• Aucun résultat trouvé

Physique statistique hors ´ equilibre – Examen

N/A
N/A
Protected

Academic year: 2021

Partager "Physique statistique hors ´ equilibre – Examen"

Copied!
6
0
0

Texte intégral

(1)

Universit´ es Paris 6, Paris 7 & Paris-Sud, ´ Ecole Normale Sup´ erieure, ´ Ecole Polytechnique Master 2 iCFP – Parcours de Physique Quantique

Physique statistique hors ´ equilibre – Examen

Lundi 9 janvier 2017

R´ edigez vos r´ eponses ` a cette partie de l’examen sur une copie S´ EPAR´ EE.

Vous pouvez r´ ediger en fran¸ cais si vous le souhaitez.

Subject : Non-intersecting fluctuating lines and 1D fermions

Introduction : We study a model of non-intersecting Brownian lines in dimension 1 + 1, using the mapping on the 1D fermion gas. We consider lines going only forward in one of the two spatial directions (axis τ in Fig. 1), which can thus be considered as a fictitious “time”. Each line is described by a displacement field x n (τ ) ∈ R for τ ∈ [0, L]. Such models appear in various contexts : directed polymers, fluctuating interface models (for atomic terraces, wetting problems,..), etc.

Single line : The basic ingredient of the model is the functional for the elastic energy E[x(τ )] = c

2 Z p

2 + dx 2 − dτ

' c 4

Z L 0

dτ (∂ τ x(τ )) 2 ,

where it is assumed that distorsions are small, |∂ τ x| 1. The positive coefficient c is the elastic coefficient (the stress). The Gibbs measure exp

− E[x(τ )]/T is identified with the Wiener measure for the Brownian motion with diffusion constant D = T /c. Such a Brownian trajectory can be conveniently described by the Fokker-Planck equation

∂ τ P τ (x|y) = D ∂ x 2 P τ (x|y) , (1) where P τ (x|y) is the conditional probability for the Wiener process (free Brownian motion).

Part C is almost independent of parts A and B.

τ

1

y

2

y

N

y

1

y

2

y

N

... x

x

x

2

(τ) x (τ)

1

(τ)

N

...

...

0 L

y

Figure 1 : N non-intersecting elastic lines (directed polymers, fluctuating interfaces,...).

A. Properties of a single line.– We first study the fluctuations of a single line.

1/ Eq. (1) describes the single line. Recall its solution on R .

Hint : The initial condition is lim

τ→0

P

τ

(x|y) = δ(x − y).

(2)

2/ Fluctuations.– We consider the situation where the line issues and ends at the same point, say x(0) = x(L) = 0. Justify

hx(τ ) 2 i = A L

Z

R

dx P L−τ (0|x) x 2 P τ (x|0) (2)

What is the normalisation constant A L ? Compute the variance hx(τ ) 2 i and plot it neatly as a function of τ ∈ [0, L]. Where are the fluctuations maximum ?

B. N non-intersecting lines.– We now consider N lines (x 1 (τ ), · · · , x N (τ ))

def

= ~ x(τ ), still described by the free diffusion, with however some constraints for non-intersection

τ P τ (~ x|~ y) = D ~ ∇ 2 P τ (~ x|~ y) with x 1 < x 2 < · · · < x N , (3) where ∇ ≡ ~ (· · · , ∂/∂x i , · · · ).

1/ Recall the general form for the decomposition of the propagator over the spectrum of the operator H N = −D ~ ∇ 2 , denoted {E n , Ψ n (~ x)} n=0, 1,···,∞ (with R

d N ~ x Ψ n (~ x)Ψ m (~ x) = δ n,m ).

Hint : Since the Fokker-Planck operator is Hermitian here, we can simply use the mapping of (3) onto the imaginary time Schr¨ odinger equation with Hamiltonian H

N

.

2/ We assume that the N lines start from ~ x(0) = ~ y and reach the same set of values at final

“time”, ~ x(L) = ~ y (like in Fig. 1). Express formally the average of an observable hA(~ x(τ ))i at

“time” τ in terms of the propagator P τ (~ x|~ y). Argue that for τ and L − τ → ∞, this average can simply be written in terms of the ground state Ψ 0 (~ x) of the operator H N .

C. Compressibility and 1D fermionic gas.– We study the response of N 1 lines to an external perturbation. In the limit L → ∞, their statistical properties are controlled by the ground state Ψ 0 (~ x) of the operator H N . Thus, we can use the mapping between (3) and the Schr¨ odinger equation for a gas of N free spinless fermions of “mass” m = 1/(2D) = c/(2T ) (we set ~ = 1), where T is the original temperature of the lines and c their elastic constant.

We now consider the fermions, described by the Schr¨ odinger equation i∂ t Ψ(~ x, t) = H N Ψ(~ x, t) : fermions occupy 1D plane waves ψ k (x) = 1

L e ikx of energy ε k = 2m k

2

. In a box of finite size L , the wave vector k is quantised (it is sufficient to remember that P

k → L /(2π) R dk).

1/ We consider a single fermion : the density operator is ˆ n(x) = δ(x − x) and its Fourier ˆ component ˆ n q = e −iqˆ x , where ˆ x is the position operator. Give the matrix element h ψ k |ˆ n qk

0

i.

2/ We now consider the fermion gas. A perturbation is introduced under the form of a scalar external potential V (x, t) = (1/L ) P

q V q (t) e iqx . The corresponding Hamiltonian is H ˆ pert (t) =

N

X

i=1

V (ˆ x i , t) = Z

dx V (x, t) ˆ n(x) = 1 L

X

q

V q (t) ˆ n −q ; (4) from now on ˆ n q = P

i e −iqˆ x

i

is the density operator of the fermion gas. Its response is h n(x, t)i ˆ V = hˆ n(x)i +

Z

dt 0 dx 0 χ(x − x 0 , t − t 0 ) V (x 0 , t 0 ) + O(V 2 ) (5) i.e. hˆ n q (t)i V = n L δ q,0 + R

R dt 0 χ e q (t − t 0 ) V q (t 0 ) + O(V 2 ), where n = hˆ n(x)i is the mean density.

a) Express χ e q (t) as an equilibrium correlation function.

b) Write its Fourier transform χ(q, ω) = e R

R dt χ e q (t) e iωt as a sum over contributions of plane

waves. Analyse its analytical structure. Interpret the position of the poles of the integrand.

(3)

3/ Static compressibility.– The connection to the problem of fluctuating lines is made more transparent by considering the static response χ(q)

def

= χ(q, e 0) at zero temperature.

a) We denote by f (ε k ) the Fermi-Dirac distribution. Show that χ(q) = −

Z dk π

f (ε k )

ε k − ε k+q (6)

where the integral is a Cauchy principal part integral (see appendix).

b) The zero temperature fermion gas involves the Fermi energy ε F = 2m 1 k F 2 controlling the number of fermions N. Compute explicitely χ(q).

c) Kohn anomaly.– Show that lim q→0 χ(q) is related to the density of states at Fermi level ρ 0 = πk m

F

. Analyse χ(q) for q → 2k F and q → ∞. Plot neatly −χ(q)/ρ 0 as a function of q/(2k F ).

Express k F and ρ 0 as a function of the parameters for the fluctuating lines (n, T and c).

d) Bonus.— Argue that the length of the lines L coincides with the inverse temperature of the fermions β ferm = 1/T ferm (go back to question B.2). Discuss the validity of the T ferm = 0 approximation, and then that of the small distorsion approximation made in the very begining (express the two conditions in terms of the parameters of the lines). For finite T ferm , what is the value of χ(2k F ) (in terms of T, n and c) ?

Appendix :

• Convention for Fourier transform in dimension d : ϕ q =

Z

Vol

d d x e −iqx ϕ(x) & ϕ(x) = 1 Vol

X

q

ϕ q e iqx −→

Vol→∞

Z d d q

(2π) d ϕ q e iqx (7)

• In pratice, Cauchy principal part integral R

dx x−x ϕ(x)

0

= Re R

dx x−x ϕ(x)

0

±i0

+

can be calculated with

− Z b

a

dx ϕ(x) x − x 0

= lim

η→0

+

Z x

0

−η a

+ Z b

x

0

dx ϕ(x) x − x 0

, (8)

where ϕ(x) is a regular function and x 0 ∈ [a, b] (if x 0 ∈ / [a, b], this is the usual integral).

• We recall that the grand canonical average of a commutator of many body operators, sums of one particle operators, of the form ˆ A = P N

i=1 a ˆ (i) and ˆ B = P N

i=1 ˆ b (i) , is h[ ˆ A , B]i ˆ = X

α

f α h ϕ α | ˆ a, ˆ b

| ϕ α i = X

α, β

(f α − f β ) a αβ b βα , (9) where the sum runs over one particle stationary states. a αβ = h ϕ α |ˆ a| ϕ β i is a matrix element of the one particle operator and f α = f (ε α ) denotes the occupancy of the individual eigenstate.

The problem is inspired by the article :

P.-G. de Gennes, Model for fibrous structures with steric constraints, J. Chem. Phys. 48(5), 2257–2259 (1968).

Solutions sur la page du cours : cf. http://lptms.u-psud.fr/christophe_texier/

(4)

Christophe Texier January 2017 Non-intersecting fluctuating lines and 1D fermions – Solutions

A. Single line.

1/ Solution is P τ (x|y) = 1

4πDτ exp

(x−y) 4Dτ

2

. Note that here, D has dimension [D] = [x] 2 /[τ ] = [length]

2/ Averaging takes the form

hx(τ ) 2 i = R

R dx P L−τ (0|x) x 2 P τ (x|0) P L (0|0)

(if one removes the x 2 in the numerator, one gets obviously h1i = 1, as required). The Gaussian integral gives the fluctuations of the Brownian bridge, hx(τ ) 2 i = 2D τ 1 − L τ

. Fluctuations are maximum at the middle τ = L/2, where the line is less constrained hx(L/2) 2 i = DL/2 = T L/(2c). Discussion : fluctuations increase with temperature and decrease with the stress. Sca- ling with the length is δx ∼ √

L as expected.

Dimensional analysis : [T ] = [energy] and [c] = [force] = [energy]/[length]. Ok.

B. N lines.

1/ The decomposition of the propagator over the spectrum of H N = −D ~ ∇ 2 is P τ (~ x|~ y) = h ~ x | e −τ H

N

| ~ y i =

X

n=0

Ψ n (~ x) Ψ n (~ y) e −τ E

n

.

2/ Averaging of a function of coordinates at “time” τ has the same structure as for one line.

Assuming a discrete spectrum, for long lines we write P τ (~ x|~ y) ' Ψ 0 (~ x) Ψ 0 (~ y) e −τ E

0

, hence hA(~ x(τ ))i =

N ! R ∞

−∞ dx 1 R ∞

x

1

dx 2 · · · R ∞

x

N−1

dx N P L−τ (~ y|~ x) A(~ x) P τ (~ x|~ y)

P L (~ y|~ y) (10)

' N ! R ∞

−∞ dx 1

R ∞

x

1

dx 2 · · · R ∞

x

N−1

dx N Ψ 0 (~ y) Ψ 0 (~ x) e −(L−τ)E

0

A(~ x) Ψ 0 (~ x) Ψ 0 (~ y) e −τ E

0

Ψ 0 (~ y) Ψ 0 (~ y) e −LE

0

.

Finally

hA(~ x(τ ))i '

τ & L−τ→∞ N ! Z ∞

−∞

dx 1

Z ∞ x

1

dx 2 · · · Z ∞

x

N−1

dx N |Ψ 0 (~ x)| 2 A(~ x) . (11) which corresponds to average in the ground state of H N . The factor N ! is required if we choose to normalise the eigenstates as R

R

N

d~ x |Ψ n (~ x)| 2 = 1.

From non-intersecting lines to 1D fermions : The constraints for non-intersection can be des- cribed by wave functions which vanish at coinciding points Ψ(· · · , x i , x i+1 = x i , · · · ) = 0.

Moreover, being restricted to the sector x 1 < x 2 < · · · < x N , we can construct a basis of wave functions antisymmetric under exchange of particles, i.e. fermionic eigenstates.

Remark : from the point of view of the diffusion, the Dirichlet boundary condition implies that normalisation is not conserved, i.e. R

d~ x P τ (~ x|~ y) ∼ e −E

0

τ decays as time grows. However, dividing

by P L (~ y|~ y) in (10) implies that averaging is taken only over non-intersecting trajectories which

have survived.

(5)

C. Compressibility and 1D fermionic gas.– We now study the 1D fermion gas (the begining of the part C reproduces a result of the lectures).

1/ Matrix elements for the one-body density operator : h ψ k |ˆ n q | ψ k

0

i = R dx

L e −ikx−iqx+ik

0

x = δ k

0

,k+q .

2/ Compressibility.– We apply one of the main results of the lectures : χ e q (t) = − i

L θ H (t) h[ˆ n q (t), n ˆ −q ]i (12) in terms of many-body density operator. The correlator can be expressed in terms of one-body matrix elements (relation of the appendix) :

χ e q (t) = − i

L θ H (t) X

k, k

0

(f k − f k

0

) |h ψ k |ˆ n qk

0

i| 2 e i(ε

k

−ε

k0

)t where f k ≡ f (ε k ) = 1/ e β(ε

k

−µ) + 1

. A Fourier transform over time finally gives the formula obtained in the exercice session (TD 8) :

χ(q, ω) = e 1 L

X

k

f k − f k+q

ω + ε k − ε k+q + i0 + (13)

where the 0 + in the denominator is a regulator. It ensures that singularities of the response function are in the lower complex plane of the frequency (causality). Besides we recognise the energy of a particle-hole excitation ε k+q − ε k .

3/ Static compressibility.– For zero frequency, we could drop the regulator which is not needed χ(q) = L 1 P

k

f

k

−f

k+q

ε

k

−ε

k+q

. We prefer to use the symmetry of the spectrum ε k = ε −k and split the sum (which requires to keep the regulator)

χ(q) = 1 L

X

k

f k − f k+q

ε k − ε k+q + i0 + = 1 L

X

k

f k

ε k − ε k+q + i0 + − 1 L

X

k

f k ε k−q − ε k + i0 +

= 1 L

X

k

f k

ε k − ε k+q + i0 + + 1 L

X

k

f k ε k − ε k+q − i0 + finally, using Ω±i0 1

+

= PP 1 ∓ iπ δ(Ω),

χ(q) = 2 L

X

k

PP f k

ε k − ε k+q = − Z dk

π

f k ε k − ε k+q

The Fermi-Dirac distribution at T ferm = 0 is f k = θ H (k F − |k|). Writing ε k+q − ε k = m q (k + q/2) we find

χ(q) = − m q π −

Z +k

F

−k

F

dk

k + q/2 = m q π ln

q − 2k F q + 2k F

(14) We obtain easily the limiting behaviours :

χ(q) ' −ρ 0 ×

 

 

1 for q k F

1

2 ln |q−2k 4k

F

F

| for q ∼ 2k F 2k

F

q

2

for q k F

(15)

where ρ 0 = πk m

F

is the DoS per unit volume at Fermi level. The logarithmic singularity is called

the Kohn anomaly. It is due to a particle-hole excitation where the particle and the hole are

close to the two Fermi points.

(6)

0.0 0.5 1.0 1.5 2.0 2.5 0.0

0.5 1.0 1.5 2.0 2.5

q  2k F

- Χ 0 H qΡ 0

Figure 2 : Zero temperature compressibility with logarithmic divergence at q = 2k F . The mapping on the problem of fluctuating lines with density n is

 

 

k F = π n m −→ 2T c ρ 0

2

c T n

Bonus :

• If we come back to the formula for the averaging (10), we see that the length L of the lines play the same role as the inverse temperature for the fermions. This is more clear is one traces over the initial/final positions

hA(~ x(τ ))i

R d~ y

−→ Tr

e −L H

N

A(~ x) Tr

e −L H

N

(16)

• The temperature of the fermions is mapped onto the inverse of the length T ferm = 1/L and the Fermi energy ε F = k 2 F /(2m) = π 2 n 2 T /c. The limit of the degenerate gas (T ferm ε F ) corresponds to a high temperature regime for the lines T n 2 L/c.

• In the very begining, we have assumed small distorsions ∂x/∂τ 1. Velocity is of the order of the Fermi velocity v F = k F /m = 2πnT /c. Small velocity v F 1 requires small temperature for the lines T c/n.

We conclude that the validity of the T ferm = 0 calculation is n 2 L/c T c/n.

• For finite temperature, we expect that the Kohn anomaly is smothed by the fermion tempe- rature as χ(2k F ) ∼ −ρ 0 ln |ε F /T ferm |. Using the mapping onto the parameters of the lines we get

χ(q = 2πn) ∼ − c

T n ln n 2 T L/c

(17)

Références

Documents relatifs

We apply a renormalization group decimation technique to the system of 2N spins on two lines with one interaction costant (K 1 ) between spins in the same line and another (K 2

Thus each mathematical symbol (G, L,. .) has a single definite meaning inside each exercise, but might have different meanings from an exercise to another.. It is not required to

represents an irrational number.. Hence, large partial quotients yield good rational approximations by truncating the continued fraction expansion just before the given

The maximal inequality (1.1) also implies a capacity estimate in Wiener space (Corollary 4.2), as well as a uniform ratio ergodic theorem for Brownian motion in Wiener space that

In this case, the laser pump excites the titanium film and, through a thermoelastic process, leads to the generation of picosecond acoustic pulses that propagate within the copper

If an abstract graph G admits an edge-inserting com- binatorial decomposition, then the reconstruction of the graph from the atomic decomposition produces a set of equations and

In particular we use the special Markov property of the Brownian snake and the connection between Brownian snake and PDE The next section is devoted to the presentation of the

Then it seems likely that for β &gt; 1/2 there is a Ricci flat K¨ ahler metric on the complement of the curve with standard cone singularities of cone angle 2πβ away from the origin,