• Aucun résultat trouvé

Determinantal quantum theory of d.c. electrical conductivity

N/A
N/A
Protected

Academic year: 2021

Partager "Determinantal quantum theory of d.c. electrical conductivity"

Copied!
24
0
0

Texte intégral

(1)

HAL Id: jpa-00246573

https://hal.archives-ouvertes.fr/jpa-00246573

Submitted on 1 Jan 1992

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Determinantal quantum theory of d.c. electrical conductivity

A. Fortini

To cite this version:

A. Fortini. Determinantal quantum theory of d.c. electrical conductivity. Journal de Physique I, EDP

Sciences, 1992, 2 (5), pp.625-647. �10.1051/jp1:1992170�. �jpa-00246573�

(2)

Classification Physics Abstracts

03.80 72.10B

Determinantal quantum theory of d.c. electrical conductivity

A. Fortini

Universitd de Caen-ISMRa, LERMAT (*), 14032 Caen Cedex, France

(Received 22 July J99J, revised J2 December J99J,

accepted

24

January

J992)

Rksumk. Le calcul

quantique

de la conductivitd en courant contblu d'un

systbme

d'dlectrons est reconsiddrd h l'aide d'une solution amdlior£e pour la matrice densit6.

L'dquation

d'dvolution utilisde est d'abord dcrite comme un

systbme

1bldaire dans

l'espace

de Liouville des stats

quantiques

discrets, rdsolue h l'aide de ddterrninants de Cramer et convenablement

adaptde

ensuite aux

problbmes

de transport. Le calcul de la moyenne

statistique

de la densit6 de courant,

en

pr6sence

d'un

potentiel

de diffusion, est effectu6 h la limite

therrnodynamique,

en utilisant une

propr16t6 caract6ristique

des matrices de collision, initialement introduite par van Hove sous le

nom de

sulgularit6

&. Contrairement aux m6thodes

pr6cddentes,

bas£es sur les

6quations cindtiques

rdsolues par

perturbations,

la th60rie de Kubo, ou l'utilisation de fonctions m6moire de type Mod, la rdsolution d'une

6quation int6grale,

ddjh incluse dans les forrnules de Cramer, n'est

plus

n6cessaire. Parmi ses autres avantages

majeurs,

la m6thode est

capable

de r6soudre la

divergence

bien connue de la conductivitd lorsque la

frdquence

tend vers zdro. Elle fournit aussi

une

expression explicite

d'un temps de relaxation effectif, construite

pr6cis6ment

h partir des

s6quences

de transitions

diagonales

responsables des effets de

self-dnergie

dans la thdorie de van Hove. Cette

expression qui

ne suppose que la distribution de

phase

aldatoire des stats

d'6quilibre

initiaux, peut Etre utilisde pour obtenir des

d6veloppements

maniables en

puissances

du potentiel de collision. Son domaine de validitd n'est limitd que par l'effet des collisions sur la

statistique d'6quilibre, qui

se r6vble

ndgligeable

dans les limites du critbre de Peierls h/r « p, off rest le temps de relaxation et p

l'dnergie

de Fermi. Finalement, la

comparaison

avec la thdorie

616mentaire montre que le temps de relaxation est

proche

de

l'expression classique,

mais ne

pourrait s'y

identifier

complbtement,

h l'ordre le

plus

bas des collisions,

qu'en ignorant

les

divergences apparaissant

au-delh du second orate.

Abstract. The quantum calculation of the d-c- electrical

conductivity

for an electron system is reconsidered

by using

an

improved

form of the

long-time

solution of the

density

matrix. The

starting

time-evolution

equation

which is written as a linear system in the Liouville space of discrete quantum states, is first solved in terms of Cramer's determinants and next

suitably adapted

to transport

problems.

The calculation of the statistical average of the

many-body

current

density, in the presence of

scattering

potential, is carried out in the

thermodynamic

limit,

using

a

characteristic property of collision matrices introduced earlier by van Hove as the so-called 3-

singularity.

In contrast to

previous

methods based on kinetic

equations

solved

by perturbations,

Kubo's

theory,

or Mori type memory functions, the resolution of an

integral equation, blitially

(*) Associd au C.N.R.S. URA N° 1317.

(3)

included in Cramer's formulae, is no

longer

required.

Among

its

major advantages,

the method succeeds u1

overcoming

the well known zero

frequency divergent

behaviour of the

conductivity.

It also

yields

an

explicit

expression of an effective relaxation time which is

just

constructed out of those

diagonal

transition sequences

responsible

for

self-energy

effects in van Hove's

theory.

This

expression

which

only

assumes the randomization of the

phases

of initial states can be used to get tractable

expansions

in powers of the collision

potential.

Its range of

validity

is limited by the effect of collisions on

equilibrium

statistics, which is shown to be

negligible

within the Peierls criterion, h/r « p, r

being

the relaxation time and p the Fermi energy.

Finally,

the

comparison

of the present

approach

with the

elementary

transport

theory

reveals that the relaxation time is close to the standard

expression,

even

though

the latter would be

exactly

recovered, in lowest order of

collisions,

only

on

disregarding divergent

terms

arising beyond

second order.

1. Introduction.

The

quantum theory

of transport processes, and

especially

the

problem

of electrical

conduction,

have received a great deal of attention until recent years. Since the earliest

Boltzmann

transport equation

which was solved with the

help

of the old random

phase

approximation (RPA),

various

techniques

have been

investigated.

The direct resolution of the

density

matrix motion

equation by

means of

perturbation

series has

given

rise to the so-

called method of

«

quantum

kinetic

equations

» first

developed by

Kohn and

Luttinger [I],

then

by Argyres

and co-workers

[2, 3a, 3b],

who introduced

appropriate

refinements to

remove the

unsatisfactory repetition

of the WA at any time. The Kubo

theory

based on the

calculation of correlation functions of current

density [4]

and combined with an extensive use of Green's

functions, diagrams

and

projection techniques [5-7],

represents an altemative

approach which, surprisingly,

has known a

widespread

use, in

spite

of its

highly

abstract and fornlal structure. The fundamental work of Mori

[8]

has

generated

a third kind of

approach

known as the « memory function method »

[9-12],

with much effort to overcome several

difficulties

remaining

in the

previous

ones.

The merit of the old Boltzmann

equation mainly

relies on its remarkable and

meaningful simplicity.

At the lowest order of

scattering

and within the one-electron

approximation,

it is

thought

to

give

the

right

result known as Drude's

fornlula,

and is often

regarded

as a test for

more elaborated

quantum-mechanical theories,

taken within the same

validity

conditions.

This is the reason

why

a main concem of

existing

theories is often to

clarify

their connection

with the Boltzmann

theory

result

[1, 13-16].

In

fact,

as stressed

by Argyres [2, 3], beyond

the lowest

order,

all methods

require

the

solution of an

integral equation. Expressions

of

conductivity

derived from

projection

techniques

which are claimed to avoid such a

resolution,

were shown to be unreliable because of the

remaining

low

frequency divergences. Proper handling

of these

divergences requires

the summation of an infinite number of tennis

equivalent

to

solving

an

integral equation.

Similar

efforts,

in the memory function scheme, to

by-pass

the

integral equation

resolution

[9, 10]

were also shown

by Argyres

to become incorrect at low

frequency

and inconsistent with

Drude's fornlula

[3].

Dramatic

improvements

in the

quantum

transport

theory

were

accomplished by

van Hove

in a famous series of papers

[17a-17e].

As is well

known,

a decisive step of van Hove's work

was to

recognize

a

general

and characteristic

property

of collision matrices in most

applications,

the so-called

3-singularities

in the continuous spectrum, I,e. in the ther-

modynamic

limit of a

large

system. This

mainly

stems from the

spatial

extension of the collision

perturbation

in

macroscopic

systems, and leads to

separate

out an

overwhelmingly

predominant diagonal part

in collision transition sets, at any order.

Among important

(4)

consequences, this

separation

entails the

preeminence

of tennis in

(A ~t)~

in the

asymptotic

behaviour

(with

A

denoting

a diniensionless

parameter

characteristic of the

strength

of the collision

potential),

which in tum entails the elimination of interference effects in the

density

matrix thus

avoiding

the

repeated

use of WA. It also results in the occurrence of a

complete

set of

asymptotically perturbed

states

including self-energy

effects

[17b].

Van Hove's ideas were next

injected

into various theoretical

approaches

of transport

problems,

and

particularly

those derived from Kubo's formalism

[5-7,

18-2

Ii. Recently,

Loss

[7a] pointed

out the interest of

dealing directly

with the

density

matrix in the Liouville space, for

extracting

from

perturbative expansions,

the A

~t

limit

intimately

connected with the 3-

singularities.

He also

published

an

application

of his method to the calculation of

conductivity,

in the lowest three orders

(A

~, A

',

A of collisions.

The determinantal method

previously developed by

the author for

solving

the time-

dependent Schr6dinger equation

of the evolution

operator [22a]

and the

density

matrix

[22b],

is

particularly

well suited to find the

long-time

limit of the response, so

providing

us with a

drastically

new means for

obtaining

a

steady-state

solution in

quantum transport problems.

In the

present

paper the method will be

applied

to the electrical

conductivity calculation,

in which its

major advantages

will be

clearly

exhibited. The resolution of an

integral equation

will be

avoided,

convergence will be ensured in any case, and an

explicit

fornlulation of a Drude-like relaxation

time,

suited to the

conductivity problem,

will be derived.

Furthernlore,

van Hove's

theory

will tum out to have a determinant contribution in the derivation of the main

result,

and the connection with Drude's formula to lowest order will be discussed in

detail.

The paper will be

organized

so as to outline the salient features of the

method, leaving

out of consideration irrelevant details such as discrete

quantum

numbers like

spin indices,

or

anisotropy

effects.

Further,

since the low

frequency

limit was a

highly

controversial

topic,

it

will be restricted to the d,c, case, which needs

simpler

mathematics. Extension to the

harmonic case, where no

divergence

occurs, will offer no additional

difficulty

and can be

rejected

in

subsequent publications.

To make the paper

self-containing,

the determinantal solution of the

density

matrix will be

bliefly

recalled in section 2. The calculation of

conductivity

will be carried out in section 3 and

given, first,

in a

general

form

capable

of

including many-body effects,

in order to allow further

developments

and

applications.

The subsections

3.4,

3.5 will be more

especially

concemed with the

important application

of most

important

results to the free electron case,

leading

to a somewhat

improved

relaxation time

definition, carefully compared

with the standard

theory. Finally,

section 4 will be devoted to a

discussion of the

ability

of the method to overcome some basic

difficulties,

as

compared

with

previous

ones.

2. Determinantal solution of the

density

matrix

equation.

We shall consider an electron system in a

periodic

lattice

potential,

described

by

the

unperturbed

Hamiltonian

Ho,

the

eigenstates

of which are denoted

by b,

c, with

eigenvalues

s~

=

hw~,

s~

=

hw~,

Predominant collision processes are described

by

the

coupling

Hamiltonian

V,

and the

coupling

of electrons

(of charge e)

with an extemal electric field of

magnitude E, applied along

the

x-direction,

is

represented by eEx,

in the scalar

jauge.

The field is assumed to be switched on at time t = 0.

Let us

briefly

recall that the so-called « determinantal » method

[22c-22d]

aims at

finding

a

trace

conserving

the solution to the

Schr6dinger equation

of the

many-body density

matrix

p

(t),

written in the fornl

dp/dt

=

(ih)~ glo

+ V +

eEx,

p

i (p pa)

s~.

(I)

JOURNAL DE PHYSIQUEI -T 2, N' 5, MAY IW2 25

(5)

pa = p

(0)

stands for the initial

equilibrium density

matrix.

Very

slow relaxation processes which are not included in V can be

described,

in a way introduced

by

Lax

[23], through

a

simple phenomenological

relaxation rate s~.

They involve,

in

particular,

the

coupling

with the heat-bath and ensure the overall coherence of the

theory by allowing dissipation

in the

purely

elastic case, while

improving

mathematical convergence. In the linear response

appxoximation

we will be restricted to,

hereafter,

s~ is small

enough

for

having

no

significant

effect on the result and

thus,

will be

regarded

as a

vanishingly

small

quantity.

Then the

Laplace-transformed density matrix, namely R(v)

<

p(t),

satisfies the trans- forn1ed

equation

:

(v

+

sr)R(v) +I*-iiHo+

v +

eEx, R(v)i

=

po(i

+

e/v) (2)

which will be rewritten in the Liouville space

i~

= iJc~ @

I(~,

where iJc~ is the Hilbert space

spanned by

the

unperturbed eigenstates

of

Ho,

and

I(~,

the dual space

[24]. i~

is sustained

by

a tetradic

representation,

the basis of which is defined

by

all

couples

of

Ho eigenstates (b, c), (bi, ci), Operators

in

I,~

such as

R(v)

become vectors, written

R(v)

in

i~,

with

components

R~~ =

<cb R(v)>

c

<cip(t)i b>

,

while superoperators such as the commutation kemel in

equation (2)

become operators, We shall write for any operator Y of

ix (or

lf in

i~)

lKvlf=ih~~~v,Y]; lK~lf=ih~~[eEx,Y] (3)

with the

following

matrix elements

(cb[ lKv [ci bi)

= lKv)~ ~' =

ih~

~(V[~ 8)~ 8j~ V~~)

,

etc.

(4)

so that

ih-

ICI lv, Yi bl

= ~Kvl~

~ Y~~

b~ = I*-

(vl~ 8(~

81~

V(~) Yll

=

ih- i

(vj~ Yj Yj v(') (5)

Summation over

repeated

indices will be

implicit throughout. Diagonal

matrix elements of V will be

ignored

without serious loss of

generality.

As

previously pointed

out

by

Kohn and

Luttinger ii,

their

only

effect is to shift the

unperturbed energies by

a constant. Once

suitably completed by

a

chronological ordering

rule

[22d], expression (4) provides

us with a mechanical means for

calculating

the

repeated

action of

lKv.

Such tetradic

representations

were

already

introduced in the Liouville space with somewhat different conventions and notation

[7a, 7b, 20].

On the

quantized

state basis of

i~, equation (2)

is

conveniently

rewritten as a linear system :

~~

+ ~<b~

(I~V

i ~ +

I~E

l ~~)

~~

hi ~ l~b~ PO

~(l

+

El" )

,

(6)

where the

d~~'s

stand for the matrix elements of the

following diagonal superopeator

d

=

(v

+

e~)

i +

~x~, (7)

denoting

the

identity

superoperator. Then the solution of

(6)

can be first written in Cramer's

fornl,

(6)

~mb R~i

=

fi po1, (8)

expressed

in tennis of the deternlinant D and the

ml (column)-cb (line)

minors of the matrix in the left-hand member of

(6).

In the

thermodynamic

limit

equation (6)

becomes an

integral

Fredholm

equation

of the second

kind,

the solution of which is then

given by

the limit of

expression (8) [25].

As

previously

shown in references

[22c, 22d],

the solution

(8)

contains unlinked sets of transitions

(associated

with

unphysical

sequences which do not

properly

start from the « initial state »

cb)

which can be eliminated in a

systematic

way,

through

the fornlal division of the upper and lower determinants of each fraction of the sum in

(8), by

the

diagonal

minor

D[( pertaining

to the

starting

state cb

(see

Ref.

[22c])

~m

D7i (Dl()~

~

~fl ~

(~cb)-

~0 b

(9)

cb cb

Note that the form

(9)

cannot

change

the value of the

rigorous

solution

(8),

but

only

the

approximate expressions

of the numerator and the denominator which will be retained in

practice. Indeed,

the deternlinant

quotients arising

in

(9)

can be

given

a more

explicit

form in tennis of

geometric

series

[22a],

e,g.,

Dft (D[()~

=

(ml (I

+

Q~~

d~ lK

)~ cb)

,

for

ml

~ cb. The occurrence of the

complementary projector Q~~

is

simply

connected with

the exclusion of the initial state row cb in the definition of the minor

Dft,

and the

subsequent

division

by

the minor

D[(. Separating

out the

particular

ternl

ml

=

cb, expression (9)

then becomes

~~

8ft (ml

lK

(I

+

Q~~

d~ lK

)~ cb)

~ ~~

d,~

+

(cb[

lK(1 + 4ll~~ d~ lK

)~ [cb)

~'

with lK

=

lKv

+

lK~,

and

ml

# cb in the upper

angular

bracket.

In transport

theory, expression (10)

is recast into a more suited fornl

[22d], including

some

simplifications, namely

the

neglect

of natural transition widths

(tennis containing

lK~ in the

denominator)

with

regard

to collision widths

(tennis

in

lKv alone),

and linearisation. This amounts to

drop

the field kemel lK~ in lK

everywhere

except in the first-order-in-E ternl in the

numerator of

(10).

The

leading

result is

given by equations (23)

and

(27)

of reference

[22d] (1),

which will be rewritten in the

following

compact fornl

nmb (nZb)-

I

pi

=

~~

~~

ih~ eE

(z[ [x, po][b) (ll)

d~j,

D(D(b)~

The

long-time

limit

v -

0+

(with

s~ -

0)

will

always

be

implied

and assumed to be taken after the

thernlodynamic

limit. On

carrying

out the deternlinant

divisions,

one obtains :

pi

=

~~ ~~ ~"'~ ~~~

~ ~~ ~

~~~~

~~ ~~~~

ih~ eE

(z[ [x, po] [b) (12) d-j,

+

(zb

IK

(I

+

Qz~

d- IK

)~ (zb)

(I)

Due to an error of

sign,

the +

sign

must be

changed

in in the numerator of

equation (20)

of

reference [22d], which entails change of

sign

in some subsequent

equations

derived therefrom.

(7)

The kemel IK now stands for

IKv

in which the redundant

subscript

V is

dropped.

The

particular

ternl

ml

= zb for which the numerator of the fraction reduces to I

(see Eq. (

II

))

is

again separated

out

and, therefore,

the index restriction

ml

# zb

imposed by

the

projector Qz~

holds in the

angular

bracket

of

the numerator.

It is worth

recalling

too, for further

discussion,

that if we carry out the division of the minor

by

the deternlinant in

(8),

instead of

keeping

the initial fractional fornl as done in the above

fornlulae,

we obtain

R~ I =

Dl't

D~ I po

[Id,

~ =

(ml (I

+ d- IK

)~ cb)

po

[/d~~ (13)

leading strictly

to the

perturbation

series of the

density matrix,

upon

expansion.

Of course the

same result could be obtained from

expression (10)

as

well, showing

that the division of the

numerator

by

the denominator in the deternlinantal fornlulae would

just

have the effect of

leaving

the numerator alone without the index restriction

imposed by Q~~ [or Qz~

in

(12)]

therein.

3. Calculation of the electrical

conductivity.

We now

proceed

to the calculation of the d.c.

conductivity

from the statistical ensemble average 3 of the current

density,

a

=

Tr

(pi), (14)

where J denotes the current

density

operator.

Making

use of our basic result

(12), equation (14) gives

3~ ieJ~ 8)

8

f

dji~

ill

lK

(I

+

Qz~

d~ lK

)~ zb)

' E h

d-~

+

(zb

IK

(I

+

Qz~

d~ IK

)~ zb)

~~ ~~'

~°~ ~~

~~~~

with11

~ zb in the

angular

bracket of the numerator.

Anisotropic

effects which would add

nothing

new to the

physical

features of the

theory

are

ignored

and so «~ is

simply

denoted

by

«.

Here, b,

c, z,

,

I,

m, represent

many-body

states which can be constructed out of individual Bloch-states

specified,

as usual,

by

a band index n, the electron momentum k and a

spin

index. As

already mentioned,

the

spin plays

no

particular

role in the present

study

and will be

ignored

for

simplicity.

We shall also limit ourselves to the one-band

approximation

(interband

transitions more

specifically

relate to the

optical

response which

requires

a

separate study).

For later use, we shall write out the current

operator

J in tennis of

single- particle velocity

v

=

p/m

in the band under

consideration,

which is

diagonal

in the Bloch

representation.

In a second

quantization scheme,

we have :

~~

~~i~k~k

~ ~k' ~~~~

k

~

where flJ denotes the volume of the system and

cl,

c~ the customary creation and annihilation

operators.

The commutator

[x, pal

in

(15)

will also be

given

a second

quantization

fornl in the n, k

representation

IX,

p01

~

i

IX,

p0fi~'C$k'

Cnk

(17)

n'k'nk

which,

in tum,

requires

the

knowledge

of the related

one-particle

matrix elements, In our intraband

approximation,

the latter reduce to

[x, po]~'

= i

(V~,

+ V~

)(nk'[po[ nk)

,

(18)

(8)

as was

already

mentioned

by

Chester and

Tellung [18].

The demonstration is

briefly

recalled in

Appendix I,

and detailed discussion can be found in the review paper

by

Blount

[26].

Notice the limit of

equation (18)

when k'- k :

[x, po]$

= I V~

(nk[po[ nk) (19)

So far the result

(15)

is rather formal. To set up a more tractable

expression

we rust have to

remove

divergences occurring therein,

and to work out an

explicit

form of the matrix

elements of the commutator

[x, po].

3.I REMOVAL oF DIVERGENCES.- We rust

recognize

in

(15)

the well known

divergent

behaviour due to the factor

dji~ becoming

very

large

in the

long-time

limit. This

primarily

occurs because the

physical quantity,

the statistical average of which we have to

calculate,

is

diagonal

in the

unperturbed representation.

A second reason is that we are in d,c,

regime.

At finite

frequency

w, v would be

replaced by

v + iw in

dir

and so, no

divergence

would occur.

However,

the fraction in

(11) (or equivalently

in

(12))

is

necessarily

finite in any case,

irrespective

to the average

quantity

which is to be

calculated,

or the

frequency

range of interest. It follows hat this

divergence

should be lifted

by

means of

L'Hospital's

rule before

going

to the

thermodynamic limit,

after

having singled

out a similar behaviour in

dji'

in the denominator. In

Appendix

II this is done in a

systematic

way for both the

dj~'

and

dji' divergences, leading

to

expression (II.4),

written in a closed form thanks to the

following

definition of the relaxation parameters

T~i

=

iii

IK

(I

+

Qz~ Qii

d~ IK

)~ (zb)

=

iii ~ z

IK

(- Qz~ Qii

d~

IK)" zb) (20)

i

Using (II.4),

the

conductivity (15)

can

finally

be rewritten in the fornl

j b

~b

j f

~fb

" ~

~i fi

(d~ l~r=/)

~fifi ~~'

~°~~

~~~~

in which the

dfi~ divergences

are now removed.

In the calculation of « from

(21)

we shall thus have to deal with the matrix elements of the

T's,

I,e.

expansions

of the fornl

(20).

Let us write out for definiteness the

expansion

of the ternl in

J~((

to lowest

(second)

order

J,( J~~/[x, po][

=

(ih~ ~)~J~i(V$~ 8(~

8$~

V(~)dj (~(V)~ 8( 8)~ V(~)[x, po][

=

=

-2j~i( ~'i

~~~i~~

~°i~

~j vi ix, Poi~ vi vi ix,joi~

vi

+

~i

i~~

°(~ ~'i ~"

dci

b

dd

16

-1>1

The

general

ternl of any order involves

products exemplified by

V~

V" V~

Vz[x, pot VI V~' V~.. V~ Vildqj, d,,n, (22)

The

d~,,,

s,

d,jj,

s... represent the energy denominators taken between internlediate states

appearing

somewhere in the sequence.

Unsignificant

indices are omitted.

To

proceed further,

we have to take more

realistically

into account the

properties

of sets of collision operators, such as those

arising

in

(22)

or in the

expression

of pa matrix elements below

(see (29)).

As

already mentioned,

this was

previously

done

by

van Hove

[17a-17e]

(9)

through

the introduction of the characteristic

3-singularity,

valid in the

thernlodynamic

limit

where the relevant

quantum

numbers become

quasi-continuous variables,

in all cases of

interest such as collisions with random scatterers, interaction with

quantized fields,

etc. This fundamental

property

holds for

products

of matrix elements of the

type (22),

as well.

Here,

the denominators d~,,~ will

only impose

some condition on the

energies

of certain

couples

of

states, upon

integration,

without

affecting

the essential of Van Hove's arguments

leading

to the

predominence

of

diagonal

contributions

[17b,17d].

In a more

explicit fornl,

the 3-

singularities

entail that among all

possible

groups of successive V

operators

of the kind shown in

(22)

and

subgroups therein,

those which are

diagonal

are much

larger

than the

off-diagonal

ones

(by

a factor of the order of the number of

particles

in the

system).

It follows that in

subsequent

summations any

diagonal

set will be at least of the same order as the sum of the

off-diagonal

ones. For

example,

the matrix elements of the

product

V...

A~VAI V,

where A i,

A~,

are

diagonal,

can be

split

as follows into a

singular

part

involving

a 3-function and a

regular part

which does not

(b'(V. A~ VAT V[ b)

=

F~(b)

8

(b'- b)

+

F~(b', b). (23)

Upon integration,

this becomes

ldb'(b'[V. A~ VAT Vi b)

=

F~(b)

+

db'f~ (b', b),

where the first ternl on the

right-hand

site is at least of the same order as the second one.

Now,

due to the very

3-singularity property,

a new kind of

divergence

arises every time a denominator like

d~~,

in

(22), vanishes,

I,e, if n

= m.

Indeed,

in such a case the

3-singularity

of the related mm

diagonal

subset

represents

a very

large (macroscopic)

number of tennis with the same

vanishing

denominator

d~~

in the

long-time

limit. In

fact,

these

divergent

tennis tum out to cancel. Let us consider the

general (n

+

I)

th order ternl of fornl

(22)

in

expansion (20).

On

starting

from the initial zb element and

going

to the last one,

it,

the occurrence of the successive

divergent diagonal

sets can be

pictured by writing

a segment up to some qq

state, in the

following simplified

form

(see Appendix III)

~ ~- l j~qp ~ l

j~pa'~a

qq~ qq Pq PP ap' a',

where

X$,

represents the matrix elements of the undetailed

subproduct

in the

right-hand side, including

the

starting

element

[x, pa[ (the

Q's which

simply impose

that qq,

pp'#

zb are

omitted).

It is shown in

Appendix

m that S~~ cancels term

by

term with S~~, except when qq

represents

the final

it

state of the

(n

+ I

)

th order set under

consideration,

since then the

I- dependent

factor

J(

comes into

play.

The related

dji' divergence, however,

has

already

been

lifted

by

means of

L'Hospital's rule, leading

to

equation (21).

It may be concluded that the

general

term of fornl

(22)

in

expansion (20)

represents a well defined finite

quantity

: the

singular part

of its

diagonal

subsets associated with

divergent

denominators will be

eliminated,

whilst in the

thermodynamic

limit the

regular

part cannot entail

divergence

but

only

a

Cauchy-type integration [25],

which is handled

by

means of the well-known rule

- w8

(w~~, )

I

,

(24)

f

+

I(1~

+

w~~,)

Wqq,

P

where v

=

f +17~

(7~ m

0),

the limit

being

taken for

f,

7~ - 0. The

subscript

P indicates the

principal

value.

(10)

3.2 MATRIX ELEMENTS oF THE COMMUTATOR

[x, po].

The

equilibrium density

matrix elements po

[

will be written in the usual

grand

canonical scheme

po = exp

i- p (Ho

+ V qn

)i/Tr

exp

i- p (Ho

+ V qn

)1 (25)

where

p

=

I/kB

T is the inverse of the thermal energy. The Fermi energy ~ is fixed whereas the number of

panicles

is not, this is more convenient because all electron dishibutions on the

available states are then allowed without any restriction. In the second

quantization

representation,

Ho

~n

=

£ (e~

~

) c(

c~

(26)

k

The calculation of matrix elements of po between the

eigenstates

of

Ho

will be facilitated if

we first recast

(25)

in a more convenient form. To this end use will be made of the

exponential

of the sum of non

commuting

operators

given

in Bourbaki

[27] (see Appendix IV), yielding

exp

i- p (Ho

qn + V

)i

= exp

ip (Ho

qn

)i

x

lm L~

V

x exp

p (Ho

l~n

)

+

z (- p

)~ + j exp

i- p (Ho

l~n

)1

,

(27)

n ° ~'

where L stands for the customary Liouville

operator

defined

by Lx~

V = @Io,

Vi.

We shall

simply

denote

by

S the first two

exponential

factors in

(27), describing

the

perturbative

effect of collisions on the statistics S

= exp

ip (Ho

l~n

)i

exp

i- p (Ho

qn

) p

exp

(- pLxo) Vi (28)

The

equilibrium density

matrix can now be rewritten iri the form S exp

i- p (Ho

l~n

)1

~° ~~

Tr

(S expi- p (Ho

l~n

)1)

'

where the

unperturbed part

is well

separated

out. The derivation of the second

quantization expression

of the

unperturbed

Boltzmann factor from

(26)

is

straightforward

and leads to the usual

expression

of po, in the absence of collision

(S =1) [28]

exp

p £ (Ho ~n) c(

c~

po(V

~

=

0)

=

Tr exp

p £ (Ho

~n

) c(

c~

~

~

fl

~~k

~~

+ ~~P ~~

~

~~k ~ ~~ ~~ ~~

=

fl a~ (30)

~

l + exp

[- p (8k

l~

)I

k

where

&~

=

ia(e~)

denotes the

occupation operator

~°k "

(1 fk )

CkC~ +

fk C(

Ck

(31)

in terms of the

single-state

Fermi function

f~

=

I/(exp §3 (e~

~

)]

+ l

).

In

(30)

use has been

(11)

made of the property that

Tr

leXp(-P £(H0~$~n)C( kjj

~

fl (I +eXp[-P(8k~$~)1).

k k

Now,

the collision factor S

splits,

upon

expansion,

into sequences of transitions

generated by increasing

powers of the

arguments

in

(28),

which will

again

exhibit

singular diagonal

parts. Due to the

property (23),

these contribute to the same order of all

nondiagonal

ones.

But,

as was discussed in detail

by

van Hove

[17d], nondiagonal

elements of initial states entail interference effects which cancel out for an

overwhelming majority

of cases and so,

they

can be

ignored

without loss of accuracy. In another way, this amounts to assume the RPA

restricted to the initial states of the

equilibrium density

matrix po.

The effect of collisions will thus be described

by

means of the

diagonal

part of S, the

expansion

of which in lowest order is calculated in

Appendix

IV

S~

= I +

£

S~

(k'k) cl

c~,

cl,

c~

(32)

k'k

with

S~(k' k)

~~P(- p

e~,

~

)

i ~

El,

~

~

$ IVl'

~

~'k

(33)

(e~,~

= e~,-

e~).

From

(17), (19)

and the above

expressions

from

(29)

to

(33),

the

commutator matrix elements of interest in

(21)

will in tum reduce to

diagonal

ones of the

form

[x, po]

= I

£ c(

c~ V~

fl

&~, +

O( IV ~), (34)

k

~

k'

where the

leading

term on the

right-hand

side is of zeroth order in collisions. The relative contribution of the correction in

IV

~ associated with the S factor

(32-33)

will be first

ignored

in what

follows,

and discussed later on, in Section

3.5,

in connection with the Peierls criterion.

3.3 RELAXATiON PARAMETERS. The f~ matrix elements as

given by (20)

represent many-

body

relaxation

parameters

associated with V

collisions,

in a

general

and

rigorous form,

where all electrons are assumed to be concemed at once in collision processes. In lowest

order,

I.e. the second one since V has no

diagonal elements, J~((

for instance is written as

J~ll

=

ibb11~Q~~

d- 1~

bb)

+

= h-

~(vt, al'- at, VI') di [, (vi' 3t, al'vi, )

+

V$ Vi V~ V~'

=

h~~

' + ' +

(35)

Iic,b Iibb,

with ci

bi,

# bb.

Obviously, expression (35)

is real. More

generally,

any

J~((

defined in

(20)

is a real

quantity.

This

simply

results from the

conjugation

relations

(j~C,b2~~ ~b,C2

~~ ~

c2b, b~c, > cj hi hic,

(which straightforwardly

derive from the definitions

(3)

and

(7))

when

applied

to each matrix element of d~

~li

iri

expansion (20),

at any order.

(12)

Assuming

a

single-particle

collision

potential,

and

using

the second

quantization expression

V

=

£

V~~

cl

c~,

(35)

takes on the more detailed form

k~k,

' ~ '

f~bb

~

£ fi-2 vk, vkJ

~ ~.

(36)

bb k~ k, ~ ~

~~~, I,I, k,

k~

where the

k;

and ky summations now run over all

occupied

dud

empty single-states

in

b,

respectively.

3.4 CONDUCTIVITY AND RELAXATiON TIME. The

foregoing expressions

of the

[x, po]

and

J~'s matrix elements enable one to calculate the

conductivity

from

(21). Assuming

the WA of initial states as discussed

above,

and

substituting J~

=

euJiJ,

we have

i~2

~ ~+[j

jx, p~j(

"

~

~Xb ~Xf

@

J~hh ~~~~

ii fib

When

only

a few electrons are correlated to one

another,

the form

(35)

or

(36)

of the J~'scan be fiirther

simplified.

To work out an illustrative test, and for the sake of

comparison

with the standard

theory, below,

we shall assume

completely independent

and

isotropic

electrons in what follows.

Then,

V

coupling only

occurs between

many-body

states which differ from one another

by

a

single-state occupation,

all other states

being regarded

as

quenched

at

equilibrium.

The set of b states can then be

approximately split

into

uncoupled

subsets of this kind and their related contribution to the determinants D and

Df(

in

(11)

are factorized into

independent

blocks which

separate

upon

simplification,

the

D((

minor

being

next redefined

accordingly

in each block. All of the above determinantal

procedures

remain in each

subspace.

In

fact, only

blocks in which

occupations

are close to statistical

equilibrium

are relevant.

Doing

so,

J~()

becomes a

unique

function of the

single

state k under

consideration,

which we shall

simply

denote

by

r

(k),

and

J~()

a function r

(k', k)

of the

single

states k' and k. Of course

dynamical exchange

effects are somewhat

oversimplified by

this

way.

Assuming

for

simplicity

a

weakly occupied

band the intermediate

occupation

factors can be

ignored,

so

that,

once reduced to one-electron

transitions, expression (36)

becomes in the

two lowest orders

l~(k)

=

2 Re

h~

2

~~'

~~~

ih-3 ~~2 ~~~

~~'

&,~~

d,

,

d,

~

~

,

(k2, ki

# k

). (38)

Similarly,

one would find from

(20) r(k', k)

=

~ ~~

fi-2 ~~' ~~'

~ ~- 3

~k' vk, ~k

~ l

~ l

~ ~~~~

lij k

~' ~ ~'

~iI I

I~II'

I~I I I~I k

Iii

j

Iii

k

Notice that in the

integrations

over intermediate states

arising

in

expansions (38)

or

(39),

the energy denominators

obviously

may pass

through

zero in the

integration

range. From the discussion of section 3.I it follows that the

divergent

terms due to the

3-singularities

of the related

diagonal

subsets cancel and can thus be

ignored.

This means that

only

the

regular

parts

of these

diagonal

subsets are to be

considered, leading

to a finite result

through

Cauchy's type integration (24).

Of course, all

diagonal

subsets are not

necessarily

associated

(13)

witli

vanishing denominators,

but one can be sure that no furtlier

divergence

of tliis kind may

occur. In all cases

using

van Hove's

techniques

may

help

in

finding

the

leading contributions,

such as saw-tooth or

symmetrical

transition sets, as was done in

previous

similar calculations

[7, 18, 19].

In the

grand

canonical

scheme,

the summation over b in

(37)

runs over all

possible

electron

occupations

of individual states.

Any given

term in the

expansion (39)

contributes the

conductivity (37) through

a definite set of electron states, e.g.

k, ki,

k' for the above third order term. It is

readily

seen that all other electron states which are not involved will be washed out. Let indeed b' denotes such a restricted

many-body

state

(I.e. excluding k, ki,

k' in the

preceding example).

From

(34),

their contribution to

(37)

will be

clearly

factorized in the form

I ~l (1 fk~ ) fl fk,

~

ll (1 fk"

+

fk" )

~

l

(40)

b' k~ k, k"Eb'

Next,

introducing

in

(37)

the

single

electron velocities from

(16),

and

making

use of

(34) (the

effect of collisions in the commutator

[x, po]

is

postponed

for later

discussion),

we obtain

"£j[~jtiiii~j~(£~

~~~~

This

expression immediately suggests

the

following

definition of an effective

conductivity

relaxation time

by

~~~~ -

&

Ii 16 [[,j~ l~~~~

~~~~~~~

~~

~ =

~~

£ T(k) V~(Vk/k)

~

~~

l~ ~~~~ ~~ ~~~

~~~~

fi%

~

~

k

~~~

which has the form of the result

given by

the standard transport

theory,

For the sake of a more detailed

comparison

with

previous theories,

we are

going

to work out an

explicit expression

of

r~k),

in second

order,

in our

isotropic

one-electron

model, assuming purely

elastic collisions. Then v~ and k are

parallel

vectors which can be

specified by

their

magnitude

and their

angular polar coordinates, 0,

p, with

respect

to any

given

direction.

From

(38),

we first have

q~

(V~~(~d~ki

r(k)

= 2 Re

~ ~

(v

-

0+) (2 w)

h

(v

+

iw~ ~)

q~

(V~'(~k)dki

~

~2

fi2 dw

~

~ ~~°k>k~ ~~°k> ~~~

~l ~~l ~~

l

0

~, pi are the

angular

coordinate of

ki

with respect to k, and use has been made of

(24)

in the limit v -0+ The 3-function

imposes ki

=

k,

and the

integration

over k~ is carried out

through

the introduction of the

density

of states in the

band, p(w~)

=

~lJk~dk/2 w~dw~,

leading

to

r(k)

= wh~ ~ p

(w~,) (V~'(~

sin 0~

d01. (44)

(14)

r is a function of the

only magnitude

of k.

Similarly,

the second-order term in

(39)

is

easily

transformed into

r~k~, k)

= 2 Re

vl'(

~/h2

d;

~ = 2 grin- 2

vl'(

~ a

(w~,

~

(45)

The calculation of r

(k)

from

(42), (44)

and

(45)

is

straightforward.

We use the

subscript

E when the

angular

coordinate are referred to the field direction

(x-axis)

and denote

by 0', p'

the

angular

coordinates of k' with

respect

to the k direction

(Fig. 1). (42) gives

q~ u~, cos

0)

2 w

Vl'

~

k,2

dk,

~~~ ~ ~~ ~

(2

W)~ Vk CDS

0E

fi~

l~(k')

dw k'

~

~~~'~

~~

~' ~~ ~' ~~'

~'~

k

o

Fig.

I.

-Angular

coordinates of the electron momentum k, k' witli respect to each otlier (9') and to tile direction of the

applied

field

E(e~, 9i).

Energy

conservation

imposes

v~,

= v~ and the

angular

coordinates are linked

by

the well- known

trigonometric

relation

cos

0)

= sin

0)

sin 0 cos

(p~ p')

+ cos

0~

cos 0'

Upon

substitution into the above

equation,

the term in cos

(p~ p')

vanishes

through

integration. Making

use

again

of the

density

of states p

(w~, ),

and

taking

into account that

r(k')

=

r(k)

if k'

=

k,

we are left with

l~

r

(k )

= r

~(k )

I + wh~ ~ r~

~(k ) VI'

~ p

(w

~,

)

cos 0' sin 0' do'

(46)

o

This

expression

does not

markedly

differ from the usual one. Since the ratio

£cos0'r(k',k)/r(k)

is

noticeably

smaller than I because of the

integration

in

k'

sin 0'cos

0',

one can rewrite

(41) approximately

as

« >

£ z

~~

~~/~

~~, ~~

=

£ z

rB

(k) vi (()

>

~4?)

k

l~(k )

+

£

(Vk'x~~kx ~

~,

Références

Documents relatifs

The problem we are interested in here is about a semilinear wave equa- tion with data given on two transversely intersecting null hypersurfaces.. Many problems

Intermittency and nonlinear parabolic stochastic partial differential equations. A local time correspondence for stochastic partial

For example, a large discrete category made concrete over Set by an ar- bitrary one-to-one functor to singleton sets is fibre-small and has suitable weak

As proved in Hough et al [9], the distribution of a (finite) determinantal process N can be viewed as a mixture of densities of some determi- nantal projection processes..

Our numerical algorithm is based on the coupling of a finite element solution of the wave equation, an exact controllability method and finally a Fourier inversion for localizing

In this mean-eld case, Graham and Meleard (1997) prove some stochastic approximations of the solution of the mollied Boltz- mann equation by interacting particle systems and obtain

In this situation we prove the existence of a weak solution and the existence ..œuLuniqueiLess jof_a smooth solution when the size of the domain increases fast enough with the

¥ If Universe is filled with cosmological constant – its energy density does not change.. ¥ If Universe is filled with anything with non-negative pressure: the density decreases as