(L x ) L2

we compute the average value y (x)y (x ) = as well as

(y )2 = (y (x) y (x ))2 = When |x x| L, the uctuation (y )2 is (y )2

kT |x x|(L |x x|) L

kT |x x|

(6)

One can also derive these last two results without appealing to normal modes, by discretizing the integral which gives L and by using the identity (A.45) on multiple Gaussian integrals. Indeed, the problem is equivalent to that of a random walk on a straight line x(t) (exercise 6.2), with the following correspondence: x y , t x et D kT /2, where D is the diusion coecient. Relation (6) is the analogue of: (x)2 = 2D|t t|. The uctuation is largest for |x x| = L/2, and in that case (y )2 1/2 L. The interface width grows as L, and is not a constant. B. 1. The energy associated with a plane interface, corresponding to the ground state, is 2JN . Let us observe that in the preceding situation N is the length of the interface, in units of the lattice spacing. In a general conguration where the interface is not a straight line, the interface length is
N

L= and the energy

p=1

(1 + |yp |)
N

H = 2J L = 2JN + 2J
N

p=1

|yp |

2. Let us dene hp = 2J (1 + |yp |) and H = function factorizes

hp : the steps are independent variables and the partition

p=1 N

Z=
p=1

Tr ehp = N

15 where represents the partition function of a single step =

+ y =

e2J (1+|y|) = e2J coth J

From the knowledge of (or of Z ), one gets and f using (3.2) and (3.6) = 2J 1 + 1 sinh 2J f = 2J 1 ln coth J

To each of the extreme regimes corresponds a very dierent behaviour of the interface 1. J 1: the steps behave as independent random variables and the interface is very irregular 2J + kT
+

f 2J kT ln(kT /J )

One then nds a kind of equipartition theorem with = 2. In this classical limit =

dy e2J (1+|y|) = e2J

1 J

2. J 1: the interface is in its ground state, meaning that it is a straight line. 2J f 2J

3. The constraint is obeyed in two situations y = p, so that P (|y | = |p|) = 2e2J (1+|p|) (7)

Setting = e2J = coth J and K = J in (7), one easily shows the two following results |y | y2 = = 1 1 1 = 2 K sinh 2J 1 1 1 2 = 4 K 2 2(sinh J )2

The rst result allows one to check = 2J (1 + |y | .

2 2 2 2 4. The central limit theorem tells us that because y = 0, whence y = qy . Moreover y = y

(y )2 = q y 2 =

q 2(sinh 2J )2

As in the continuum model (part A), one nds that (y )2 q |x x|.

Problem 3.8.5 Debye-H uckel approximation

1. The densities n (r ) are proportional to the probabilities of nding a charge q located at position r as measured from the ion at the origin of coordinates. They are thus proportional to the Boltzmann weight q (r ) n (r ) exp kT since the potential energy of an ion with charge q is q (r ). The normalisation is xed thanks to the condition that, if r , (r ) 0 and n(r ) n. The condition of weak deformations, in other words the fact that there are only small deviations with respect to the ideal gas case, corresponds to a regime

16 where the modulus of the potential energy, |q | is very small with respect to the kinetic energy, kT . Therefore4 n (r ) = neq(r) n (1 q (r )) (8) Let us note that in the general denition (3.75) of the density, the origin of coordinates can be taken arbitrarily, in contrast to the present situation where the position r which appears in the densities n (r ) is measured from a specic ion. In question 4, where we shall give the expression for the pair density, we shall see very clearly that n (r ) must be interpreted in terms of conditional probabilities. 2. The second form of the Poisson equation illustrates very clearly the meaning of the ionic cloud: a central charge, q (r ), which modies locally the charge distribution around it q (n+ (r ) n (r )). Using (8), the Poisson equation becomes 2 (r ) and its Fourier transform reads p2 = so that = 2nq 2 q 1 q + = 2+ 0 kT 0 b 0 1 q 2 0 p + b 2 q er/b 40 r 1 er/b 4b2 r 2nq q (r ) + (r ) 0 kT

Using (3.161), the inverse Fourier transform of the preceding equation gives (r ) = whence n+ (r ) n (r ) =

3. The charge Q(R) contained in a sphere with radius R is d3 r (n+ (r ) n (r )) = q 1 1 b2

R

Q(R) = q 1 = qe
R/b

0r R

(1 + R/b)

dr rer/b

When R one expects that this charge tends to zero because the solution is electrically neutral. One checks this to hold true for R b, which allows us to interpret the Debye length b as measuring the extension of the ionic cloud: far beyond the Debye length, the central charge is completely screened. 4. The expression (3.162) of the average potential energy is obtained directly from the general relation (3.84), by identifying separately the contributions of the densities associated with the two kinds of ion, 3 3 n+ and n . For the sake of simplicity, let us consider one of the two pair densities: n+ 2 (r , r )d r d r is interpreted as the mean number of pairs of identical charges in d3 r d3 r around r and r . However, assuming that an ion is located at r = 0, then n+ (r )d3 r represents the mean number of pairs of ions with identical charge which one may build around the position r with the central ion. Hence
2 q(r) n 2 (r , r ) = nn (r = r r ) = n e

The last equality gives the pair correlation function (3.77) of the problem.
4 The expansion of the exponential is clearly not valid for r 0. However we only need it to be valid on the scale of the average distance between ions: q ( r ) kT , with r n1/3 . As we shall see in question 5, the approximation is valid provided the density of potential energy is small with respect to the density of kinetic energy. If that were not the case n (r ) would not be given simply by a local Boltzmann weight!

17 The expression (3.162) of the potential energy is clearly of the form (3.84), once the result is multiplied by a factor of two in order to take into account the fact that two signs are possible for the central charge Epot =
1 q2 n+ d3 r d3 r 2 (r , r ) n2 (r , r ) 40 V |r r | 1 q2 d3 r n+ (r ) n (r ) = nV 40 V r

q2 = nV 40 b2

dr er/b = nV

q2 40 b

5. The average potential energy between two ions (taken with the same sign to simplify the discussion) is q2 q2 40 r 40 n1/3 The condition of weak deformations also reads q2 kT 40 n1/3 so that 40 kT 1 = 8b2 nq 2 n2/3

which leads correctly to the condition r b; this last condition implies that the present model is only valid if the volume b3 contains a large number of charges. In an equivalent way, one may observe that the density of potential energy nq 2 /(40 b) is very small with respect to the density of potential energy (3/2)nkT provided b r .

Problem 3.8.7 Beyond ideal gas: rst term of virial expansion

1. One deduces (3.166) from the expression (3.86) of the pressure P = and from /V = (1/N )(/v ). 2. Low values of the temperature enhance the eect of the attractive part of the potential, while high values enhance the repulsive part. The average number of molecules in a sphere of radius r0 is N1 3 nr0 1 if one chooses n to be arbitrarily small. Since the molecules are assumed to be independent, the probability P (N1 ) of nding N1 molecules in a sphere with radius r0 is a Poisson distribution. The probability of nding two molecules in the sphere is then N1 2 /2 N1 . If one draws around each molecule a sphere with radius r0 , one sees that the average number of molecules whose distance is less 3 3 3 than r0 is N nr0 /2. This number is negligible if N nr0 = n2 V r0 1, namely if one considers a suciently small volume of gas at a xed density. This condition being satised, let us assume that the molecules 1 and 2 are located at a distance less than r0 .The probability of nding in the gas another pair of molecules 3 2 whose distance is less than r0 is then (N nr0 ) and thus negligible. The integrand in (3.167) is dierent from zero only if |r1 r2 | < r0 , while for all the other pairs one can take this distance to be larger than r0 , and the integral can be evaluated as I V (N 2) d3 r1 d3 r2 eU (|r1 r2 |) 1 V (N 1) d3 r eU (|r|) 1 1 ln ZK 1 ln ZU 1 ln Z = + V V V

Since the number of pairs is N (N 1)/2 N 2 /2, one deduces (3.168) and (3.169) with B (T ) = 2
0

dr eU (|r|) 1 r2

(9)

18
3 using polar coordinates. As B (T ) r0 , one has (N 2 /V )B (T ) 1 and

ln ZU

N2 B (T ) = N f (v, T ) V

One checks the extensivity of the free energy and obtains P v/kT from ln ZU by using (3.166). 3. If the gas is suciently diluted, v b and P = so that by identication with (3.170) B (T ) = b a kT kT kT a v b v2 v 1+ b v a v2

4. In the case of a dilute gas, the probability of nding a molecule at a distance r from another molecule is proportional to the Boltzmann weight exp[U (r)], because the potential energy of the set of the two molecules is U (r). This is not true for a dense gas, because this potential energy is modied by the presence of neighbouring molecules, but this eect is negligible for a dilute gas. The normalisation limr g (r) = 1 demands that g (r) exp[U (r)]. Putting this value of g (r) in (3.172) and using polar coordinates, one nds 2n2 3 P = nkT r dr U (r)eU (r) 3 0 One then integrates by parts, taking as a primitive of U (r) exp[U (r)] the function 1 U (r) e 1

so that the pressure is given by a convergent integral. The nal result is P = nkT 2n2 which coincides with (3.170). 5. 1. Hard sphere gas
0

r2 dr eU (r) 1

B (T ) = 2

r2 dr =

2 3 3

The rst virial coecient is positive and independent of temperature. A naive reasoning relying on excluded volume considerations would lead to a coecient 4/3 in the van der Waals equation, instead of the correct value 2/3; actually it is the number of pairs of colliding molecules which is the relevant quantity. 2. In this case the virial coecient is given y B (T ) = 2 3 3 3 ) (eu0 1)(r0 3

When T 0, exp(u0 ) is large and the attractive part of the potential dominates: B (T ) < 0. On the contrary, when T , the repulsive part of the potential dominates and B (T ) (2 3 )/3; the potential energy is then negligible with respect to the the kinetic energy, except if the repulsive part is concerned. One can then predict a change of sign of B (T ).

19 3. Lennard-Jones potential B (T ) = = = 2
0

dr r2 exp 4u0

12

12

1
6

2 3 3 f

dx x2

exp 4 1

u0 kT

1 x

1 x

u0 kT

= 3 f

If one draws B (T )/ 3 as a function of , one should obtain a universal curve, independent of the type of molecule. However, at low temperatures, and for the lightest gases, hydrogen and helium, quantum eects may not be neglected and one observes deviations with respect to the universal curve which relies on the classical theory.

Problem 3.8.8 Theory of nucleation

1. Using the results of subsection 3.5.4, one easily shows that (E T0 S 0 N ) 0 (10)

2. If the system exchanges work with the external medium, the expression of energy conservation becomes E = Q + W + 0 N so that using once more Q T0 S W (E T0 S 0 N ) In the case of a reversible transformation we have Stot = (E T0 S 0 N ) W = 0 and one checks Wmin = (E T0 S 0 N ) 2. The entropy is an increasing function of the energy, as is clearly illustrated by the curve Stot (Etot ). 3. in the vicinity of point D (or C ), one may use a linear approximation Stot dStot dEtot Wmin = 1 Wmin T0

When going from D to B , one goes from a situation where the system is in equilibrium with the reservoir, and where the number of congurations is eq , to an o-equilibrium situation where the number of congurations is . Therefore the probability of such a uctuation is P
q = e(Stot Stot )/k = eStot /k eWmin /kT0 q

4. One works in a container of volume V . Before the droplet is formed, the gas occupies the whole volume, and thus (see (3.133)) Jin = P2 V After formation of a droplet of volume V1 , Jn = P2 (V V1 ) P1 V1

20 The transformation gas-droplet is made at constant temperature and chemical potential. Indeed the supersaturated vapour plays the r ole of the reservoir (gure 3.24), and the droplet that of the system A. The minimum work which is necessary in order to form the droplet is thus given by (3.173), when the surface energy is taken into account Wmin = J + A = V1 (P2 P1 ) + A 5. In the case of a spherical droplet, V = (4/3)R3 et A = 4R2 , one has dWmin = 4R2 (P2 P1 ) + 8R dR and R is given by R = 2 P1 P2 (11)

One observes on gure 3.36 that a droplet disappears if it is formed with a radius less than the critical radius R . One must overcome a potential barrier of value Wmin (R ) in order to obtain the droplet nucleation. This is the meaning of the expression activation energy which is given to the critical value Wmin = Wmin (R ) 3 16 (12) Wmin = 3 (P1 P2 )2 Wmin 16 3 3 (P1 P2 )2

3/(P1 P2 ) R

Figure 1: Curve giving Wmin (R). 7. Clearly the probability of nucleation is given by P eWmin /kT Taking (12) into account, one understands that droplet formation is more and more likely as the pressure dierence between the liquid and gaseous phases increases. 8. One must necessarily have 1 (P1 , T ) = 2 (P2 , T ) otherwise one would observe a ux of molecules: , T ) = 2 (P , T ). Therefore, at xed temperature at the transition 1 (P ) = 1 (P1 ) 1 (P 1 (P + P1 ) 1 (P ) = v1 P1 = ) 2 (P2 ) 2 (P ) 2 (P + P2 ) 2 (P v2 P2

21 where we have used (/P )T = v which is deduced from the Gibbs-Duhem relation (3.105). We can now rewrite the denominator of (11) as P1 P2 = P1 P2 = = P2 v1 P2 (v2 v1 ) v1 2 1 1 = (2 (P2 , T ) + 1 (P2 , T )) P T P T v1

and one then obtains (3.174), namely an expression for the critical radius which does not involve the knowledge of the pressure inside the droplet.

22

Solutions for chapter 4

Exercise 4.6.1 High temperature expansion and the Kramers-Wannier duality
1. One rst proves the identity exp (KSi Sj ) = cosh K + Si Sj sinh K by noticing that there are only two possibilities, Si Sj = 1 and Si Sj = 1, and that the identity is valid in both cases. Then, expanding the expression of the partition function, one readily nds Z (K ) = (cosh K )L
C i,j

(1 + Si Sj tanh K )

(13)

Let us represent graphically the product of Ising spins

n1 nN ni S1 SN Si

by drawing a line on the lattice between two Si s if they are nearest neighbours. The contribution of the spin product to the partition function will be non zero only if there is an even number of lines (0, 2 or 4) leaving from all sites. If a term in the expansion corresponds to an odd number of lines leaving any site, then it gives a vanishing contribution. Thus the contribution of the preceding spin product will be non zero only if the lines form a closed polygon on the lattice. 2. Let us count the number of polygons for the rst values of b, without paying attention to the boundary conditions. 1. b = 4. The polygon is a square with perimeter 4 (in units of the lattice spacing). Since this square can be placed in N dierents ways on the lattice, (4) = N . 2. b = 6. The polygons are rectangles with perimeter 6, whose largest side can be either horizontal (horizontal rectangle) or vertical (vertical rectangle). Then (6) = 2N . 3. b = 8. The leading term in N is obtained by placing two squares with b = 4 on the lattice. This can be done in N 2 /2 ways. Other possibilities (i.e. rectangles with perimeter b = 8) give only a factor proportional to N . Thus (8) = N 2 /2. It is left to the reader to show that the exact result, using periodic boundary conditions, is (8) = N (N 5)/2. 4. b = 10. The leading term is given by a square with perimeter 4 and a rectangle with perimeter 6. This rectangle can be either vertical or horizontal, so that (10) = 2N 2 . 5. b = 12. The leading term corresponds to three squares with perimeter 4, giving (12) = N 3 /3!. The rst terms in the high temperature expansion of the partition function are then, setting tanh K = x ZN (K ) 2N (cosh K )L 1 + N x4 + 2N x6 + + 2N 2 x10 + 1 3 12 N x + 6 1 2 8 N x 2

(14)

The free energy FN is proportional to the logarithm of ZN ; we have, to order x12 FN ln ZN N x4 + 2N x6 + 1 2 1 1 2 8 N x + 2N 2 x10 + N 3 x12 2 6 2 1 1 N x4 + 2N x6 + N 2 x8 + N 3 x12 + 2 3

= N (x4 + 2x6 + ) + N 2 x12 +

23 At least to order x10 , we have thus checked the extensivity of the free energy, but our estimates of (b) are not precise enough to give the coecient of N in the x8 term and to eliminate the N 2 factor in the x12 term. 3. Let us ip all the spins of the dual lattice inside a given polygon of the original lattice. It is clear that there is a one-to-one correspondence between the links on the polygon and the number of broken links on the dual lattice: each broken link on the dual lattice crosses a side of the polygon once and only once. Starting from a conguration with all spins up, the energy of a conguration with n broken links will be En = KL + 2nK because each broken link costs an energy 2K , hence the partition function is, in the low temperature region Z (K ) = 2 exp(KL) 1 +
n=4,6,8

(n) exp(2nK )

(15)

where the factor 2 comes from the fact that we could have chosen to start from a conguation with all spins down. Note that (13) converges for temperatures T > Tc1 , while (15) is convergent for T < Tc2 . with Tc2 Tc1 : The temperatures Tc1 and Tc2 are dened as those temperatures which limit the radius of convergence of the expansions (13) and (15) respectively. 4. From the high temperature expansion we have Z (K ) = 1 + 2N (cosh K )L

b=4,6,8,...

while the low temperature expansion gives for a temperature T

(b)(tanh K )b

Z (K ) = 1+ (n) exp(2nK ) 2 exp(K L) n=4,6,8 The two ratios are equal if one chooses to relate K and K by exp(2K ) = tanh K This gives a relation between the partition function at two dierent temperatures. If there is a unique critical temperature where both the high and low temperature expansions diverge, Tc1 = Tc2 Tc , we obtain this critical temperature Kc as exp(2Kc ) = tanh Kc or sinh2 Kc = 1 Note that the assumption that there is a unique critical temperature is a strong one: one easily nds models where this assumption is not satised, see for example problem 7.9.3.

Exercise 4.6.5 Shape and energy of an Ising wall

1. The equation of motion for M d2 M = V (M ) dz 2 is immediately deduced from the identities (A.59) and (A.60). In the mechanical analogy, the conservation of kinetic energy is 2 1 dx + U (x) = const 2 dt The particle starts at t = with zero velocity from a point of abscissa x = M0 and arrives at t = + with zero velocity at the point x = M0 . At the points x = M0 , the kinetic energy vanishes and

24 the potential energy is U (M0 ), so that the constant in the preceding equation is U (M0 ). Transposing this result to the original problem leads to 1 2 dM dz
2

= V (M ) V (M0 )

(16)

Of course, one could prove (16) directly from the equation of motion exactly as one proves energy conservation in classical mechanics. One checks that dierentiating (16) with respect to z gives back the equations of motion, while the boundary conditions are limz M = M0 and lim|z| (dM/dz ) = 0.
2 2. Dividing equation (16) by |r0 |M0 , one gets

1 2 2|r0 |M0

dM dz

1 u0 2 2 4 4 2 (M M0 ) + 4!|r |M 2 M M0 2M0 0 0 u0 1 = 2 4 4!|r0 |M0 4M0

Now the zero eld magnetization is

2 M0 =

6|r0 | u0

whence

This yields the equation for f 2 df dz

2

1 1 1 = (f 2 1) + (f 4 1) = 1 f2 2 4 4

Taking the square root of the preceding equation and choosing the positive sign (the negative sign would also lead to an equally acceptable solution, corresponding to a decreasing function of z , one gets df dz = 2 1f 2 from which one deduces f (z ) = tanh z z0 2 (17)

Taking z0 = 0 simply xes the position of the wall at z = 0. 3. The surface tension is the dierence per unit area of the energy in the presence of a wall (rst bracket in the following equation) and the energy without any wall (second bracket)
+

dz

1 2

dM dz

+ V (M ) V (M0 )

Using (16) allows us to cast the expression for in the form

+

=
2 = M0

dz
+1

dM dz

2 M0

dz

df dz

df M2 df = 0 dz 2

+1

(1 f 2 )df =

2 2 M0 3

Exercise 4.6.6 The Ginzburg-Landau theory of superconductivity

1. The magnetic eld B vanishes with the vector potential. Furthermore, decomposing into real and imaginary parts = 1 + i2

25 we get | |2 = |1 + i2 |2 = (1 )2 + (2 )2 so that ( N ) becomes N = d3 r a| |2 + b| |4 +

2

2m

(1 )2 + (2 )2

which is exactly (4.88) with a suitable redenition of the parameters a, b and . 2. The covariant derivative D = i q A transforms in the following way under a local gauge transformation exp(iq (r ))(i q A) exp(iq (r )) = i q A(r ) + q (r ) = i q A (r ) which shows the invariance of ( N ) under local gauge transformations. 3. Let us give some details on the computation of functional derivatives. We begin with 1 = 1 2 d3 r B 2 = 1 2 d3 r ( A)2

We use the identity, easily proved from the techniques given at the end of 10.4.1 (V W ) = W V V W to write 1 as 1 = = = d3 r( A) ( A) = d3 r B ( A) (18)

d3 r A ( B ) + ( A B ) d3 r A ( B ) A (d2 S B )

where we have used the divergence (Greens) theorem in going from the second to the third line. The second term in the last line is a surface term, and neglecting this term for the moment yields 1 A(r ) = B (r ) (19)

Let us now compute the functional derivative with respect to A of 2 2 = = 1 2m 1 2m d3 r d3 r i q A

2 2

| |2 + i q (A ( )2 A + q 2 | |2 A2

from which one easily deduces 2 A = i q q2 ( ) ( ) + | |2 A2 2m 2m (20)

One recognizes the familiar expression for the probability current of the wave function in the presence of a magnetic eld. It remains to deal with 3 = d3 r B H

26 The variation of 3 is given by 3 = = = d3 r H A d3 r ( A H ) + d3 A (d2 S H ) d3 r A ( H )

We have used Greens theorem and the fact that H = 0 inside the superconductor. The surface term derived a few lines above combines with that obtained previously to give the nal boundary condition n (B H ) = 0 (21)

where n is a unit vector normal to the boundary of the superconductor: thus the tangential component of (H B ) is continuous at the boundary between the normal and superconductor phases. The variation with respect to leads to much easier calculations and we only quote the nal results 1. The equations of motion 0 = B 2. The boundary conditions n (B H ) = 0 4. In a uniform situation and zero magnetic eld | |2 = a 2b a| |2 + b| |4 = a2 4b (25) n (i q A) = 0 (24) = 1 i q A + a + 2b| |2 2m 0 q 2 i q0 ( ) A | |2 2m m (22) (23)

The dierence of Gibbs potentials between the normal phase in the presence of an external eld and in the absence of such a eld is
H

N (H ) N (H = 0) =

dH B (H ) =

H2 20

while S (H ) = S (H = 0) since B = 0 in the superconductor phase. The phase transition between the normal and superconductor phases occurs when the Gibbs potentials are equal. This denes the critical eld Hc (T ) 1 S (H = 0) = N (H = 0) H 2 (T ) 20 c Now one uses (25) S (H = 0) N (H = 0) = so that
2 Hc (T ) =

a2 4b

0 a2 2b

The critical eld vanishes linearly at T = Tc since a(T ) = a(T Tc )

27 5. Let us assume a semi-innite geometry where the normal-superconductor boundary is the plane z = 0. Then the order parameter obeys | (z = 0)| = 0 lim | (z )| = |a| 2b

We may choose to be real without loss of generality. As in exercise 4.6.5, we dene a dimensionless function f (z ) f (z ) = b (z ) 2|a|
2

f (z = 0) = 0

lim f (z ) = 1

The equation of motion for (z ) is d2 + a + 2b 3 = 0 2m d z 2

from which one deduces the corresponding equation for f 2 2 which is integrated to 2 df dz
2

d2 f f + f3 = 0 dz 2

1 1 = f 2 + f 4 + const 2 4

For z , df /dz 0 and f 1, which means that the constant is 1/4. One nally gets 2 df dz
2

1 (1 f 2 )2 4

(26)

which is easily integrated as in the preceding exercise and allows one to obtain the z -dependence of the order parameter |a| z (27) (z ) = tanh 2b 2 Starting at z = 0 from a vanishing order parameter, the order parameter grows from zero toward its bulk value |a|/(2b) over a length , the coherence length. 6. We now assume = cst and A = 0. In other words, we consider the situation deep inside the superconductor, at distances from the boundaries. The electromagnetic currrent in (25) becomes (r ) = q 2 nS A m (28)

so that the equation of motion (23) becomes in the Coulomb gauge A = 0 2 A 0 q 2 nS A=0 m (29)

This equation denes a second characteristic length of the superconductor, the London penetration length 1/2 1/2 m 2mb = (30) = 0 q 2 nS 0 q 2 |a| In order to make explicit the fact that is indeed the penetration length, let us rst express (29) in terms of the magnetic eld by taking the curl of both sides 2 B 1 B=0 2

28 In a one-dimensional situation where B is perpendicular to the z -axis, B = B u , u z = 0, one nds 1 d2 B 2B = 0 2 dz z Thus the magnetic eld vanishes in the bulk of the superconductor. B (z ) = B0 exp whose solution is

Exercise 4.6.11 Critical exponents to order for n 1

1. We write 4 = (2 2 ) + 2
2

= 2

+ 2 2 (2 2 ) + (2 2 )2

We may omit 2 2 , which adds a constant to H . Furthermore the term u0 2 ( 2 )2 1 n and may be neglected. With these simplications, the Hamiltonian reads H= dD r 1 1 u0 2 2 ((r ))2 + r0 2 (r ) + (r ) 2 2 2n (31)

and is now of Gaussian type. The coecient of 2 is 1 u0 2 r0 + 2 n and one may write immediately the correlation function from (4.171) ij (q ) = G On the other hand ij q + r0 +
2

u0 2 n dD q Gii (q ) (2 )D

(32)

2 i =
i=1 i=1

i (0)i (0) = n

which leads to the consistency equation

where we have used

= nKD
0

q D1 dq u0 2 q 2 + r0 + n

(33)

dD q SD q D1 dq = = KD q D1 dq (2 )D (2 )D

We may check in (33) the consistency of our assumption 2 n. The rst two terms in H are of order n. As 4 is of order n2 , one must choose the combination u0 /n so that all terms in H are of the same order in n when n . (q = 0), 2. From the uctuation response theorem, the inverse susceptibility (T ) is proportional to 1/G so that u0 2 (T ) = r0 (T ) + n

= r0 (T ) + u0 KD
0

q D1 dD q q 2 + (T )

(34)

29 The critical temperature Tc corresponds to an innite susceptibility, and thus to (Tc ) = 0. This leads to

0 = (Tc ) = r0 (Tc ) + u0 KD
0

q D1 dq q2

(35)

Subtracting (35) from (34) we obtain

(T ) = r0 (T Tc ) u0 (T )KD

q 2 (q 2
0

q D1 dq + (T ))

(36)

Let us make the change of variables q = x . The equation for (T ) becomes

/

(T ) = r 0 (T Tc ) + [(T )](D2)/2 u0 KD where is the ultraviolet cut-o.

xD1 dx x2 (x2 + 1)

(37)

3. The integral in (35) is divergent for D 2, and the theory can only be valid for D > 2. For 2 < D < 4, the integral in (37) is convergent at innity, and one may write (T ) = r 0 (T Tc ) + C(D2)/2 whence (T ) (T Tc )2/(D2) Indeed, for 2 < D < 4, 2/(D 2) < 1 and one may neglect (T ) on the LHS (38). The critical exponent is then 2 2<D<4 = D2 (q ) 1/q 2 , so that = 0. Finally For T = Tc , G (T ) ((T ))1/2 (T Tc ) D2 = = To summarize, for 2 < D < 4 and with an error O(1/n) =0 = 1 D2 = 2 D2
1

(38)

1 D2

which obeys the scaling laws (4.166). Note also that the results agree with the n limit of (4.203). For D > 4, one recovers, as one should, the mean-eld exponents. This can be seen as follows. Since the integral over q is ultraviolet divergent and infrared convergent even when = 0, we may write the integral in (36)

u0 KD
0

q D1 dq u0 KD q 2 (q 2 + (T ))

q D5 dq =
0

u 0 K D D 4 = C () D4

The equation for (T ) becomes (T ) = r 0 (T Tc ) C ()(T ) so that (T ) (T Tc ), and = 1. The only eect of the interaction is to renormalize the value of r 0 , leaving the critical exponents unchanged.

Exercise 4.6.12 Irrelevant exponents

30 1. From the denition u0 = g0 and (4.185) we deduce g We now write the integral in (39) as
g

u0 = g exp
g

dg

1 + (g , ) g
g 0

(39)

dg
0

1 + (g , ) g

=
0

dg

+ (g g )

dg

1 + (g , ) g (g g )

The second integral is convergent at g = g , so that it may be expanded in powers of g

g

A(g ) = exp
0

dg

1 + (g , ) g (g g ) g g g

= 1 + c1 g + c2 g 2 +

Performing the integrals in (39) yields u0 g A(g ) from which one deduces (g g ) g g = u0 or g = g 1 namely The dierence between the value g of g at the xed point and g is controlled by denition by the highest irrelevant exponent 3 . 2. We want to keep nite and are not allowed to take the limit 0 of the various expressions. We start from the expression of (g, ) given in 4.5.3 (g, ) = = with B () = This gives the value of g g = On the other hand, from the results of 4.5.3 and
0 / /

(40)

[g A(g )]

u0 g A(g )

(41)

g g 1 + O( )

g +

3 (/2) 2 g + O(g 3 ) 2 (4 )D/2 3 g + B ()g 2 + O(g 3 ) 2 (/2) (4 )D/2 2 3B ()

( + 1)( 1) u d u = (u + 1) ( ) (/2) 2(4 )D/2 (1 /2

we get for (r0 r0c )

r0 r0c = 2 1 + u0

31 This leads to the following expression of Z (g ) 1 d Z (g ) = 2 (r0 r0c ) 2 d from which one deduces (g ) = (g, ) dZ = dg 3 g + B ()g 2 2 B () 1 = gB () 2 2 =1+ gB () 2

u0

Subsituting the value of g in this expression leads to 1 1 (g ) = = if D = 3 3 3 For D = 3 and using (4.196), the critical exponent becomes = instead of = 7/12 = 0.583 from the -expansion. 3 = 0.6 5

Exercise 4.6.13 Energy-energy correlations

1. The only temperature dependence in the Landau-Ginzburg Hamiltonian (4.96) is hidden in r0 , as r0 = r 0 (T T0 ). We may write the partition function (4.95) in zero magnetic eld as Z= 1 dP [] exp r0 2 dD r2 (r ) dP [] exp 1 2 dD r 2 (r ) (42)

Indeed, since we are only interested in the singularity structure of the physical quantities, we do not pay attention to the proportionality factors, so that we may make the replacement r0 = r0 (T T0 ) by noticing that r0 (T T0 ) = r0 k 1 1 0 r0 0 2 k0

It is important to notice that the factor dP [] is temperature independent dP [] = 1 1 1 r0 2 D(r ) exp []2 + u0 4 2 4! 2 k0 (43)

The average values are by denition = Thus the average energy E is E= ln Z 1 = 2 dD r 2 (r ) 1 Z dP [] exp 1 2 dD r 2 (r ) () (44)

We obtain the specic heat by dierentiation C= dE 1 1 dE 1 d2 ln Z = = 2 = 2 2 dT kT d kT d 4kT 2 dD r dD r (r )(r )

c

(45)

This is of course an example of the uctuation-response theorem.

32 2. Using translation invariance, with u = r r and the scaling properties of 2 2 C = V dD u 2 (u)2 (0) c 4kT 2 V V D2 dD u u2 2 4kT 2 u < 4 kT
c

for u <

Note that the last result follows also directly from from dimensional analysis. On the other hand, from the very denition of the critical exponent and the scaling law (4.165) C |T Tc | = |T Tc |2+D 2/ D and by identication of the exponents we conclude that = d2 = D 1 Z 1 1 2 (46)

3. In the case T > Tc , where (r ) = 0, we have for the connected correlation function of (r1 )(r2 )
c

dP [] (r1 )(r2 ) exp

dD r 2 (r )

Then, by dierentiation with respect to (r1 )(r2 )

c

= + =

1 Z

1 Z Z

(r1 )(r2 )

dP [](r1 )(r2 )

1 2 (r3 ) dD r3 exp 2
c

1 2

dD r 2 (r )

1 dD r3 (r1 )(r2 ) 2 (r3 ) 2 Integrating the preceding equation over r1 and r2 gives dD r1 dD r2 (r1 )(r2 )
c

1 2

dD r1 dD r2 dD r3 (r1 )(r2 )2 (r3 )

(47)

Within a numerical factor, the LHS of (47) is the temperature derivative of the magnetic susceptibility , which is the integral of the correlation function t (2) V 2 1/ V D2d +1/ t t Let us evaluate the RHS of (47). From translation invariance The integral over u and v is limited by u, v < . On the other hand, assuming a scaling behaviour of g (u, v ) at T = Tc with scaling exponent p g (u, v ) = p g (u, v ) suggests the change of variables u = u , v = v I = V dD u dD v g (u, v ) = V 2D
u ,v < 1

(r1 )(r2 )2 (r3 ) = g (r1 r2 , r1 r3 ) = g (u, v )

dD u dD v g (u , v )

= V 2D+p By identication with the LHS we get

dD u dD v g (u , v ) V 2D+p

1 = (d2 + 2d ) Thus the scaling exponent of 2 c is nothing other than 2d + d2 . If we are interested in the average value of N s and M 2 s, it is clear that the following scaling law will hold at the critical point p = 2d D + (r1 ) ((rN )2 (r 1 ) 2 (r M )
c

= (N d +Md2 ) (r1 ) (rN )2 (r 1 ) 2 (r M )

33

Solutions for chapter 5

Exercise 5.6.4 Non-degenerate Fermi gas
1. We have shown in exercise 2.7.2 that the density of states not only depends on the dispersion law (p), but also on the dimension of space. For example in space-dimension two () = mS =D 2

2. With a uniform density of states, one immediately obtains F = N D E= 1 N F 2 (48)

The classical approximation is valid when z = e 1 (section 5.1.2); in such a case

N =z
0

d ()e = zDkT

Then the classical limit corresponds to N DkT (49)

The order of magnitude of the maximal energy is kT . The quantity DkT gives the number of levels between = 0 and = kT and provides an estimate of the number of accessible levels. The condition (49) is nothing other than the validity condition of the Maxwell-Boltzmann statistics in the particular case of a uniform density of states, namely that giving the small probability of occupation of the levels. 3. Let us make the change of variables y = x. Then, using 1, we determine immediately the dominant terms which we were looking for
+

1 1 = dx x e +1

dy
+

1 1 e ey + 1 y 1 2 e +1

(50)

x 1 dx x = 2 e +1

dy

ey

(51)

In order to demonstrate (5.86), we are going to use the symmetries of the integral. We call f (x) the Fermi function written in terms of the variable x = and we set (x) = S (x) + A (x) f (x) = fS (x) + fA (x) From f (x) = 1 f (x), we deduce fS = 1 2 fA (x) = f (x) 1 2 S,A (x) = 1 (x) (x) 2 1 fS,A (x) = f (x) f (x) 2

As the integration domain is symmetrical with respect to x = 0 we may eliminate the integrands which

34 are antisymmetric in x x in the calculation of (5.86)

dx (x)f (x) = 2
0

dx S (x)fS (x) + 2
0

dx A (x)fA (x)

=
0 0

dx S A + 2

dx A (x)f (x)
0

dx (x) +
0

dx

(x) (x) ex + 1

4. The chemical potential is given by

N =D

1 =D dx x e +1

1 dx x +D e +1

dx

1 ex + 1

Using (4.86) with (x) = 1, one obtains N = D + O(e ) while, from (48) one gets = F + O(e ) The average energy reads E = D

(52) x ex + 1

1 dx x +D e +1

dx

The rst integral is identical to that which appears in the calculation of the chemical potential; the second one features once more (5.86) with (x) = x, and one uses (51) in order to be able to use (A.51). One nally gets for the energy 2 1 D(kT )2 + O(e ) (53) E = D2 + 2 6 5. As a general rule, when one applies the Sommerfeld expansion of the Fermi distribution to a generic test function g () (g () = () for N and g () = () for E ), one obtains its Taylor expansion in terms of the (small) parameter (kT /)2 . The peculiarity of the two dimensional situation, in which the density of states is uniform, leads to a Taylor expansion which is necessarily nite. Indeed

= D
0

d + all other terms vanish 2 d + D(kT )2 6

= D
0

d ( ) + all other terms vanishing

The calculation of the preceding integrals leads indeed to the results (52) and (53). Clearly the corrections of order e cannot be taken into account in an expansion in powers of (kT /)2 . Exercise 5.6.5 gives, in the case of a Bose gas, a still more spectacular illustration of the inuence of the space dimension on the behaviour of a quantum gas. Let us also remark that the techniques developed in this exercise allow one to give an alternative proof of Sommerfeldss formula (5.29).

Problem 5.7.4 Quark-gluon plasma

35 A. 1. Because the -meson has three dierent charge states, the density of states of the -meson gas reads V 3V 2 () d = 3 4p2 dp = d (2 )3 2 2 For a boson gas with a vanishing chemical potential, the grand partition function is given by (see (5.7))

ln Q =

d () ln 1 e

(54)

V = 2 2

3 2 V 3 d = T e 1 30

One uses the integral (A.53) to obtain the nal result. From the equation of sate of an ultra-relativistic ideal gas 1 P V = T ln Q = E 3 and using (54), one easily nds the pressure and the energy density P (T ) = (T ) = 2. cv (T ) et s(T ) are obtained from (56) cv (T ) = s(T ) =
0

2 4 T 30 2 T4 10

(55) (56)

2 2 3 d = T dT 5
T

dT

cv (T ) 2 2 3 = T T 15

A direct calculation gives the energy density n(T ) = 3 2 2

d
0

2 3 (3) 3 T = 1 2

(57)

One notes that the proportionality of cv , s and n to T 3 is the consequence of a simple dimensional argument. 3. From (56) et (57) one can express the mean particle energy as (T ) = (T ) 4 = T n(T ) 30 (3)

Lets use once more a reasoning which was already developed in the discussion of (5.67). From n and , one may build two length scales. One of them, d n1/3 T 1 , is a measure of the mean distance between particles, the other one, p1 1 T 1 , provides an order of magnitude of quantum eects. If one compares these two typical distances, one observes that, whatever the temperature, /d 1. The -meson gas can never be treated as a classical gas. B. 1. The reader is referred to subsection 5.1.3. 2. Because massless particles are ultra-relativistic, the density of states of this Fermion gas reads () d = V 2 d 2

36 Expression (5.105) is obtained thanks to the change of variables x = in

ln Q =

d () ln 1 + e () + ln 1 + e (+)

(58) 1 1 + e () + 1 1 + e (+)

V = 3 2

d 3

For = 0, one gets (see appendix A.5.2) ln Q = 7 2 V 3 T 180 P (T ) = 7 2 4 T 180 n(T ) = 3 (3) 3 T 2

The entropy is deduced from the grand potential J = T ln Q s(T ) = 1 J V T =

V

7 2 3 T 45

3. Let us call respectively I et I+ the two integrals between square brackets in the second line of (58). One can cast the quantity T ln Q into the form V T ln Q = 2 3

d 3 [I + I+ ]

The condition T = 0 implies that the two integrals are to be calculated in the limit . Let us assume > 0

lim I = ( ) lim I = 0

lim I+ = 0 lim I+ = (( + ))

if > 0 if < 0

From these equation it follows that

V T ln Q = 2 3

d 3 =
0

V 4 12 2

T ln Q 1 P () = 4 = V 12 2 1 4 () = 3P (T ) = 4 2 1 T ln Q 1 n() = = 2 ||3 V T,V 3 The entropy of a fully degenerate Fermi gas of course vanishes. 4. Let us start from (5.105) and set
+

I1 =
0

dx

x3 e
(x )

+1

I2 =
0

dx

x3 e(x+ ) + 1

We do not seek to compute separately I1 and I2 , but rather we seek to cast them into a form where their

37 sum has a simple expression

I1 =

(y )3 + dy y e +1 dy (y )
3

dy
0

(y + )3 ey + 1
+

1 1 y e +1
+

+
0

dy

(y + )3 ey + 1

I2 = The sum I1 + I2 is then

dy
0

(y )3 + ey + 1

dy
0

(y )3 ey + 1

I1 + I2 =

dy (y ) + 2

y3 dy y + 62 e +1

dy
0

y ey + 1

(59)

The rst integral in (59) is elementary integral, the other two are given in appendix A.5.2. Inserting (59) in (5.105), one nally gets the expected result (5.106). C. 1. In order to estimate the contribution of the quark gas to the plasma pressure, one can simply use the results for the Fermion gas; it is only necessary to take properly into account the degeneracy linked to the three colour states 7 2 4 T Pquark = 2 3 180 For the contribution of the gluon gas, one simply uses the expression derived previously for the -meson gas 1 2 4 Pgluon = (2 8) T 3 30 The gluons have spin one, but since they are massless, their spin degeneracy gives only a factor two. The factor 1/3 comes from the fact that gluons have only one charge state. On gathering all contributions, one gets the expression (5.107) of the pressure of the quark-gluon plasma. 2. The additional term proportional to the volume in the expression of the free energy leads to a constant term in the pressure F 37 2 4 Pplasma = T B = V T 90 3. Figure 2 shows the behaviour of the pressure as a function of temperature for the -meson gas and for the quark-gluon plasma. The most stable phase is that which minimizes the free energy, or equivalently, that which has the highest pressure. The transition temperature Tc is xed by the equality of the pressures 2 4 37 2 4 Tc B = T 90 30 c a condition leading to 45 B 1/4 144 Mev 17 2 This rough estimate of Tc is in qualitative agreement with the results from lattice quantum chromodynamics. Tc = 5. The entropy densities of the two phases are dierent 2 74 T 3 P T ln Q 45 = = s(T ) = 2 T V T 2 T 3 15
1/4

for the plasma for the meson gas

38 P

mesons

plasma

P (T )

Tc B Pplasma (T )

Figure 2: Curves giving P (T ) for the -mesons and the quark-gluon plasma. At the transition, the entropy variation does not vanish and is given by s = 68 2 3 T 45 c

The phase transition is thus rst order and is accompanied by a latent heat per unit volume = Tc s.

Problem 5.7.5 Bose-Einstein condensates of atomic gases

1. One uses the fact that the potential is slowly varying over distances . The total energy is = so that 1 2 2 p + p2 y + pz + U (r ) 2 x
2 2 n2 x + ny + nz

px =

h nx etc.

2m ( U (r )) h2 Counting the number of states in the box is done exactly as in the U = 0 case, making the substitution U (r ). In order to get the total density, it suces to integrate in the whole space; however one 2 2 should limit the integration to the region where U (r ), because (n2 x + ny + nz ) 0. In the second method, one starts from the single-particle Hamiltonian H (p, r ) = The semi-classical approximation reads () = p2 d3 pd3 r U (r ) 3 h 2m 2 p2 = 3 d3 r d(p2 ) p U (r ) h 2m 2 (2m)3/2 h3 d3 r U (r ) ( U (r )) p2 + U (r ) 2m

(60)

In the case U (r ) = 0 in a volume V , U (r ) = + outside V , (60) is in agreement with the familiar result () = 2 (2m)3/2 V h3

39 2. In the case of a harmonic potential, (60) becomes () = 8 2 (2m)3/2 h3

dr r2

1 1 m 2 r2 m 2 r2 2 2 1 m 2 r2 2 m 2 r2 2 2 r2 m 2

8 2 (2m)3/2 = h3 8 (2m) h3
2 3/2

dr r 2
2 m2

0 q

dr r2

On setting u = m 2 r2 /2, one can feature the integral (5.108) and obtain () = 1 1 2 2 ( )3 (61)

3. As in (5.68), one can isolate the contribution of the ground state 0 = 3 /2 N= with N1 z e0 z

+
0

() = N0 + N1 z 1 e 1 () e (0 ) 1

d
0

Indeed, at the transition, the chemical potential takes its maximum value = 0 : the ratio N0 /N can tend toward a nite value only if z = exp(0 ). It is more convenient to choose the ground state energy as a reference energy, and thus to redene the chemical potential as follows = 0 = 3 2 z = e

The critical temperature Tc (c = 1/kTc) is then xed by the condition

N = N1 =
0

() e c 1
3

namely, taking into account (61), N= In the case T Tc , z = 1 and N = N0 + kTc (3)
3

(62)

kT

(3)

(63)

On comparing (62) et (63), one obtains the expression (5.109) of the ratio N0 /N . 4. With the data given in the text N (3)
1/3

= 1.86 1029

in MKS units, which leads to a critical temperature Tc = 1.35 K.

40 5. At T = Tc particles occupy excited states and

N=
0

() e 1

The potential U (r ) being slowly variable, one can dene, by using (60), a local density of states as n(, r ) = 2 (2m)3/2 h3 U (r ) ( U (r)) e 1

(5.110) is obtained by integrating n(, r) over . To establish this formula, we have assumed, as in question 3, that the chemical potential is zero when one chooses the ground state energy to be zero. One notices that for U (r ) = 0, the density in the center of the trap corresponds to the density (5.70) at the critical temperature. With the data given in the text and Tc = 1.35 K, one nds for the density n(r = 0) = . 2mkTc h2
3/2

(3/2) = 8.6 1019 atoms.m3

Problem 5.7.6 Solid-liquid equilibrium for helium-3

A. 1. From the expression (5.64) of the specic heat in the Debye model, one deduces that of the entropy per atom 3 4 S T = 4 k = N 5 TD and that of the energy per atom (5.112) thanks to f /T = . For a value of the temperature T = 1 K, kT 0.1 meV, while (T /TD )3 103 . The last term of (5.112) is then on the order of 6 103 meV and it is negligible compared to the other two terms which are on the order of the meV. 2. The entropy of a system of N disordered spins 1/2 is kN ln 2. One must add to f the term T = kT ln 2, because, in the solid, interactions between nuclear moments are negligible with respect to thermal motion, except when T reaches values on order of a mK. 3. To go from the free energy F to the Gibbs potential G, one must add P V ; as G = N , one must add vs P to f . This expression of does not obey the third law s k ln 2 for T 0, and it does not vanish when T = 0. In fact the spins begin to be ordered below 2 mK: solid helium-3 becomes anti-ferromagnetic below this temperature and the entropy vanishes5 B The logarithm of the grand partition function is given by V (2m)3/2 ln Q = 2 2 3 One integrates by parts 1 V (2m)3/2 P = ln Q = V 3 2 3

d
0

ln (1 + exp( ( )))

d
0

3/2 exp( ( )) + 1

5 This nuclear anti-ferromagnetism occurs for an exceptionally high temperature in the case of solid helium-3, because of an exchange interaction of quantum origin. In general nuclear anti-ferromagnetism, due to a magnetic interaction between nuclear moments, occurs only at the K level.

41 At low temperatures, the integral is computed to the order (kT /)2 thanks to the Sommerfeld formula (5.29) with the result P = = 2 (2m)3/2 5/2 15 2 3 B5/2 1+ 5 2 8 1+ 5 2 8
2

kT

kT F (P )

B=

2 (2m)3/2 15 2 3

The replacement of by F (P ) is correct to this order in T . This substitution allows one to obtain as a function of P 2 kT 2 A = B 2/5 = AP 2/5 1 4 F (P ) At T = 0 F (P ) = AP 2/5 = F (P0 ) P P0
2/5

and one nally obtains (5.114). Let us compute the compressibility6 at T = 0 (T = 0) = ln n 3 1 n = = n P P 5P

Using the expression (5.22) of the pressure yields (T = 0) = 3 5m 1 1 3m (3 2 n)1/3 = 2 (F ) = 2 2 2 2 2 2 / 3 5 3 n n (3 n) n

C. 1. The expressions of and v are derived from (5.111) v = = = 2kb T kT A P P0 P P0

1

2kb 1b kT A

kT A
2

(64) A P0 P P0
1

A = P P0

(65)

Dierentiating the entropy with respect to T allows one to obtain the specic heat cv cP = T As n = 1/v P (1) one gets (T = 0) = 2kb T kT A

1 ln n = P P

2. The transition line is given by the equality of the chemical potentials of the liquid and solid phases u0 + P 9 D kT ln 2 + vs P = A 8 P0

1b

kT A

(66)

Dierentiating this equation with respect to T , or equivalently, using Clapeyrons formula (3.97), one gets dP k ln 2 2bk (kT /A) s = = (67) dT vs v vs (4A/5P0 ) Writing (dP/dT = 0) leads to ln 2 b 12.5 (meV)1 = A 2kTm
6 Using

the Gibbs-Duhem relation gives a more direct proof of (5.117).

42 3. One may cast (67) into the form k ln 2(1 T /Tm ) dP = dT vs v

4. A =

hence (vs v ) 0.25 1029 m3 , in good agreement with the value given in the introduction: (vs v ) 0.22 1029 m3 .
5 4

v P0 1.1 meV, A/k 13 K, while

b=

ln 2 A 14 2Tm k

For T = 1 K, b(kT /A)2 0.08. In the case of an ideal Fermi gas with specic volume v , one nds F 0.55 meV, and 2 b 4.5 (meV)1 = A 4 F while 2/5 2 (2m)3/2 2/5 A= P0 2.4 meV 15 2 3 using the value = 2/5 of the ideal Fermi gas. The numerical result of the ideal gas and that of the actual liquid dier by a factor 2-3. The Debye frequency follows from the value of A

leading to D 0.75 meV, that is TD of 16 K, and a sound velocity cs =

9 D = u0 vs P0 + A 8 9 K, of the same order of magnitude as the experimental value D 2 (6 ns )1/3 102 m.s1

6. Because of the third principle, one must have s (T = 0) = (T = 0) = 0, and (dP/dT ) in (67) must vanish, since vs v = 0. The transformation being adiabatic and quasi-static, the entropy is constant and is represented by an horizontal line on gure 5.21. The method relying on the Pomeranchuk eect does not allow one to cool down below a temperature 1 mK. In the solid, the spins can be identied: they form a paramagnetic system for T > 2 mK, and the entropy per atom of the solid is dominated by the term k ln 2. In the same range of temperatures, the entropy of the liquid tends to zero, because Paulis principle implies that the ground state is unique. D. 1. One nds numerically that He B/k = 0.78 102 K, and He B/kT = 0.78 for T = 0.01 K, whence 2 Sz = tanh He B 0.65 kT

5. As (s ) 0 for T Tm , the quantity of heat which is absorbed in in the solidliquid transformation is Q = T (s ) 0.

2. As the term k ln 2 is absent in the entropy of the fully polarized solid, s 0, and from (67), the pressure is an increasing function of T and does go through a minimum as in the unpolarized state. The Pomeranchuk eect is no longer present. As expected, the polarized solid is more organized than the polarized solid.

Problem 5.7.7 Superuidity for hardcore bosons

1. Expressing the ai s and the a i s in terms of the spin operators Si gives for the Hamiltonian H = t
+ S+ i Sj + Si Sj

Sz i +
i

ij

1 2 Sz i (68)

N = t 2

ij

+ S+ i Sj + Si Sj

43 2. The occupation number operator at site i is

z ni = a i ai = Si +

1 2

(we have used a system of units where

= 1). We have for the spin system

i

1 Sz i | = | 2

Sz i |+ =

1 |+ 2

so that the action of the occupation number operator ni on the states | ni | i = 0 ni |+

i

is

= |+

Since S+ i |+ i = 0, the only possible eigenvalues of ni are ni = 0 or ni = 1. If site i is in the state | i , then the site is not occupied, and it is occupied if the state is |+ i . From these results, the total number of bosons in the ground state |0 vanishes Nb |0 = Nb
i

| i = 0

3. Because of the relation between the occupation number ni and the z -component of the spin operator Si 1 Sz i = ni 2 the average value of
i

Sz i given by Sz i =
i i i

ni

N N = Nb 2 2

Let us compute the average value of that for i = 0 The operator exp(i Sy i)

Sz i in the state (5.123). We remark (see footnote 34, chap. 5) |

i

= ei()Si |

(69)

rotates spin i by an angle around Oy

z i S i = Si cos Sx ei Si Sz i e i sin
y

(70)

Thus, with = Sz i = =
i S i | ei Si Sz i e
y y

1 1 x |Sz cos i cos Si sin | = cos = 2 2

i

where we have used |Sx | = 0. We thus obtain the average value Sz i =

i

Sz i in the state | (71)

N cos 2

from which follows the relation between Nb and N N N cos = Nb 2 2 In terms of the density = Nb /N this relation reads sin2 = 4(1 ) (72)

44 4. The number of bosons in the condensate is N0 = | a (k = 0) a(k = 0)| = 1 N 0| ei Sl

l
y

1 N

ij

|a j ai |
y

S+ j Si q

ij

ei Sq |0

This can be simplied by noting that operators referring to dierent sites commute, so that 0| ei Sl
l
y

S+ j Si q

i Si i Sj e |0 ei Sq |0 = 0|ei Si ei Sj S+ j Si e

We use once more the fact that exp(i Sy ) is a rotation operator ei S S ei Sl = Sx cos + Sz sin iSy and the property |Sx | = |Sy | = 0 to derive at once
y y 1 1 |ei S S ei Sl | = sin = sin 2 2 y y

so that N0 = or in terms of the superfuid density

N N sin2 = [4(1 )] 4 4 0 = (1 )

(73)

In general 0 < , unlike the ideal gas case where 0 = (remember that we are working at zero temperature). When 1, the bosons rarely collide, so that the system is approximately free and we regain 0 = . 6. The grand potential is given from (5.132) and the expression (68) by J = N 2
z |Si | t + | S+ i Sj + Si Sj |

(74)

ij

The second term in (74) is evaluated from the results of question 3 and the third from those of question 4 J = N N cos tN sin2 2 2 (75)

7. The minimization equation dJ /d = 0 gives at once the value of cos = 4t (76)

from which we derive the mean eld values of , 0 and J = 1 + 2 8t 0 = 1 4 8t

2

J = N

t+

2 16t

(77)

8. The state | T is obtained from the ground state |0 by a site dependent rotation of ( ) around Oy followed by a rotation of around Oz . Clearly, the rotation around Oz does not aect the calculation of the average value of Sz i , so that we may use the result of question 3
T

Sz i |

N cos 2

45 On the other hand, from the transformation properties of Sx and Sy under a rotation around Oz (or from a simple matrix multiplication) z z eiS S eiS = ei S and we get
T T |S+ i Sj | |S+ j Si | T T

= =

1 sin2 ei(i j ) 4 1 sin2 ei(i j ) 4 = 2 N sin2 1 + cos 2 L

This leads to
T ij + |S+ i Sj + Si Sj | T

where we have used the fact that there are N nearest neighbours in the x direction as well as in the y direction. Gathering all the preceding results gives the grand potential JT = N tN N cos (1 + cos ) sin2 2 2 2 2t(1 + cos ) (78)

The minimization with respect to gives the following value of cos =

9. In the limit 1 the preceding value of becomes cos 4t 1+ 1 ()2 4

It is then straightforward to compute the dierence (JT J ) per site 1 J (JT J ) = = t(1 )()2 N N But J can also be written in terms of the superuid density s and the velocity v as J 1 = ms v 2 N 2 Comparing the two expressions of J /N and using v = 2t leads to s = (1 ) (79)

10. The condensate density 0 and the superuid density s are thus identical within the mean eld approximation 0 |MF = s |MF This equality does not persist if one includes the eect of spin waves. If one takes spin waves into account, the analytical results are in excellent agreement with those of numerical simulations.

46

Solutions for chapter 6

Exercise 6.4.5 Lord Kelvins model of earth cooling
1. In the region x 0, the temperature (x, t) = 0, while in the region x < 0, the temperature (x, t) obeys the diusion equation 2 (80) =D 2 t t The most general solution of (80) is (without taking into account boundary conditions) (x, t) = A + B 4Dt
+

dx g (x ) exp

(x x )2 4Dt

(81)

in terms of an arbitrary function g (x) and of two constants A and B . This result is easily proved by showing that (81) obeys (80) thanks to 2 2 = 2 x (x x )2 and the fact that 1 x2 exp 4Dt 4Dt

obeys (80). For t = 0 we have x0 x<0 (x, t = 0) = (x, t = 0) = 0 A + B g (x) = 0

These conditions imply that one may choose g (x) = 1 and A + B = 0 for x < 0, and they allow us to compute (x = 0, t) B (x = 0, t) = A + =0 2 whence A = 0 B = 2 0 The explicit expression for (x, t) (for x 0) follows from (81) (x, t) = 0 + 0 Dt
0

dx exp

(x x )2 4Dt

(82)

2. The gradient is easily computed by making the change of variable u = x x 0 (x, t) = 0 + Dt with the result for x = 0 x

du exp

u2 4Dt

(83)

x=0

0 = Dt

(84)

3. The numerical result for the age of the Earth is t = 8.5 1015 s = 2.7 108 y 4. In three dimensions, we may use the Laplacian in spherical coordinates 2 f (r) = 1 d2 [rf (r)] r dr 2

47 valid for a function f (r) which is regular at the origin. The function (r, t) = r(r, t) obeys the equation 2 (85) =D 2 t r As in Question 1, the general solution of (85) may be written in terms of three constants A, A and B as B (r, t) = Ar + A + 4Dt
+

dr exp

(r r )2 4Dt

(86)

from which we derive A = 0 (because must be nite at r = 0) and g (r) = r so that B (r, t) = A + r 4Dt
+R

r dr exp
R

(r r )2 4Dt

(87)

The symmetric integration from R to +R ensures that (r, t) be nite at r = 0. Furthermore we have (r = 0, t) = A + B 1 + O(eR
2R
2

/(Dt)

so that A + B = 0 . Let us now compute (R, t), with the change of variable u = r r (R, t) = B A+ R 4Dt du (R u) exp u2 4Dt

2 B 1 + O(eR /(Dt) ) = A+ 2 so that A + B/2 = 0. As in question 1, A = 0 and B = 20 . The nal result is then

0 (r, t) = 0 + r Dt

r +R

du (r u) exp
r R

u2 4Dt

The reason for the symmetric integration from R to +R, which may seem surprising at rst sight, may be understood by going to Fourier space. Assume that (r, t = 0) = 1 for r R. Then the Fourier is transform (k, t = 0) =
r R

For R Dt, the second term in the square bracket is exponentially negligible, while the third one gives a relative error O(Dt/R2 ).

from which we derive the expression of the gradient 2 0 = 1 + eR /(Dt) + r r=R Dt

2R

u2 u du 2 exp R 4Dt

dr e

i k r

4 = k

r dr sin kr
0

Let us give some intermediate steps (r, t) = 4 1 r

R

d3 k ikr Dk2 t 1 e e (2 )3 k r dr
0 R

r dr sin kr
0

dk [cos k (r r ) cos k (r + r )] eDk (r r )2 4Dt exp

1 r 4Dt

r dr exp

(r + r )2 4Dt

48 The symmetric integration is recovered by making the change of variable r r in the second term of the square bracket.

Problem 6.5.2 Hydrodynamics of the perfect uid

1. Let us assume that P depends only on z and consider in the uid a parallelepiped with edges parallel to the axes, the area of the horizontal faces located at z and z + dz z et z + dz being dS . The force on the parallepiped is dP dP dz = dV dS [P (z + dz ) + P (z )] dS dz dz In the general case, the force is given by P dS = P dV

where we have used Greens theorem7 . The z -component of the force is indeed P , and Eulers equation is the consequence of Newtons law applied to a uid element. , s = ns 2. From = n, h = nh and from (3.93) one derives nd s = ds s dn and then d T s = T ds + (h )dn 1 = T ds + (h T s)dn n 1 = T ds + (h T s)d

3. Using the summation convention on repeated indices t 1 2 u 2 = = = which gives (6.104). 3. One obtains from (6.103) s =T + t t h T s t 1 2 u u + u 2 t t 1 2 u (u ) u u ( u ) u P 2 1 2 u u + P u + P ( u ) 2

Adding this equation to (6.104) and using the entropy conservation law (6.105) and that of the mass (6.64) yields K + P u (k + ) + j t = = One rewrites ( u )
7 If

P ( u ) T (su )

h T s (u ) h T s ( u ) T u ( s) u ( )

( u ) = ( u ) u
R f (r )n dS = R dV n f (r ).

n is an arbitrary xed vector, Greens theorem gives

49 and one uses once more (6.103) to compute and to derive the nal result (6.107). This equation is that of energy conservation, and it shows that the energy current is E = 1 2 u + h u = ( + P ) u 2

This expression is a particular case of (6.83) when P = P and when the heat current E is zero. The fact that this result features the enthalpy density can be explained as follows: the energy ux through a closed surface S is E dS = u dS + P u dS The rst term is a convection term, while the second term corresponds to the work done by the pressure on the uid element inside S .

Problem 6.5.4 Isomerization reactions

1. Since the number of molecules is conserved, the balance equation for the number of molecules of a given form reads dNi i = =N (kij Nj kji Ni ) dt
j =i

2. Let us consider an ideal gas made of N = N1 + N2 molecules of two dierent kinds, at a xed pressure. From (2.90b), the gas entropy is S = kN1 ln T 5/2 T 5/2 + kN2 ln + const P1 P2

where P1 = (N1 /N )P (P2 = (N2 /N )P ) represents the partial pressure of the molecules of type 1 (2). The entropy can be cast into a form which features explicitly the mixing term S = kN ln N T 5/2 kN1 ln N1 kN2 ln N2 + const P N T 5/2 0 0 0 0 kN1 ln N1 kN2 ln N2 + const P

Only the mixing term depends on the values of N1 and N2 and, at equilibrium, S0 = kN ln whence
0 0 0 0 S S0 = k N1 ln N1 + N2 ln N2 N1 ln N1 N2 ln N2

The relation (6.112) is derived in the same way. 3. On expanding to second order the logarithms in (6.112), one establishes the following inequality S S0 = k 2 x2 i Ni0 (88)

One checks on this expression that S S0 0, where the equality only holds at equilibrium. The anities are easily deduced from (88) : xi i = k 0 Ni i , transport equations are obtained by making explicit the anities in the kinetic Because ji = x i = N equations ji =
j =i

kij Nj kji Ni = =

j =i

kij xj kji xi Ni0 kji i 1 k

0 Nj kij j j =i

1 k

j =i

50 One then identies the response coecients Lii = and one checks Onsagers relations 1 0 1 Lij = Nj kij = Ni0 kji = Lji k k Because of the conservation of the total number of molecules, the three transport equations are not independent. Assuming, to x ideas , N2 and N3 to be independent, the reader will easily express the transport equations relating j2 j1 and j3 j1 to the anities. 1 k Ni0 kji
j =i

1 0 kij Lij = Nj k

51

Solutions for chapter 7

To be completed soon

52

Solutions for chapter 8

Exercise 8.5.4 Calculation of the collision time
1. Let us evaluate the collision time of a target particle with momentum p1 by computing the number of collisions per unit time in a phase space volume d3 r d3 p1 . The ux of incoming particles with velocity v2 is |v2 v1 |f (p2 )d3 p2 , and, from the denition (8.7) of the cross-section, the number of collisions per unit of time is f (p1 )d3 r1 d3 p1 d3 p2 d (|v2 v1 |, )|v2 v1 |f (p2 ) The number of collisions per unit of time suered by a particle with momentum p1 is the inverse of the collision time 1 (p1 ) = = d3 p2 |v2 v1 |f (p2 ) d (|v2 v1 |, ) (89)

d3 p2 |v2 v1 |f (p2 )tot (|v2 v1 |)

The average collision time is obtained from averaging over p1 (one must remember that one calculates the mean number of collisions, and the average which must be taken is that of 1/ (p1 )) 1 = whence (8.150). 2. This distribution function is indeed correctly normalized according to (8.1), because d3 p f (p) = p2 dp dp f (p) = n 1 (p
1)

1 n

d3 p1

1 (p
1)

f (p1 )

(90)

As (p1 ) does not depend upon the direction of p2 , one may choose p1 Oz . If denotes the angle between Oz and p2 2 2 (v2 v1 )2 = 2v0 (1 cos ) = 4v0 sin2 2 assuming the cross-section to be velocity independent. By substituting the values of |v1 v2 | and of f in (8.150) 1 (p
2)

= = =

tot

n 2v0 4
1

dp1 sin

2 2

tot nv0 4 tot nv0 3

1

d(cos ) sin

The collision time and the mean free path are given by = 3 4nv0 tot = 3 4ntot (91)

3. Let us introduce the center-of-mass momentum P = p1 + p2 of the two colliding particles, and let us make the change of variables 1 1 p2 = P p p1 = P + p 2 2 where p is the reduced momentum. We get from (6.117) 1 1 = tot n d3 P d3 p f 1 P +p f 2 1 P p 2 2|p| m

53 and then f 1 P +p f 2 1 P p 2 = n2 1 2mkT

exp

P2 4mkT

exp

p2 mkT

The rst exponential features the total mass 2m and the reduced mass m/2. The integrals second one the are performed thanks to the change of variables P = 2 P et p = p / 2 and by using the denition of the mean velocity for a Maxwell distribution, v = 1 2mkT
3/2

d3 p

|p| p2 exp m 2mkT

Problem 8.6.2 Electron gas in the Boltzmann-Lorentz model

A. Since the cross-section does not depend on the angles, the collision term (8.35) is expressed as a function of (v ) = 1/(nd v (v )) as follows C [f ] = nd v d (v ) (r, p) = 1 f (r, p) f (r, p , ) f 4 (v )

because, by assumption (p = (p , )) (r, p , ) = 0 d f One then derives the expression for f = (v )(v )f0 f and the expressions for the currents from (8.49). When does not depend on v , one can extract from the integrals in (8.61) and (8.62) the factor v (v ) = = 1/(nd ), which is the mean free path of the electrons determined by the collisions with the scattering centers. This leads to the expressions for the currents d3 p v f0 E = d3 p v f0 N = 3 3 B. 1. The expressions for the currents and for the coecients Lij are obtained as in section 8.2.5 by writing f0 as a function of (1/T ) and of (/T ). Let us make in LN N the change of variables = p2 /2m, md = pdp 8 m LN N = d f0 3 k 0 so that = 1 for LN N . For LEN one nds an analogous expression with = 2, while = 3 for LEE . Using the denitions given in the text, one computes the integrals in the form LN N LN E = LEN LEE = = = nT m 2 nkT 2 = 2kT LN N m 6 2 3 k T = 6(kT )2 LN N m

2. From these expressions, the diusion coecient is derived thanks to (6.27) D= k LN N = kT n m

54 and the thermal conductivity coecient thanks to (6.50) = 1 2 = 2 nk 2 T L L L EE N N EN T 2 LN N m

3. See in section 6.2.2 the discussions which follow equations (6.57) and (6.61). From (6.57) el = and the ratio /el is k2 =2 2 T el q 4. When an external force is present8 , f0 is given by f0 = ( + V ) V f0 by (F = V in (8.32)) and f = v r f0 (r V ) p f0 f v taking into account p f0 = vf0 = v ( + V ) f0 f v This equation shows that the results in the presence of an external force are deduced from those without . There is no change in the external force through the substitution + V in the expression for f expression for LN N : LN N = LN N , while LEN LEE = = 3km 3km
d3 p p( + V )f0 = L N E + V LN N 2 d3 p p( + V )2 f0 = L EE + 2V LEN + V LN N

q2 nq 2 LN N = T m

one nds

5. One remarks that p f0 = v (df0 /d), whence el = q = q 2 d3 p v f d3 p df0 q2 v E (E v ) = v d 3m d3 p p df0 d

The change of variables p leads to el = 8q 2 m 3 d 8q 2 m df0 = d 3kT d f0 () = q2 LN N T

for a Maxwell-Boltzmann distribution where (df0 /d) = f0 . C. 1. On using (5.19), F = for T = 0 and pF = (2m)1/2 n=
8 By

8 (2m)3/2 3 h3

convention = r unless otherwise stated.

55 2. The number of possible microscopic states in d3 r d3 p is 2 3 3 d r d p h3 where the factor 2 comes from the spin degrees of freedom, and one has f d3 r d3 p states. The probability occupation of a state is then h3 f /2. The probability for the state p to be accessible is (1 h3 f /2). The integral in the collision term will be proportional to f 1 h3 h3 f f 1 f 2 2 = f f

and the collision term remains unchanged9 . 3. Given the Sommerfeld formula (5.29) f0 f0 that is f0 = 4. For LN N , s = 1 and LN N = = 2T h3 f0 2 1 2 3 ( ) + 3 ( ) h 6 f0 2 2 3 ( ) 3 ( ) h 6 + T 1 T ( ) + 2 ( ) 6 2

whence c1 = 2T /h3 and c2 = 2 /(6 2 ). From here on, the computation is identical to that of part B. d ( ) = 0. There only remains the contribution of ( ) 16T m = 3 h3

d ( ) =

16T m T n nT F = = 3 h3 pF m

We have used the relation between the chemical potential and the density, as well as = F vF . Only the electrons located close to the Fermi surface p = pF contribute to transport, because the electrons inside the Fermi sea (p < pF ) cannot be thermally excited. The reasoning is analogous to that used in section 5.2.2 for the specic heat.

5. The diusion coecient is given by (6.27) D= and the electrical conductivity by (6.57) el = q2 q2 LN N = n T m F 1 T n LN N =
T 2 F 3 m

This last result is identical to that of the classical gas, provided one replaces by F , the collision time on the Fermi surface; on the contrary, the result for D is quite dierent from the classical one. 6. The electronic density is n = 8.4 1028 m3 , which implies F = 2.1 1014 s.

7. The coecient LEN is given by LEN = 16T m 3 h3

d 2 ( ) + 2 3 2 2

2 ( ) 6 2

= LN N

1+

9 On the contrary, the collision term in the Boltzmann equation would be modied, leading to the so-called LandauUhlenbeck equation.

56 while LEE is LEE = = 16T m 3 h3 2 LN N

d 3 ( ) + 2 2 2

2 ( ) 6 2

1+

These results allow us to compute the thermal conductivity coecient = 2 LN N T2 1+ 2 2 2 1+ 2 3 2 2

2

1 2 2 1 2 2 k LN N = k nT F 3 3m

Comparison with the electrical conductivity el = q 2 LN N /T leads to the Franz-Wiedemann law 2 k2 = T el 3 q2 The factor 2 in question B.3 has become 2 /3. The method followed in question B.5 leads once more to the electrical conductivity using 2 df0 3 ( ) d h because 2 8q 2 m q2 16q 2 m el = 3 = LN N d ( ) = h 3 3 h3 T

Problem 8.6.5 Electrical conductivity in a magnetic eld and the quantum Hall eect
A. Given the assumptions in this problem, the Boltzmann-Lorentz equation reads ] = nd vtot (v ) f (F p )f = C [f 1. In the case where B = 0, and neglecting terms of second order with respect to f0 , (92) yields = q (p)(E p )f0 f We remark that (f0 ) = p f0 = p from which follows f0 f0 = v (92)

= q (p)(v E ) f0 f On the other hand, the electrical current density is given by el = qN = q that is el = q 2 d3 p (p)v (v E ) d3 p v f (93)

f0 4 q 2 = 3 m2

d p p4

f0 E

At T = 0 the equilibrium distribution for an electron gas is f0 () = 2 (F ) h3

57 so that

2 m 2 f0 (p pF ) = 3 (F ) = 3 h h pF

(94)

One nally obtains for the electrical conductivity el = 2 4 3 p h3 3 F q2 m F

The term between square brackets is nothing other than the electron density. 2. In the presence of a magnetic eld, the full Lorentz force (8.167) must be taken into account in (92). However, if B and E are collinear, the magnetic force will average to zero, since, on average, the electrons move in the direction of E : the conductivity will remain unchanged. In the general case, the magnetic force appears as a correction term in (92). Indeed, since f0 v , the term (v B ) f0 = 0. It is thus which appears in the form (8.169) of the Boltzmann-Lorentz equation. the perturbative term f Even without knowing its explicit general form, the values taken by C in two limiting cases B = 0 and E = 0 are easily obtained. Indeed, if B = 0, one reverts to the preceding case, and thus C = q (p)E ; if E = 0, the electrons suer only the Laplace force q (v B ) which averages to zero because of the isotropy of the scattering, and thus C = 0. 3. By making C explicit in (8.169), one gets qv E f0 f0 q (v B ) p v C = 1 f0 vC

Let us examine the gradient term p v C f0 =vC 2 f0 1 f0 v+ C 2 m

The rst term is eliminated on taking the scalar product with v B . Using the invariance of the mixed product under circular permutations, one nally obtains qB f0 C f0 v qE C = v m (p) Equation (8.170) follows because this relation must be satised whatever the value of v . From the two vectors E and B , one builds the natural basis (E, B, B E ) on which one can decompose the vector C . The coecients of this decomposition are given by (8.170) q q 2 B m E+ + q BE m B+ + q m (B E ) = 0

where we have used B (B E ) = (B E )B B 2 E . In the case under study, the electric and magnetic elds are perpendicular. The preceding system leads to = q 1 + 2 2 =0 = q q m 1 + 2 2

, one goes easily from (8.170) to (8.169). Knowing the expression of C , and thus of f 4. Taking (94) into account, relation (8.171) can also be written as q p = 2 E + ( E ) (p pF ) f h3 1 + 2 2 pF

58 Putting this result in the expression for el (93), one gets el =

1 nq 2 F 2 (E + F E ) m 1 + 2 F el = 2 (E + F E ) 1 + 2 F

The only non vanishing components of the conductivity tensor are xx = yy = xy = yx el 2 1 + 2 F el F qB el = = F 2 2 1 + 2 F 1 + 2 F m

(95)

The minus sign in xy = yx follows from the Onsager relations in the presence of a magnetic eld. B 1. Equation (8.173) is nothing other than Newtons law, in which, in addition to the Lorentz force, there also appears a friction force which is introduced in order to take collisions into account. This friction force governs the dynamics for short times, and vanishes exponentially with a characteristic time scale given by the parameter . el In this model, the electrical current densities are simply given by j = qn v . One thus obtains the same conductivity tensor as above, with the only dierence that the Boltzmann-Lorentz model was able to x the value of the characteristic time
el 2. Using (8.172) and (95), the condition jy = 0 gives

EH =

yx Ex = F Ex yy

In this situation, one easily checks that the conductivity does not depend on the Laplace force
el jx = xx

xy yx yy
2 xy 2 xx

Ex Ex

= xx 1 + = el Ex
el One gets for jx as a function of EH el jx =

nq el EH = B EH F EH l B = el jx l d nd q (96)

and thus RH =

The fact that the Hall resistance does not depend on the relaxation time shows that this resistance arises because a stationary situation is reached in the transverse direction. As a consequence of the Laplace force, charges accumulate on the surface of the metal. The potential dierence which follows from this accumulation of charges generates a transverse electric eld, which, after a time on the order of F , reaches an equilibrium value EH . In the framework of the Drude model, this means that the friction force vanishes, and that Fy = 0, that is qEy q vx B = 0 so that
el qnEy jx B=0

One recovers the preceding expression for RH .

59 C. 2. In two dimensions, the level density is uniform (see exercise 2.7.2) () = 2mS h2

2. Since the level density is uniform, the number of states in an interval ,which corresponds to the gap between two consecutive Landau levels, is given by g = () = mS qB = S h h

3. Taking the degeneracy into account, there are g electrons up to level with a surface density nS = qB h

Setting nd = nS in relation (96), one gets for the quantum Hall coecient RH = h q 2

The preceding reasoning is not enough to justify the existence of plateaus in the Hall resistance. A general argument relying on gauge invariance allows one to to show rigourously the existence of those plateaus (see Laughlin, [71]).

Problem 8.6.6 Specic heat and two-uid model for helium II

A. 1. For T TD the calculation of the internal energy features the integral (A.53) E
phonon

= =

V (2 )3

V (kT )4 pc = d p pc e 1 2 2 ( c)3
3

dx
0

ex

x3 1

2 V (kT )4 30 ( c)3

One deduces the specic heat by dierentiating E phonon with respect to T

phonon CV =

1 E phonon V T

=
V

2 2 k 4 3 T 15 3 c3

2. For T < 5000. Hence, replacing the Bose distribution by the 8.5, so that exp((p)) > 1 K, (p) > Maxwell distribution is quite justied. Figure 8.10 shows that rotons can only exist in a limited range of values of p centered around p0 , p0 < p< p0 + . The integral which occurs in the calculation of the roton contribution to the internal energy depends only upon a limited integration range V = 2 2
p0 +

roton

e 3

p0

d p p2 +

(p p0 )2 2

exp

(p p0 )2 2

However, since p2 0 /2 100, the Gaussian factor in the integral exhibits a narrow peak around its mean 2 2 value, which allows one to replace p2 by p2 ) and to 0 (the error being on the order of /(p0 ) 10

60 integrate from + on p : E
roton

V 2 2

e 3

dp p2 0 +

(p p0 )2 2

exp p2 2 1+ 1 kT 2

(p p0 )2 2

V p2 0 e 2 2 3 V 2 2 V 2 2 p2 0 3

dp

p2 2

exp
+

exp kT kT

2kT

dx 1+

kT 2 x exp x2

p2 0 exp

2kT

The third line is obtained by setting x = /2(p p0 ). The Gaussian integrations are performed with the help of (A.37). In the temperature range which concerns us here, the term kT / 1/10 in the last bracket could be neglected. The specic heat is obtained by dierentiation of the internal energy. It reads at the leading order in T
roton CV

k 2 2

p2 exp 3 0

kT

2kT

kT

1+

kT 3 + 4

kT

B. 1. Each atom of the uid has in the reference frame R a momentum qi = qi mv

N

phonon roton For T 0, CV which varies as T 3 dominates over CV which vanishes exponentially. However, phonon 1 roton for T = 1 K, CV 3.5 kJ.K is smaller by a factor 3 than CV 11 kJ.K1 .

H =
i=1 N

1 (qi mv )2 + 2m 2
2 qi 2m N

i=j

U (ri rj ) 1 2
N

=
i=1

qi
i=1

v+

mv 2 +
i=1

1 2

i=j

U (ri rj )

(97)

1 = H P v + M v2 2 If the uid is in its ground state with energy E = E0 and momentum P = 0 in the reference frame R, going form R to R leads from (97) to an energy E in R E = E0 + 1 M v2 2 (98)

Assume that an elementary excitation is created in the uid with momentum p and energy (p) in R, going from R to R implies, again from (97) E = E0 + (p) p v + 1 M v2 2 (99)

because the uid energy in R is E0 + (p) and its momentum P = p. Subtracting (99) from (98) gives (8.178) (p) = (p) p v As the gas of elementary excitations is in equilibrium with the walls of the container, one must use the energy in the reference frame R in order to write the equilibrium distribution n (p, v ) = 1 e (p) 1 = n( p v )

61 In the reference frame R, the momentum density of quasiparticles is given by g= d3 p p n( p v ) (2 )3

If the ow velocity is small, namely if p v , then n( p v ) n() p v dn() d

The rst term on the RHS does not contribute to the momentum density and g= The nal result (8.179) g= 1 3 d3 p 2 dn p v (2 )3 d d3 p dn() p (p v ) 3 (2 ) d (100)

is obtained thanks to the symmetry argument leading to (8.60). From g = n v and using (8.179), one deduces 1 d3 p 2 dn n = p 3 (2 )3 d For a phonon gas where = cp, one integrates by parts the previous equation with the result 4 4 n = 2 3c (2 c)3 The energy density in R reads, to lowest order in v = d3 p 4 n() = 3 (2 ) (2 c)3

d 3 n()

(101)

d 3 n()

(102)

4 3 c2 The dispersion relation of the phonon gas is identical with that of a photon gas, in which case we have shown that the pressure is one third of the energy density (see (2.29) or (4.35)) n = P= The phonon equation of state is thus P= 1 3

On comparing (101) and (102), one gets

c2 n 4 Each phonon propagates with a velocity c along a direction p which depends on its momentum: p = p/p. The expression for the energy current is jE = d3 p cp n( p v ) c2 (2 )3 d3 p 2 dn 1 p ( p v) p = c2 (2 )3 d 3 d3 p 2 dn p ( p v) p 3 (2 ) d d3 p 2 dn() p v = c2 g (2 )3 d

= c2

Taking into account E = (c2 g ) = c2 (n v ), the continuity equation for the energy reads + c2 (n v ) = 0 t

62 Dierentiating this equation with respect to time yields a wave equation for the energy density with speed c/ 3. Indeed 2 + c2 (n v ) 2 t t 2 c2 2 t2 3 = 0 = 0 (103)

where we have used Eulers equation (6.70) which becomes for v 0 1 (n v ) = P = t 3

Equation (103) shows that since T 4 , any local heat excess will propagate with velocity c/ 3: The second sound waves may thus be interpreted as temperature waves. We have just exhibited one of the remarkable properties of Helium II, namely that it can transport heat not through diusion but through wave propagation, a mechanism which is much more ecient than diusion. Second sound is a sound wave in the gas of elementary excitations. The previous considerations can be immediately transposed to the case of a photon gas, and in fact to any gas of massless particles at thermal equilibrium. If one replaces the sound velocity by the light velocity, the preceding calculation shows that the sound velocity in any ultrarelativistic dilute gas is c/ 3.

Problem 8.6.7 The Landau theory of Fermi liquids

I Static properties 1. At zero temperature, the chemical potential is by denition = E (N + 1) E (N ) where N is the number of particles and E (N ) the ground state energy for N particles. Then is the energy needed to add one particle in the lowest available momentum state, namely pF , and = (pF ). In other words, is the extra energy of the system if one adds a quasiparticle on the Fermi surface. 2. The interaction term between two particles can be non zero only if the distance between these two particles is less than the range a of the interaction. The probability for this to happen is a3 /V . Then the expression of (p) is, as it should be, independent of V . As |f (p)| 1, the order of magnitude of the second term in the expression for (p) is () . It is independent of V , because () V , and is of order . If f (p) is isotropic, we can take the z -axis along p. If ( , ) denote the polar and azimuthal angles of p , we have d3 p P (cos )f (p ) =
l l

p dp d cos( ) d P (cos )f (p ) 0 d3 p f (p )

since the only non zero contribution comes from the = 0 term. We remind the reader of the standard result for Legendre polynomials
1

d(cos )P (cos )P (cos ) =

1

2 2 + 1

P0 (cos ) = 1

P1 (cos ) = cos

2V d3 p f (p ) h3 and comparing the previous two equations, one derives N = (p) = (p) + 0 N

On the other hand

63 A higher order term in the expansion of (p) would be of the form 2V h3

2

d3 p1 d3 p2 (2) (p; p1 , p2 )f (p1 )f (p2 ) 2 ()(2) 2

and thus of higher order in . (p) (p)] is of order 3. From the denition of f (p) we may write, using the fact that [ f (p) = f (p) f0 (( (p) (p)) + (p)) f (p) f0 ((p)) ( (p) (p)) = f (p) + ((p) ) 4. The excitation at temperature T , kT , is f (p) = = f0 (p, T ) f0 (p, T = 0) = 2 (kT )2 ( (p) ) 6 1 ( (p)) 1 + exp[( (p) )/kT ] 2V h3 f0 (p)

d3 p (p, p )f (p )

where we have used the Sommerfeld approximation (5.29); here must be identied with kT , since the Fermi distribution is rounded o around on a distance of order kT (see gure 5.2). We have neglected the variation of with T , as this assumption will be shown later on to be self-consistent. Note that one can write the zero temperature equilibrium distribution either as f0 (p, T = 0) = ( (p)) or f0 (p, T = 0) = ( (p)), since there are no quasiparticles at equilibrium and zero temperature: f (p)|T =0 = 0. The interaction term in the quasiparticle energy is proportional to E given by E =

d(cos ) d P (cos )
0

p dp ( (p ) )(kT )2 p dp ( (p ) )(kT )2
2

d(cos ) d P (cos )
2

(kT ) () We have made use of the fact that the replacement of (p ) by (p) in the -function is consistent with our approximations: | | is on the order of kT , which plays the r ole of . Hence we may replace by in the calculation of thermodynamic quantities if we neglect terms of higher order in T . For example we may compute the term proportional to T of the specic heat. Let us rst obtain the explicit expression for the density of states () = = = 2V h3 8V h3 d3 p ((p) )

p2 dp ((p) ) =

8V dp h3 d

p=pF

p2 F

V m p F 8V m 2 p = F h3 p F 2 3

The term proportional to T in the specic heat is obtained following 5.2.2: one has only to replace the free mass m by the eective mass m and one nds CV = V k2 T 2 k2 T m pF () = 3 3 2

64 Let us show the consistency of our approach. Following again 5.2.2, we easily show that k2 T T so that it is consistent to neglect the T -dependence of the chemical potential if we limit ourselves to terms of order zero and one in T . 5. At zero temperature, when the chemical potential varies from to + d, the excitation f (p) is by denition f (p) = ( + d (p)) ( (p)) d ( (p) ) It is crucial to understand that in going from to + d, we have added quasiparticles and we must use (p) and not (p). On the other hand we also have from question 2 f (p) = f (p) + 0 N ((p) ) since the error made in replacing (p) by (p) in the -function is of higher order in . Integrating over phase space leads to ()d = N + ()0 N = N (1 + 0 ) or, equivalently () N = 1 + 0 The T = 0 compressibility is obtained form (1.42) in terms of the density n = N/V = 1 n n2 = 1 N n2 V

T =0

T =0

Finally one uses the expression for () in terms of the eective mass m of the preceding question to derive the nal form of the result m p F 1 = 2 2 3 n (1 + 0 ) II Boltzmann equation 6. From the conservation of the number of quasiparticles and Liouvilles theorem, we obtain in the absence of collisions f + p r f (r ) p f = 0 t 7. Let us evaluate the various terms in the above collisionless Boltzmann equation. First r (r, p, t) = while p f = p f0 = p ( (p)) p ((p) ) = vp ((p) ) m 2V r h3 d3 p (p, p )f (r, p , t)

Inserting these results in the Boltzmann equation and adding the collision term, we get 2V f + vp r f 3 r t h d3 p (p, p )f (r, p , t) vp ((p) ) = C [f ]

Note that we must take f0 as the r -independent equilibrium distribution, and not a r -dependent local equilibrium distribution, because we want to have r f = r (f0 + f ) = r f

65 We can now combine the last two terms in the right hand side by using the denition of f (p) in question 3 and obtain f + v r f = C [f ] t 8. The dierence n = n n0 between the equilibrium density n0 and n is obtained by integrating f over p 2 n(r, t) = 3 d3 p f (r, p, t) h Thus, integrating the Boltzmann equation from the previous question over p and using the conservation of the number of particles in a collision, which implies d3 p C [f ] = 0 we get the continuity equation n + r = 0 t =

2 d3 p vp f h3 The second term in p is due to the fact that the moving quasiparticle drags along with it particles of the surrounding medium. Another expression for the current is obtained if one expresses f in terms of f (r, t) = 2 h3 d3 p vp f (r, p, t) + vp ((p) ) 2 2V h3 h3 and we may write (r, t) = with p = vp + 9. In order to compute the integral Ip Ip = d3 p vp ((p ) )(p, p ) 2V h3 2V h3 d3 p (p, p )f (r, p , t)

provided the current is dened as

The second term in the preceding equation is rewritten by exchanging the integration variables p and p d3 p f (r, p) 2 h3 d3 p vp ((p ) )(p, p ) d3 p p f (r, p, t)

d3 p vp ((p ) )(p, p )

we take the z -axis along p and denote by ( , ) the polar angles of p (or vp ) vp = vp (sin cos , sin sin , cos ) Only the component along p survives the integration Ip = p p dp d(cos )d
l 2

P (cos ) cos ((p) )

This equation is simplied thanks to the identity for Legendre polynomials (see question 2)
1

d(cos )Pl (cos ) cos =

1

2 l1 3

with the nal result

1 h3 1 vp 1 d3 p ((p ) ) = vp 1 () 3 3 2V We thus have two possible expressions for the current Ip =

66 (r, t) = (r, t) = 2 h3 2 h3 d3 p d3 p p m 1+ 1 () f (r, p, t) 3

p f (r, p, t) m

and comparison of the two expressions leads to the relation between the mass of the free particle m and the eective mass m 1 1 1 1 = 1 () = 1 + 1 m m 3

Problem 8.6.8 Calculation of the coecient of thermal conductivity

1. Using the relation between pressure and density for an ideal gas n(z ) = P = P (z ) kT (z )

allows one to write the local equilibrium distribution (8.115) in the form f0 (z ) = P (z )p2 [ (z )]5/2 exp 2m (2m)3/2

Let us compute the drift term of the Boltzmann equation ((p) = p2 /(2m)) f0 = z 5 f0 (p)f0 2 = z T (z ) 5 T kT (p) f0 2 z

so that the drift term is, with cP = 5k/2 Df = vz T f = (cP kT ) f0 z T (z ) z

2. We multiply the linearized Boltzmann equation ((p) cP T )pz = by f01 (p1 ) and integrate over p1 to get d3 p (p)pz f0 (p)(p) = Tm 4
4

mT

d3 pi W f02
i=2

d3 pi W f01 f02 ()2

i=1

where we have used the condition that the uid is at rest d 3 p pz f = d3 pf0 (p)(p)pz = n pz = 0

and the symmetry properties of the Boltzmann equation.

E 3. The energy current jz is E jz =

d3 p f vz =

d3 p (p)vz f0 (p)

T z

From the denition (6.18) of the coecient of thermal conductivity, we can identify = 1 m d3 p (p)pz f0 (p)(p)

67 With |X = ((p) cP T )pz = L| and from the denition of the scalar product F |G we can write = |X = |L| We dene the positive semi-denite scalar product (, ) (, ) = (|L|)= T 4
4

d3 pi f01 W f02
i=1

Then, restricting ourselves to functions orthogonal to the zero modes, we dene [] [] = (, )2 (, )(, ) = (, ) = (, ) (, )

4. From the symmetry properties of the problem, we may write (p) = A(p)pz . In order for the average velocity of the gas to vanish, we have seen previously that the function A(p) must obey d3 p f0 (p)(p)pz = d3 p p2 z A(p)f0 (p) = 0

This condition also ensures that (p) is orthogonal to the ve zero modes. Writing A(p) = A(1 p2 ) we determine from the condition 0=
2 d3 p p2 z (1 p )f0 (p)

p4 dp (1 p2 ) exp 5 2 5 2 m 1
5/2

p2 2m 7 2 m
7/2

23/2 = 23/2 with the result = /(5m). 5. Starting from

25/2 5 m 2

|X = ((p) cP T )pz = L| we obtain |X = = 1 m 1 m d3 p ((p) cP T )pz f0 (p)A(1 p2 )pz

2 d3 p (p)p2 z f0 (p)A(1 p )

where we have used the result of the preceding question. The explicit calculation of |X gives |X = = = An 2 m2 2m
3/2 2 d3 p p2 p2 z (1 p ) exp 3/2

p2 2m 9 2 m
9/2

4 An 3 2 m2 An 2

2m

25/2

7 2

7/2

27/2

68 Let us nally compute |L| . We rst evaluate

2 2 2 = (1 p2 1 )p1z + (1 p2 )p2z (1 p3 )p3z (1 p4 )p4z 2 2 2 = p2 1 p1z + p2 p2z p3 p3z p4 p4z

where we have taken momentum conservation into acount. Using the center-of-mass variables p and P , we have for example 1 1 2 2 pz + Pz P p2 1 p1z = p + p P + 4 2 so that = 2 (p P )pz (p P )p z and |L| = A2 T 2
4 i=1

d3 pi W f01 f02 (p P )pz (p P )p z

Now we can write the square backet of the preceding equation as

[. . .]2 = (pi Pi )(pj Pj )p2 z + (pi Pi )(pj Pj )pz 2(pi Pj )(pi Pj )pz pz 2

Let us take the angular average over the direction of P at xed p, p Pi Pj and [. . .]2 = 1 2 P ij 3

1 2 2 2 2 p pz + p p z 2(p p )pz pz 3 It remains to take an angular average over the directions of p and p when the angle between p and p is xed 2 2 4 2 [. . .]2 ang,P ,p ,p = P p (1 cos ) 9
ang,P

Gathering all preceding results, using the change of variables (p1 , p2 ) (p, P ) and d3 p3 d3 p4 W leads to |L| = = 4A2 T 2 n2 9m 32A2 T n2 25 2m m 3
3

2p m

d (p, )

d3 P P 2 exp m
1/2

P 2 4m

d3 p p5 exp

p2 m

(p, )(1 cos2 )

tr

where in obtaining the last line we have assumed the cross-section to be independent of p. We nally obtain | |X |2 25 k mkT = = |L| 32 m tr

69

Solutions for Chapter 9

Exercise 9.7.1 Linear response: The forced harmonic oscillator
1. From our convention for the time-Fourier transform f ( ) = dt eit f (t)

we have t i , and the equation of motion for the forced oscillator becomes in Fourier space
2 x( ) = 2 i + 0

f ( ) m

The susceptibility is by denition ( ) = The poles of ( ) are located as follows 1. Underdamped case: < 20 = i 2. Overdamped case: > 20 = i 4 2 1 1 20 2 1i 2
2 4 0 1 2 1/2

1 x( ) = 2) f ( ) m( 2 i + 0

1/2

2. In the strong friction limit 0 / 1, The susceptibility ( ) is large only for

2 0 =

0 0

2 Thus we may neglect 2 with respect to 0 and write

( ) = We also have

1 1 2 i = 1 i m 0 ( ) =

2 0

1 2 m0

where the second expression holds in the strong friction limit. 3. Let us write the response as x(t) = and the average work dW dt
T T

1 1 2 2 2 2 2 2 m ( 0 ) + + 2

dt (t )f (t t )
T

1 = T

1 dt f (t)x (t) = T

dt f(t)x(t)
0

The boundary terms from the integration by parts may be neglected as they tend to zero in the limit T . We thus have dW dt
T

1 = 4T

i f e
0

it

it f e

dt

i (tt ) dt (t ) f ei(tt ) + f e

70 On performing the t-integration and taking the large T limit, only the terms which are -independent survive, and the result of the t-integration is
i 1 |f |2 eit eit = |f |2 sin t 4 2

Then, using (t) = 2i(t) (t), we obtain dW dt = |f |2 2 |f | 2

2

i |f |2 dt (t) sin t = 2 dt (t) eit eit =

dt (t) sin t

1 |f |2 ( ) 2

where we have used the symmetry property (t) = (t) in obtaining the last line. One can also use the method of exercise 9.7.6.

Exercise 9.7.6 Dissipation

We write ( ) = ( ) + i ( ) where and are in general, but not always, real functions. In all cases they are related through a Hilbert transfrom P d ( ) ( ) = ( ) = ( ) ( ) = ( ) (104)

which shows that the parities of and are opposite

In the usual case is an odd function of , so that the upper sign holds in the preceding equation. Using the summation convention over repeated indices, we write Ai (t) = and dW dt to obtain dW dt 1 = 4T
T

dt ij (t )fj (t t ) 1 T
T 0

dt fi (t)Ai (t)

i fi eit fi eit dt

ij (t ) fi ei(tt ) + fi ei(tt ) dt

After performing the t-integration, we are left with dW dt i it it ij (t )dt fi fj e fj fi e 4 i = f f ij ( ) fj fi ij ( ) 4 i j i [ij ( ) ji ( )] = fi fj 4 =

Now, time translation invariance combined with hermiticity yields the symmetry properties
ij ( ) = ji ( ) ij ( ) = ji ( )

71 so that we are left with

dW dt dW dt

= =

i 2i f ij ( ) fj 4 i 1 2
fi ij ( )fj i,j

and the nal result in the form

Exercise 9.7.9 Strong friction limit: the harmonic oscillator

1. We rst take the time-Fourier transform of the equation of motion which gives the relation between X ( ) and F ( ) F ( ) 2 2 i + 0 X ( ) = (105) m and we then dene the function XT (t) as XT (t) = X (t) if |t| T 2 XT (t) = 0 if |t| > T 2

From the Wiener-Kinchin theorem (9.170) we get

XT ( )XT ( ) FT ( )FT ( )

= T Cxx ( ) = T CF F ( )

Given the relation (105) between X ( ) and F ( ), we deduce that Cxx ( ) = 1 CF F ( ) 2 )2 + 2 2 m2 ( 2 0 CF F ( ) = 2A and one obtains Cxx ( ) = 1 2A 2 )2 + 2 2 m2 ( 2 0 2A/m2 + 22

If the autocorrelation of the force is given by (9.125), then

2. In the strong friction approximation, the autocorrelation function Cxx ( ) is peaked around = 0. One notes that Cxx ( = 0) = 1 4 0 Cxx
2 0 4 0

(106)

1 4 2 0

2 so that Cxx ( ) is peaked at = 0 with a width 0 / . Then the width x of Cxx (t) is of order of

2 0

One should note the following hierarchy in the strong friction limit
2 0 0

, we deduce 3. From P = m2 X Cpp ( ) = m 2 Cxx ( )

72 and Cpp ( ) = ( 2

2A 2 2 )2 + 2 2 0

The autocorrelation function Cpp ( ) has maxima around = 0 and we note the following remarkable values 2 1 0 1 1 = 2 Cpp Cpp ( ) 2 Cpp (0 ) = 2 2 2 Thus the width of Cpp ( ) is of order and the autocorrelation time p 1/ . In the strong friction limit x 1 p 2 0

This result means that the particle acquires its limit velocity over a short time on the order of p , and then it evolves much more slowly under the combined inuence of the harmonic force and friction over a time on the order of x . The momentum distribution is almost Maxwellian for t p , because the particle has enough time to reach equilibrium in momentum space. 4. Neglecting the inertial term in the equation of motion leads to
2 X ( ) = i + 0

F ( ) m

and one gets directly the strong friction limit (106) of the autocorrelation function. The corresponding equation of motion in t-space is 2 = 0 + F (t) X m
2 2 One can read directly on this equation that x = /0 . The x-space viscosity is = 0 / .

Problem 9.7.1 Inelastic light scattering from a suspension of particles

1. Let us start from the standard Maxwell equation in the presence of a polarized material B = c2 A = From the decomposition of the polarization P = 0 ( 1)E (r, t) + P (r, t) we get 1 P E E 1 P + = + 0 t t t 0 t and the equation for A becomes c2 ( A) 2 A = 1 P E + t 0 t 1 P 2A + t2 t 0 t A t (107) E 1 P + 0 t t

= where we have used

E = Choosing the gauge

+A= 0 c2 t

73 leads to the equation for the vector potential 2 A 1 P 2A = 2 c2 t2 0 c t (108)

The equation for the scalar potential is obtained from E = or 2 + A 1 P = t 0 c2 t 1 P 0

Using the gauge condition (107), we easily obtain the equation for 2 1 2 = ( P ) 2 2 c t 0

2. One checks at once that the expressions for and A = Z A= Z c2 t

are consistent with the gauge condition (107). The following partial dierential equation for Z 2 Z 2Z 1 = P 2 2 c t 0 (109)

gives back the equations for and A. The expression for E as a function of Z is E= 2Z + ( Z ) c t2 (110)

3. Let us take the time-Fourier transform of Z (r, t) Z (r, ) = Then equation (109) reads, with k = /c 1 P (r, ) 0 (111) dt eit Z (r, t)

2 + k 2 Z (r, ) =

The varying polarization is located in a small region in space, while we are interested in measuring the scattered electromagnetic eld at large distances. We shall use a technique which is familiar from scattering theory, by introducing the Green functions G (r ) 2 + k 2 G (r ) = 4 (r) It is easily checked that the functions eikr r are solutions of the preceding equation. An easy proof is as follows. The Laplacian of a spherically symmetric function is 1 d2 [rf (r)] 2 f (r) = r dr2 G (r ) =

74 provided f (r) is regular at r = 0. We thus write 2 The rst function is regular at r = 0 2 while 2 1 d2 eikr eikr 1 = eikr 1 = k 2 2 r r dr r 1 = 4 (r) r eikr 1 1 eikr = 2 2 r r r

Equation (111) is solved by using the Green function with outgoing waves 1 Z (r, ) = 40 eik|rr | P (r , ) d r |r r |
3

(112)

Now, since P is conned to a small region in space, r r , we may use the approximation |r r | r r r so that (112) becomes Z (r, ) 1 eikr 40 r d3 r eikr P (r , )

(113)

In order to compute E , we write P = P e 0 , where e 0 is the polarization of the incident electric eld. We observe that eikr ik eikr eikr e0 r ) e 0 = ( 2 r r r and the second term may be neglected since kr 1. Similarly e 0 r ) so that the electric eld is proportional to 2 e 0 k 2 r e 0 r ) c2 We nally get the expression for E E (r, ) 2 eikr 40 c2 r
( P (r , ) k ) d3 r eikr k

ik eikr eikr k 2 ( e0 r ) r r r

2 2 e )) = e ) e 0 k 0 k 0 e 0 k 2 c c2

4. In order to prove the Wiener-Kinchin theorem, we start from the result of exercise 9.7.8
T /2 T /2 +T

I (T ) =
T /2

dt1
T /2

dt2 g (t) = T
T

dt g (t) 1

|t| T

We dene
T /2

XT ( ) =

dt e

it

XT (t) =
T /2

dt eit X (t)

75 Then
T /2 XT ( ) XT ( ) T /2

=
T /2

dt1
T /2 T /2

dt2 X (t1 )X (t2 )

= T
T /2 T /2

dt eit

|t| T

XT (t)XT (0)

T
T /2

dt eit XT (t)XT (0) = T ST ( )

5. The electromagnetic power is given by the ux of the Poynting vector 0 E B = 0 c|E |2 k But P being parallel to e 0 , one can write P = e 0 P and use ( ))2 = 1 ( )2 = ( )2 = sin2 (k e0 k e0 k e0 k . From the Wiener-Kinchin theorem where /2 is the scattering angle, cos = e 0 k |E |2 = while E (r, ) E (r, ) = = 1 40 1 40 c c
4

1 T

d E (r, ) E (r, ) 2

1 )2 d3 r d3 r eik(r r ) P (r ) P (r , ) ( e0 k 2 r 1 )2 P (k, ) P (k, ) ( e0 k r2

dP /(d d ) is the ux of the Poynting vector per unit frequency and through a surface r2 d, so that 0 c dP = d d (40 )2 6. P (k, ) = = so that dP 0 c 2 2 E0 = d d (4 )2 c
4

1 )2 P (k, ) P (k, ) ( e0 k 2T

e 0

dt d3 r ei(kr t) ei(k0 r0 t) n(r, t)

0 E0 n(k k0 , 0 ) = 0 n(k , ) )2 ( e0 k n(k , )n(k , ) 2T

where we have used the fact that n(r, t) is a real function. Using once more the Wiener-Kinchin theorem n(k , )n(k , ) = V T Snn (k , ) We thus get the power per unit volume of target 0 c dP = d d 32 3 c
4

)2 2 E 2 Snn (k , ) ( e0 k 0

76 7. The expression for Snn (k , ) stems from a hydrodynamic approximation relying on the continuity equation and Ficks law. In this case one knows from (9.40) that (k , ) = (k ) Dk 2 + (Dk 2 )2

and from the classical form of the uctuation-dissipation theorem (9.28) , we get Snn (k , ) = 2 Dk (k ) 2 + (Dk 2 )2
2

8. One can write the emitted power in two dierent forms dP 0 c = d d (4 )2 and dP d d = = The coecient A is thus A= Taking into acount d eit 1 |t| = e 2 2 + 2 2 we obtain for the autocorrelation function of E E (r, t) E (r, o) ei0 t eDk
2

2 2 E0 sin2

Dk 2 (k ) 2 + (Dk 2 )2

0 cr2 ) E (kr, ) E (kr, 2T 0 c SEE (k , ) 2

4 2 2 E0 sin2

1 8

1 (k ) A(0 )

A(0 )

9. Since E is a Gaussian random function, its moments of order 4 can be expressed in terms of its moments of order 2. Taking into account that E is parallel to a xed direction and that E (t) E (t) = 0 [E (t) E (t)] [E (0) E (0)] = E (t) E (t) E (0) E (0) + E (t) E (0) E (0) E (t) This leads to I (t)I (0) = A2 (1 + eDk and taking the Fourier transform A2 2 ( ) + 2 4Dk 2 + (2Dk 2 )2
2

it is much easier to measure a deviation on the order of 500 Hz with respect to zero, than to measure the same deviation with respect to 1014 Hz.

Problem 9.8.2 Light scattering from a simple uid

1. One sets in equation (6.91) g = mnu, = mn and neglects the advection term (u )u which is quadratic in u. Furthermore, let us examine the deviation (u) from equilibrium (u) = u + u

77 The rst term in this equation is of second order in the deviations with respect to equilibrium and may be neglected. A similar reasoning holds for the spatial derivatives, so that we obtain the Navier-Stokes equation, valid to rst order in the deviations from equilibrium t g + P 1 mn + 2 ( g ) g=0 3 mn

In the expression (6.83) for E , one may replace P by P since the neglected terms give contributions which are quadratic in u, and one may also replace in u the energy density by the energy density in the rest frame of the uid, which is denoted by in the present problem. 2. Equation for gL : from 0 = ( gL ) = ( gL ) 2 gL +P g T mn

we immediately get (9.195). We write the expression for the energy current E = from which we derive the continuity equation t + Combining with (9.193) we get (9.197). 3. The derivation follows that of (9.35), with D /(mn). Writing the response function of the momentum density )T (k ) L (k ) ( k k k g g = k we get for T (see 9.40) T (k, ) = 2 k 2 /(mn) T (k ) 2 2 4 + k (mn)2 +P ( gL ) 2 T = 0 mn

P P n + n n Let us rst examine the second term on the right hand side of the preceding equation. From P = P and S = P S S =

4. We start from

1 E V S =

T V

We get P
n

V P T S

To transform the second term of P we start from V d = T dS ( + P )dV For d = 0 we get 0=T whence S P P V
V

dP + T

S V

dV ( + P )dV + P V

+ P P T S

78 where we have made use of (1.36) in the last step. Finally we use V = n n V to obtain P n = V ( + P ) P nT S + P n

Collecting all contributions we get the nal expression for P P = P n

S

n +

V P T S

5. Let us give the interpretation of q . At xed N we have = so that +P n n = = +P n 1 E E 2 V + V V V n V T 1 (E + P V ) = S V V E V = 1 V E 2 E V V n = n V V

We see that q is indeed related to the entropy variation per unit volume. 6. The determinant of the matrix M is z det M = kmc2 T ik 2 n S k m z + ik 2 DL 0 0 V P k T S n z + ik 2 mncV z bk igk 2 ak z + idk 2 0 0 fk z + ihk 2

where we have simplied the notations in order to compute the determinant by setting a g = = 1 m T n b = mc2
S

d = DL

f =

V P T S

h= mncV

7. The expression for the determinant reads det M = z + ihk 2 z (z + idk 2 ) c2 k 2 + iaf gk 4

In the low temperature approximation we may neglect the last term in the preceding expression. The zeroes of the determinant are then obvious 1. z = O(k 2 ) or more precisely z = ihk 2 = i k 2 = iDT k 2 mncV

Because CP CV in the low temperature limit, we may write DT = mncP

79 2. z = ck + , | | ck One solves the equation for z to rst order in z = ck i DL k 2 2

In the general case, one must evaluate, recalling that the sound velocity c2 = (P/)S , = mn af g c2 = = n P V T T P P S T n
n

V P T S

Remember that we are working at constant N , which means that constant n implies constant V . Then the Maxwell relation T P T P = = S V V S S n V S implies af g V T T = c2 T P S V S while S T T V T = = P S P T S P CP T P where we have used (1.28). From an analogous calculation T V so that = T P CV T

af g V T P = 2 c CP CV T CP CV = T P T

V T V T

We nally use
V P

to obtain

af g V (cP cV ) = (CP CV ) = c2 CP CV mncP cV

where CV,P = mnV cV,P , mcV,P being the specic heat per particle. We may now revert to the calculation of the zeroes of the determinant. 1. Solution z = O(k 2 ). Neglecting terms of order k 4 we get z i h because h af g c2 k2 = i k2 mncP

af g = c2 mncP

2. Solution z = ck + . One solves to rst order in 2 idk 2 i or af g 2 k c2 cP 1 cV

i = k 2 DL + 2 mncP

80

Problem 9.8.4 It o versus Stratonovitch dilemma

1. Preliminary result. In the evaluation of the Chapman-Kolmogorov equation, one must compute the following integral I= dy f (y ) (x y g (y )) x = [y + g (y )]y=y0 Dierentiating this equation with respect to x gives 1= Using the standard formula (x y g (y )) = and writing y0 = f (x) we get 1 dy (y y0 ) = |1 + g (y )| dx
y =y 0

The argument of the delta-function vanishes for y = y0 given by

dy (1 + g (y )) dx

y =y 0

(y y0 )

The value of y0 is given by the equation

(x y g (y )) = f (x) (y f (x)) y0 = x a(y0 ) B D(y0 ) )

2 We solve this equation by iteration to order and B (remember that B is formally of order

y0

x a(x) B D1/2 x B D1/2 (x) x a(x) B D1/2 (x) +

1 2 B D (x) = f (x) 2

Since f (x) = 1 a (x)

D (x) 1 2 1 B 1/2 D (x) + B 2 D (x) 2

the Chapman-Kolmogorov equation (9.144) leads to J = 1 D (x) 1 B 1/2 + B 2 D (x) 2 D (x) 2 1 2 1 2 D (x)) (y x) + B D (y x) (y x) + a(x) + B D1/2 (x) B 2 2 dy P (y, t|x0 ) 1 a (x)

2 = 2 we get for the average over all realizations of b(t) Using the average values B = 0 and B

= =

P (1 a (x) + D (x)) a(x)P (x) + 2D (x)P (x) + D(x)P (x) 2 [a(x)P (x)] + 2 [D(x)P (x)] P (x) x x P 2 = [a(x)P ] + 2 [D(x)P ] t x x

where P (x) = P (x, t|x0 ). This leads at once to the F-P equation

2. We now have X (t + ) y = a(y ) + D1/2 [y + (1 q )(X (t + ))] a(y ) + D

1/2 t+ t t+ t

b(t )dt b(t )dt

(1 q )D (y ) (X (t + ) y ) (y ) + 2D1/2 (y )

81 Expanding to order and taking the average value over all realizations of b(t) yields X (t + ) y = a(y ) + (1 q )D (y )
t+ t

dt
t

t+

dt b(t )b(t )

= (a(y ) + (1 q )D (y )) In other words, the only modication with respect to question 1 is in the drift term a(y ) a(y ) + (1 q )D(y ) so that the F-P equation becomes P t = = 2 [a(x)P (x)] (1 q ) [D (x)P (x)] + 2 [D(x)P (x)] x x x 1q q D (x) (D (x)P (x)) [a(x)P (x)] + x x x

Problem 9.8.5 Kramers equation

1. Let us compute the relevant moments of p and x. 1. p p(t + ) p(t) = (F (x) p) + Taking an average over all values of f gives 1 1 p = lim [p(t + ) p(t)] = F (x) p 0 2. x x(t + ) x(t) = = so that 1 m
t+ t t t+

f (x, t )dt

1 p(t )dt = m
t+

t+

dt
t t

p+
t

p (t )dt

p 1 + m m

dt p +
t t

dt f (x, t ) + O(2 )

1 p x = m

3. p2 (p(t + ) p(t))
2

= (F (x) p)2 2 + 2(F (x) p)

t+ t+ t

t+ t

f (x, t )dt

+
t

dt dt f (x, t )f (x, t )

Taking the average, the rst term is of order 2 , the second one vanishes and the third one is equal to 2A so that 1 p2 = 2A 4. xp. It is easy to see that this moment is of order 2 .

82 5. x2 It is also easy to see that this moment is at least of order 2 . Taking into account the results of exercise 9.7.11 with x1 = p and x2 = x, we have in the notations of this exercise p A1 = F (x) (x)p A2 = A11 = A(x) m from which the F-P equation follows p 2 P = ([F (x) p]P ) P + 2 (A(x)P ) t p x m p We have taken from the beginning x-dependent and A. One remarks that this choice does not imply any modication in the F-P equation since the p-derivative does not act on an x-dependent function. With A(x) kT (x) = m we arrive at p 2P P = (x) + + F (x) (pP ) + mkT 2 t m x p p p

2. Let us dierentiate (x, t) with respect to time = t dp P = t dp p P P 2P F (x) + (x) (pP ) + mkT (x) 2 m x p p p

An integration by parts shows that many terms vanish, for example dp F (x) P = [F (x)P (p)] = 0 p

because the distribution P (p) vanishes at innity. One is thus left with = t dp p P j = m x x

In the strong friction limit, the particle reaches almost instantaneously its limiting momentum p given by p =0 F (x) p(x) = (x) The momentum follows a Maxwell distribution centered at p = p(x), whence P (x, p; t) (x, t) [p p(x)]2 1 exp 2mkT (x) 2mkT (x)

3. Let us multiply both sides of the F-P equation by p and integrate over p dp p2 P + m x dp F (x)p P = (x) t dp p (pP ) + (x) p dp mkT (x) 2P p2

Once more some terms vanish after an integration by parts and we are left with m where we have used dp p j (x, t) + K (x, t) F (x)(x, t) = (x)mj (x, t) t x (pP ) = p dp (pP ) p = p dp (pP ) = mj

83 K is of course related to the average kinetic energy of the particle 1 p2 K= 2 2m Setting 2 = mkT we have 1 2 2 dp p2 exp (p p)2 2 2 = 1 2 2 dp (p + p)2 exp p 2 2
2

= p2 (x) + 2 (x) = p2 (x) + mkT (x) so that K (x, t) (x, t) kT (x) + p2 (x) m

In the strong friction limit, the particle reaches its limiting speed in a time on the order of the velocity correlation time 1/ , and then it evolves slowly with a characteristic time 1/ 1/ j |j | |j | t Neglecting j/t, we obtain an approximate expression for j j (x, t) = and if kT (x) p2 (x)/m K (x, t) = [kT (x)(x, t)] = [m (x)D(x)(x, t)] x x x where we have used Einsteins relation which gives the x-dependent diusion coecient in x-space D(x) = The nal expression for the current is then j (x, t) = 1 F (x)(x, t) m (x)D(x)(x, t) m (x) x j + =0 t x 4. If F (x) = 0 and is x-independent, the expression for the current reads j (x, t) = and the F-P equation is 1 [A(x)(x, t)] m2 2 x kT (x) m (x) K 1 F (x)(x, t) m (x) x

and the F-P equation for (x, t) reads

1 2 = 2 2 2 [A(x)(x, t)] t m x

The corresponding Langevin equation is x = with the It o prescription. A(x) (t) m (t) (t ) = 2 (t t )

84 On the contrary, if A is x-independent the expression for the current reads A A (x, t) j (x, t) = m (x) x m (x) and the F-P equation is A A = (x, t) t x m (x) x m (x)

The corresponding Langevin equation is A x = (t) m (x) with the Stratonovitch prescription.

(t) (t ) = 2 (t t )

Two dimensional Ising model Black: 10X10, Red: 16X16, Green: 24X24, Blue: 36X36

0.5

E/Ns
1.5 2

2
T

30

two dimensional Ising model Black: 10X10, Red: 16X16, Green: 24X24, Blue: 36X36

20

10 0

3
T

2.5

two dimensional Ising model

Triangle: Exact Tc=2.2692; Line is a linear fit, intercept: 2.26

2.45

2.4

Tc(L)
2.35 2.3 2.25

0.02

0.04

0.06
1/L

0.08

0.1

0.12

two dimensional Ising model, (/)exact = 1.75

100

Squares: Simulation, Line: Power law fit, slope =/=1.782

(Tc(L))

10

10
L

100

0.05

two dimensional Ising model Black: 10X10, Red: 16X16, Green: 24X24, Blue: 36X36

0.04

0.03

L
0.02 0.01 0

25

15

5
L
1/

15
(TTc(L))

25

35

45


1
two dimensional Ising model Black: 10X10, Red: 16X16, Green: 24X24, Blue: 36X36

0.8

0.6

<m>
0.4 0.2 0

3
T

two dimensional Ising model: Intersection at T=2.27 Black: 10X10, Red: 16X16, Green: 24X24, Blue: 36X36

1.5

L <m>

0.5

2
T

1.5

two dimensional Ising model: Used =0.125 and =1 Black: 10X10, Red: 16X16, Green: 24X24, Blue: 36X36

L <m>
0.5 0

10

5
L
1/

0
(T2.27)

10


2
two dimensional Ising model Black: 10X10, Red: 16X16, Green: 24X24, Blue: 36X36

1.5

0.5

2
T

Two dimensional Ising model: L=10, 16, 24, 36 Line is a fit: C=a + b ln(L), a=0.1135, b=0.522 2.0

1.8

C(Tc)

1.6

1.4

1.2 2.0

2.5

3.0

3.5

4.0

Ln(L)


two dimensional Ising model Black: 10X10, Red: 16X16, Green: 24X24, Blue: 36X36

0.6

0.4

S
0.2 0

2
T

Quantum SHO: =0, =2, m=1, =0.05 is fixed Line: Exact, points simulation

<E>

Quantum SHO: =0, =2, m=1, =0.05 is fixed Line: Exact, points simulation

<n>
0 0

Quantum SHO: =0, =2, m=1, =20 Line is a fit: Intercept=1.036, exact=1.037

1.1

1.05

<E>

0.95

0.9

0.05

()

0.1

0.15

0.008

Quantum SHO: =0, =2, m=1 Solid: =20, dashed (both): =0.33, long dashed: =0.25

0.006

P(x)

0.004

0.002

0 x


0.01
Quantum AHO: =2, m=1, =20, Lt=400 Bottom to top: =0, 1, 2, 3, 4

0.008

0.006

P(x)
0.004 0.002

0 x

0.8

Kinetic energy: Virial (circles) and direct (star)

0.7

Kinetic energy
0.6 0.5

10
0

=2, =0,1,2,3,4. E= 1.98, 2.156, 2.289, 2.374, 2.47

Top to bottom: =0,1,2,3,4: G() ~ exp([E1E0])

10

10

0 1 2 3 4

E1 2.98 3.198 3.3668 3.4866 3.6126

G()
10
3

10

10


0.02
Quantum AHO: =2, m=1, =8, Lt=400 Bottom to top: =20, 5, 1, 0.1 and 0.1

0.015

P(x)

0.01

0.005

0 x