SERC School on Magnetism and Superconductivity 06

The Kondo Eect

Vijay B. Shenoy
([email protected])

Centre for Condensed Matter Theory Indian Institute of Science

VBS

The Kondo Eect 0

SERC School on Magnetism and Superconductivity 06

Overview
What is Kondo eect? Motivation Why do this? Plan of the Lectures

VBS

The Kondo Eect 1

SERC School on Magnetism and Superconductivity 06

Kondo...

This is Jun Kondo...but, whats his eect? Something that happens in metals...

VBS

The Kondo Eect 2

SERC School on Magnetism and Superconductivity 06

Metals, Wonder Materials!

What is a metal? Many electron system with a Fermi Surface Key: Gapless excitations Linear T specic heat Temperature independent magnetic susceptibility ... Key: Protected from repulsive interactions by Pauli Fermi liquid

VBS

The Kondo Eect 3

SERC School on Magnetism and Superconductivity 06

Resistivity in Metals

Almost constant at low temperatures...all way to linear at high temperatures

VBS The Kondo Eect 4

SERC School on Magnetism and Superconductivity 06

Resistivity in Metals...Theres More!

Increases with impurity content Has some universal features...

VBS The Kondo Eect 5

SERC School on Magnetism and Superconductivity 06

The Kondo Eect!

There is a resistance minimum in some cases!

VBS The Kondo Eect 6

SERC School on Magnetism and Superconductivity 06

When does this happen?

Impurity susceptibility: Fe in Cu

Happens when impurity is magnetic! Whats the big deal? Why bother?
VBS The Kondo Eect 7

SERC School on Magnetism and Superconductivity 06

Why bother?

Universal relation between dimensionless magnetic part of resistivity and temperature!!

VBS

Universal relations observed for other properties (susceptibility etc.) The Kondo Eect 8

SERC School on Magnetism and Superconductivity 06

Not Impressed? How about Quantum Dots?

Regions that can hold a few hundred electrons! Can drive a current through these! This is Nano!
VBS The Kondo Eect 9

SERC School on Magnetism and Superconductivity 06

Universal Stu in QDs too!

Universal relation between dimensionless conductance and temperature!

VBS The Kondo Eect 10

SERC School on Magnetism and Superconductivity 06

Theres More!

Dimensionless resistivity vs. temperature in Kondo metals and dimensionless conductance vs. temperature show the same universal function!!!!
VBS The Kondo Eect 11

SERC School on Magnetism and Superconductivity 06

Why Bother?
These results cannot just be happenstance! ... The Kondo problem is a cornerstone in physics...it gave rise to a lot of physics! ... Two Nobel Prizes! ... Dynamical Mean Field Theory (DMFT): Revival of interest in quantum impurity problems

VBS

The Kondo Eect 12

SERC School on Magnetism and Superconductivity 06

Plan of the Lectures

When is an impurity magnetic? Anderson Model The sd model: Low energy physics Whence resistivity minimum? (High T ) The Kondo Problem! Variational approach to the Kondo ground state (Low T ) Interpolation Scaling/RG ideas Local Fermi Liquid ideas

VBS

The Kondo Eect 13

SERC School on Magnetism and Superconductivity 06

Prerequisites
Must know Band theory of metals (One e physics) Essential Stat. Mech. No harm if you know Second quantization Greens function Great if you know Diagrammatic perturbation theory!

VBS

The Kondo Eect 14

SERC School on Magnetism and Superconductivity 06

How do we model an impurity in a metal?

Host metal A set of Bloch states |k ( - spin index) Hamiltonian Hm = k (k )c k ck Bandwidth 2D, i. e., D (k ) D Density of states ()

Impurity with d-levels Impurity d-level |d with energy d Hamiltonian Hd =

c d d cd

Host-impurity Hybridization Hybridization : Hh =

k Vk ck cd + Vk cd ck

Hamiltonian H = Hm + Hd + Hh specied when (k ), d and Vk are given

VBS The Kondo Eect 15

SERC School on Magnetism and Superconductivity 06

The Impurity Question

Given that we have the host metal at some lling, say the chemical potential is mid-band ( = 0), what happens to the impurity d-level? Is the impurity level occupied? Is the impurity magnetic, i. e., does it retain a spin?
Recap: What happens in a free ion?

Answer the question in a simple model: Host () = 0 constant...at band Vk = V , a constant Now, a three parameter model (0 , d and V ) And, is given to be zero
VBS The Kondo Eect 16

SERC School on Magnetism and Superconductivity 06

The Impure Answer: Physics Arguments

From physics arguments Impurity level will be occupied if d < (Of course!) But an electron cannot stay at the impurity for ever...due to hybridization...the impurity state picks up a lifetime Estimate of lifetime ( = 1): =
1

The impurity level behaves like a virtual state with i energy eigenvalue Ed = d This implies a Lorentizan density of state for the impurity 1 d () = ( d )2 + 2

|V |2 0

VBS

The Kondo Eect 17

SERC School on Magnetism and Superconductivity 06

nical Digression: The Resolvent Greens Functi

Hamiltonian: H The Greens operator (function) (i H )G(t t ) = (t t )1 t In frequency language G+ ( ) = ( + H )1
Recall:
1 x+

1 x+i

=P

Why bother? Well, 1 () = Im(TrG( )), 1 a () = ImGaa ( )

`1
x

i (x)

The Greens function: route to calculate density of states!

VBS The Kondo Eect 18

SERC School on Magnetism and Superconductivity 06

Back to Impurity Density of States...

Hamiltonian: H = Hm + Hd +Hh (Perturbation Hh )
H0

Notation, G0 = ( H0 )1 , G0d =
Gd ( ) = G0d P =

1 d

The impurity Greens function (after a bit of algebra)

1 G0d =
|Vk |2 k (k)

X |Vk |2 1 , Ed = d + P , (Ed i) ( k ) d k X |Vk |2 (d (k))

k

1 P

|Vk |2 k (k)

Physics: Impurity level forms a virtual state

VBS The Kondo Eect 19

SERC School on Magnetism and Superconductivity 06

Impurity Density of States...and...!!

In the simple three parameter model
D d 2 , Ed = d + 0 |V | ln d + D = 0 |V |2

This implies a Lorentizan density of state for the impurity

1 1 d () = ImGd () = ( Ed )2 + 2

The occupancy of the virtual level

nd 1 1 tan1 = d nF () d () |{z} 2
T =0

Ed

But.. Sz nd nd = 0! This model says impurity can never have a moment...Whats the missing physics?
VBS The Kondo Eect 20

SERC School on Magnetism and Superconductivity 06

The Anderson Model

Anderson, PR 124, 41 (1961), realized the moment formation has origins in strong correlations! (contribution from Wol/Fridel) Introduced local Coulomb correlations on the impurity
Hd = X

d c d cd + U nd nd

U is the d orbital Coulomb repulsion energy The Anderson Model

H = X
k k

(k)c k ck

cd ck c + V Vk c d k k

d c d cd + U nd nd +

Can we see that this is likely to produce a moment?

VBS The Kondo Eect 21

SERC School on Magnetism and Superconductivity 06

Take a Moment...See the Moment

Impurity spin operator S = Now, S S =
1 c d cd 2 3 5 ( n + n ) d d 4 4 nd nd

The impurity Hamiltonian can now be written as Hd =

3 5 (d + U ) cd cd U S S 4 4
d

Thus, if U is large enough, and if we can arrange things such that nd = 0, the impurity will maximize S S ...will pick up a moment! So, what competes with this?...Kinetic energy (hybridization)!

VBS

The Kondo Eect 22

SERC School on Magnetism and Superconductivity 06

The Hartree-Fock Theory

Hartree-Fock (Mean Field) treatment of Coulomb
HF Hd =

d c d cd + U ( nd nd + nd nd nd nd )

Note: nd are unknown (variational) parameters HF Anderson Hamiltonian

H
HF

X
k

(k)c k ck

X
k

cd ck c + V Vk c d k k

(d + U nd ) cd cd U nd nd + | {z } d

This is a one particle Hamiltonian that we have already solved...apply our results to each spin channel...
VBS The Kondo Eect 23

SERC School on Magnetism and Superconductivity 06

The Hartree-Fock Solution

Four parameter Anderson model (d , 0 , V, U ) at chemical potential Greens function
(Ed i) D d = d + 0 |V |2 ln , + D +U nd d | {z }
d

Gd ( ) =

1 = 0 |V |2

Ed

DOS etc.

d () nd

= =

1 ( Ed )2 + 2 1 1 E d tan1 2

VBS

The Kondo Eect 24

SERC School on Magnetism and Superconductivity 06

The Hartree-Fock Solution

The occupancy condition enables us to determine nd
nd nd = = d + U nd + U n 1 1 d d tan1 2 1 1 tan1 2

Question: When is it possible to have nd = nd ? Need to nd a condition on d , 0 , V, U ! This will tell us when an impurity can be magnetic... The condition will depend only on two parameters...(d )/U and /U

VBS

The Kondo Eect 25

SERC School on Magnetism and Superconductivity 06

The Hartree-Fock Solution

If the impurity develops a moment, then the DOS picture will look like
Ed d() Ed

d()

So moment denitely forms when

d < with |d |, AND d + U > with |d + U |

Physics: (i) Impurity level must be occupied by only one spin species (ii) The hybridization should be small enough to hold this spin species at the impurity

VBS

The Kondo Eect 26

SERC School on Magnetism and Superconductivity 06

The Hartree-Fock Solution

Can work out the exact condition for a special case... |d | = U 2 ...the symmetric Anderson model... Here, nd + nd = 1 when impurity develops a moment Exercise: Show this. If we call m = 1 2 ( nd nd ), then from occupancies 1 calculated earlier 2m tan1 2mU Exercise: Show this. Condition for moment to exist:
U

>1

Physics: Coulomb must beat kinetic energy if a moment is to exist! U > 0 2 |V |2 ...host metals with small density of states at the Fermi level encourage impurities to be magnetic!
VBS The Kondo Eect 27

SERC School on Magnetism and Superconductivity 06

The Hartree-Fock Solution

T = 0 phase diagram of the Anderson model The shaded region corresponds to the broken symmetry phase...Exercise: The boundary represents a quantum phase
VBS
transition...of what order?

The Kondo Eect 28

SERC School on Magnetism and Superconductivity 06

The Hartree-Fock Extras

Is in general agreement with available experiments Does the impurity polarize the conduction electrons? What happens to the conduction electron density of states?....Nothing!! This is called the Anderson compensation theorem. Exercise: Prove this theorem. Signatures in specic heat and susceptibility
argue, then work out the signatures. Exercise: First

There is an entirely dierent (and a more beautiful) way to arrive at this using the Fridel sum rule which states that the number electrons localized on the impurity is equal to the phase shift of the conduction electrons at the chemical potential (modulo factors of )...you will work this out in the problem set.
VBS The Kondo Eect 29

SERC School on Magnetism and Superconductivity 06

The Hartree-Fock Critique

How good is the Hartree-Fock solution? Sources of discomfort Not a true ground state of the Anderson Hamiltonian!! Quantum uctuations are important... Phase transition in a nite system! 1 Correlation time scale U must be much larger 1 than life time of state for HF solution to be 1 1 good...Thus U , or U ...alas, thus HF solution may not be so good in the magnetic regime!! ...
VBS

We now know that there is a lot more to the story than Hartree Fock! The Kondo Eect 30

SERC School on Magnetism and Superconductivity 06

Low Energy Physics

Focus on systems which do show an impurity moment Key idea: Large U supresses charge uctuations on the impurity...the impurity charge degree of freedom is quenched... The only degree of freedom that remains at the impurity is its spin...low energy excitations are only the spin excitations What is the eective low energy Hamiltonian for the system? Naturally, such a Hamiltonian should contain only the spin operator of the impurity... Such a Hamiltonian is called the sd Hamiltonian...

VBS

The Kondo Eect 31

SERC School on Magnetism and Superconductivity 06

The sd Hamiltonian
The Hamiltonian H=
k

(k )c k ck + J s S

1 c s= N k,k k ck is the conduction electron spin at the impurity site (N is number of sites); J is an exchange term...
i J X h c k c k c k ck S z + c k c k S + c k ck S + 2N
k,k

Notice the spin ip scattering terms... Whence J ? How is this related to the Anderson Hamiltonian?
VBS The Kondo Eect 32

SERC School on Magnetism and Superconductivity 06

Whence sd? The Physics

Discuss this in the context of the symmetric Aderson model (|d | = U/2) Large U requires that nd = 1 always... This is bad for kinetic energy...electron is strongly localized So, when Coulomb is not looking, the impurity...electron hops on to the band, or a band electron hops on to the impurity site to gain kinetic energy... This leads to an antiferromagnetic exchange interaction between the local impurity spin and the conduction electron spin a the impurity site
VBS The Kondo Eect 33

SERC School on Magnetism and Superconductivity 06

Whence sd? The Physics

Initial Virtual Final

Band V V Impurity

Energy U/2

Thus J
VBS

|V |2 U ,

and is antiferromagnetic in nature!

Now for a more detailed calculation to get the constant! The Kondo Eect 34

SERC School on Magnetism and Superconductivity 06

Whence sd? The Formal Stu

Several ways: Projection, Canonical Trans. etc.. Projection Technique: Break up the Hilbert space into three subspaces... corresponding to nd = 0, nd = 1 and nd = 2...we are looking to develop an eective Hamiltonian in the nd = 1 subspace.. State | = |0 + |1 + |2 The eigenvalue problem H | = E | can now be cast as
0 H00 B B H10 @ 0 H01 H11 H21 0 H12 H22 10 | 0 CB C B | 1 A@ | 2 1 | 0 C B C = E B | 1 A @ | 2 0 1 C C A

H00 , H11 , H22 are Hamiltonians in the 0, 1, 2 subspaces, H01 etc are hopping
operators from one subspace to another; obviously H10 = H01

VBS

The Kondo Eect 35

SERC School on Magnetism and Superconductivity 06

Whence sd? The Formal Stu

A bit of algebra gives H11 + H10 (E H00 )1 H01 + H12 (E H22 )1 H21 |1 = E |1
ef f H11

A bit of thought gives H01 =

k

Vk c ), k cd (1 nd

H21 =
k

Vk c d ck nd

And, after a few ten sheets of paper, some lost sleep and hair...we may get..

VBS

The Kondo Eect 36

SERC School on Magnetism and Superconductivity 06

Whence sd? The Formal Stu

ef f ; two terms ...the additional part to H11 in H11
X Jkk i h c k ck c k c k S z + c k c k S + c k c k S + +
sd

k,k

k,k

{z like interaction

{z } | Potential scattering 1 1 Vk + Jkk = Vk d + U (k ) (k) d V Vk 1 1 k Vkk = 2 (k) d d + U (k )

Vkk c k ck

Note Jkk is antiferromagnetic for states near the chemical potential.. Approximation, set Jkk = model...
VBS J 2N ,

and Vkk = 0...get the sd

Can we now explain the Kondo eect...i.e., the The Kondo Eect resistivity minimum? There are only two parameters in 37

SERC School on Magnetism and Superconductivity 06

Electrical Transport: Recap

The famous formula =
ne2 m

is the transport life time...all transport physics is in .. is obtained from the T matrix t 1 = 2 d3 k |Tk ,k |2 (1 cos kk ) ((k ) (k )),
1

k kf as a

Strategy: Treat sd as perturbation, and obtain series in J But, before that, what is the T matrix?

VBS

The Kondo Eect 38

SERC School on Magnetism and Superconductivity 06

The T Matrix
Consider H = H0 + V where V is an external potential (like the sd interaction term) Roughly, the T matrix represents an eective potential seen by the bare states Recall, G = ( H )1 = G = G0 + G0 V G Denition T G0 = V G, implies
T = V + V G0 V + V G0 V G0 V + ... = V (1 G0 V )1 Think of the connection to the path integral...

Physics: The rate of transition from |a to |b induced by V is given by 2 | a|T |b |2 (a b )...generalization of the Fermi golden rule!
VBS
The T matrix is connected to the S matrix, for any two states |a , |b , a|S |b = ab 2 a|T |b (a b )

The Kondo Eect 39

SERC School on Magnetism and Superconductivity 06

Resistivity in the sd model

In the sd model, we expect lifetimes to be spin dependent, i. e., we need to calculate and There is an added technical complication: if we set out to calculate Tk,k , we have to trace over the impurity spin states...we will adopt a notation that Tk,k is a 2 2 matrix,...we calculate such matrices and in the nal step take the trace of the impurity states...this will naturally handle the intermediate impurity states We shall use diagram techniques for this calculation, useful to rewrite sd interaction as
i J X h c k c k c k ck S z + c k c k S + c k ck S + 2N
k,k

VBS

The Kondo Eect 40

SERC School on Magnetism and Superconductivity 06

T matrix elements to 1st order in J

JSz k k JS+ k k

Diagrams for the rst order process

1 Tk ,k = 1 J 2N 1 Sz , Tk ,k = J 2N

S+
1 2+S S J 2 Sz +

2+S S J 2 Sz + and similarly,

Thus, if we stop at 1st order in J for T , we get 1 1 1 2 + S S+ + S+ S J 2 = + J 2 2 Sz independent of temperature!! Bad news!!

VBS

The Kondo Eect 41

SERC School on Magnetism and Superconductivity 06

T matrix elements 2nd order in J - Direct terms

2 Diagrams for Tk ,k
JSz JSz k q k JS JS+ k q k

JSz JSz q

JS+ JS q

VBS

The Kondo Eect 42

SERC School on Magnetism and Superconductivity 06

T matrix elements 2nd order in J - Flip terms

2 Diagrams for Tk ,k
JS+ JSz k q k JSz JS+ k q k

JSz JS+ q

JS+ JSz q

VBS

The Kondo Eect 43

SERC School on Magnetism and Superconductivity 06

T matrix elements 2nd order in J

Matrix elements up to 2nd order in J
(2) Tk k

J J2 2 = Sz + 2 Sz (P ((k) + H ((k))) + S S+ P ((k)) + S+ S H ((k)) N N J2 J (2) S+ + 2 [(Sz S+ + S+ Sz )(P ((k)) + H ((k)))] Tk k = N N

What are P and H P ( ) =

q

(1 nF ((q ))) H ( ) = (q )

nF ((q )) (q )

Note that for our simple two P ( ) + H ( ) 0 for !! Even otherwise, P ( ) + H ( ) 0 is independent of temperature!
VBS The Kondo Eect 44

SERC School on Magnetism and Superconductivity 06

T matrix elements 2nd order in J

Thus, using some spin algebra S S =
(2) Tk k

3 4

2S Sz z

1 X nF ((q )) J (1 2Jh( )) h( ) = = N N q (q )

In our case
h ( ) 0 Z nF () d , D
D

For T > 0 h( = 0) 0

kB T D

1 d = 0 ln

kB T D

Note that h( ) is logarithmically divergent as and T 0...the key to this divergence is the intermediate spin ip scattering in the direct process! It is the quantum nature of the spin that rears its head! Physics: Spin ip scattering leads to a logarithmically divergent inverse lifetime!
VBS The Kondo Eect 45

SERC School on Magnetism and Superconductivity 06

And, the Kondo Result for Resistivity

From the T matrix elements, obtain the lifetime (), and use it in the formula for conductivity
(T ) Z d() () dnF () d

We get resistivity R
kB T Rsd (T ) = c R0 R1 ln D

Total resistivity R = AT 5 + Rsd (T ) Minimum occurs at TM = Bc

VBS
1 5

c concentration

This one-fth power law has been experimentally veried! The Kondo Eect 46

SERC School on Magnetism and Superconductivity 06

And Experiments

In a certain temperature regime near the resistivity minimum, the 2nd order T matrix theory works quite well!
VBS

Are there signatures in other properties?

The Kondo Eect 47

SERC School on Magnetism and Superconductivity 06

Magnetic Susceptibility
We know that the susceptibility drops (compared to the Curie behaviour as temperature is decreased...) How to we understand this from the sd Hamiltonian? Imagine an applied magnetic eld...if there were no sd interaction, then the spin will point in the direction of the eld... Now turn on the sd interaction...what happens? The spin ip scattering from the conduction electrons acts to reduce the moment pointing along the direction of the magnetic led...we expect susceptibility to fall with reducing temperature...

VBS

The Kondo Eect 48

SERC School on Magnetism and Superconductivity 06

Magnetic Susceptibility
In fact, a little thought will tell you that the spin ip process (which was the key in giving raise to the upturn of resistivity) is also the key player here, and contributes to a logarithmic correction to the susceptibility 1 (T ) C0 + (2J0 )2 ln kB T kB T D

Note that the above result includes the conduction band polarization..

The logarithm is rearing its head again...this is in general agreement with experiments in a certain temperature range...but there is a problem is going to lower temperatures...
VBS The Kondo Eect 49

SERC School on Magnetism and Superconductivity 06

What have we learnt?

We learnt the following from low order perturbative calculations Spin ip scattering is the key to much of the physics seen Becomes increasingly eective at lower temperatures...this is due to the fact that the thermal kicks that the impurity and the conduction electrons experience destroy the coherent spin ip scattering at higher temperatures... In the temperature regime where the spin ip process just begins to rear its head, we notice that all properties show logarithmic temperatures corrections... The simple low order perturbative calculation leave us clueless vis-a-vis low temperature experiments...

VBS

The Kondo Eect 50

SERC School on Magnetism and Superconductivity 06

How do we chill-out?
The obvious thing to do is to extend the perturbation theory to all orders and hope for the best... The obvious is technically a bit tight! Abrikosov was a key player in working out the techniques required to sum the most singular spin ip diagrams...it has now been done for various responses... Results for all orders in J
R (T ) J2 1 2J0 ln 1 0 kB T 2 (2J0 ) ln 1 @ D A 1+ k T kB T 1 2J0 ln B
D

kB T D

(T )

VBS

Great! But...

The Kondo Eect 51

SERC School on Magnetism and Superconductivity 06

Chill! The Kondo Problem!

...everything blows up on our face at kB TK = De the Kondo scale This is TROUBLE! Things are blowing up at a FINITE temperature TK ... Interestingly, ALL calculated properties blow up at TK ... This is therefore the key energy scale in the problem...but what could such an energy scale be associated with? Note the nonanalyticity in J ... How to get around this? May be start from T = 0...
VBS The Kondo Eect 52
1 2J 0

SERC School on Magnetism and Superconductivity 06

Aim for the Ground! Hope for the Sky!

Since the coupling is antiferromagnetic, it is reasonable to assume that the ground state is a singlet...total spin zero state! Mattis proved that the ground state is a singlet...but did not nd the ground state wavefunction... Given that we know the ground state is a singlet, can we guess the wavefunction? We make one such guess, and learn something!

VBS

The Kondo Eect 53

SERC School on Magnetism and Superconductivity 06

Cooper meets Kondo!

The true singlet ground state might be complicated...we ask a dierent question analogous to Coopers question in the superconductivity context... Given a lled Fermi sphere |F S and an impurity spin, what happens if I add one more electron ? We may imagine the ground state of the system to be the undisturbed Fermi sphere and a singlet that forms between the added electron and the impurity...This will be consistent with Mattis... The wavefunction is
1 X a k ck |F S ; c k |F S ; | = 2 k>k
f

with

VBS

surface (exactly as in the Cooper calculation)...

k>kf

|ak |2 = 1 Note that the added electron lives outside the Fermi

The Kondo Eect 54

SERC School on Magnetism and Superconductivity 06

Cooper meets Kondo!

We need to determine ak ...how? The optimal ak s are the ones that minimize the ground stated energy Need to calculated Eg ({ak }) = |H | ...a bit of calculation shows ( is set to zero)
Eg ({ak }) =
k>kf

(k)|ak |2 +

3J 2N

k,k >kf

a k ak

Minimizing subject to the condition using a Lagrange multiplier E , gives 3J ((k ) E )ak 4N
VBS

k>kf

|ak |2 = 1

ak = 0
k >kf The Kondo Eect 55

SERC School on Magnetism and Superconductivity 06

Cooper meets Kondo!

Resulting secular equation
Z D 4 1 1 1 X 4 3J 0 d = = 0 = E = De 3J N k>k (k) E E 0
f

The gain in energy does look like the Kondo scale, but it is larger than kB TK ... Physics: The added electron has localized around the impurity and formed a singlet with it...There is, of course, increase of kinetic energy due to localization, but gain in exchange energy...all of it eventually results
0 ...Note that for any in a non analytical gain of De antiferromagnetic J no matter how small, the singlet we have assumed will be lower in energy... 4 3J

VBS

The Kondo Eect 56

SERC School on Magnetism and Superconductivity 06

Cooper meets Kondo!

Electron localized around the impurity (Kondo cloud)
vf De4/3J0

| =
1 ( 2

Length scale of the cloud But is this the real story?

VBS

vf De
4 3J0

...

The Kondo Eect 57

SERC School on Magnetism and Superconductivity 06

Better Variations...
Energy gain De
4 3J 0

is less than kB TK

Variational state is not an eigenstate of the sd Hamiltonian! sd interaction can create particle-hole excitations... Yoshida considered a variational singlet state with all possible particle hole excitations, and concluded that the ground state has an energy gain of kB TK It contains a Kondo cloud that screens the impurity spin by forming a singlet...it is true many-body state... The dressed Kondo impurity scatters any electronic excitations with the maximum possible phase shift (/2) and the T = 0 resistivity is the maximum possible resistivity permitted by unitarity...
VBS The Kondo Eect 58

SERC School on Magnetism and Superconductivity 06

arming up from T = 0...Finding the connection.

Knowing the ground state is good...but knowing the nature of excitations above the ground state is key... There has been a lot of eort to calculate properties, but most will work only for T TK ...

On one hand we have a theory that works well for T TK , and on the other hand another theory for T TK ...

What about a smooth connection between the two? Anderson provided a key insight as to why it is dicult to obtain a smooth connection by usual techniques...This work contained in it certain ideas that were later key to the complete solution of the Kondo problem by Wilson...

VBS

The Kondo Eect 59

SERC School on Magnetism and Superconductivity 06

Poor Mans Approach to the Kondo Problem

Anderson asked the following question: What are the most important degrees of freedom in the Kondo problem? Can one write out an eective Hamiltonian just for these degrees of freedom? Clearly the key degrees of freedom at low temperatures are the states near the chemical potential... They can undergo scattering from the impurity that may be direct or spin-ip... He considered the Hamiltonian of the form
X
k

(k)c k ck

k,k

c J z ck ck c k ck S z + J c S + J c k k ck S + k

This is an anisotropic sd Hamiltonain with Jz , J ...

VBS

The Kondo Eect 60

SERC School on Magnetism and Superconductivity 06

Poor Mans Approach to the Kondo Problem

What is the eective Hamiltonian in the band reduced by D as shown?
111 000 000 111 000 111 000 111 000 111

111 000 000 111 000 111 000 111 000 111

=0

We expect to get a Hamiltonian with the same form as the sd but with new renormalized values of the J s... z .... Interpret Jz : it is How can we nd the new J s? Let us say that Jz goes to J the amplitude for scattering from |k to |k ... Now if k, k belong to the reduced band, the amplitude of this scatteringy will be a sum of direct scattering in the band (Jz ), and scattering from k to k via all intermediate states which lie in the shaded part of the band (call this Jz )... Clearly we need only to worry about spin ip scattering..

VBS

The Kondo Eect 61

SERC School on Magnetism and Superconductivity 06

Poor Mans Approach to the Kondo Problem

Flow equation for J s best seen by diagrams
J S z Sz J = Jz Sz + J S+
2 S S+ J = J

J S+ + J S

D 0 D

D dJ z D 2 2 2 S+ S 0 + J = 20 J = = 20 J D D dln D

With a similar equation for J , we have dJ z 2 = 20 J dln D dJ = 20 J Jz dln D

VBS The Kondo Eect 62

SERC School on Magnetism and Superconductivity 06

Poor Mans Approach to the Kondo Problem

2 J2 = C! The solution Jz

Key point: Suppose we start with Jz = J = J > 0 (antiferromagnetic), then the coupling constant J ows to innity...cannot do any pertubation theory
VBS The Kondo Eect 63

SERC School on Magnetism and Superconductivity 06

Poor Mans Approach to the Kondo Problem

2J 0 = De 0 !! Thus Note that for our sd-model, De the Kondo temperature is preserved along the trajectory!! Thus the key energy scale remains invariant 1 2J 1

Changing D is like changing T , and hence we expect T all properties to be universal functions of T K In many ways, this is the solution of the Kondo problem! But the full solution by Wilson is a treat in itself...

VBS

The Kondo Eect 64

SERC School on Magnetism and Superconductivity 06

Wilsons RG Appoach to Kondo Problem

Key observation: The problems are caused by a b b logarithmic divergence...consider a d = ln a
1 We ask why is 0 d = ? To see this, break up [0, 1] into intervals [(n+1) , n ] with > 1...Thus
[0, 1] =
[

[(n+1) , n ]

n=0

1 0

Z n X X d d ln = !! = = ( n +1) n=0 n=0

The divergence occurs because all the logarithmic intervals contribute equally... To study the Kondo model, Wilson considered a simplied Hamiltonian
H= Z
1 1

dk kc k ck + J s S

VBS

The Kondo Eect 65

SERC School on Magnetism and Superconductivity 06

Wilsons RG Appoach to The Kondo Problem

Starting from H = R1
1

dk kc k ck + J s S, and using log descretization

reduced to
X 1 1 H = (1 + ) n (c +n c+n cn cn ) + J s S 2 n=0

cn are electron operators to the right (left) of (k = 0) Now Z 1 1 1 s = f0 f0 f0 = dk ck 2 2 1 1/2 X 1 1 (1 ) n/2 (c+n + cn ) 2 n=0

f0

Thus, the impurity couples equally to states of all energies! This is the problem!

VBS

The Kondo Eect 66

SERC School on Magnetism and Superconductivity 06

Wilsons RG Appoach to The Kondo Problem

By sheer genius (bordering on subterfuge!) Wilson mapped this Hamiltonian to

H=
n=0

fn+1 (fn

1 + fn+1 fn ) + J f0 f0 2

This is a 1-D semi-innite chain with the rst site interacting with the impurity, and an exponentially falling hopping between neighbours!
0 J S 1 2 n n+1 n

VBS

does...what do the operators f We know what f0 n do?

The Kondo Eect 67

SERC School on Magnetism and Superconductivity 06

Wilsons RG Appoach to The Kondo Problem

around the impuritiy...f1 is a wave packet which peaks at a larger distance from the impurity... and so on!

f0 corresponds to a spherical wave packet localized

Electrons can hop from one wavepacket state to the neighbouring wave packet states...
The states fn can be obtained from c n ...via Lanczos tridiagonalization of the kinetic energy!

Why all this? Impurity now couples only to one state!

VBS The Kondo Eect 68

SERC School on Magnetism and Superconductivity 06

Wilsons RG Appoach to The Kondo Problem

Wilson then used a numerical renormalization group technique to diagnolize the Hamiltonian which involves the following step Dene
HN = (N 1)/2
N 1 X n=0

1 f ) + J + f fn n (fn f f0 S n+1 n +1 2 0

Why do this?: N is like taking T 0, note that (N 1)/2 J which is the eective exchange coupling at N (temperature T ) goes to innity as N (T 0).... Dene a transformation f + f HN +1 = 1/2 HN + fN N +1 fN N +1

VBS

The Kondo Eect 69

SERC School on Magnetism and Superconductivity 06

Wilsons RG Appoach to The Kondo Problem

Take J = 0 to start with, and ask what happens for large N ? You will see with a bit of thought that you will get two types of spectrum depending on if N is even or odd.. For = 2 Wilson showed that the eigenvalues are 1 even N : 0, 1.297, 2.827, 4 2... 2 2.... odd N : 0.6555, 1.976, 4, 8, ..., 2 Thus RG transformation HN +2 = R[HN ] has two xed points one corresponding to even number N and another odd number N ...

VBS

The Kondo Eect 70

SERC School on Magnetism and Superconductivity 06

Wilsons RG Appoach to The Kondo Problem

Now start with a very small J = 0...and focus on N even... Until N becomes large enough so that Jef f = (N 1)/2 J 1, the eigenvalues will look like those of even N ... For even larger N the Jef f ...what does this mean...the site 0 will from a singlet with the impurity and completely drop out of the Hamiltonian...i. e., the site 0 will decouple from the chain since it fully couples with the impurity!! This means although there are N is even, the spectrum of HN will be similar to that of odd N! When you start with a tiny J , there is a range of N (high T ) where the behaviour is same as that of the J = 0 xed point (this is where perturbation theory works)... and as N is increased it ows to the J xed point (the Kondo singlet)!
VBS

J = 0 is an unstable xed point, J = is a stable xed point...Best visualized by lifting a gure straight out of Wilsons

The Kondo Eect 71

SERC School on Magnetism and Superconductivity 06

Wilsons RG Appoach to The Kondo Problem

RG ow in the Kondo problem...all well and good..what about quantitative things?

VBS The Kondo Eect 72

SERC School on Magnetism and Superconductivity 06

Wilsons RG Appoach to The Kondo Problem

Using the fact that T N , and using an iterative scheme that correctly calculates the low energy excitations of the Hamiltonians at large N , Wilson obtained the universal function for susceptibility...

VBS

The Kondo Eect 73

SERC School on Magnetism and Superconductivity 06

Wilsons RG Appoach to The Kondo Problem

So what determines the ow towards the strong coupling xed point? Wilson argued that the ow is governed by the leading irrelevant operators... Based on symmetry arguments Wilson proposed that the leading irrelevant operators are of the form
2 + N/2 (a(f1 HN = H f2 + f2 f1 ) + b(f1 f1 1) ) | {z } Leading Irrelevant Operators

Note that the second irrelevant operator looks like an onsite interaction at site 1!!

Now using a combination of perturbation theory and numerics, Wilson showed that
W = Rimp Rf ree electron =2 where R = T Cv W is the Wilson ratio

VBS

The Kondo Eect 74

SERC School on Magnetism and Superconductivity 06

Nozi eres Local Fermi Liquid Picture

Nozi eres made very a insightful observation that near is that of a non the xed point Hamiltonian H interacting Hamiltonian, and near the xed point T 0, there is just a local interaction on the site 1...

2 = 2n n (n + n 1), f 1) Noticing that (f1 1 1 1 1 1 he realized that the system would behave like a Fermi liquid... He now used Fermi liquid arguments; the scattering phase shift of a quasiparticles of energy is a functional of the number density of all quasi-particles n
[, n ] = 0 () + X f, (, )n ()

0 corresponds to equilibrium (n0 () = nF ())

VBS

The Kondo Eect 75

SERC School on Magnetism and Superconductivity 06

Nozi eres Local Fermi Liquid Picture

Expanding about the chemical potential (= 0)
0 () = 0 +

and taking f, = f as energy independent, and that the quasiparticle scatters o only an opposite spin quasiprticle, the second term can be written as
,

f, (, )n () = f

We thus have () = 0 + + f X

Note that 0 = /2, and that there are two parameters and f ...

VBS

The Kondo Eect 76

SERC School on Magnetism and Superconductivity 06

Nozi eres Local Fermi Liquid Picture

Nozi eres argued that the phase shift will not be aected by small changes in the chemical potential...gives the condition
+ f 0 = 0 This leaves only one parameter!

Noting that the impurity density of states is given by 1 d () = , we get (accounting from spins), d () =
2 ,

thus

Cimp CF E

2 ...

Now by the Fridel sum rule magnetization 1 m= ( ) in presence of a magnetic eld B ...m
(22f 0 ) B

i.e.,

imp F E

VBS

This simple argument gives W = 2!!

The Kondo Eect 77

SERC School on Magnetism and Superconductivity 06

What about the Anderson Model?

The Anderson model was solved by NRG techniques by Krishnamurthy et al.

VBS

The Kondo Eect 78

SERC School on Magnetism and Superconductivity 06

An Important Piece in AM...

The impurity develops a resonance at the chemical potential for T < TK ...the Abrikosov-Shul resonance aka Kondo resonance

VBS

The Kondo Eect 79

SERC School on Magnetism and Superconductivity 06

To Conclude...
Kondo problem is solved, but it taught us... Correlation physics Mean eld theory Projection technique (Low energy physics) Transport calculations Perturbative (Diagram) calculations RG ideas Future... Kondo lattice etc.. Correlations at the nanoscale..

VBS

The Kondo Eect 80