Lecture 4

03/02/23, Th.

Today

Reading

1. Non-interacting system
(1)H0=αεαaαaα

The observables

(2)A=αβAαβaαaβB=αβBαβaαaβ

Then

χAB0(t)=iθ(t)αβγηAαβBγη[aα(t)aβ(t),aγaη]0

Noting the identity

[aαaβ,aγaη]=δβγaαaηδαηaγaβ

We find

(3)χAB0(t)=iθ(t)αβAαβBβα(f0(εα)f0(εβ))ei(εαεβ)t/

In the frequency domain

(4)χAB0(ω)=dteiω+tχAB0(0)=αβAαβBβαf0(εα)f0(εβ)ω+εαεβ+i0+

where

(5)ω+=ω+i0+.

The convergence factor i0+ is introduced as part of the adiabatic switch-on perturbation, to ensure that in far past the perturbation is off and the system is in an equilibrium state governed by H0. The response function χAB0(ω) then describes the response to a periodic perturbation like

(6)F(r,t)=Fei(qrωt)+Fei(qrωt)

so that

(7)δA(ω)=χAB0(ω)F

 

Next, we consider the density response function. The density operator is

n(r)=σψσ(r)ψσ(r)=1Vσkkakσakσei(kk)r=[dq]eiqrσkσakσak+qσ

so

(8)n(q)=kσakσak+qσ

 

Think of

δn(r,t)=id3rdt[n(r,t),n(r,t)]0{Fei(qrωt)+c.c.}=idtθ(tt)[n(r,t),n(q,t)]0F+c.c.

So for the density response commensurate with the external fields

(9)δn(q,t)=idtθ(tt)[n(q,t),n(q,t)]0F

This means

(10)χ0(q,tt)=iθ(tt)[n(q,t),n(q,t)]0

So that

(11)χ0(q,ω)=s[dk]f0(εks)f0(εk+qs)ω+εksεk+qs+i0+

This is the so-called Lindhard function.

 

Digression on spin response

When spin is considered, we need to consider both spin densities ns(r)where s,s=↑,. We define the total density and spin polarization, respectively, as

(12)n(r)=n(r)+n(r)m(r)=n(r)n(r)=Sz(r)

Then the spin-resolved density–density response function is defined as

(13)χss(r,r,tt)=i[ns(rt),ns(rt)]θ(tt)

Clearly,

(14)χnn=ssχss

What we call the longitudinal spin response: δm to the Zeeman field

(15)χS=χmm=ssssχss,with s=±1 for /.

A conclusion that one can draw immediately is that the spin–density response functions vanish in the paramagnetic state due to the obvious additional symmetry χ↑↑=χ↓↓. Thus, spin and density responses are perfectly decoupled in a paramagnetic electron liquid: this is an exact result.

For a non-interacting electron gas, we have χ↑↓=χ↓↑=0.1 So

(16)χS0(q,ω)=χ0(q,ω)

Oftentimes, we need to describe systems or excitations for which spin is no longer a good quantum number. For this purpose we need the spin density (in units of 2) is defined as

(17)Sα(r)=ssψs(r)σssαψ(r)

where σα are the Pauli matrices

(18)σx=(0110),σy=(0ii0),σz=(1001).

which have following properties

(19)σασβ=δαβ+iγϵαβγσγ.

Additionally

(20)σ±=12(σx±iσy)

So

(21)Sα(q)=k,ssaksσssαak+qs

Then we may define the transverse spin-spin response function χS+S. For the non-interacting electron gas

(22)χS+S0(q,ω)=σ[dk]f0(εk)f0(εk+q)ω+εkεk+q+i0+

 

 

2. The Lindhard function

The Lindhard function in Eq. (11) contains some essential information of how a many-electron system responds to external fields. We'd like to inspect its structure after analytically computing it. Since spin is good quantum number in this case, we can compute the spin Lindhard function defined as

(23)χs0(q,ω)=[dk]f0(εks)f0(εk+qs)ω+εksεk+qs+i0+.

Now we separate the above into two sums over the occupied states

(24)χs0(q,ω)=[dk]f0(εks)ω++εksεk+qs+[dk]f0(εks)ω++εksεk+qs.

where the second term comes from the replacement kkq.

Using the dispersion relation for a free-electron gas

[dk]f0(εks)ω++εksεk+qs=m2kFσq[dk]f0(εks)ω+qvFsq2kFskkFscosθ[T=0]=mkFσd12q01dxxd1dΩd(2π)df0(εks)ω+qvFsq2kFsxcosθ

where cosθ=k^q^, x=x/kFs, Ωd is the solid angle in d-dimensions.

Now define a function

(25)Ψd(z)01dxxd1dΩdΩd1zxcosθ

which is antisymmetric: Ψd(z)=Ψd(z)

We find

(26)χs0(q,ω)=g0skFsq[Ψd(ω+qvFσq2kFσ)Ψd(ω+qvFσ+q2kFσ)]

 

image-20230302093546225

Table 4.1 from Electron Liquids.

 

 

staticLindhard

Static Lindhard functions for d=1,2,3. Lines are analytical results, and dots are from an adaptive MC integration

 

The most important feature in the static Lindhard function is the singularity at q=2kF. As is clearly seen in the figure above, at this point χ0 diverges logarithmically for 1-dimensional system, and has discontinuity in the derivatives for 2- and 3-dimensional systems. A singularity in a function can produce long-range features in the Fourier spectrum, which in this case is the density variation in position space. It turns out, these integrals can be again evaluated analytically. Take the 3d case for example

(29)χ0(r,0)=12πng(0)sin(2kFr)2kFrcos(2kFr)(2kFr)4

with kF3=3π2n (spin degeneracy included). Suppose a positive point test charge (physically, a charged impurity) is placed at origin, it is seen that the response function above corresponds to the electron density modulation caused by a point test charge at origin. The density modulation has a net surplus at near the test charge. The over screening of the first electron shell leads to depletion of the next shell. And this oscillatory density modulation decays as cos(2kFr)/r3 for large r. Thus at large distance the test charge is completely screened. This spatially damped density oscillation with a characteristic period π/kF caused by external potential is called the Friedel oscillation.

 

In the long-wavelength limit

(30)limq01ddk(2π)2fk+qfkωk+qωk=ddk(2π)2fkεk=ddk(2π)2δ(εFεk)=gs(0)

This result is totally reasonable since the susceptibility is directly related to the phase space volume that allows for zero-energy excitation, which should be none but the number of states on the Fermi level in the q0 limit. Therefore, χ0 vanishes if there is a band gap.

 

 

 

 

Notes and references


1 Not true if there is interaction, in which case we only have χ↑↓=χ↓↑.