You are on page 1of 86

Le

ture I 1

Le ture I: Broken symmetry, quasi-parti les & olle tive modes


Before leaping into ondensed matter phenomena, it is perhaps useful to develop some
perspe tive on the subje t in general. In this the rst le ture, I would like to dis uss
some of the guiding prin iples of ondensed matter physi s. Most of the thoughts are not
original and, therefore, perhaps they are useful!
. What are the guiding prin iples of ondensed matter physi s?1
a tually, high-energy physi s too!
 \Adiabati ontinuity" ;
 Con ept of quasi-parti les and olle tive modes;
 Importan e of (broken) symmetry
 and (perhaps most importantly) Universality
To understand why, let us ontemplate a...
. \Theory of Everything" (on Earth) | TOE
All matter an be des ribed S hrodinger equation
i ~t = H^
involving a many-body Hamiltonian of ele trons and ions

Xz }|X { zX X}| { X
^e
H ^i
H
N 2
p^
N
e2 M
P^2 N
(Ze) 2 N;M
Ze2
^=
H i
+ + I
+
i
2me i<j
jr
i r
j j I
2mi I <J
jR R
I J j i<J
jr R
i J j
(well, minus the ele trodynami eld...)
In prin iple, H^ des ribes metals, insulators, semi- ondu tors,
and everything else besides! | it as pre ise as it is useless
How, then, an one explore theoreti ally the behaviour of a many-body system...?
. Adiabati ontinuity and the quasi-parti le on ept
The development of ondensed matter physi s has, to a large extent,
hinged on the \unreasonable" su ess of non-intera ting theories...
viz. free ele tron theory of metals, Debye theory of solids, et .
...yet perturbation a e ted by intera tions (and the host environment)
are rarely small (see TOE) f. metals...
1
terminology of \ ondensed matter" attributed to Anderson and Heine(?) | (a) indi ates a broader remit
than \solid state", (b) referen e to universality impli it, and ( ) resonates with funding agen ies...

Le ture Notes O tober 2004


Le ture I 2

Apparent impoten y of intera tions attributed to prin iple of adiabati ontinuity:


 The quantum numbers (p; , et .) that hara terise a many-body system are deter-
mined by the fundamental symmetries (translation, rotation, parti le ex hange, et .) |
mu h more robust than \material parameters"
 While symmetry is maintained, the elementary ex itations of an intera ting system
an be tra ed ba k \adiabati ally" to those of the bare parti le ex itations in the non-
intera ting system, f. single-parti le states in a quantum well

λ
0 1

 Elementary ex itations | quasi-parti les | of intera ting system mirror ex itations


of non-intera ting system
. \Quasi-parti le orresponden e is embodied in Landau's Fermi-liquid theory (see next
le ture):
In the non-intera ting system, a single ele tron is indexed by momentum p and spin
 . In the intera ting system, the passage of ele tron is impaired by other parti les.
Yet, while the intera tion remains weak, even though the bare parti les may be ome
strongly \dressed" by their intera tion with the parti les in the ba kground, the elemen-
tary ex itations of the system an be lassi ed by \quasi-parti les" whi h share the same
quantum numbers (p,  ) as the bare parti les
More formally, the non-intera ting system provides a referen e state from whi h the
true ground state of the system an be inferred from perturbation theory (albeit of, perhaps,
in nite order!) Indeed, radius of onvergen e of perturbation theory extends beyond region
where perturbation is small
. Importan e of (broken) symmetry
However, being ontingent on symmetry, prin iple of adiabati ontinuity and, with
it, the quasi-parti le orresponden e, must be abandoned (or, at least, revised) at a
phase transition | here, intera tions a e t a substantial rearrangement of the many-
body ground state.
In the symmetry broken phase, a system may | and frequently does | exhibit ele-
mentary ex itations very di erent from those of the non-intera ting phase

Le ture Notes O tober 2004


Le ture I 3

These elementary ex itations may be lassi ed as new spe ies of quasi-parti les with
their own hara teristi quantum numbers or they may represent a new kind of ex itation
| a olle tive mode | engaging the olle tive motion of many bare parti les
e.g. when ions or ele trons ondense from a liquid into a solid phase, translational
symmetry is broken and the elementary ex itations | phonons | involve the olle tive
motion of many individual bare parti les
e.g. in the fra tional Hall liquid, the elementary quasi-parti le ex itations, a omposite
of many individual bare ele tron degrees of freedom, arry fra tional harge and exhibit
fra tional statisti s!
. Universality
Phenomenology above lends itself to a heirar hi al perspe tive of ondensed matter
physi s: ea h phase is asso iated with its own \non-intera ting" referen e ground state
with its own hara teristi quasi-parti le ex itations | a produ t of the fundamental
symmetries that lassify the phase.
Providing one stays within a given phase, one an draw on the prin iple of adiabati
ontinuity to infer the (usually benign) in uen e of intera tions from e.g. phenomenology
(as with FLT) or perturbation theory
. Yet, the heirar hi al pi ture brings two further profound impli ations:
 Firstly, within quasi-parti le pi ture, the underlying \bare" or elementary parti les
remain invisible;2 ( f. the fra tional Hall uid)
 se ondly, while the apa ity to on eive of new types of intera tions is almost un-
bound, the freedom to identify free (i.e. non-intera ting) theories is strongly limited,
and onstrained by the spa e of fundamental symmetries, i.e. drawing on the prin i-
ple of ontinuity, one an therefore anti ipate a substantial degree of \universality"
in properties in low-energy properties of ondensed matter.
e.g. photons, phonons, and antiferromagneti spin waves are all lassifed by the
same (relativisti ) low-energy theory
For many, this uni ation/ lassi ation is the primary goal of ondensed matter
physi s
All bran hes of ondensed matter physi s, quantum or lassi al, hard or soft, bene t
from this heirar hy.
. Therefore, as ondensed matter theorists, what is our goal? And what is the aim of
these graduate le tures?
2
To quote from P. W. Anderson's now-famous arti le More is di erent, (S ien e 177, 393 (1972)), \the ability
to redu e everything to simple fundamental laws does not imply the ability to start from those laws and re onstru t
the universe"

Le ture Notes O tober 2004


Le ture I 4

 Condensed matter \ab initio": although a symmetry-based lassi ation of physi s


is on eptually appealing, sometimes the devil lies not in phenomenology, but in
\detail" | anoni al behaviours may be ontingent on hemisty (e.g. is Yb inter-
al ated graphite a super ondu tor? The phenomenology of a super ondu tor wont
help answering this question!)
In these ases, the on ept of adiabati ontinuity often provides a means to justify
density fun tional theory s hemes to explore the TOE Hamiltonian
 Condensed matter \the Russian way": (a) isolate the \ anoni al" Hamiltonian of
the quantum system (i.e. the simplest one that aptures the essential physi s);
(b) apply methods of quantum eld theory or diagrammati perturbation theory to
un over the low-energy stru ture | foolproof but not \fool proof"!
 Condensed matter \the lazy way": Contingent on the on ept of universality is the
(perhaps miserable) realisation that there is (almost) \nothing new under the sun".
The majority of important low-energy or \quasi-parti le" theories have been identi-
ed, lassi ed, and \ anonised" in the literature.
The ingenuity of the ondensed matter physi ist lies, in part, in re ognising and
thereby pro tting from unexpe ted onne tions (or mappings) between seemingly
di erent physi s systems (e.g. the illuminating analogy that exists between polymer
dynami s and parti le physi s identi ed and harvested by Sam Edwards and others!)
Our sights this term are set on the \lazy" | we will fo us on phenomenology lassifying
many di erent themes within quantum ondensed matter:
 Broken symmetry, quasi-parti les & olle tive phenomena [1℄
 Landau's Fermi-Liquid theory [1℄
 Strong orrelation & the Mott transition [1℄
 Quantum magnetism [1℄
 Bogoliubov theory of weakly intera ting Bose gas [1℄
 Correlated system in one-dimensional & Luttinger Liquid theory [2℄
 Ele tron-phonon intera tion & the BCS theory of super ondu tivity [2℄
 Atomi ondensation phenomena [4℄
 Anderson lo alisation & orrelated lassi al insulators [2℄ (; Lent)
 Quantum Hall e e ts [2℄ (; Lent)

Next term, we will attempt a more te hni al synthesis fo using on appli ations of quan-
tum eld theory in ondensed matter

. Reading List:

Le ture Notes O tober 2004


Le ture I 5

P. W. Anderson, More is Di erent, S ien e 177, 393 (1972)


R. B. Laughlin and D. Pines, The Theory of Everything, Pro . Nat. A ad. of S i. 97,
28 (2000). http://large.stanford.edu/rbl/essays/p01apr99.htm
P. W. Anderson, Basi Notions of Condensed Matter Physi s, (Benjamin, Menlo Park,
CA, 1976).

Le ture Notes O tober 2004


Le ture II 6

Le ture II: Landau's Fermi-Liquid Theory

...a triumph of phenomenology!

.Referen es:
Original papers: L. D. Landau, The Theory of the Fermi Liquid, Sov. Phys. JETP 3,
920 (1956); Os illations in a Fermi Liquid, 5, 101 (1957)
D. Pines and P. Nozieres, The Theory of Quantum Liquids, vol. 1, Addison-Wesley 1989

. Motivation: Why does the free ele tron theory work so well as a des ription
of `real' metals and Fermi systems (viz. 3He, 6Li, et .) at low temperatures?
Is the intera tion energy small? With d the average ele tron separation, 34 d3 e = 1,
~ 2 2
hk:e:i  2md2
; hp:e:i  4"e d
0

i.e. hhpk::ee::ii  2"


me2
2
d=
d
r ,
s aB | ele tron Bohr radius
0~ Ba

Intera tions should dominate if2 r  1 (i.e. e small)


s
Indeed, for r  O(10 ), ele trons known (via QMC) to rystallise (Wigner)
s

Ceperly, ...
For metals, typi al values:
Be : 1:87  r  Cs : 5:62
s

not high enough to rystallize but Coulomb intera tion is strong


...so why does free-ele tron theory work so well?

Le ture Notes O tober 2004


Le ture II 7

Answer lies in prin iple of adiabati ontinuity


. (As dis ussed in rst le ture)
Landau: while the fundamental symmetries remain unbroken, there will be a one-to-
one orresponden e between new and old states | new states an be labelled by the same
quantum numbers
. Appli ation to Fermi systems
Ground state of non-intera ting Fermi gas is Fermi sphere
with ele trons lled up to the Fermi surfa e

kf kf

Elementary ex itations orrespond to parti le-hole ex itations


with allowed quantum numbers xed by Fermi surfa e and Pauli prin iple
k | empty state below E | hole
F
k0  0 | newly o upied state above E | parti le
F

Adiabati ontinuity 7! same true of intera ting system: Elementary ex itations are
quasi-parti les and quasi-holes about a lled quasi-parti le Fermi surfa e
Corollary: onsisten y of quasi-parti le pi ture relies on well-de ned parti le-like ex-
itations | i.e. long lifetime...
. Consisten y?
Lifetime of quasi-parti le of energy   E = j  j above Fermi surfa e?
k k F
Fermi's Golden rule:
1 ( ) = 2 X j V j 2 Æ (  ) -q
k-q

ij j

e ~ nal states j k

kf

Taking into a ount energy and momentum onservation, k’

1 ()  2 Z dk0 Z dq jV (q)j2Æ(E + E


q
k’+q

+k q E E )k 0
q k 0
k
e ~
Restri tion that ex itation energies fE > 0g presents severe onstraint
i
on momentum sum (exer ise)
1 ()   jV j2 2 ( T 2 at nite T)
e ~ 3F

Le ture Notes O tober 2004


Le ture II 8

for slow de ay, require 1=e  1; onsistent for =  1 F

. Physi al Consequen es:


Entropy: as with non-intera ting system, depends on allowed on gurations
f. Pk = Gk !=Nk !(Gk Nk )!, et .

S = kB
X p ln p = k X [n ln n + (1 n ) ln(1 n )℄
i iB k k k  k   
k
P
i

P
As usual, minimising S on nk subje t to onstraint k Ænk = 0
and k k Ænk = 0, obtain Fermi-Dira distribution

nk = 1 + e( k1  )=kB T

Energy: transfer to quasi-parti le pi ture may


(i) result in modi ed single-parti le spe trum, and
(ii) energy an depend on o upan y of other quasi-parti le states

kf

Setting nk = n k + Ænk , where nk is thermal equilibrium distribution


1 X f (k; ; k00 )Æn
   

qp = 
k  +
k F k0  0
V k0 0

Landau parameters f (k; ; k00) an depend both on angle kk between k and k0 as 0

well as relative spin orientation


Intera tions e e t both f (k; ; k0; 0) and (linearising around  ) F

2
k F ' ~mk (k F
kF )

(N.B. m 6= m is intera tion, and not latti e, e e t)


For spheri al Fermi surfa e, f (k; ; k0; 0) depend only on kk 0

i.e. separable into symmetri and antisymmetri spin omponents,


f (k;  ; k0  0 ) =
X [f P ( os  ) + f P ( os  )0℄
s a
` kk ` kk 0
` ` 0

Legendre polynomials: P0( os ) = 1, P1( os ) = os , et .


Le ture Notes O tober 2004
Le ture II 9

Quasi-parti le Hamiltonian
H =
X(  F )Ænk +
1 X Ænk f (k;  ; k0 0 )Ænk 
k
k
2V kk 0
 0
0 0

. Experiment
(1) Spe i heat: Only hange re e ted in modi ation of DoS through e e tive mass
mk
m :  (0) = 2 2 ,
F

 ~
2 
Cv = 3  ( )kB2 T = m3~k2
F
F
kB2 T

with Pk(Ænk" + Ænk# ) = 0, and


(2) Pauli spin sus eptibility: response to external magneti eld

M= 
X (Æn Æn )
k" k# B
k

an infer
X Æn X Æn = 2M
k" = k#
B
k k

kF kF

kF
∆E
kF

Shift in density distribution isotropi ; shift in energy


X
E =  B + V1 [(f0 + f0 )Ænk" + (f0 f0 )Ænk# ℄ =  Mf0a
BB
s a s a
B
k
B V

polarisation
z }| {
Therefore, noting M =  V  (0)E B

2 V  (0) 2  (0)
M= B; =
B B
1 +  (0)f0 1 + F0 a a

where F0 =  (0)f0 is (dimensionless) \Landau parameter"


a a

When F0a 7! 1, divergen e (i.e. transition to ferromagnetism)

Le ture Notes O tober 2004


Le ture II 10

(3) Compressibility: K = 1 V
V P
=
involves isotropi hange in density (exer ise)
1 n
n2 

V  (0) 1 1 = ~2k2 (1 + F )
K= ; v 2
= F s
n2 1 + F0 Kmn 3m m
s s0

F0 positive | requires more energy to distort Fermi surfa e


s

. Landau parameters are not independent:

Galilean invarian e | uniform translation of Fermi sphere


an be interpreted through F1s as quasi-parti le/quasi-hole ex itation
Mass transfer by quasi-parti les | no. quasi-parti les equals no. parti les
X Z dp p n =!X Z p
dp m n
qp

p p
 
p

translates to ondition (exer ise)


m
= 1 + F31
s

. What happens to bare ele trons?


j kqp i = Z 1 2 j kel i + p=h ex itations

= :


n qp T=0 n el.

kF kF
i.e. Fermi-surfa e dis ontinuity
Spe tral fun tion A(k; !)| probability of nding ele tron k at energy ! | broadened

A(k, ω)
2
ω

k F (k−kF )/m* ω

Le ture Notes O tober 2004


Le ture II 11

. Appli ations
0 bar 27 bar
3 He
m =m 2:8 5:26
s
F0 9:28 68:17
s
F1 5:39 12:79
a
F0 0:696 0:76
Metals: ompelling eviden e for Fermi-Liquid theory from heavy fermion materials
e.g. In UPt3 , m =mband  17 (measured by de Hass-van Alphen
and on rmed by spe i heat = C=T )

Le ture Notes O tober 2004


Le ture III 12

Le ture III: Strong Correlation: Mott Transition

. Referen es:
N. F. Mott, Metal-Insulator Transitions, Taylor-Fran is 1990
P. W. Anderson, Con epts in Solids, Harwood Publishing
P. W. Anderson, Lo al moments and lo alized states, Rev. Mod. Phys. 50, 191 (1978)
. Band theory of metals:
Completely lled and empty bands give insulating behaviour while partially lled
bands ondu t ; metals
H. Bethe, Theory of the di ra tion of ele trons in rystals, Ann. Phys. 87, 5 (1928).
A. Sommerfeld, An ele troni theory of the metals based on Fermi's statisti s. I. General
onsideration, ondu tivity and surfa e emission phenomena, Z. Phys. 47, 1 (1928).
F. Blo h, The ele tron theory of ferromagnetism and ele tri al ondu tivity, Z. Phys. 57,
545 (1929).
Band pi ture onsolidated with realisation that small band-gap materials ould be made
;
metalli by thermal ex itation of arriers or hemi al doping semi- ondu tots
A. H. Wilson, Pro . R. So . London, Ser. A133, 458 (1931); Pro . R. So . London,
Ser. A134, 277 (1931).
R. H. Fowler, Pro . R. So . London, Ser. A140, 505 (1933); Pro . R. So . London,
Ser. A141, 56 (1933).
However, early on, it was found that some ompounds (e.g. Ni0) do not respe t this
lassi ation:
J. H. de Boer and E. J. W. Verway, Pro . Phys. So . London, Se . A49, 59 (1937).
Some materials whi h should involve a half- lled band (i.e. \one ele tron per site")
behave as insulators
Following suggestion by Peierls,
R. Peierls, Pro . Phys. So . London, Ser. A49, 72 (1937).
\it is quite possible that the ele trostati intera tion between the ele trons prevents
them from moving at all. At low temperatures the majority of the ele trons are in
their proper pla es in the ions. The minority whi h have happened to ross the poten-
tial barrier nd therefore all the other atoms o upied, and in order to get through
the latti e have to spend a long time in ions already o upied by other ele trons.
This needs a onsiderable addition of energy and so is extremely improbable at low
temperatures."

Le ture Notes O tober 2004


Le ture III 13

it was argued by Mott that strong ele tron orrelations drive system insulating
N. F. Mott, Pro . Phys. So . London, Ser. A62, 416 (1949).
Qualitatively, behaviour an be understood from a \simple" model of strongly orrelated
ele trons:
. The Mott-Hubbard Model
Consider latti e \tight-binding" Hamiltonian:
H^ 0 = t X yi j + h: : = X knk
hij i k

e.g. simple 2d ubi latti e k = 2t( os kxa + os ky a)


together with lo al \on-site" intera tion:
H^ I = U X ni" ni#
i
generally (and here) U > 0 ! Coulomb repulsion
Control parameters: U=t, temperature, ele tron density, (and dimensionality)
. Expe ted properties of the half- lled system (viz. n = n" + n# = 1):

(i) U=t  1 expe t Fermi-liquid behaviour ( aveat: perfe t \nesting" symmetry an


indu e weak transition to insulating spin or harge density wave | see below; also
depends on dimensionality)
(ii) U=t  1 expe t ele trons to lo alise (one per atomi orbital). Sin e harge ex-
itation involves site double o upan y (energy ost U ), there is a harge gap ;
Mott insulator.
. How does insulating phase (ii) develop from metalli phase (i)? | the Mott transition

. Three prin iple and omplementary ideas expounded in the literature

(a) Mott-Hubbard: rooted in a modi ed band pi ture


U-4td
LHB UHB LHB UHB
DoS

Energy Energy

and not dire tly linked to any kind of magnetism...


(b) Slater: nesting symmetry ; development of (antiferro)magneti order; splits the
band and reates insulating phase (J. C. Slater, Magneti E e ts and the Hartree-
Fo k Equation, Phys. Rev. 82, 538 (1951).)

Le ture Notes O tober 2004


Le ture III 14

magnetic
BZ

( ) Brinkman and Ri e: seamless onne tion to Fermi-liquid theory through divergen e


of e e tive mass (W. F. Brinkman and T. M. Ri e, Appli ation of Gutzwiller's
Variational Method to the Metal-Insulator Transition, Phys. Rev. B 2, 4302 (1970).)
m
m
 1 U rit U

. Whi h pi ture is orre t? To some extent, all \phases" realized in experiment:


e.g. V2O3

T Paramagnetic Ins.
(Mott-Hubbard)

Fermi-liquid
(Brinkman-Rice)

AF insulator
(Slater)

Pressure (cf. t/U)

At present, reliable theory largely unavailable


it remains one of the great outstanding problems of ondensed mattter!
Some(?) intuition provided by \exa t" solution in limit of in nite dimension (but poor
des ription of magnetism)...
...and exa t solution in one-dimension
. Phase diagram away from half- lling?

Naively, one expe ts a jump in the hemi al potential as one rosses the transition:
Le ture Notes O tober 2004
Le ture III 15

U-2dt

...with transition to metal when U < 4dt


However, bands hange with density (it is not a \rigid band system"). In thermody-
nami limit, the jump (if it exists) may not orrespond dire tly to the Hubbard gap (e.g.
as seen in the d = 1 system, mid-gap states ould emerge)

In 1d: obtain a Luttinger liquid for n 6= 1 (see later le tures)


In 2d: it has been suggested (notably by Anderson) that the doped Mott insulator
may, indeed, not be a Fermi-liquid at all!
. Nature of Mott insulating state (n = 1; U=t  1)?

Suppose one ele tron per site ; massive 2N spin degenera y


| lifted at se ond order of perturbation theory...
In limit U=t  1, an treat
H^ 0 = t X yi j + h: :
hij i

as perturbation on lo al Hubbard intera tion H^ I = U Pi ni" ni# :


E = h 0 jH^ 0 1 ^ H^ 0 j 0i
0 HI

What intuition is provided by \two-site" problem?:


Le ture Notes O tober 2004
Le ture III 16

same spin: "1 "2 ! frozen (Pauli ex lusion)


1
^ 
^ z 0 }|HI { ^ "1 #2
opposite spin: "1 #2 H!0 01 "#2 !
H0
#1 "2
note sign hange due to ex hange
...in operator form, using the identity,
S^ 1  S^ 2 =
1 y y 1 n^ n^
2 1 2 1 2 4 1 2
obtain (exer ise)
^ = J S^ 1  S^ 2 J ; J = 4t2 > 0
H
4 U
i.e. an antiferromagneti ex hange intera tion of spins
eigenstates:
8
< j 1""i Sz = 1
S = 1 triplet p
: j ##i
2
(j "#i + j #"i) Sz = 0
Sz = 1

S = 0 singlet p (j "#i
1 j #"i) Sz = 0
2
ground state is a singlet with energy E = 34J
. Note that, unlike a ferromagneti intera tion (viz. J < 0), the lassi al ground state
(j "#i or j #"i) is not, by itself, an eigenstate | onsequen e of non-trivial zero-point
energy
. Generalised to many-site system, e e tive Hamiltonian of insulating phase given (up to
irrelevant onstant) by the quantum spin 1=2 antiferromagneti Heisenberg Hamiltonian
^ = J X S^ i  S^ j
H
hij i

. Nature of ground state?


For a time, it was thought that zero-point u tuations might destroy AF ordered phase
even at T = 0!
In fa t, ground state usually ordered but depends sensitively on latti e geometry, and
dimensionality (see next le ture)
Doping away from half- lling? Latti e t J Hamiltonian
^s | proje tion onto singly o upied sub-spa e
P

^ = t X P^s yi j P^s + J X S^ i  S^ j


H
hij i hij i

Le ture Notes O tober 2004


Le ture III 17

phase diagram? (lesson from uprates?)


. Summary: In the half- lled system, strong ele tron intera tion leads to
 Mott transition to insulating phase;
 formation of lo al magneti moments;
 with a quantum AF ground state ( f. tenden y of Hund's rule in atoms whi h
favours ferromagneti orrelations | exer ise: why the di eren e?)

A ording to lassi ation in le ture 1., development of magneti state, whi h involves
breaking of ontinuous spin symmetry, should be a ompanied by appearan e of low-energy
olle tive ex itations | spin waves. Goal of next le ture is to explore nature of spin wave
ex itations in the magneti state

Le ture Notes O tober 2004


Le ture III 18

Le ture Notes O tober 2004


Le ture III 19

Le ture Notes O tober 2004


Le ture III 20

Le ture Notes O tober 2004


Le ture III 21

Le ture Notes O tober 2004


Le ture IV 22

Le ture IV: Quantum Magnetism

. Referen es:
D. C. Mattis, Theory of Magnetism , Springer
A. Auerba h, Intera ting ele trons and quantum magnetism , Springer
. We have seen that residual intera tions in magneti insulators
an lead to an ex hange Hamiltonian for lo al moments:
X
H^ = J S^ m  S^ n
hmni
Sign of J depends on material parameters:
Coulomb intera tion favours ferromagnetism J < 0 ( f. Hund's rule)
and superex hange pro esses favour antiferromagnetism J > 0
What are the low-energy olle tive ex itations of the magneti system?

. Classi al ground states:


Ferromagnet: all spins aligned along a given dire tion
Antiferromagnet: (where possible) all neighbouring spins antiparallel
. Quantum ground states:
Ferromagnet: as lassi al
Antiferromagnet: spin ex hange intera tion (viz. Sm+ Sn ) ; zero point u tuations
. Formation of magneti ally ordered state breaks ontinuous spin rotation symmetry. So
should, by Goldstone's theorem, be a ompanied by low-energy olle tive ex itations |
spin waves. How an we explore ex itation spe trum?

. Elementary ex itations?
For large spin S , helpful to swit h to a representation
in whi h deviations of spin from g.s. are presented as bosons:
jS z = S i j n = 0i
j S z = S 1i j n = 1i
j S z = S 2i j n = 2i
 
jS z = S i j n = 2S i
i.e. n (\hard ore") bosons per site

Le ture Notes O tober 2004


Le ture IV 23

Mapping useful when spin ex itation involves n  2S


. Mapping of operators: S^z ; S^ = S^x  iS^y ?
with ~ = 1, operators obey spin algebra
[S^+ ; S^ ℄ = 2S^z ; [S^z ; S^ ℄ = S^
f. bosons: [a; ay℄ = 1; n = aya
A ording to mapping, S^z = S ay a; therefore, to rst approximation in S  1,
S^ ' (2S )1=2 ay ; S^+ ' (2S )1=2 a
In fa t, exa t equivalen e provided by Holstein-Primako transformation
 y a 1=2
y a
S^ = (2S ) a 1
1 = 2
; S^+ = (S^ )y; S^z = S aya
2S

. Ferromagneti Heisenberg spin S hain

N 1 2

Quantum ground state: all spins aligned along, say, z -dire tion,
S^iz jg:s:i = S jg:s:i for all i
In `spin-wave' approximation
N 
X 
1
H^ = J S^mz S^mz +1 + (S^m+ S^m+1 + S^m S^m++1 )
m=1
2
Xn o
= J S2 S (ay a
m m a y a ) + S (a a y
m+1 m+1 m
y
m+1 + am am+1 ) + O (S )
0

m
X 
= J S2 2Saym am + S aym am+1 + h: : + O(S 0 )
m

p.b. . S^m+N = S^m and am+N = am


Hamiltionian bilinear in bose operators; diagonalised by dis rete Fourier transform,
N B:Z:
1 X 1 X
ak = p e an ;
ikn
an = p e ikn
ak ; [ak ; ayk ℄ = Ækk 0

N N
0

n=1 k
P
ne = NÆkk
noting i(k k0 )n 0

B:Z:
X
H^ = JNS 2 + !k ayk ak + O(S 0); !k = 2JS (1 os k) = 4JS sin2 (k=2)
k

Le ture Notes O tober 2004


Le ture IV 24

f. free-parti le spe trum


Higher order orre tions ; spin-wave intera tions
How does this ompare to the AF spin system?

. Antiferromagnet Heisenberg spin S hain


X
H^ = J S^ m  S^ m+1
m

Classi al g.s. no longer an eigenstate | nevertheless, lassi al Neel state


serves a referen e for spin-wave expansion
Convenient to implement anoni al transformation in whi h spins on one
sublatti e, say B , are rotated through 180o about the x-axis,
i:e: S^Bx 7! S^Bx ; S^By 7! S^By ; S^Bz 7! S^Bz

X 1 ^+ ^ +

H^ = J S^mz S^mz +1 (S S + S^m S^m+1 )
m
2 m m+1

applying spin wave expansion: Sm ' (2S )1=2 aym , et .


Xh i
H^ = NJS 2 + JS 2aym am + am am+1 + aym aym+1 + O(S 0)
m

; pro esses that do not onserve parti le number!


P
Fourier representation: am = N 1=2
ke
ikm a ,
k et .
  
X 1 k ak + O(S 0)
H^ = NJS (S + 1) + JS ( ayk a k)
k
k 1 ay k
where k = os k.
Bogoliubov transformation: ( f. Lorentz boost)
    
k osh k sinh k ak
=
y k sinh k osh k ay k
preserves ommutation relations, N.B. metri g = z
o -diagonal terms removed by setting osh(2k ) = (1 k2 )1=2 = j sin kj
X  
H^ = NJS 2 + 2JS j sin kj ky k + 21 + O(S 0)
k

; linear (i.e. relativisti ) ex itation spe trum


Le ture Notes O tober 2004
Le ture IV 25

. Magnetisation and sublatti e magnetisation


Ferromagnet:
X X X
h N1 S^iz i = S h N1 ayi ai i = S h N1 ayk ak i
i i k
Z Z
dk 1 kB T 1
=S
(2 )d e !k 1
 J
kd 1 dk
k2
6 0 in d  2, i.e. no long-range order
divergent for T =
Antiferromagnet:(sublatti e magnetisation)
At T = 0, do quantum zero-point u tuations destroy long-range order?
X X 1 X
hg:s:j N1 hg:s:j N1

S^iz jg:s:i = S ayk ak jg:s:i = S (1 k2 ) 1=2
1
i k
N k

diverges only in d = 1!
. Frustration
AF ex hange intera tion on bipartite latti e Neel ordering;
whi h, in d > 1 survives zero-point u tuations
However, suppose that latti e is not bipartite (or spin intera tions long-ranged)
;
frustration of Neel order, e.g. triangular latti e
Can zero-point u tuations lead to spin liquid phase...?

Le ture Notes O tober 2004


Le ture IV 26

Le ture Notes O tober 2004


Le ture IV 27

R. Coldea et al. , Spin Waves and Ele troni Intera tions in La2 CuO4 , 2001

Le ture Notes O tober 2004


Le ture V 28

Le ture V: Bogoliubov Theory of weakly intera ting Bose gas

Referen es:
e.g. Abrikosov, Gor'kov and Dzyaloshinski, Methods of Quantum Field Theory in
Statisti al Physi s, Dover
The theory of super uidity is better postponed to the eld integral dis ussion next term.
However, a mi ros opi theory of the weakly intera ting Bose gas an be developed within
a se ond quantised pi ture. Here we dis uss the e e t of weak intera tions on the nature
of the Bose-Einstein ondensed state and the spe trum of low-lying ex itations around the
ground state.
. Weakly intera ting Bose gas
Consider N bosons on ned to a volume L , density n = N=L , d d

In non-intera ting system, at T = 0, all bosons ondense


into the lowest energy single-parti le state jg:s:i = (ay0) j
i N

How is ground state in uen ed by weak intera tion?


H^
z Z }| {
I

X ~2 k2 Z
H^ = ayk ak +
1 dx dx0 ay(x)ay (x0 )V (x x0 )a(x0 )a(x)
k
2 m 2
V (x x0 ) | pairwise parti le intera tion
In Fourier basis:
H^ =
1 X
V (q) ayk+q ayk ak ak +q
2L 0 0
I
d
k;k ;q
0

ak = L d=2 R dx e kxa(x),
and V (q) = dx e qxV (x).
i
R i

Suppose intera tion weak | in ondensed phase, we may assume that the lowest-lying
single parti le state is still ma ros opi ally o upied, i.e. N0 =N = O(1)
Sin e N^0 = ay =0a =0 = O(N )  1, the ommutator
k k

a0 ay0 ay0 a0 = 1
is small ompared to a0 and ay0 themselves, i.e. to a good approximation, p
eld operators an be repla ed by an ordinary -number N0
Then, taking V (q) = V onst., i.e. a onta t intera tion
momentum onservation identies the ontributions:
 
^
H =
V 2 V X y y 1 
y y 0
I
2L N0 + L N0 k6=0 ak ak + a ka k + 2 a k ak + aka k + O(N0 )
d d

Le ture Notes O tober 2004


Le ture V 29

f. AF Hamiltonian in spin-wave approximation


. Physi al interpretation of omponents:
 V aykak represents the `Hartree-Fo k energy' of ex ited parti les intera ting with on-
densate | note that the onta t nature of the intera tion disguises the presen e of
the dire t and ex hange ontributions
 V (a kak + aykay k) represents reation or annihilation of ex ited parti les from the
ondensate; in the approximation, total no. of parti les is not onserved
P
Final part of approximation involves noting that N = N0 + k6=0 aykak ,
and using this identity to trade N^0
  
V nN X 0  Vn
H^ = + k + V n ayk ak + a k ak + ayk ay k
2 k6=0
2
n = N=V | total density and 0k = ~22 k2 m

As with spin wave theory, Hamiltonian diagonalised by Bogoluibov transformation:


    
ak = osh k sinh  k
ay k y k
k

sinh k osh k
Left as exer ise to show that, when
 0   0 
2
osh k = 21 k + V n
+1 ; 2 1
sinh k = 2 k + V n
1
k k

H^ =
V nN 1 X(0 + nV k ) +
X
k ky k ;
h 
k = 0k + V n 2 (V n)2
i1=2
2 2 k6=0 k k6=0

Spe trum of low-energy ex itations s ales linearly as (k) ' ~ jkj


where the velo ity = (V n=m)1 2 =

High-energy spe trum is free parti le-like, i.e. when k  k0 = m =~,


e e t of intera tion is to displa e parti les from ondensate even at T = 0
. Ground state wavefun tion: de ned by ondition k jg:s:i
Now Bogoluibov transformation an be written as k = Uak U 1 where (exer ise)
" #
X k
U = exp y y
k6=0
2 (aka k ak a k )

Sin e g.s. of ideal Bose gas given by akj0 i = 0 (all parti les in k = 0 state),
new ground state is j i = U j0 i | proof:
V

k
z }| {
0 = ak=0 j0i = U 1 Uak U 1 U j0 i
Le ture Notes O tober 2004
Le ture V 30

. Depletion of ondensate by intera tion


Z
N N0 1
= N hg:s:jakakjg:s:i = N sinh k = n (2dk)3 sinh2 k = 312n k03
1 1
X X
y 2
N k6=0 k6=0

i.e. a. one parti le per ` oheren e length'   1=k0


Re ast using s attering length of onta t intera tion V = 4~2a=m,
N N0
= 8 (na3 )1 2
p =

N 3 
. Ground state energy

E0 =
V nN 1 X(0 + nV k
(nV )2 )
2 2 k6=0 k 20k
(where extra term ontrols unphysi al divergen e UV divergen e
required for onta t potential approximation)
 
E0 n2 V 128
L
= 2 1 + 15p (na )
d
3 1 2 =

. Experiment
When ooled to T  2K , liquid 4 He undergoes transition to super uid state
In Helium, steri intera tions are strong | at higher energy s ales
an important se ond bran h of ex itations known as rotons appear

Le ture Notes O tober 2004


Le ture V 31

Le ture Notes O tober 2004


Le ture V 32

M. R. Andrews, et al. Phys. Rev. Lett. 79, 553 (1997).

Le ture Notes O tober 2004


Le ture VI 33

3
Le tures VI: Correlated Quantum Spins in One Dimension

. Referen es:
Original papers:
 H. Bethe, Z. Phys. 71, 205 (1931), for an exa t diagonalisation of the S = 1=2
Heisenberg hain; method not useful for orrelation fun tions.
 A. Luther, Phys. Rev. B 14, 2153 (1976); A. Luther and I.Pes hel, Phys. Rev.
B 12, 3908 (1975), for mapping of S = 1=2 Heisenberg hain to free massless
bosons, and orrelation fun tions.
Books and Reviews:
 A. O. Gogolin, A. A. Nersesyan, and A. M. Tsvelik, Bosonization and Strongly
Correlated Systems (Cambridge University Press, 1998).
 H. J. S hulz, G. Cuniberti, and P. Pieri, Fermi Liquids and Luttinger Liquids,
ond-mat/9807366.

. Notable features
(a) The S = 1=2 antiferromagneti hain never orders, however at T = 0, it is riti al,
i.e., orrelation fun tions de ay algebrai ally.
(b) The S = 1 hain develops a spin gap ) un orrelated over long distan es.
. The Heisenberg S = 12 hain
X 1 

H =J 2 S +S
i j + Si S + + S z S z
j i j (1)
hij i

 At any given site i, the spin operators Si ; Si+ anti ommute,


fSi+; Si g = 1;
prompting us to make an analogy with fermions,
= iy; Si = i ; Siz = iy i 21 :
Si+ (2)
Unfortunately, the analogy stops here for spins at di erent sites ommute whereas fermions
at di erent sites anti ommute, e.g.,
f i ; jyg = 0; [Si+; Sj ℄ = 0 i 6= j:
3
Le ture notes prepared by Vikram Tripathi

Le ture Notes O tober 2004


Le ture VI 34

Another angle : The spin operators Si ; Sj+ behave as annihilation and reation opera-
tors for bosons at di erent sites (i 6= j ). The presen e or absen e of a boson at a site an
be interpreted as spin-up and spin-down respe tively. Of ourse, you are not allowed to
put two bosons at any site be ause you an only have either spin up or spin down, and
not both. Thus the bosons have a hard- ore hara ter. And hard- ore bosons prevent
double-o upation at any site whi h is a property of fermions also sin e two identi al
fermions will never o upy the same site.
What should be done? We need to modify the analogy in Eq.(2) appropriately so
that an extra negative sign is introdu ed when ommuting spins at di erent sites. The
transformation was obtained by P. Jordan and E. Wigner, Z. Phys. 47, 631 (1928).
. (1928) The Jordan-Wigner transformation in 1D

Write
" #
X
Si+ = y exp
i i ^
nj
j<i
" #
X
Si = exp i ^
nj i; ^ =
nj
y
j j:
j<i
h i
The Jordan-Wigner fa tors exp i Pj<i n^ j introdu e just that extra negative sign
be ause of relations su h as
exp[inj ℄ j = j ; and j exp[inj ℄ = j e =
i
j:

 Physi al meaning of the transformation : One way to interpret the transformations


is to onsider fermions to whi h a single ux quantum is atta hed. When two su h
` omposite-fermions' are ex hanged, their relative displa ement is rotated by . This is
what the Jordan-Wigner phase fa tor ounts.
Consider a state with two parti les labelled A and B at sites i and j . Let B be to
the right of A and let there be n parti les in between. Ex hanging A with B involves
ex hanging A with ea h of the n parti les to its right, then an ex hange with B , and
nally and ex hangen of B with n parti les to its left. That results in a umulation of an
extra fa tor of ( 1)  ( 1)  ( 1) = 1.
n

. (1989{1993) The Jordan-Wigner transformation in 2D and 3D

De ning moments: E. Fradkin, Phys. Rev. Lett. 63, 322 (1989); D. Elizer and G. W.
Semeno , Phys. Lett. B. 286, 118 (1992); L. Huerta and J. Zanelli, Phys. Rev. Lett.
71, 3622 (1993).
In 2D, the transformation (due to E. Fradkin) is
" #
X
Si+ = y exp
i i arg(j i) nj :
j 6=i

Le ture Notes O tober 2004


Le ture VI 35

The transformation in 3D was obtained by L. Huerta and J. Zanelli: Phys. Rev. Lett.
71, 3622 (1993).
 The S = 1=2 Heisenberg hain in 1D an now be re ast as
N     
H =J
X 1  1 1
i i+1 + i+1 i +  (3)
y y y y
i+1 i+1
hi=1i 2 i i
2 2
Observe that  = 0 orresponds to non-intera ting fermions. The `kineti ' term may
be diagonalised going over to the momentum representation,
y = p1 X eik j y (k);
j
N k
the Hamiltonian takes the form
( os(k) ) y(k) (k) + J2N eiq y(p + q) y(k ) (k) (p) + J N4  :
X X
H =J q
k p;q;k
It is remarkable that the ex itations of a hain of Heisenberg spins oupled antiferro-
magneti ally in the XY plane are the same as non-intera ting ele trons! This feature is
pe uliar to 1D be ause of the lose similarity between hard- ore bosons and fermions: they
are related to ea h other through a ounting fa tor, the Jordan-Wigner phase, and ount-
ing doesn't ost energy. In higher dimensions, the Jordan-Wigner phase ontributes to the
low energy ex itations, hen e the ex itation spe trum is no longer that of non-intera ting
fermions.
(a) Consider  = 0: The XY hain
E
∆=0

−π π
k
0

Figure 1: Ex itation spe trum for an XY hain. At half lling (indi ated as bold urve) the
average polarisation hSz i = 0.
Z 
 The ground state energy E = 2J N (dk) os(k) = NJ :
=2
 It's also possible to al ulate the orrelation fun tions:
( ) = h0jS0x Srxj0i = 41 h(S0+ + S0 )(Sr+ + Sr )i0
C r

= 14 hS0 Sr+i0 + . .
* " # +
1 Xr
= 4 0 0 ry exp i ni 0 + . .:
i=0

Le ture Notes O tober 2004


Le ture VI 36

Repla e the phase fa tor by its mean value, hnii = 1=2:


C (r ) =
1
0 y 0  eir=2 + . .
4 X0 r
= 41N
0 eikr (k) y(k) 0  eir=2 + . .

Z k
1
= 4 (dk) eikr  eir=2 + . .

1  
= 4r sin r2  eir=2 + . . = 8r1 sin(r):

) Os illatory, i.e., antiferromagneti orrelation. Note the slow power-law de ay of


( ). So the 1D XY hain at T = 0 is not quite ordered (for C (r) de ays), nor is it
C r
fully disordered (for then C (r)  exp( r= ) sin(r)).
(b) Consider  = 1 or less: Ferromagneti Heisenberg hain

( os(k) ) y(k) (k) + JN ) (k) (p) + J N4  :


X X
H =J eiq y (p + q ) y (k q
k p;q;k

. If   1, then the `kineti ' term is always positive ) we want an empty band as
the ground state. An empty band also diagonalises the se ond term in the Hamilto-
nian.
) Ground state is fully polarized. Ex itations are spin waves.
( ) Case of arbitrary  > 1: need to understand intera ting fermions in 1D, i.e., a
Luttinger Liquid. To learn more about this ase, refer to the book by Gogolin,
Nersesyan, and Tsvelik.

Le ture Notes O tober 2004


Le ture VII 37

4
Le ture VII: Luttinger Liquid Theory

. Referen es:
Original papers:
 Sin-itiro Tomonaga, Remarks on Blo h's Method of Sound Waves applied to
Many-Fermion Problems, Progress of Theoreti al Physi s 5, 544 (1950).
 J. M. Luttinger, Journal of Mathemati al Physi s 4, 1154 (1963).
 F. D. M. Haldane, `Luttinger liquid theory' of one-dimensional quantum u-
ids:I. Properties of the Luttinger model and their extension to the general 1D
intera ting spinless Fermi gas, Journal of Physi s C: Solid State Physi s 14,
2585 (1981).
Books and Reviews:
 A. O. Gogolin, A. A. Nersesyan, and A. M. Tsvelik, Bosonization and Strongly
Correlated Systems (Cambridge University Press, 1998).
 H. J. S hulz, G. Cuniberti, and P. Pieri, Fermi Liquids and Luttinger Liquids,
ond-mat/9807366.

. Why is 1D spe ial?


In 1D, parti le-hole ex itations with a given momentum have the same kineti energy.
This holds even in ase of intera ting ele trons. The only requirement is that the spe trum
is linear about the Fermi points.
E

δq
kF
q δE

q
δE
q
δq

(a) (b)

Figure 2: (a)For a xed q in 2D, a parti le-hole pair an have both a high and low energy of
ex itation. (b) In 1D, the ex itation energy has a single value for a given Æq.

. How does this pe uliarity a e t the physi s?


4
Le ture notes prepared by Vikram Tripathi

Le ture Notes O tober 2004


Le ture VII 38

Intera tions are usually expressed in density: V (x x0 )^(x)^(x0 ). In se ond quantised


form this looks like
X X
V (x x0 )^(x)^(x0 ) = V (q ) yk+q yp q k p
x;x0 k;p;q
X
 V (q )^q ^ q ; (4)
q

where X
^q = yk+q q (5)
k

is a harge density wave ex itation with momentum q.


So if the parti le-hole ex itation (q) has a well-de ned kineti energy, then it means
that (q) is an eigenstate:
H^ ^(q )j0i = Eq ^(q )j0i: (6)
 The statement is a tually stronger. A ording to Blo h, the low energy states of a
1D ele tron system are ompletely exhausted by its density ex itations.
. The Luttinger model
E E

δ q’

EF EF

δq
k k
−k k −k k
F F F F
(a) (b)

Figure 3: Violation of one-to-one orresponden e. (a) Ex itations involving momentum hange


of the order of 2kF (ba ks attering). (b) Parti le-hole ex itations with large energy.

An idealised model of 1D ele trons an be written down whi h ex ludes pro esses that
violate this one-to-one orresponden e of q and eigenstates of H^ , and renders the model
exa tly solvable even with strong intera tions, is alled the Luttinger model. Haldane
(1981) onje tured that all intera ting 1D metals are adiabali ally ontinuable from the
Luttinger model leading to the on ept of a Luttinger liquid. In the Luttinger model we
(a) Linearise the spe trum.
Le ture Notes O tober 2004
Le ture VII 39

kF
k
Left branch Right branch
η = −1 η = +1
k
k F− q

(say on the η = + branch)


k
−kF kF

Filled states

(a) (b)

Figure 4: The Luttinger model. (a) Linearizing and adding an in nity of lled states (bold
lines) with negative energy. (b) Illustration of nk q and nq . Their di eren e appears in the
ommutation relation of densities  (q).

(b) Add an in nity of lled states with negative energy.


Ignoring spin for the moment|write the density as
X
 (q ) = y;k+q ;k :
k

This operator behaves like a boson:


[ ( q);  (q0)℄ = Æ Æqq qL
2 : 0 0 0

By simply substituting the expression for  (q) in terms of ele trons, the ommutator
gives X
Æ Æqq (nk q nk ):
0 0

gives zero (just put k~ = k q) but with an in nite number


P
For a nite bandwidth the k
of lled states we have X qL
(nk q nk ) =
2 :
k

This is how it works in the ground state but it just relies on there existing some k 0
below whi h all states are lled (in the ground state k = kF ) so it is quite generally true.
0
We make a set of bosons now:
r
= 2 y ;
X
by
+;q
Lq k +;k+q +;k
r
by ;q
2 X y ;
= Lq (7)
;k ;k +q
k

Le ture Notes O tober 2004


Le ture VII 40

so that the bosons b;q satisfy the familiar Bose ommutation relations,
[b;q ; by q ℄ = Æ Æqq0:
0 0 0

Sin e there is one-to-one orresponds between density u tuations and the low-energy
eigenstates of the Hamiltonian, we may express the original problem of ele trons in terms
of these bosons:
H =
2vF X  (q) ( q)
0
L  
q>0; = 1
X
= vF q byq bq : (8)
q>0; = 1
 Introdu ing intera tions (g ology)
+ + + +

− − + +
g2 ρη (q) ρ− η (−q) g4 ρη (q) ρ η (−q)

Figure 5: The g ology of intera tions. The diagram to the left shows intera tion of left movers
with right movers. The diagram to the right denotes intera tion of the same spe ies.

H=
X
vF q byq bq +
1 X g  (q ) ( q ) + 1 X g  (q ) ( q ):

q>0; = 1
2L q> ;  
0 =

1
2L q> ;   
2
0 = 1
4

Now express this in terms of the bosons:


X h qg y qg  y y i
H= y
vF q bq bq +
2 bq bq + 4 bq b q + bq b :
4 2
q

q>0; = 1

The Hamiltonian is quadrati and so an be solved exa tly via a Bogoliubov transfor-
mation (i.e., a rotation that permits mixing of b and by) to a new basis of bosons ,
y :
r  
H=
X
q  vF +
g  g  y 1 2 2

2 q q + 2
4 2

q> ; 0 = 1
2
So the intera ting 1D ele tron model looks
q exa tly like the Hamiltonian for waves on a
stret hed string with wave velo ity v = vF + g4 g2

. 2
2
2
2

. Can we represent single ele trons in terms of bosons?


To be really useful, we would like to know how to write a single ele tron in terms of
the bosons ) then we an use our solution to write orrelation fun tions.
Le ture Notes O tober 2004
Le ture VII 41

It turns out that you an do this (Luther and Pes hel0 74 and Haldane081). Details will
not be provided here. Refer to the review by S hulz I have ited in the reading list at the
beginning of this le ture. One de nes eld operators (essentially going ba kwards in our
se ond quantization treatment of the string) whi h are usually expressed in terms of 
or b; by and not the Bogoliubov elds ; y. This means the ele tron is truly lost in the
many-body physi s ontained in the Bogoliubov elds.
De ne the following two Bose elds in real spa e orresponding to displa ement and
the anoni al momentum:
i X 1 Æjqj= iqx
(x) = e
L q=6 0 q
[ (q) +  (q)℄ Nx ;
2
+
L

(x) = L1 e [ (q)  (q)℄ + NJ ;


X
jj
Æ q =2 iqx
+
6
q =0

where N = N + N and J = N N and Æ is a small number to ensure onvergen e.


+ +

You may verify that the anoni al ommutation relations are satis ed:
[(x); (x0 )℄ = iÆ(x x0):

 Comparison with a stret hed string


Compare the Hamiltonian
Z  
uK u
H= dx
2 (x) +
2K (2
r )2 ;
R h i
with a stret hed string: H = dx 1
2
(x) + (r) : Here
2 1
2T
2

r r
u=
g 2
vF + 4
 g 2 2vF + g g :
2 ; K=
2 4 2

2 2vF + g + g
4 2

 Expression for ele tron operator


 Z x 
1
p U exp ikF x  i(x) + i (x )dx :
0 0
 (x) = Ælim
! 2Æ 
0

. In luding spin: spin- harge separation


So far we have just onsidered spinless fermions in 1D. Adding spin enables us to
see that the spin and harge of an ele tron ompletely separate. There is no way to
adiabati ally ontinue from non-intera ting ele trons to the resulting ex itations. Spin-
harge separation o urs without expli itly spin dependent for es.
Le ture Notes O tober 2004
Le ture VII 42

+ −

g
1

+ −

Figure 6: New kind of intera tion involving spin. An intera tion on the left bran h s atters
from one on the right bran h and a spin ip o urs. This term annot be expressed as a pair of
bosons.

Adding spin to the ele trons adds an additional label . It also allows a new type of
intera tion [See Fig.6℄ This term annot be expressed as a pair of bosons, so I will ignore
it for the moment.
X h qg  y y  qg i
H= y
vF qÆ bq bq + y
4 bq b q + bq b q + 2 bq bq
2 4
0 0 0 0

q>0;; 0  = 1 

Now writing
p1 byq" + byq#
 
byq = : harge
2
byqs
1
= p byq" byq#

s: spin
2
Then we have
X h  qg  y y  i
H= vF q byq bq + byqs bqs +
4 bq b q + bq b
2
q + qg y
2 bq bq ;
4


q>0  = 1

whi h an again be diagonalised by a Bogoliubov transformation:


" r   #
y + 1 + qv y
X  g   g 
q  vF +
2 2
H= q
4 2
F qs qs :
q> ; 
0 = 1
2 2 q
2
For the q < 0 bran h, q should be repla ed with q above.
 The harge and spin densities move with di erent velo ities!
These ex itations exhaust the Hilbert spa e of the Hamiltonian|this is the omplete
solution.
Spe tral fun tions
Rather than write down the expression for the spinful ele tron, it's perhaps more
useful to quote some results. Compare the spe tral fun tion A(!) of a Fermi liquid and
a Luttinger liquid:
Le ture Notes O tober 2004
Le ture VII 43

Α(ω) Α(ω) Α(ω)

ω ω ω
~2 kF (k kF )
~2 kF (k kF ) vs k v k
m m

Figure 7: (Left to right) Spe tral fun tions for non-intera ting ele trons, Fermi liquid and a
Luttinger liquid respe tively
(a) (b)
ne (k) ne
T=0

Infinite slope but no jump

k k k k
F F

Fermi liquid Luttinger liquid

Figure 8: Ele tron distribution fun tion in 1D for (a) a Fermi liquid and (b) a Luttinger liquid.
In a Fermi liquid, the distribution fun tion has a nite dis ontinuity at kF while in a Luttinger
liquid, the distribution fun tion is ontinuous although its slope is in nite at kF .

The T = 0 ele tron distribution fun tion


. The Luttinger liquid hypothesis
You might wonder what happened to the possible g intera tion shown in Fig.6. Also, 1
the ele tron spe trum is not quite linear whi h may lead to deviations from Luttinger
liquid behaviour. Haldane (1981) suggested that in 1D, a new type of adiabati ontinuity
might operate: any intera ting 1D metal is adiabati ally ontinuable from the Luttinger
model. (Formally, g an be shown to be irrelevant). All of the physi s is determined
1
by three parameters: v ( harge velo ity or holon velo ity), vs (spin velo ity or spinon
velo ity), and K , whi h ontrols power laws of the ele tron orrelation fun tion. The
fourth possible parameter, K  1 for physi al intera tions.
. Appli ations

Le ture Notes O tober 2004


Le ture VII 44

 Quantum wires
 Organi s: TTF{TCNQ (CDW instabilities)
 SrCuO (observation of spinon and holon dispersion in photoemission experiments
2
by Z.X.Shen et al.)

Le ture Notes O tober 2004


Le ture VIII 45

Le ture VIII: Ele tron-phonon intera tion and Cooper instability

. History:
 1911 dis overy of super ondu tivity
 1951 \isotope e e t" | lue to ( onventional) me hanism
 1957 BCS theory of onventional super ondu tivity
 1986 Dis overy of un onventional super ondu tivity in uprates
 ???? awaiting theory?
. (Conventional) me hanism: ex hange of phonons an indu e (non-lo al) attra tive
pairwise intera tion between ele trons
Ele tron pairs an ondense into ma ros opi phase oherent state
with energy gap to quasi-parti le ex itations
. How does phonon ex hange mediate intera tion...?
k−qσ

Ele tron-phonon intera tion: kσ

H^ H^
zX }| X0
{ zX  }| {
1

^
H =
y
k k k +
y
~!q aq aq + y
Mq k q k aq + h: : +
k q k ;q

simpli ity: assume Mq = M onstant


E e tive Hamiltonian: eliminate phonons by anoni al transformation
H^ 7 ! H^ 0 = e S H^ eS ( f. se ond order perturbation theory)
^ ^

S ^ S ^ S^℄ + 1 [[H;
^ + [H; ^ ^ ^
e =H
2 S ℄; S ℄ +   
^ ^
e H

\S hrie er-Woolf transformation"


Terms 1st order in M eliminated by setting H^ + [H^ ; S^℄ = 0, viz.
1 0

h i
^ 0 = H^ + 1 H^ ; S^ + O(M )
H
3

20 1

In basis of non-intera ting Hamiltonian: jni = jfkgi :


jfqgi :, fEng = fk + ~!q g
el ph

hnjH^ jmi = hnjS^H^ H^ S^jmi = (Em En)hnjS^jmi


1 0 0

^
hnjS^jmi = hEnjH jEmi ; 1

m n

Le ture Notes O tober 2004


Le ture VIII 46

X  
i:e: hnjH^ 0jmi = hnjH^ jmi + hnjH^ jpihpjH^ jmi 1 +E 1 + O (M )
3
0 1 1
Em Ep n Ep
p

In operator form,
 
^ 0 = H^ + X jM j 1 1
2
H + ( y y
k0 +q0 k0 0 k q k
0
2 kk0 q
(k k q ) ~!q k0 +q k0 ) ~!q
X ~!q
= H^ + jM j (
0
2

k q )
2
~2 !q2
y
k0 +q0 k0 0 k q k :
y
kk q k 0

translates to an attra tive (i.e. negative) intera tion for ex itation energies
in range jk k qj < ~!q, and repulsive otherwise
N.B. onsisten y: trun ate !q sum at Debye frequen y, !D ' q vS max
maximum quantum that an be ex ited among latti e modes
. Cooper instability

Consider two ele trons propagating above a lled Fermi sea:


Is a weak pairwise intera tion V (r r ) suÆ ient to reate a bound state of pair?
1 2

Consider variational state


z
spin }|
singlet {spatial symm: g = g
zX }| k { k
(r ; r ) = p1 (j " i
j # i j " i
j # i) gk e
ik  r r ( 1 2)
1 2
2 1 2
jkjk
2 1

Applied to S hrodinger equation: H^ = E


X X ik  r
gk [2k + V (r r )℄ eik r r = E
1 gk e 2
( 1 2) ( 1 r2 )
k k
R
Fourier transforming equation: L d
d (r 1 r 2 )e 
ik0 (r1 r2 )

X Z
Vkk0 gk0 = (E 2k)gk; Vkk0 = dr V (r)ei k ( k0 )r
k0
If we assume
n
Vkk0 = 0 V fj k F j; jk
otherwise
0 F jg < ! D

X
V gk0 = (E k gk )
k0
X 1 X X
V
E 2k gk0 = gk
k k0
k
X 1
V
k
E 2k = 1
Le ture Notes O tober 2004
Le ture VIII 47

X Z  
V
1 '  ( )V
F +!D
d
= ( 2 ) ln 2F +
 F V 2~!D E
=1
k
2k E
F
F 2 E 2F E
In limit of weak oupling, i.e.  (F )  1
2~!D e ' 2
2
 (F )
E F

 i.e. pair forms a bound state (no matter how small intera tion!)
 energy of bound state is non-perturbative in  (F )
P
Radius of pair wavefun tion: g(r) = kg (k)eikr
R P R F +!D
drr2 jg (r)j2
hr i = R drjg(r)j2 = Pk jrjgk(gk()kj)2j = 43 (2
2 4d
2 vF 2
2
' R F +!D F (2 E )4
d
vF
E ) 2
k F (2 E )2
F

if 2F E  k T , T  10K, v  10 m/s,  = hr i  10 A,


B F
8
0
2 1=2 4

i.e. other ele trons must be important


. BCS wavefun tion
two ele trons in a paired state has wavefun tion
(r r ) = ( j " i
j # i j # i
j " i) g ( r r )
1 2 1 2 1 2 1 2

with zero entre of mass momentum


Drawing analogy with Bose ondensate, let us examine variational state
Y
N

(r    r N ) = N
1 2  (r n r n )
2 2 1
n=1

P 
Is state ompatible with Pauli prin iple? Using g(r 1 r2 ) = k gk e
ik (r1 r2 )
or, in se ond quantised form,
Z Z X y y
dr1 dr2 r2 ) ay y
Ld Ld
g (k)eik r ( 1
" (r1 )a# (r2 )j
i = gk k" k# j
i
k

then, of the terms in the expansion of


" #
Y
N
X y y
j i= gkn kn " kn # j
i
n=1 kn

those with all kns di erent survive

Le ture Notes O tober 2004


Le ture VIII 48

Generally, more onvenient to work in grand anoni al ensemble


where one allows for (small) u tuations in the total parti le density
Y
j i= (uk + vk yk" y k# )j
i
k

i.e. statisti al independen e of pair o upation


Normalisation demands uk + vk = 1 (exer ise) while onstraint in parti lePnumber
2 2

leads to k vk = 2N 2


In non-intera ting ele tron gas vk = 01 jjkkjj >
< kF
kF

expe t intera tions to smear Fermi surfa e


to balan e kineti and potential energies from pairing
Furthermore,
!1=2  1=2
X
h j(N^ N ) j i=2
2
uk vk N 1=2 !D
F
N
k

To determine the variational parameters vk, one an use a variational prin iple, i.e. to
minimise
h jH^ F N^ j i
. BCS Hamiltonian: However, sin e we are interested in both the ground state energy,
and the spe trum of quasi-parti le ex itations, we will follow a di erent route and explore
a simpli ed model Hamiltonian
X X
H^ = k yk k V yk " y k # k# k"
BCS 0 0

k kk0

in a mean- eld approximation | formally equivalent to variational analysis (where ap-


propriate)

Le ture Notes O tober 2004


Le ture IX 49

Le ture IX: BCS Super ondu tivity

. Re all: We have seen that phonon ex hange ; a pairing intera tion whi h renders
a single pair of ele trons unstable towards the formation of a bound state (Cooper)
Motivated by this onsideration, we have proposed a many-body generalisation
of the pair state in the form of the variational BCS state
Y
j i = (uk + vk yk" y k#)j0i
k

in whi h the anomalous average operator yk" y k# a quires a non-zero expe tation value
Therefore, sin e we expe t u tuations in the pair operator yk" y k# to be small,
we may set
z
small
}| {
yk" y k# = bk + yk" y k# bk

where bk = h j yk" y k# j i so that the BCS Hamiltonian...


^ N^ = X k yk k V X yk " y k # k# k"; k = k 
H   0 0

k kk0
an be expanded as
X 
^ N^ ' X k yk k
H   V b
k k # k " + bk k" k#
0 0
y y
0 b bk k0
k kk0

Then, if we set V Pk bk  , obtain Bogoliubov or Gor'kov Hamiltonian


^ N^ ' X k yk k X  k # k " +  yk" y k# + jj2
 
H  
V
0 0

k k

Bilinear in fermion operators, H^ N^


diagonalised by anoni al transformation
yk" = uk ky " + vk k#
k# = vk ky " + uk k#

where anti ommutation relation requires u2k + vk2 = 1


Taking  real (for simpli ity), Hamiltonian diagonalised if
2kukvk + (v2 u2k) = 0 k

With this hoi e


^ N^ = X(k k2 + 2 ) 2 + X k2 + 2 ky k
q q
H  
k
V k

Le ture Notes O tober 2004


Le ture IX 50

Bogoliubov quasi-parti le ex itations, reated by ky , have minimum energy 


known as the energy gap ; rigidity of ground state


Ground state wavefun tion identi ed as va uum state of algebra f k ; ky g, 


i.e state whi h is annihilated by all the quasi-parti le operators k .



Condition met uniquely by the state  


Y Y
j i  k" k#j
i  uk + vk yk" y k# j
i
k k
. Self- onsisten y ondition P
X X V  X 1 h j k y k j i
 = V bk = V h j k" k#j i = 2
y y p   

k k k k2 + 2
BCS gap equation
At T = 0, h j P ky k j i = 0 and
  
Z
V X
1= 2 p 2 2 = 2 1 V  ( ) F
d p
1 !D

k k +   2 + 2
!D

if !  ,  ' 2! exp[ 1= ( )V ℄


D D F

Ground state: In limit  ! 0, vk2 7! ( k), and the ground state ollapses
F
to the lled Fermi sea with hemi al potential  F

As  be omes non-zero, states in the vi inity of the Fermi surfa e rearrange into
a ondensate of paired states | involves population of single-parti le states with
energy k >  | but, total energy of ground state is lower than non-intera ting
F
Fermi sea however small V
Ex itations: It is very important to distinguish quasi-parti le states and `ex itations'.
Quasi-parti le states are the eigenstates of the BCS Hamiltonian with a positive
and a negative energy bran h. For  ! 0, the quasi-parti les evolve into ordinary
ele trons and holes.
By ontrast, the energy of ex itations (as reated by the operators ky ) is always
positive. An ex itation an be either the reation of a quasi-parti le at positive


energy or the elimination of a quasi-parti le (the reation of a quasi-hole) at negative


energy. In the ground state, all negative-energy quasi-parti le states are lled.
Density of quasi-parti le ex itations near FermiZ surfa e
() =
1 X Æ( p + 2 ) = d 1 X Æ(  ) Æ( p + 2 )
k k
L d
k
L d
k
| {z }
 ( )
XZ 1 Æ   2 ℄1 2 
s[2 =

 d
[ + ℄
= 2  (
2 1=2
 ) ( 2 2 )1 2 ;
=1s 0
 2

=

Spe tral weight transferred from Fermi surfa e to interval [; 1℄


. Postpone phenomenology of super ondu tor until eld theory next term

Le ture Notes O tober 2004


Le ture XIV 52

5
Le ture XIV: Anderson Lo alisation

. Referen es:
Original papers:
 P. W. Anderson, Physi al Review 109, 1492 (1958). 6

 F. J. Wegner, Z. Physi s B 25, 327 (1976).


 E. Abrahams, P. W. Anderson, D. C. Li iardello and T. V. Ramakrishnan,
Physi al Review Letters 42, 673 (1979). [AALR℄
7

 L. P. Gorkov, A. I. Larkin and D. E. Khmelnitskii, JETP Letters 30, 248


(1979). 8

 A. I. Larkin and D. E. Khmelnitskii, Soviet Physi s Uspekhi 136, 536 (1982).


[LK℄
 B. L. Altshuler, D. E. Khmelnitskii, A. I. Larkin and P. A. Lee, Physi al Review
B 22, 5142 (1980). [AKLL℄
Reviews:
 P. A. Lee and T. V. Ramakrishnan, Reviews of Modern Physi s 57, 287 (1985).
 B. L. Altshuler, A. G. Aronov, D. E. Khmelnitskii and A. I. Larkin in Quantum
Theory of Solids, edited by I. M. Lifshits (MIR Publishers, Mos ow, 1983). 9

[AAKL℄
 Y. Nagaoka and H. Fukuyama, Eds., Anderson Lo alization, (Springer, New
York, 1982).
 D. J. Thouless, Physi s Reports 13, 93 (1974).

. Why is disorder important in metals?


Disorder, even when weak, makes a qualitative di eren e to the transport properties
of metals. The Boltzmann des ription of harges s attering o impurities predi ts a
residual resistivity that in reases with disorder. This simple pi ture is insuÆ ient to
explain experimentally observed behaviour even for weak disorder. When disorder is
weak, deviations are seen from Ohm's law in the sample-size dependen e of resistan e.
These deviations are espe ially remarkable in one and two dimensions.
Physi ally, repeated s attering of an ele tron from impurities auses the ele tron wave-
fun tion to quantum-interfere with itself. Su h interferen e may even ause it to be ome
lo alised whi h leads to an exponential in rease of resistan e with system size. This is
ompletely unexpe ted in the semi- lassi al treatment of ele tri al ondu tion.
A phenomenon analogous to lo alisation in metals is observed ommonly: a sta k of
transparen ies is pra ti ally opaque while individual sheets are transparent.
5
Le ture notes prepared by Vikram Tripathi
6
DiÆ ult reading.
7
Also known as the Gang of Four.
8
Perturbation theory, diagrammati analysis.
9
Hard to nd. Can read LK above instead.

Le ture Notes O tober 2004


Le ture XIV 53

Figure 9: Simple physi al pi ture of lo alisation due to AAKL (1983) and Bergmann [PRB
P
25, 2397 (1982)℄. The Green fun tion G(r; r0 ; E ) an be onstru ted as a sum of amplitudes,
p Ap e
ip (r;r ) , of paths between r; r 0 : In the presen e of impurities, most paths arrive with a
0

random phase, with the ex eption of self-interse ting paths.

. How does interferen e en ourage lo alisation?


Consider the probability to go from r to r0 , i.e., jG(r; r0 ; E )j . If the amplitude for the 2

kth path is Ak eik , the probability is


* +
X 2 X X X


ik
Ak e = A2k + 2 Aj Ak h os(j k )i = A2k ;
k k 6
j =k k

sin e the average over the osine vanishes. However if r = r0, then there are two paths k
and k that are time-reversed ( lo kwise-anti lo kwise) analogues of ea h other (loop part
1

in Fig.9), so Ak = Ak = Ak ; k = k = k . This ontributes 4Ak to the return prob-


2
2

ability, as opposed to 2Ak in the absen e of interferen e. Even though the enhan ement
1 2 1 2
2

is large, the number of su h time-reversed pairs is a small fra tion of the total number of
possible paths. Nevertheless, total return probability is enhan ed by interferen e!
 Enhan ement of probability to return is tantamount to de rease of ondu tivity
The ele tron is a wave pa ket of transverse dimension F and longitudinal dimension
vF ' , where ' is the phase-breaking time, assumed to be mu h larger than the mean
10

free time for elasti ollisions  = l=vF . Sin e thed=ele tron motion is di usive for times
t   , the ele tron is present in a volume (Dt) where D is the di usion onstant 2

( = e N D). In time t the wave pa ket harts out a volume


2
vF tdF , and the probability 1

that it will interse t its ourse in a further time dt is vF dF dt=(Dt)d= . Thus the relative
0
1 2

hange in ondu tivity is


 = v d Z ' dt : 1

 F F
 (Dt)d= 2

In three dimensions, the orre tion is


 = 2vF F  p1 p1  = 2vF F  1 1  ;
3D
2 2

 D=
3D  
3 2
D L L' D '
10
The hara teristi time for the ele tron wavefun tion to lose its phase memory due to inelasti ollisions. One
inelasti ollision might not be suÆ ient and ' an ex eed inelasti s attering time.

Le ture Notes O tober 2004


Le ture XIV 54

p
where LD = D is the di usion length and L' = pD' p is the phase-breaking length.
Note that the orre tion be omes a onstant at large L' = D'.
In two dimensions, the orre tion is logarithmi :
 = vF F ln( = ) = 2vF F ln(L =L ):
2D
' ' D
 2DD D

In one dimension, interferen e e e ts are most signi ant as L' in reases:


 = p2vF (p p ) = 2vF (L L ) :
1D
' ' D
1D D D

Su h orre tions to metalli ondu tivity are termed \weak lo alisation". If the phase
breaking time ' is temperature-dependent, then the weak lo alisation orre tions also
11

a quire a temperature dependen e (see Fig.10).

Figure 10: Resistivity hange as a fun tion of log10 T for PdAu lm [Dolan and Oshero , 1979℄.

 A magneti eld suppresses interferen e e e ts


We dis ussed how return probability is enhan ed be ause of interferen e of time-
reversed paths. Any looping path (with its time-reversed ounterpart) ontributes to
enhan ement of the return probability. In presen e of a magneti eld, a loop and its
time-reversed ounterpart a quire ex ess Aharonov-Bohm phases,
I
' = (e=~ ) A  dl = (e=~ )HS;
where H is the omponent of the magneti eld normal to the loop and S is area of the
loop. In time t   , the area harted may be estimated as S  Dt. The Aharonov-
Bohm phase di ers from loop to loop, and on e it ex eeds 2, we may disregard larger
11
This is possible if phase breaking is due to inelasti pro esses that are temperature-dependent.

Le ture Notes O tober 2004


Le ture XIV 55

loops whi h will not ontribute onstru tively to the return probability. That means a
(magneti ) length s ale LH = (eH=~ ) = = . Independently, loops that are larger than
1 2

the phase-breaking length L' = (D') will be dephased. The e e t of a magneti


1 2

eld be omes signi ant when LH be omes the shorter phase-breaking length, i.e. when
LH < L' . [AKLL, 1980; Hikami, Larkin, Nagaoka, 1980℄ Sin e L' an be quite large in
lean metals, the hara teristi magneti eld required to suppress interferen e e e ts an
be quite small, of the order of tens of gauss.
Suppression of interferen e by a magneti eld redu es lo alisation and therefore in-
reases ondu tivity. Weak lo alisation orre tions to magnetoresistan e are thus positive.
The estimate for the ondu tivity is
Z '   
 (H; T )  (0; T ) dt eHDt
 (0; T )
= vF F
d 1

(Dt)d= 1 os ~ : 2
0

This expression an be simpli ed. In two dimensions,


Z eH
 (H; T )  (0; T ) vF F ~ D' dz
2D

 (0; T )
2D
= D z
(1 os z):
2D 0

The result is
8 2
2D (H; T ) 2D (0; T ) < vF F eH
~ D' ; eH
~ D'  1
:
4D

2D (0; T ) 
vF F
D
ln eH
~ D' ; eH
~ D'  1:

. Basi on epts of lo alisation


 Anderson (1958) argued that if disorder is strong enough, the ele tron wave fun -
tion may get lo alised, j (r)j  exp( jr r j= ): In one dimension, an arbitrarily weak
disorder lo alises all states [Mott and Twose, 1961℄. The traditional view had been that
0

s attering by impurities auses the wavefun tion to lose phase oheren e over the mean
free path l, nevertheless the wavefun tion extends throughout the sample.
 Instead of varying disorder, let us vary the energy of the ele tron. States deep in
the band tail are lo alised as they are formed from lo alised orbitals bound in the rarely
o uring deep potential u tuations. States near the entre of the band are likely to be
extended for weak disorder. Below a ertain energy E (mobility edge), all states will be
lo alised. [Mott, 1967℄.
Q: As one rosses the mobility edge, does the metal-insulator transition happen on-
tinuously, or not?
 Mott (1973) argued that a dis ontinuous transition to the insulating state o urs
on e the mean free path l  kF . In the metalli state, kF l  1. Thus the minimum
1

Le ture Notes O tober 2004


Le ture XIV 56

metalli ondu tivity an be estimated as


 3D
= e2 n
( k l )


= 1 e2 1 in3D; andauniversal
min
~ kF2
F
kF l1 3 2 ~ a
0:1 e~ in2D:
2
min
2D


 The s aling theory of lo alisation whi h is the main subje t of this le ture ontests
the existen e of  in both 3D and 2D. Instead it predi ts that the metal-insulator
transition is a ontinuous one in 3D and all states are lo alised in 2D.
min

. S aling theory of lo alisation


S aling theory des ribes lo alisation by onsidering the behaviour of the dimensionless
ondu tan e g = G=(e =~) as a fun tion of system size L  l, or of other s ale variables.
2 12

Delo alised phase: The ondu tan e of a d dimensional hyper ube of size L > l and
ondu tivity  is given by Ohm's law,
g (L) = Ld : 2

Lo alised phase: DC ondu tion o urs by ele tron hopping from an o upied to an
uno upied state with nearly the same energy. However lo alised states lose in
energy are very far apart in spa e, so the tunnelling amplitude is exponentially
small. Quite reasonably,
g (L)  e L= :
( is the lo alisation length, l    L). Note that this is a very non-Ohmi
size-dependen e.
 Single parameter s aling
AALR (1979) argued that
d ln g
(g ) =
d ln L
= Lg dL
dg

is a fun tion of g alone. Physi ally, this means that the hange in e e tive disorder as
the size of the sample is in reased a bit is determined by its value at the previous length
s ale, and the only measure of this e e tive disorder is the ondu tan e.
The physi al motivation behind su h a pi ture is as follows. Consider the energy
levels of a d dimensional hyper ube of dsize 2L in terms of the energy levels of the 2d
onstituent blo ks of size L. If Æ = (N L ) is the mean level spa ing in the onstituent
1

blo ks, and t is the overlap element between the blo ks, then one expe ts the ratio t=Æ to
0

be a good measure of lo alisation. If the wavefun tion is lo alised, t will be insensitive


to the boundary onditions and is therefore exponentially small (so is t=Æ). If however
12
Note that the ondu tivity expressed in units of e2 =~ is not a dimensionless parameter.

Le ture Notes O tober 2004


Le ture XIV 57

t=Æ is large, the eigenstate of the 2L ube will be spread over all the 2d ubes of size L,
and is therefore extended. Therefore sensitivity to boundary onditions (or the ratio t=Æ)
appears to be the single parameter that ontrols the nature of the eigenstate as the system
doubles its size. [Thouless, 1974℄. It appears plausible that the sensitivity to boundary
onditions of the larger blo k depends on the sensitivity to boundary onditions for the
onstituent blo ks, i.e., (t=Æ) L is a fun tion of (t=Æ)L.
2

 S aling fun tion (g)


(a) Large ondu tan e g  g :
(g ) = (d 2):
(b) Small Condu tan e g  g :
(g ) = ln(g=g ):
( ) Perturbation theory gives deviation from Ohm's law:
g
(g ) = (d 2) :
g

Figure 11: S aling fun tion (g) = d ln g=d ln L in various dimensions.


For ele trons, g = 1= . In two dimensions, integrating the s aling fun tion (g) from
2

the lower length s ale l to the system size L gives the deviation of s ale dependen e of
the ondu tan e from Ohm's law:
 
g (L) = g (l)
1 ln L k l
; g (l ) = F :
 2
l 2

Le ture Notes O tober 2004


Le ture XIV 58

The logarithmi orre tion to ondu tan e be omes omparable with the Boltzmann on-
du tan e g(l) at the lo alisation length L =  : 2D
lo

  le g l = lekF l= :
2D 2 () 2
lo

A two-dimensional disordered metal is non-Ohmi at all length s ales and for arbitrarily
weak disorder!
In three dimensions, if g > g , then (g) is positive, i.e., ondu tan e in reases as the
size of the ondu tor is in reased. Thus we have a metal. If g < g , (g) is negative,
whi h means the ondu tan e de reases as the size in reases. This is an insulator. At the
riti al value of ondu tan e g , there is no hange in the ondu tan e as the sample size
is in reased!
I summarise below the s ale-dependen e in various dimensions when the ondu tivity
is large.
 
 (L) =  (l) ~e l L
2 1 1
D=3: 3D 3
 
 (L) =  (l) ~e ln Ll
2
D=2: 2D 2

 (L) =  (l) ~e (L l).


2
D=1: 1D

Observe that the above s aling is the same as we obtained from onsiderations of interferen e-
enhan ed return probability!
. Next le ture
We found that in the lo alised phase, the d ondu tivity at zero temperature vanishes.
The ele trons are lo alised on the s ale of the lo alisation length  . At a nite temper-
ature, the ondu tivity is nonzero, and the me hanism of transport is ele tron hopping
lo

between di erent lo alised sites. What is the temperature dependen e of su h hopping


ondu tivity?

Le ture Notes O tober 2004


Le ture XV 59

13
Le ture XV: Hopping Condu tivity in Disordered Insulators

. Referen es:
Original papers:
 N. F. Mott, Philosophi al Magazine 19, 835 (1969).
 V. Ambegaokar, B. I. Halperin and J. S. Langer, Physi al Review B 4, 2612
(1971).
 A. L. Efros and B. I. Shklovskii, Journal of Physi s C: Solid State Physi s 8,
L49 (1975).
Reviews:
 A. L. Efros and B. I. Shklovskii in Ele tron-ele tron intera tion in disordered
systems, eds. A. L. Efros and M. Pollak. North-Holland, Amsterdam 1985.

. Lo alised states in disordered insulators


In the last le ture we studied how quantum-interferen e e e ts in disordered metals
an abruptly lead to ele tron lo alisation when the Fermi level rosses the mobility edge. 14

Similarly in amorphous semi ondu tors, a band of lo alised states may exist where there
are gaps in the band-stru ture of the ordered material. [see Fig.12℄ In either ase, if the
15

Fermi energy falls in the lo alised band, the d ondu tivity will vanish at zero tempera-
ture. At nite temperatures, transport in su h insulators takes pla e by hopping between
lo alised sites. Su h transport is a very nontrivial ompetition of inelasti s attering with
thermal a tivation and long-range Coulomb intera tion.
. A tivationless hopping ondu tivity
Denote by N the average density of states per unit volume near the Fermi energy. If
the lo alised ele tron wavefun tions overlap strongly, then an inelasti ollision an enable
a transition between states with di erent energies. The distan e between these states is
of the order of the lo alisation length. Thus inelasti ollisions lead to a di usion with
the oeÆ ient
 2
D : lo

in
The ondu tivity is proportional to the di usion oeÆ ient. The situation may hange at
lower temperatures where inelasti ollisions involve very small energy hanges, or when
onsidering states with mu h shorter lo alisation lengrh. This will be the subje t of the
rest of the le ture.
13 Le ture notes prepared by Vikram Tripathi
14 Just as a reminder, the s aling theory of lo alisation predi ts a mobility edge in three dimensions but not in
one or two dimensions.
15 The lo alisation length in a dirty metal would be mu h larger than the mean free path, while the amorphous
semi ondu tor represents a situation where the lo alisation length is mu h smaller than the distan e between the
impurities.

Le ture Notes O tober 2004


Le ture XV 60

ρ(Ε)

Extended

Localised

E
Bandgap of clean
semiconductor

Figure 12: Extended and lo alised states in disordered semi ondu tors.

. Mott variable-range hopping (VRH) ondu tivity


DC ondu tivity of amorphous Ge in the temperature range 60K  T  300K obeys
the law
 (T ) / exp[ (T =T ) = ℄:
0
1 4

Similar temperature dependen es are observed in many other amorphous semi ondu tors
for a range of temperature. Mott (1969) explained this in terms of a pi ture where harge
is transported by the thermally-ex ited hopping of ele trons between randomly pla ed
potential wells (impurities).
 Physi al pi ture of Mott VRH
The number of states per unit energy range in a spheri al volume of radius R is of the
order of NR d . In other words, on the average, one state may be found in the energy range
E (R) = (NR  d ) . Therefore thermally assisted hopping of a harge over a distan e
1

R is suppressed by a Boltzmann fa tor exp[ (E (R)=kB T )℄. Clearly this probability is
larger if the hopping distan e R is large. On the other hand, tunnelling in the insulator is
exponentially suppressed by a fa tor exp[ (R= )℄, where  is the lo alisation length.
Combining these two fa tors, the probability p(R) to hop a distan e R is
lo lo

R E (R)
  
p(R) = exp + :
lo kB T
The dominant ontribution to the ondu tivity omes from hops of size R = R Mott
for
whi h p(R) is maximised:
 1
dp(R)

 d +1
= 0 ) R = k T N
d

Mott
: lo

dR R RMott=  B
The resulting optimum hopping probability is
"   1 # 1
p(R ) = exp
Mott
T 0
+1
Mott d
; where T
2 d+1 d
=  N k : Mott

T 0 d
lo B

Le ture Notes O tober 2004


Le ture XV 61

The 3D ase (d = 3) explains the experimentally observed behaviour (T ) / exp[ (T =T ) = ℄.


0
1 4

 What did we assume?


(a) N  is onstant and nite at the Fermi energy.
(b) We ignored inhomogeneities in the impurity distribution. Inhomogeneities reate
bottlene ks in transmission whi h be ome important in the insulating phase be ause
harge annot be transferred through the system if a path annot be found between
the leads. Refer to Ambegaokar et al. (1971) for a realisti treatment where spatial
uniformity of impurities is not assumed.
More problemati is the rst assumption. Following Mott's work, the understanding
emerged [Ambegaokar et al., 1971; Efros, Shklovskii, 1975℄ that the density of states
at the Fermi energy in insulators will be suppressed in presen e of long-range Coulomb
intera tions. In a metal, the ele trons s reen the Coulomb intera tion so that the eld of
a harge does not extend beyond the Thomas-Fermi s reening length, thus the density of
states at the Fermi energy is a onstant.
. Coulomb e e ts on single-parti le density of states
The main e e t of Coulomb intera tions is already present in the stati properties su h
as the density of states. The density of states turns out to be gapped at the Fermi energy
in presen e of Coulomb intera tions. To obtain the ondu tivity, one needs to onsider,
besides Coulomb intera tion, inter-impurity hopping.
I will show that N must vanish at the Fermi energy as a power law in presen e of
long-ranged Coulomb intera tions,
NE / E d : 1

Thus Mott's VRH behaviour will be observed in a disordered insulator only if the long-
range part of Coulomb intera tion is s reened out.
Model Hamiltonian
X e2 X n i n j
H= n i 'i +
i
2 i6=j Rij :
The ele tron distribution is very nontrivial be ause the energy Ei of an ele tron on
the i site depends on the ele tron o upation of all the other sites,
th

X nj
Ei = 'i + e2 :
i6=j
Rij

If an ele tron is moved from an o upied site i to an uno upied site j , the hange in
energy of the system is
e 2
Eji = Ej Ei ;
R ij

Le ture Notes O tober 2004


Le ture XV 62

where e =Rij is the Coulomb attra tion of the ele tron-hole pair in the nal on guration.
2

These ele tron-hole ex itations are the lowest harge- onserving ex itations of the system.
Taking the ground state energy as zero, all ex ited states will have positive energy, 16

Eji > 0:

Our previous dis ussion of Mott VRH for non-intera ting ele trons had demonstrated
that for a given hopping distan e R , one state is available with an energy separation
(Ej Ei ) = (NR ijd ) on the average.ij
1

Turning the argument, given an energy separation (Ej Ei ), the hopping distan e for
Mott VRH is Rij = [N (Ej Ei )℄ =d . 1

The Coulomb intera tion energy for the ele tron-hole pair at this separation is
e
= e N (Ej Ei ) :
2 1
Eij = int 2 d
Rij

 The pun h line


The intera tion energy annot have a larger magnitude than Ej Eifor then we
annot satisfy the requirement that the ex itation energy Eji = Ej Ei + Eij > 0: In
int

other words,
1
N = NE E < d (Ej Ei )d : 1

e j i 2

Clearly,d at the lowest energies, the density of states N must vanish at least as fast as
(Ej Ei ) , else the ex itation energy will not be positive.
1

 Density of states
So far we have found thatd the density of states NE vanishes at least as fast as E d . 1

Now we argue that NE / E : 1

Suppose NE de reases faster than E d , say NE / E d ; > 0. Then the intera tion
1 1+

energy
E = e NE E =d / E =d  E asE ! 0:
int
 
2 1 1+( )

Thus for any nite and small E , the intera tion energy will be irrelevant ompared to
the non-intera ting part E . In that ase, we ould have disregarded Coulomb intera tion
at the outset. The only onsistent solution is therefore = 0, hen e
A
NE = d E d ; whereA < 1isa onstant:
1

e2
16 Eji > 0 also means that Rij > e2 =(Ej Ei ), thus the ele tron-hole \ex iton" annot ollapse.

Le ture Notes O tober 2004


Le ture XV 63

This argument for the Coulomb gap in the density of states was presented by Efros
and Shklovskii in 1975.
. Efros-Shklovskii variable-range hopping ondu tivity
Following Mott's VRH argument, the probability p(Rij ) to hop a distan e Rij is
  
Rij Eji
p(Rij ) = exp + ;
lo kB T
where
e2
= (1 A =d ) Re :
2
Eji = Ej Ei 1

Rij ij
The dominant ontribution to the ondu tivity omes from hops of size R = E for ES

whi h p(R) is maximised, s

R =
ES (1 A =d )e  :1 2
lo

kB T
The resulting optimum probability is
p(R ) = exp
ES
 q 
T =T ; where T =
4e (1 A = ) :
ES
2 1 d

 k
ES
0 0
lo B

Note that the Efros-Shklovskii VRH law is the same for any dimension, unlike Mott's
VRH.

Le ture Notes O tober 2004


Le ture VII 66

Le ture XVIII: Feynman Path Integral

. Motivation:
 Alternative formulation of QM ( f. anoni al quantisation)
 Close to lassi al onstru tion | i.e. semi- lassi s easily obtained
 E e tive formulation of non-perturbative approa hes
 Prototype of higher-dimensional eld theories
. Time-dependent S hr odinger equation
i~ j i = H^ j i t

X
Formal solution: j (t)i = e ^ ~ j (0)i = e iH t= iEn t=~
jnihnj (0)i
. Time-evolution operator
n

j (t0)i = U^ (t0; t)j (t)i; U^ (t0; t) = e ~ ^ ( ) (t0 t) i


H t
0
t

N.B. Causal
 Real-spa e representation:
Z
j ih j
dq q q Z
(q0; t0)  hq0j (t0)i = hq0jU^ (t0; t) ^ j (t)i = ( ; ) (q; t);
dq U q 0 ; t0 q; t

where U (q0 ; t0; q; t) = hq0je i


~H
^(
t
0
) jq i(t0
) | propagator or Green fun tion
t
t
 
i ~ t0
^ U (t0 t) = i~Æ(t0 t)
H

N.B.  (t0 t) = Æ(t0 t) t0

Physi ally: U (q0 ; t0; q; t) des ribes probability amplitude for parti le to propagate
from q at time t to q0 at time t0
. Constru tion of Path Integral

Feynman's idea: divide time evolution into N ! 1 dis rete time steps t = t=N
^ h ^  ~i
=
N
~
e iH t=
e iH t=

Then separate the operator ontent so that momentum operators stand to the left
and position operators to the right

e
^
iH t= ~
=e iT ^ t= ~
e
^
iV t= ~
+ O(t2 )

Le ture Notes O tober 2004


Le ture VII 67

h ^  ~i
hq j e jq i
N
iH t=
F I

' hq j^e ^ ~ e F
iT t= ^
iV t=~
^e
^
iT t= ~
e iV^
^ : : : ^e
^  ~ e ^  ~ jq i
t= ~ iT t= iV t=
I

Z Z
Insert at ^ resol. of id: = dqn dpn jq ihq jp ihp j, and using hqjpi = p 1 eiqp=~ ,
n n n n
2 ~
e
^
iV t=~
jq ihq jp ihp je ^ ~ = jq ie ( ) ~ hq jp ie
iT t= iV q t= ( )
iT p t= ~
hp j ;
hp +1jq ihq jp i = 21~ exp [iq (p p +1)℄
n n n n n n n n

and n n n n n n n

" #
^
Z Y1 N
Y N
X1 
N
qn+1
hq j e
F
iHt= ~
jq i =
I dqn
2 ~
exp
dpn i
~
t V qn ( ) + T (p +1) n pn+1
t
qn
n=1
qN =qF ;q0 =qI
n =1 n =0

p Phase
Space

tn t
qF 1 2 N

qI

i.e. at ea h time step, integration over the lassi al phase spa e oords. x  (q ; p ) n n n

Contributions from traje tories where (q +1 q )p +1 > ~ are negligible


motivates ontinuum limit
n n n

Z Z
D (q; p)
t

dt0
( )= (0)= pq_j =
z }| { z 0}| { H ( xj = )
q t qF ;q qI

Z Y1 h i 1 z }| { t0 tn

X z }| { q +1 q i
t0
Y dp
tn

^
N N N

hq j e
F
~
iHt=
jq i = I dq
2 ~
exp ~ t (V (q ) + T (p +1) p +1 t )
n
n
n n n
n n

=1 n=1
qN =qF ;q0 =qI
=0 n n

Propagator expressed as fun tional integral:


Hamiltonian formulation of Feynman Path Integral
z A tion
}| {
Z hi Z z
Lagrangian
}| {i
hq j e ^ ~
jq i = iH t=
D (q; p) exp dt0 (pq_ H (p; q ))
t

F I
~ q t ( )= qF ;q (0)= qI 0
Le ture Notes O tober 2004
Le ture VII 68

Quantum transition amplitude expressed as sum over all possible phase spa e
traje tories (subje t to appropriate b. .s) and weighted by lassi al a tion
. Lagrangian formulation: for \free-parti le" Hamiltonian H (p; q) = p2=2m + V (q)
z
Gaussian integral
}|
on p {
Z Z  Z  2 
^ jq i =
R
Dp exp
2m pq_
( ~) ( ) i p t

hq j e
F
~
iHt=
I Dqe
~
i=
dt0 t
0 dt V
0
q

( )=
q t qF ;q (0)= qI 0

p0
2
_ 7! 21m (p 1 mq_2
p2 z }| {
mq_)2
2m pq
2
Fun tional integral justi ed by dis retisation

Z  Z  
hq j e ^ ~
jq i = exp i t

dt0
_
mq 2
()
Dq
2 V q
iHt=
F I
q t( )= qF ;q (0)= qI
~ 0

  2 Y1
N= N

e =
Dq ! Dq lim Nm
N !1 2
it  ~
n =1
dqn

Le ture Notes O tober 2004


Le ture XI 69

Le ture XI: Coherent States

Dis uss element of PI onstru tion whi h demands generalisation

. Generalisation of FPI to many-body systems problemati


due to parti le indistinguishability and statisti s
Can se ond quantisation help? automati ally respe ts parti le statisti s

Require omplete basis on the Fo k spa e to onstru t PI


Su h eigenstates exist and are known as Coherent States
referen e: Negele and Orland

. Coherent States (Bosons)

What are the eigenstates of Fo k spa e operators: ai and ayi s.t. [ai ; ayj ℄ = Æij ?
Being a state of the Fo k spa e, an eigenstate ji an be expanded as
X (ay )n1 (ay2 )n2
ji = Cn1 ;n2 ; p1
n1
pn   j0i
n1 ;n2 ; 2

N.B. notation j 0i for va uum state!

(i) ayi ji = i ji? | in fa t, eigenstate of ayi an not exist:


if the minimum o upation of ji is n0 , the minimum of ayi ji is n0 + 1
P
(ii) ai ji = i ji? | an exist and given by: ji  exp[ i i ayi ℄j0i N.B.   fi g

Proof: sin e ai ommutes with all


y
aj for j 6= i | fo us on one element i

a exp(ay ) j0i = [a; exp(ay)℄j0i =


X1 n 1
X
 nn y n 1
= [a; (a ) ℄j0i =
y n
(a ) j0i =  exp(ay)j0i
n! n!
n=0 n=1

a(ay )n = aay (ay )n 1


= (1 + ay a)(ay )n 1
= (ay )n 1
+ aya(ay )n 1
= n(ay )n 1
+ (ay)n a
i.e. ji is eigenstate of all ai with eigenvalue i | known as Bosoni oherent state
. Properties of oherent state:
 Hermitian onjugation:
8i : hjayi = hji
i
 is omplex onjugate of i

Le ture Notes O tober 2004


Le ture XI 70

 By dire t appli ation of i (and operator ommutativity):


8i : ayi ji = i ji
P
 Overlap: with hj = (ji)y = h0je i iai
" #
Pi iai Pi ii X
h j  i = h0j e ji = e h0ji = exp i i
i

i.e. states are not orthogonal! operators not Hermitian


" #
X
 Norm: hji = exp i i

i

 Completeness | resolution of id. (for proof see notes)


Z Y 
di di P i i

e i jihj = 1F
i

where di di = dRe i dIm i


. Coherent States (Fermions)
Following bosoni ase, seek state j i s.t.
j i = iji;  = fig
ai 

6 j ) implies eigenvalues i anti ommute!!


But anti ommutativity [ai ; aj ℄+ = 0 (i =
i j = j i

i an not be ordinary numbers | in fa t, they obey...


. Grassmann Algebra
In addition to anti ommutativity, de ning properties:
(i) i2 = 0 ( f. fermions) but note: these are not operators, i.e. [i ; i ℄+ 6= 1
(ii) Elements i an be added to and multiplied by ordinary omplex numbers
+ i i + j j ; i ; j 2C
(iii) Grassmann numbers anti ommute with fermioni reation/annihilation operators
[i ; aj ℄+ = 0
. Cal ulus of Grassmann variables:
(iv) Di erentiation: i j = Æij
N.B. ordering i j i = j i i = j for i 6= j

Le ture Notes O tober 2004


Le ture XI 71

R R
(v) Integration: di = 0, di i=1
i.e. di erentiation and integration have the same e e t!!

. Gaussian integration:
Z Z Z Z
a
dd e = dd (1 a ) =a d d =a
Z
dd e T A = detA (exer ise)
f. ordinary omplex variables
. Fun tions of Grassmann variables:
Taylor expansion terminates at low order sin e  2 = 0, e.g.
F ( ) = F (0) + F 0 (0)

Using rules
Z Z
dF ( ) = d [F (0) + F 0 (0)℄ = F 0 (0)   F [ ℄
i.e. di erentiation and integration have same e e t on F [ ℄!
Usually, one has a fun tion of many variables F [ ℄, say  = f1 ;    N g
X1
1  n F (0)
F ( ) = j    i
n! i    j
n=0
but series must terminate at n = N
with these preliminaries we are in a position to introdu e the

P y ℄j0i i.e.
. Fermioni oherent state: ji = exp[ i i ai  = fi g
Proof ( f. bosoni ase)
a exp( ay )j0i = a(1 ay ) j0i = aayj0i = j0i =  exp( ay ) j 0i
Other de ning properties mirror bosoni CS | problem set
. Di eren es:
Pi ai i P
(i) Adjoint: h j = h0j e  h0je i i ai but N.B. i not related to i !
Z
(ii) Gaussian integration: dd e  =1 N.B. no  's

Completeness relation
Z Y Pi i i
di di e jihj = 1F
i

Le ture Notes O tober 2004


Le ture XII 72

Le ture XII: Many-body (Coherent State) Path Integral

Could formulate many-body propagator (Green fun tion), but here, onvenient to fo us on
partition fun tion.

. Quantum partition fun tion


X
(H^ N^ )
Z= hn j e j n i; = 1=kB T;  hemi al potential
fng2Fo k Spa e

Coherent state representation of Z | insert resolution of id. (fermions/bosons)


Z P Y d i d i
d[ ; ℄e i i i j ih j = 1F ; d[ ; ℄ 
 (1  )=2 i

ea h element i asso iated with one basis state, viz. ayi


| e.g. i may in lude position, momentum, spin, latti e site, et .
Z Pi i X
d[ ;
Z= ℄e i hnj ih je ^ N^ )
(H
jni
n

Elimination of jni requires identity: hnj ih jni = h  jnihnj i


Proof: E.g. jni = ay1ay2    aynj0i
hnj i = h0jan    a2 a1j i = n    2 1 h0j i = n    2 1
h jni = 1 2    n
hnj ih jni = n    2 1 1 2    n = 1 1 2 2    n n
= (  1 1 )(  2 2 )    (  n n) = h  jnihnj i
P
ommute through and erase n jnihnj
Z Pi i
d[ ;
Z= ℄e i h ^
 je (H N^ )
j i

. Coherent State Path Integral


Applied to general Hamiltonian
H^ N^ = X(hij Æij )ayi aj +
X
Vijkl ayi ayj ak al
ij ijkl

N.B. operators are normal ordered


Follow general strategy of Feynman:

Le ture Notes O tober 2004


Le ture XII 73

(i) Divide `time interval' into N segments  = =N


h  je (H^ N^ ) j i = h  je  (H^ N^ ) ^e  (H^ N^ )
^   j i
(ii) At ea h ^ insert resolution of id.
Z
d[ n ;
1F = ℄e n  n j n ih nj
n

i.e. N -independent sets N.B. ea h n is a ve tor with elements f i gn


(iii) Expand exponent in 
h i
h 0je  (H^ N^ ) j i = h 0j 1  (H^ N^ ) j i + O( )2
= h 0j i  h 0j(H^ N^ )j i + O( )2
= h 0j i [1  (H ( 0; ) N ( 0 ; ))℄ + O( )2
' e 0  e  (H ( 0 ; ) N ( 0 ; ))
0 ^
) = h h jH0 j j i i = hij i0 j + Vijkl i0 j0
X X
with H ( 0; k l
ij ijkl

similarly N ( 0; ) N.B. h 0j i bilinear in , i.e. ommutes with everything


Z N
Y PNn=1[ n ( ))℄
Z= d[ n ; n ℄e n n 1 )+ (H ( n ; n 1 ) N ( n ; n 1

n=0
N =  0 ; N =  0

Continuum limit N ! 1
N
X Z N
Y
 ! 1 ! d[ n ; ℄ ! D( ; )
n n
d;  ;
n=0 0   =n
n=0
n

omment on \small" Grassmann nos.

Z Z
Z= D( ; )e S [ ; ℄; S [ ; ℄ = d    + H ( ; ) N ( ; )
( )=  (0)
( )=  (0) 0

With parti ular example:


Z " #
X X
S [ ; ℄= d i ( ) [( )Æij + hij ℄ j ( ) + Vijkl i ( ) j ( ) k ( ) l ( )
0 ij ijkl

i.e. Quantum partition fun tion expressed as path integral over elds j ( )
Le ture Notes O tober 2004
Le ture XII 74

. Matsubara frequen y representation


Often onvenient to express path integral in frequen y domain
Z
( ) = p1 X
e i!n 
; =p 1 d ( )ei!n 
n !n
!n
0

where, sin e ( ) = ( + )


!n =
2n= ; bosons; ; n 2 Z
(2n + 1)= ; fermions
!n are known as Matsubara frequen ies
Z
Using 1 d e i(!n !m )
= Æ!n!m
0
X
S [ ; ℄= i!n [( i!n ) Æij + hij ℄ j!n +
ij!n

+ 1
X X
Vijkl i!n1 j!n 2 k!n3 l!n4 Æ!n1 +!n2 ;!n3 +!n4
ijkl !n1 !n2 !n3 !n4

. Exer ise: Establish relation between the oherent state path integral and the Feynman
path integral for the quantum harmoni os illator

Le ture Notes O tober 2004


Le ture XII 75

Le ture XII: Weakly Intera ting Ele tron Gas: Plasma Theory

. How are the properties of an ele tron gas in uen ed by weak Coulomb intera tion?
X
N
p^ 2i X e2
^ =
H
i=1
2m + i<j
jri rj j

. Qualitative onsiderations:
With r0 = n 1=3
the average ele tron separation, typi al p.e. e2 =r0 and k.e. ~2 =mr0

i.e. dimensionless ratio: rs = er mr02


 ar0 | a0 is Bohr radius
2

0 ~
2
0

i.e. Coulomb e e ts dominate at low density!


At rs  35 there is (believed to be) a transition to an ele tron solid phase (QMC)
known as a Wigner rystal ( f. Mott-Hubbard insulator)
For most metals (2 < rs < 6), k.e. and p.e. omparable;
fortunately (re all FLT) \weak oupling" theory valid even for intermediate rs
oupling theory rs  1:
. Motivates onsideration of weak
Z Z Z
^ p^ 2 1 e2
y
H = dr  (r)  (r) + dr dr0 y (r) y0 (r0 ) 0 (r0 )  (r)
2m 2 jr r j
0

- onvention on spin
Aim: to explore diele tri properties and ground state energy of ele tron gas through...
. Quantum partition fun tion:
Z
Z  tr e H^
=  (0)=   ( )
D(  ;  )e S [  ;  ℄

 (0)=   ( )
Z Z  

p^ 2
S [  ;  ℄ = d dr  (r;  )  + 2m  psi (r;  )
Z Z 0

1
+ 2 dr dr0  (r;  ) 0 (r0;  ) jr e r0j
2
0 (r0 ;  ) (r;  )
 

X i(kr !  )
Expressed in Fourier basis: (r;  ) = p 1 e 
n
k;!n ;
L3 k;!n
Z " #
X 1 X 2e2  ( ) ( )
S [  ; ℄= d k ( ) ( + k ) k  +

0 k
2 q6=0 q2 q q

R P 
where k = k2=2m and q = dr e i q r (r)  k k k+q;

Le ture Notes O tober 2004


Le ture XII 76

Neutralising ba kground ; ex lusion of q = 0 from sum


Being quarti in the fermioni elds, Z an not be evaluated exa tly
For weak intera tion, rs  1, ould expand in Coulomb intera tion:
Feynman diagram expansion ( f. Gell-Mann Bru kner theory)
Alternative | use eld integral to isolate leading diagrammati series ontributions
. General prin iple:
When onfronted with intera ting eld theory, seek de omposition of intera tion
through auxiliary eld whi h aptures low-energy ontent of the theory
In some ases, these elds will be identi ed with the elementary parti les
that mediate the intera tion (see below);
in others, these elds will en ode the low-energy olle tive modes
of the system ( f. super ondu tor)
De oupling fa ilitated using the...
. Hubbard-Stratonovi h transformation:
Z R P  
R P 2e2 d q2 ie
q6=0 8 q ( ) q ( )+ 2 (q ( ) q ( )+q ( ) q ( ))
e 0 d q=0 q2 q ( )
6 q ( )
= D e 0

. Physi ally,  represents (s alar) photon eld whi h mediates Coulomb intera tion
Z Z R R   2
 
d dr 81 ()2 +   + 2p^m +ie
Z= D(  ; ) 0 
 D e

quadrati in Grassmann elds, eld integral may be performed straightforwardly:


R
Using identity D[ ; ℄ exp[ M ℄ = detM = exp[ln detM ℄
2 3
Z Z Z spin
z}|{  
6 1 p^ 2
Z= D exp 4 d dr () + 2 ln det  +
2
 + ie 7
5
0 8 2m
Setting e = 0, photon eld de ouples from determinant
re overs partition fun tion of non-intera ting ele tron gas
Philosophy: All HS de ouplings (in luding this one) are exa t. So whi h to hoose? |
the one that exposes the low-energy degrees of freedom of the theory | usually sig-
nalled by the appearan e of some symmetry breaking (although not in this ase) and the
proliferation of massless degrees of freedom, e.g. magnetisation and super onudu tivity
. Perturbation Theory in e:
Le ture Notes O tober 2004
Le ture XII 77

De ne free parti le Green fun tion: G^ 0 = [  p^ 2


2m + ℄ 1 and expand: P
ln(1 x) = n
n=1 x =n
    h i
p^ 2
ln det  + 2m  + ie  tr ln G^ 0 1
= tr ln G^ 0 1 + tr ln 1 ieG^ 0 
ie
  2 
^ ^ 1 ^
= tr ln G0 tr ieG0  + 2 ieG0  +   
1

First order term:for onvenien e, set k  (k; !n), et .


1
p 3 k=0
G0 (k) L
X z }| { z }| { X
2tr[G^ 0℄ = 2 hkjG^ 0jki hkjjki = p 2 3 i!
1
 + 0
 =0
k L k n k

0 = 0 due to neutralising ba kground


Se ond order term:
X X
2 e2 tr[G^ 0℄2 = e2 G0 (k) pq 
= e2
2 2
G 0 (k + q ) p q (q ) q q
k;q L3 L3 q

where density-density response fun tion


(q ) = 3
2 X 1 1
L k i!n k +  i!n + i!m k+q + 

Combined with bare term, to leading order in e,


e2 D 1
(q )
z }| {
Z
1 X q2
Z = Z0 D e S [℄
; S [℄ =
2 4 e2 (q ) jq j2 + O(e4)
q

Z0 denotes partition fun tion of non-intera ting gas


. Physi ally, D (q) denotes dynami ally s reened Coulomb intera tion
1

q2
D 1 ( q ) = ( q ) 2 ; ( q ) = 1
4e2 (q)
4e q2
where (q) is the energy and momentum dependent e e tive diele tri fun tion
This result, known as the Random Phase Approximation, amounts to treating the long-
range part of the Coulomb intera tion as an \external" polarisation eld, with s reened
polarisability:
4e2 (q)
q2
Le ture Notes O tober 2004
Le ture XII 78

k+q, ωn+ ωm
q,ωm q,ωm
= + +
k,ωn + ...

= + =
4πe 2
χ( q,ωm ) 1-
q2

Diagrammati interpretation:
D (q ) =
4e2 1
q2 1 4e2
q2 (q )

Diele tri properties: (q)?


. Digression: Matsubara Summations
Basi idea: introdu e auxilliary fun tion g(z) that has simple poles at z = i!n
X I
f (i!n ) = dzg (z )f (z )
!n C

For example:
8
8 >
>
>
< ; bosons < oth( z=2); bosons
>
2
g (z ) = exp( z ) 1 ; and g (z ) =
> >
:
exp( z) + 1 ; fermions >
: tanh( z=2); fermions
>
2

Poles at Matsubara
frequencies

Pole of f

Applied to density-density response fun tion,


g(z ) f (z )
I z }| {z }| {
I  dz
1 1
C exp( z) + 1 z + z  + 0
(i) Integral exists (the integrand de ays suÆ iently fast)

Le ture Notes O tober 2004


Le ture XII 79

(ii) I = 0
Residue integral:
0 1
Z f (i!n ) g( )=(0 )

0= dz (f g )(z ) = 2i 
X z
B Res (f g )(i! ) + Res (f g}|)( {) +Res (f g )(
}| { z C
0 ) A
n
C n

Using ei !n = 1
(!m ; q) =
2 X nF (k q) n F ( k )
; nF () =
1
L3 k i!m + k q k e ( ) +1

 Stati Limit: (j!mj  kF jqj=m)


For frequen ies small as ompared to the momentum transfer
(and temperatures T  )
X q   k n F ( k ) Z
2
(0; q) ' 3 ' 2 dk
L k q   k k (2)3 k nF (k ) = 2 ();
where  () is DoS at Fermi level, i.e.
4e2
D(0; q) ' 2
1
q 1 + 2 qe2 2  ()
4

Fourier transformed, one obtains the stati s reened Coulomb intera tion
e2 jrj=TF
Vs reen (r) =
jrj e
where TF = 2  4e2 () denotes the Thomas-Fermi s reening length
At long time s ales, the Coulomb intera tion is s reened by u tuations
in the neighbouring harge
 High Frequen y Limit: (j!nj  kF jqj=m)
X q  k nF (k ) Z  
2 dk 1 qk
(!m ; q) ' ' 2 (2)3 i! 1 + im! q  knF (k )
L3 k i!m q  k k m m
Z
dk q2 q2
' 2 (2)3 m!m2 nF (k ) = m!m
2
n

where n = N=L3 is the total number density (in luding spin)


4e2 1
D (q ) = 2
q 1 + 4m!
e2 n
2
m

Le ture Notes O tober 2004


Le ture XII 80

Analyti ontinuation to real frequen ies, i!m ! ! + i0


lim 4e2 1
DR (!; q) = 2
m!=kF jqj!1 q 1 !p22
!

pole at !p = 4e2n=m orresponds to the Plasma frequen y


Generally, for jqj  kF ,
  
v F j qj
(q ) ' 2 () 1 + 2v!mjqj ln i! m
i!m + vF jqj
F
  
!m 1 v F j qj
= 2 () 1 v jqj tan !m
F

Lindhard fun tion


For j!mj  vF jqj,
 
 j!m
(q ) ' 2 ( ) 1 2 vF jqj
orre tion due to Landau damping:
Ex itation of parti le/hole pairs ; nite lifetime for 

 Ground State Energy


lim Z  e
!1
Eg:s:
:

In the RPA approximation, integ. over 


Eg:s: = Eg:s: (e = 0)
1 X ln[e2 D(q)℄; Eg:s: (e = 0)  ln Z0 = = 3n=5
2 q

High density (rs  1) expansion


 
E 'n
2:21 0:916 + 0:622 ln r 0:142 Ry:
g:s: s
rs2 rs

Le ture Notes O tober 2004


Le ture XII 81

Le ture XII: Metalli magnetism and quantum riti al phenom-


ena

. Referen es:
E. C. Stoner, Ferromagnetism, Rep. Prog. Phys. 11, 43 (1947).
J. A. Hertz, Quantum riti al phenomena, Phys. Rev. B 14, 1165 (1976).
Although we have seen that e e ts of weak intera tion in an ele tron gas an be
largely benign, we know that strong intera tion an lead to the development of magneti
orrelations in the ground state.
Consider, e.g., the latti e Hubbard model: H^ = H^ 0 + H^ U
^ 0 = X p yp p ; ^ U = U X n^i" n^i#
N
H H
p i
P
= yi i and - onvention on spin
n^ i
Dispersion p (a fun tion of the latti e geometry) is, for the present, left unspe i ed
Phase diagram of H^ is ri h, exhibiting a range of orrelated ground states
depending sensitively on density and strength of intera tion
In latti e system, ommensurability e e ts an trigger transition to
harge or spin density wave state...
while, at large U , the ele tron system an `freeze' into an insulating
antiferromagneti Mott-Hubbard phase
Aim: here we will show that, at low densities the system may adopt an itinerant
(i.e. mobile) spin polarised phase, the Stoner ferromagnet.
. Physi al argument: Capa ity of intera ting ele tron system to form a ferromagneti
phase re e ts ompetition between the k.e. and p.e. Being forbidden by Pauli ex lusion to
o upy the same site, ele trons of the same spin an es ape the lo al Hubbard intera tion.
However, the same ex lusion prin iple requires the system to o upy higher lying single-
parti le states raising their k.e. When total redu tion in p.e. outweighs in rease in k.e. a
transition to a spin polarised or ferromagneti phase is indu ed.
Operationally, it is useful to separate intera tion into hannels sensitive to
harge and spin densities,
^ U = U X(^ni" + n^i# )2 U X(^ni" n^i# )2:
H
4 i 4 i
Sin e u tuations in harge density have little e e tPon thermodynami s of low density
system, negle t their in uen e setting H^ U ' U i(S^iz )2 , where S^iz = (^ni" n^i# )=2.
Le ture Notes O tober 2004
Le ture XII 82

Exer ise: how would you rewrite intera tion in a spin-rotationally invariant manner?
Cast as oherent state PI:
Z ( Z " #)
X U X X
Z = D( ; ) exp d p ( + p) p
4 ( i  i ) 2
0 p i 

 p = p 
Hubbard-Stratonovi h de oupling:
Z Z ( Z
U X X
Z = Dm D( ; ) exp d
4 m2i ( ) + p ( + p p
i p !)
U X
+2 i mi i
i

where mi is lo al magnetisation density


Gaussian integration over Grassmann degrees of freedom:
Z ( Z  )
X X
Z = Z0 Dm exp U4 d m2i ( ) + tr ln 1 U2 m^ G^ 0
0 i 

where G^ 0 = (in p^ ) 1 and m^ = fmi( ) Æij Æ(  0 )g


. From here, one an pro eed in one of two ways:
(1) Vary a tion on m and nd saddle-point magnetisation
; self- onsistent BCS-like equation for magnetisation
(2) Fo using on magneti riti al point, seek perturbative expanision in m
| mirrors losely al ulation of RPA in weakly intera ting ele tron gas
Here we follow route (2) leaving (1) as an exer ise
Drawing on RPA expansion of intera ting ele tron gas,
and noting that terms odd in powers of m vanish (N.B. symmetry m ! m), obtain
Z
Z = Z0 Dm e S[m℄
where, setting q = (!m ; q),
1 X 1 X Y4
P
S [m℄ =
2 q v2 (q)jmq j + 4N qi v4(fqig) i=1 mqi Æ i qi;0 + : : :
2

Le ture Notes O tober 2004


Le ture XII 83

where v2(q) = U2 (1 U (q)) with the magneti sus eptibility (q) = 2(q)
(q ) =
2 X G G = 2 X nF (k ) nF (k+q )
N k 0;k 0;k+q N k i!m + k k+q

i.e. sus eptibility depends sensitively on dispersion k


Stati sus eptibility dominated by ontributions where p ' p+q
over a domain of the re ipro al spa e
i.e. regions where q is small
or where, for some non-zero q = Q, p ' p+Q | `nesting' ondition
If we assume that the spe trum varies smoothly so that nesting symmetry is absent,
the sus eptibility is maximised for q = 0
Fo using on free ele tron system k = k2 =2m, for j!nj=jqjvF small,
(q) ' ( 0; q) 
j!nj
v j qj
where v = vF with depending on dimensionality
For q small, (0; q) '  [1  2q2 +   ℄ where   1=kF . Similarly
v (fq g) 7! v (0)  u =
2(U=2)4 X[G (p)℄4  U 4  00 ;
4 i 4 0
N p

where  00 = 2 j=EF  ()


Altogether,
U 2 X j !nj
 uT
Z
S [m℄ =
4 r+  2 q2 + vjqj jm(q)j + 4N
2 dx m4 (x) +   
q

where r = 1=U 1
R R R
. Finally, res aling the magnetisation eld and u, ( dx  dd x d )
1 X  2 2 j!nj  2 u Z
S [m℄ =
2 q r +  q + vjqj jmq j + 4 dx m (x)
4

the Hertz-Millis a tion


Frequen y-dependent term des ribes damping of magneti u tuations due
to ex itation of parti le/hole pairs
Restoration of spin symmetry endows magnetisation eld with n- omponents

Le ture Notes O tober 2004


Le ture XII 84

A tion aptures e e ts of quantum dynami al u tuations


on Ginzburg-Landau expansion for magnetisation
. Mean- eld theory:
S [m℄ r 2 u 4
L3
= 2m + 4 m :
Minimising the a tion with respe t to m, one nds a Stoner transition
to a spin polarised ferromagneti phase when
r = 0; i:e: U  = 1
p
for U > U , m = r=4u
Although the mean eld theory provides a good qualitative understanding of the nature
of the transition, the Stoner riterion itself is unreliable. In the latti e system, the density
of states is typi ally set by the bandwidth, i.e.   1=t. Therefore, at the Stoner transition
where U=t  1, the system enters the strongly orrelated phase where the intera tion an
not be onsidered as a small perturbation. In this regime, the ele trons experien e an
e e tive intera tion renormalised by the s reening e e t of the harge redistribution. An
estimate for the e e tive intera tion Ue  U=(1+ U=t) shows that, in the relevant regime,
the Stoner riterion be omes repla ed by
t  1:
Typi ally, for a smoothly varying density of states,   1=t and the inequality is diÆ ult to
satisfy. In pra ti e, the Stoner transition to ferromagnetism tends to appear in materials
where there is signi ant enhan ement of the density of states near the Fermi energy.
. Flu tuations?
Mean- eld theory perturbed by both lassi al and quantum dynami al u tuations
subje t of Hertz-Millis theory (beyond s ope of le tures)
E e ts depend on proximity to riti al region
phase diagram

Le ture Notes O tober 2004


Le ture XII 84

Le ture XII: Bose-Einstein Condensation and Super uidity

. Coherent State PI of quantum partition fun tion:


Z
Z= D( ; ) e S [ ; ℄
;
b: :

with a tion
Z Z h i
g
S= d dr (r;  )( + H
^0 ) (r;  ) + ( (r;  ) (r;  ))2 ;
0 2
. Bose-Einstein Condensation
Consider non-intera ting Bose gas:
Z " #
X
Z0  Z g=0 = D( ;

) exp an ( i!n + a ) an :
an

where we assume a  0 and 0 = 0


Stability: hemi al potential   0

Number of parti les:


X 1 X
N ( ) = T = nB (a ) ;
na
i!n a +  a

where nB () = (e ( ) 1) 1
denotes Bose distribution
At riti al temperature T ,  redu es to zero. For T < T ,  remains zero and a
mi ros opi number of parti les N N1 must a umulate in the single-parti le ground
state: Bose-Einstein ondensation
X
nB (a )
=0

T <T
N1 < N
a>0

µ
TC
0
T

Le ture Notes O tober 2004


Le ture XII 85

How an the phenomenon of BEC be implemented into the fun tional integral repre-
sentation
hara teristi s of ondensate des ribed by the zero eld omponent 0 ( )

But, below the ondensation transition, its a tion apears unbound: both the hemi al
potential and the eigenvalue are zero. This means that the a tion of the zero Matsubara
omponent 0;0 vanishes.
Therefore, let us treat 0 ( ) as a time-independent Lagrange multiplier to be used to
x the number of parti les below the T :
X
S0 j=0 = 0  0+
an (i!n + a ) an ;
a6=0;n

With this de nition,


X 1
N =  F0 j=0 = T  ln Z0 j=0 = 0 0 +T = 0 0 + N1
a6=0;n
i!n a

i.e. 0 0 = N0 | no. parti les in ondensate


. Weakly Intera ting Bose Gas
Dominant ontirbution to Z from lassi al 0 se tor:
1  g 
S[ 0; 0℄ =  0 0 + ( )2
2Ld 0 0
Saddle-point analysis:
 
g 
0 + = 0:
Ld 0 0

Le ture Notes O tober 2004


Le ture XII 86

S1 S

im

im1 re
re1

. For  < 0 (i.e. above the ondensation threshold of the non-intera ting system),
the equation exhibits only the trivial solution 0 = 0. This means that no stable
ondensate amplitude exists.
. Below the ondensation thresholdp(i.e. for   0), the equation is solved by any
on guration with j 0 j =  Ld =g. Noti e that 0 0 / Ld , re e ting the
ma ros opi population of the ground state.
. The equation ouples only to the modulus of 0 . I.e. the solution of the stationary
phase equation is ontinuously degenerate: Ea h on guration 0 = exp(i);  2
[0; 2 ℄ is a solution. One ground state hosen ;
spontaneous symmetry breaking.
. Super uidity
Below T , expand a tion:
Z Z    
2 g
S [ ; ℄ = d dr (r;  )   (r;  ) + ( (r;  ) (r;  ))2 ;
0 2m 2
around mean- eld 0 = 0 = (Ld =g )1=2 =
Flu tuations:
(r;  ) = [0 + Æ(r;  )℄1=2 ei(r; ) ;
(r;  ) = [0 + Æ(r;  )℄1=2 e i(r; ) ;

where 0  2 = 0 0 is the ondensate density


f. anoni al transformation:
a(x)  ^(x)1=2 ei(x) ; ay (x)  e ^(x)1=2
^ i^(x)

Assuming anoni al ommutation relations


(x); ^(x0 )℄ = iÆ (x x0 );
[^ (x); ^(x0 )℄ = [^(x); ^(x0 )℄ = 0
[^

on rm:
^ :℄
i[;
e ^
h i z }| {
1 ^ ^
[a; ay ℄ = ^1=2 ei ; e ^ ^℄
^ i^ 1=2 i^ i^
^ = ^ e ^e = ^ ^ + i[; [; [; ^℄℄ + : : : = 1
2
Le ture Notes O tober 2004
Le ture XII 87

Using density-phase relation


Z Z  
 g2
S [Æ; ℄  d dr i  + 0 ()2  0
0 2m 2
First term has the anoni al stru ture `momentum   ( oordinate)' f. anoni ally
onjugate pair
Se ond term measures energy ost of spatially varying phase u utations
Noti e that (r;  ) = onst: does not in ur an energy ost | i.e.  is a Goldstone
mode
Third term re ords the energy ost of massive u tuations from potential minimum
Gaussian integration over Æ leads to Lagrangian formulation
Z Z  
1 1 
S [℄  d dr ( )2 + 0 ()2 :
2 0 g m
f. d-dimensional os illator; !k = jkj0 =mg
Physi al rami ations: quantum me hani al urrent operator
^j(x;  ) = i (ay (x;  ))a(x;  ) ay(x;  )a(x;  ) fun!
  : int
2m
i   
( (x;  )) (x;  ) (x;  ) (x;  )  0 (x;  )

! 2m m
i.e.  is measure of (super) urrent ow
Consider stationary phase equations
0 2
i  = gÆ + ; i Æ =  =j
m
Se ond equation ; ontinuity equation
First equation: system adjusts to spatial u tuations of density by dynami al phase
u tuation
Equations possess steady state solutions with non-vanishing urrent ow: setting
  =  Æ = 0, obtain Æ = 0 and   j = 0, i.e. below the ondensation temperature,
a on guration with a uniform density pro le an support a steady state divergen eless
(super) urrent.
Noti e that a `mass term' in the  a tion would spoil this property, i.e. within our
present approa h, the phenomenon of super urrent ow is intimately linked to the Gold-
stone mode hara ter of the  eld.
Steady state urrent ow in normal environments is prevented by the me hanism of
energy dissipation, i.e. parti les onstituting the urrent ow s atter o imperfe tions

Le ture Notes O tober 2004


Le ture XII 88

inside the system thereby onverting part of their energy into the reation of elementary
ex itations. How an dissipative loss of energy be avoided. Trivially, no energy an be
ex hanged if there are no elementary ex itations to reate. In reality, this means that the
ex itations of the system are energeti ally high-lying su h that the kineti energy stored in
the urrent- arrying parti les is insuÆ ient to reate them. But this is not the situtation
that we en ounter in the super uid! As we saw above, there is no energy gap separating
the quasi-parti le ex itations of the system from the ground state. Rather, the dispersion
! (k) vanishes linearly as k ! 0. However, there is an ingenuous argument due to Lan-
dau showing that a linear ex itation spe trum indeed suÆ es to stabilize dissipationless
transport:
Consider ow of uid through a pipe. Let us assume that the ow o urs at a uniform
velo ity V. Taking the mass (of a ertain portion of the uid) to be M , the urrent arries
a total kineti energy E1 = M V2 =2. Now, suppose we view the situation from the point
of view of the uid, i.e. we perform a Galileian transformation into its own rest frame.
From the perspe tive of the uid, the walls of the pipe appear as though they were moving
with velo ity V. Now, suppose that fri tional for es between uid and the wall lead to
the reation of an elementary ex itation of momentum p and energy (p), i.e. the uid is
no longer at rest but arries kineti energy. After a Galileian transformation ba k to the
laboratory frame we nd that the energy of the uid after the reation of the ex itation is
given by
M V2
E2 = + p  V + (p)
2
Now, sin e all of the energy needed to manufa ture the ex itation must have been provided
by the liquid itself, energy onservation requires that E1 = E2 , or p  V = (p): Sin e
p  V > jpjjVj, this ondition an only be met if jpjjVj > (p). While systems with
a `normal' gapless dispersion, (p)  p are ompatible with this energy-balan e relation
2

(i.e. no matter how small jVj, quasi-parti les of low momentum an always be ex ited),
p!0
both gapped dispersions (p) ! onst: and linear dispersions are in ompatible if V be-
omes smaller than a ertain riti al velo ity V . Spe i ally for a linear dispersion
(p) = v jpj, the riti al velo ity is given by V = v . For urrents slower than that, the
ow is ne essarily dissipationless.

Le ture Notes O tober 2004

You might also like