Linear response

theory
 Institute of Solid State Physics
Technische Universität Graz

Classical linear response theory

Fourier transforms
Impulse response functions (Green's functions)
Generalized susceptibility
Causality
Kramers-Kronig relations
Fluctuation - dissipation theorem
Dielectric function
Optical properties of solids
Impulse response function (Green's functions)

A Green's function is the solution to a linear differential

equation for a -function driving force

For instance, d 2g dg
m 2 b  kg   (t )
dt dt

has the solution

1   bt   4 mk  b 2 
g (t )  exp   sin  t t0
m  2m  2m 
 
 Green's functions

A driving force f can be thought of a being built up of many delta

functions after each other.

f (t )     t  t   f  t   dt 

By superposition, the response to this driving function is superposition,

u (t )   g (t  t ) f (t )dt 

d 2u du
For instance, m 2 b  ku  f (t )
dt dt

has the solution


1  b  t  t     4 mk  b 2 
u (t )   exp   sin   t  t    f (t ) dt 

m  2m   2m 
Green's function converts a differential equation into an integral equation
 Generalized susceptibility

A driving function f causes a response u

If the driving force is sinusoidal,

f (t )  F0 e i t
The response will also be sinusoidal.

u (t )   g (t  t ) f (t )dt    g (t  t ) F0 e i t  dt 

The generalized susceptibility at frequency  is

  i t 
u g ( t  t ) e dt 
 ( )  
f e i t
Generalized susceptibility

u
f
 ( ) 
http://lampx.tugraz.at/~hadley/physikm/apps/resonance.en.php
 Generalized susceptibility


i t 
u g ( t  t  ) e dt 
 ( )  
f e i t

Since the integral is over t', the factor with t can be put in the integral.

 ( )   g (t  t ) e  i ( t  t  ) dt 

Change variables to  = t - t', d = -dt', reverse the limits of integration

 ( )   g ( ) e i d

The susceptibility is the Fourier transform of the Green's function.

1

 i t
g (t )   ( ) e d
2
 First order differential equation

 ( )   g (t ) e  i t dt
dg
m  bg   (t )
dt b
 i
1 m
1  bt  b  ( ) 
g (t )  H (t ) exp    0 m  b 2
  
2
m  m m
m

The Fourier transform of a decaying exponential is a Lorentzian

 Susceptibility

du
m  bu  F (t )
dt
Assume that u and F are sinusoidal u  Ae i t F  F0 e i t

i mA  bA  F0

F0 b  i m
A  F0 2
b  i m b  m 2 2
b
 i
u 1 m
 
F m  b 2
    2

m
The sign of the imaginary part depends on whether you use eit or e-it.
 Susceptibility

dg
m  bg   (t )
dt
Fourier transform the differential equation

i m     b     1

1

b  i m

b
 i
1 m

m  b 2
    2

m
 Damped mass-spring system

d 2g dg
m 2 b  kg   (t )  2 m   i b   k   1
dt dt

ge t  b  b 2  4 mk
 
2m

k b
  2  i
1 m m
   2 2
Fourier transform pair  m   k 2   b 
       
 m   m 
1   bt   4 mk  b 2 
g (t )  H (t ) exp   sin  t
m  2m   2m 
 
http://lamp.tu-graz.ac.at/~hadley/ss1/crystaldiffraction/ft/ft.php
 More complex linear systems

Any coupled system of linear differential equations can be

written as a set of first order equations

dx 
 Mx
dt

The solutions have the form xi e  t


where xi are the eigenvectors and  are the eigenvalues of
matrix M.

Re() < 0 for stable systems

 is either real and negative (overdamped) or comes in complex

conjugate pairs with a negative real part (underdamped).
 More complex linear systems

resonances
Low frequency "1/f noise"
amplitude

frequency
 Odd and even components
Any function f(t) can be written in terms of its odd and even components

E(t) = ½[f(t) + f(-t)]

O(t) =½[f(t) - f(-t)]
f(t) = E(t) + O(t)
f(t) = ½[f(t) + f(-t)] + ½[f(t) - f(-t)]
 



f (t ) e  i t dt    E (t )  O (t )  cos  t  i sin  t  dt

 
  E (t ) cos  tdt  i  O (t ) sin  tdt
 

The Fourier transform of E(t) is real and even

The Fourier transform of O(t) is imaginary and odd
 b
 i
1 m
 ( ) 
m  b 2
  
2

m
even component exp( t )

sgn(t ) exp( t )
odd component

O (t )  sgn(t ) E (t )
E (t )  sgn(t )O (t )
Causality and the Kramers-Kronig relations (I)

 ( )   g ( ) e  i d   E ( ) cos( ) d  i  O ( ) sin( ) d       i    

The real and imaginary parts of the susceptibility are related.

If you know ', inverse Fourier transform to find E(t). Knowing E(t) you
can determine O(t) = sgn(t)E(t). Fourier transform O(t) to find ".
Causality and the Kramers-Kronig relation (II)
Real space Reciprocal space


1  ( )
 ( )   P d 
E (t )  sgn(t )O (t )    



O (t )  sgn(t ) E (t ) 1  ( )
 ( )  P d 
 

 

i i
  * i  , i    * 
 
Take the Fourier transform, use the convolution theorem.
P: Cauchy principle value (go around the singularity and
take the limit as you pass by arbitrarily close)

Singularity makes a numerical evaluation more difficult.

http://lamp.tu-graz.ac.at/~hadley/ss2/linearresponse/causality.php