Professional Documents
Culture Documents
Kubo Formula and Its Applications
Kubo Formula and Its Applications
3 Mar 15. Added Sect. 3 of the document about Kubo giving the same
results for non equilibrium spin distribution as the other methods we have
been using.
Introduction
This document aims to summarise those aspects of Kubo necessary for actually
doing stuff with the Kubo formalism without spending too much time going into
proofs of stuff which may be readily found in the specified sections in textbooks,
and which are things which you probably need to just go through once in your
life and just know the end result of in your day to day life as an engineer. The
textbooks referred in this document are
Bruus Many-Body Quantum Theory in Condensed Matter Physics: An
Introduction
Mahans Many-Particle Physics (Third edition)
We take it as a given that the non-equilibrium response of an operator A(t)
due to a perturbation H (t ) is given by
Z
(t)]i0
dt (i(t t )h[A(t),
H
hA(t)i =
t0
Z
R
CA,H
=
(t, t ).
t0
and H , and the hats on the operators denote that these operators are in the
interaction picture (but which I think in practice doesnt matter)
The recipe for evaluating a retarded correlation function between two operators
A and B is as follows
R
Write CA,B
in its Matsubara form (Bruus Eq. 11.20)
(t)]i0 hT A( )B( )i
i(t t )h[A(t),
H
= CAB ( )
where calligraphic script operators denote imaginary time operators, and
it was proven in Bruus Eq. 11.22 that CAB (, ) = CAB ( ).
exp(in ) )CAB ( ).
2.1
R
Suppose that we need to evaluate the correlation function CAB
(). Following
the prescription, we first calculate
CAB (, ) = hT A( )B( )i
Since CAB (
, ) = CAB (
) (Bruus Eq. 11.22) we might as well set
A, B, hT c ( )c ( )c c i
(1)
,,,
hT c ( )c ( )ihT c c i hT c c ( )ihT c ( )c i
G, (0)G, (0) G, ( )G, ( ).
2
Somehow the term on the left is always ignored, so ok Ill ignore it as well.
Substituting the second term on the last line back to Eq. 1, we obtain
CAB ( ) = A, G, ( )B, G, ( )
where we now adopt the convention of summing over repeated indices. This is
sometimes written as Tr(AG( )BG( )).
We now perform the Fourier transform
Z
CAB (in ) =
d exp(in ).
0
We can probably really work this thing out explicitly by expanding the G( )s in
the expressions in terms of their Matsubara frequency Fourier series sums and
evaluating the resulting Dirac delta functions, but integrals of this form occur
frequently enough that in practice we remember that the Fourier transformed
version of a product of two terms, one with and the other with is
CAB (iqa ) =
1X
Tr(AG(i(n + qa ))BG(in ))
n
(E.g. Bruus Eq. 11.87) where the n s are Fermionic frequencies (i.e. odd integer
multiples of /s) where qa is a Bosonic frequency (even integer multiple of
/), probably because there is a 2 difference in the arguments of the Greens
functions being transformed.
This by the way is the starting form of the Kubo formula presented in
the Indians paper, XL Qi and SC Zhangs paramagnetic wadeva-wadeva PRB
paper.
2.2
i
ij ()
1X
Tr(J i G(i(n + qa ))J j G(in )).
n
We invoke the end result, proved in Bruus Sect 11.4 and summarized in Eqs
11.57 and 11.58, that
X
1X
Res[f (zj )]n(zj ) exp(zj )
f (in ) exp(in ) =
i
j
n
(2)
where f doesnt have branch cuts (which G0 and products of G0 are examples
of, the sum over the right hand side are over the poles of f (z), n is the F-D (BE) distributions for fermions (bosons) depending on whether the n s summed
over are Fermionic / Bosonic frequencies, and the + () of the on the RHS
applies for fermionic (bosonic ) frequencies, to evaluate the Matusbara frequency
summation.
If the , , , basis is the H0 eigenbasis , G0 s is diagonal with
G0;, (in ) = (in E )1 , .
(Bruus Eqs 11.42, 11.43) In this basis and invoking Eq. 2 we have
CJ i ,J j (iqa ) =
=
=
1X
Tr(J i G(i(n + qa ))J j G(in ))
n
1 XX i
j
J, G (i(n + qa )J,
G (in )
, n
X
j
i
J,
J,
(G (E iqa )n(E ) + G (E + iqa )n(E ))
,
i
J,
n(E ) n(E )
E E + iqa
j
J,
.
where in going from the third to the last line we used n(E + iqa ) = n(E ) for
a Bosonic frequency qa . (This expression is similar to that in Bruus Eq. 11.88 )
Substituting iqa + i, we obtain
ab () =
=
=
i
ab ()
i
C a b (iqa = + i)
J ,J
i
n(E ) n(E )
a
b
J,
J,
.
E E + + i
(3)
2.3
Textbook results
Moving on from Eq. 3, we next want to obtain the real part of the DC (i.e.
= 0) conductivity. The part which most textbooks are interested inside
i
n(E ) n(E )
a
b
Re
J, J,
E E + + i
is the combination
(1) () = Im
1 n(E ) n(E )
E E + + i
Invoking
(x + i)1 = P
a
b
Re J,
J,
1
i(x)
x
we have
(1)
() =
n(E ) n(E )
a
b
(E E + )
Re J,
J,
which is similar to Marder Eq. 20.71, or the second line of Mahan Eq. 3.394
after expanding.
The presence of the (E E + ) term permits us to replace the E in
n(E ) with E + so that in taking the 0 limit, we have
n(E ) n(E + )
(E n
)/
lim
=
=
E=E
(E n(E ))
=
=
(1) ( = 0)
n(E ) n(E )
a
b
(E E + )
Re J,
J,
lim
0
a
b
(E n(E ))Re J,
J,
(E E ).
(Im just writing the superscript (1) on (1) to distinguish between various
terms identified as the real conducitivity in textbooks and papers. ) Recognising
that (E n)(E ) = exp(E )n(E ) and substituting inside gives Mahan Eq.
3.396.
2.4
Kubo in papers
2.4.1
1
2
X J,
J,
c.c.
<Ef >Ef
( )2
() =
Re
.
Im J,
J,
E E + + i
,
(2) () =
Re
Im J,
J,
E E + + i
,
X 1 n(E ) n(E )
2
a
b
(1
+ O( ) Im J,
J,
.
E E
E E
,
E E
,
disappears and can be ignored. (Ill go type out the proof some other time. )
The remaining term is then
(2) =
X n(E ) n(E )
,
(E E
)2
a
b
Im J,
J,
(4)
2.5
This expression apparently comes from PRB80, 134403 which one of the
authors of the ArXiV paper the good Dr. Ion Garate is an author of (probably
together with his sup), and in that paper (if I understood it properly) they
seem to imply, in their transition between the Eq 19 (which btw agrees with
the textbook prescription for a Matsubara frequency sum) and Eq 20 there, the
intriguing relation
X
!
A
R
lim Im(
Ga (in )Gb (i(n + ))
= GR
a (Gb Gb )
0
i=+i
n
(I think they set the Boltzmann constant to 1 so they had a T appearing in the
equation in the paper rather than the more usual 1 ) which I dont think Ive
seen anywhere else yet (although I havent really looked).
The annoying things about the above is that (i) in the ArXiV (and for
that matter, the Supplementary Online Material for the Kurebayashi Nature
Nano paper which the ArXiV paper expands on) it does not, on substituting
R/A
= (Ef Ea i), yield the
the expressions that they actually gave for Ga
expressions for that the papers gave (!!! O.o :p ), and (ii) I cant prove the
expression. Maybe there is some secret trick I dont know or have not realized
yet, but since these two conditions hold Ill just ignore whatever expressions
A
R
that are given that have the GR
a (Gb Gb ) combination.
It is also interesting to note that the clean limit of spin accumlation the
Manchon ArXiV paper, given explicitly in Eq. S11 of the Nature Nanotech
Supplementary Online Material, has the same form of Eq. 4.
Here I will attempt to reverse engineer what the authors of ArXiv and
Nature Nanotech paper actually did (instead of what their claimed used of the
A
R
GR
a (Gb Gb ) equation) from the results which they gave. What follows is
highly speculative and might be wrong.
The results of the ArXiV Eqs 3 and 5, and Nature Nanotech SOM Eq. S12,
are consistent with taking Eq. 3 and replacing i i. Writing, as a shorthand,
n n(E ) n(E ) and E E E ), we have
i n(E ) n(E ) a b
J, J,
E E + + i
"
1 #
i
n
a
b
Re
J, J, 1 +
E + i
E + i
n
a
b
J,
J,
... Re i
(E + i)2
((E)2 2 )
2(E)
a
b
a
b
...
Im(J, J, )
Re J, J, n
((E)2 + 2 )2
((E)2 + 2 )2
Re
=
(Oops to match the paper I should have replaced the J a in the equations
above with a actually. )
The ArXiv paper differs from the Nature Nanotechnology SOM
in one key aspect Eq. 3 of the ArXiV paper considered the intra-band
7
transistions in its Eq. 3 which were appararently neglected in the Eqs. (S12) of
the Nature Nanotech SOM. I think the origin of the ArXiV Eq. 3 comes from
something along the lines of
X
i n(E ) n(E ) a b
lim Re
,
J, J,
0
E E + + i
,
~1inter
~2inter
m
~ m)
~ z)
(m
((
z E)
+E
2
2m
~ m)
m
((
z E)
m
~
(m
(
z E))
2J
The form of the numerical factors for ~1inter differ from that in the paper,
~ z term for my expression for the ~ intra term.
and there is an additional E
2.5.1
X
,
(n n )
a
b
Im(J,
J,
)
(E E )(E E i)
which looks suspiciously like Eq. 4 above actually except for the additional
factor of in the denominator and i.
At this point I suspect that the paper Eq. 9 seems to be sloppily written
it gives the spin z current flowing in the x direction in response to an electric
8
field in the y direction, instead of the spin z current flowing in the y direction
due to an x electric field which the rest of the paper talks about. ll just ignore
the presence of these two and see what happens. To match the actual quantity
k
calculated in the paper we replace J a above with kmx z , and J b with my + x ..
The non-zero parts (remember that for a given ~k ,h~k, +|~k|~k, i = 0 will then
read
X
Im((ky z ), (y ), )
()
(n n )
m
(E E )2
,
Doing this matches the result in the paper for the universal intrinsic Hall
conductance given in the paper. In my actual calculation in the accompany
Mathemamatica notebook ArXiv1501.03292b.nb I actually calculated the spin
((E)2 2 )
a
b
z current for what corresponds here to the ((E)
2 +2 )2 Im(J, J, ) term with
a finite actually. The end result is independent of (which probably justifies
why its called universal); also on this level of Kubo its hky z i together which
has a finite value, hky i and h~ i are all apparently 0 when J = 0.
2.6
It turns out that the celebrated result that the SHE is kind of Berry curvature
is ultimately rooted in Eq. 4 (of this set of notes) as well.
In the 4th line of the Eq. A2 in the paper, they had a line saying lim0 i Qxy (+
i). This line is meant to be interpreted as linearising the ( i Es +Et )( 1)
term in to (Et Es ) as we have done before in this set of notes. ( The
Bernevig and Hughes TI book (Topological insulators and topological superconductors ) apparently shows that the 0th order term in goes to 0 in the leadup
to the Eq. 3.74 there which carried out the same steps. )
~
k a=6=b=
n(Ea ) n(Eb )
Im(ha||bihb|(k B)|ai
(Ea Eb )2
~
is in the
where a 6= b = refer to the eigenspinors, and the k in hb|(k B)|ai
direction of the electric field (this is the part of the current operator which has
non-zero contribution. )
Now for each value of ~k the summation over a 6= b has only two terms
a = +, b = and a = , b = +. The denominator (Ea Eb )2 = 2|vecB|2 is the
same for both terms, while the na nb part is equal in magnitude and opposite
in sign between the two terms. (At 0K one is 1 and the other 0.) We thus define
h~ (~k)i
X n(Ea ) n(Eb )
~ ~ )|ai
Im(ha|~ |bihb|(k B
(Ea Eb )2
a=6=b=
=
=
1
~ ~ )|i)
~
h|~ |+ih+|(k B
Im(h+|~ |ih|(k B)|+i
2|B|2
1 1
~ ~ )|ih|~ |+i)
~
h+|(k B
(h+|~ |ih|(k B)|+i
2|B|2 i
1
(zz )
where in going from the second to the third lines we made use of Imz = 2i
but the complex conjugate of each of the two terms being added is the other
term.
Next consider
Im(h+|[A, B]|+i
= Im(h+|A|+ih+|B|+i + h+|A|ih|B|+i h+|B|+ih+|A|+i h+||B|ih|A|+i)
= Im(h+||A|ih|B|+i h+||B|ih|A|+i)
1
=
(h+|A|ih|B|+i h+|B|ih|A|+i)
i
so that from the previous equation block before this one we recognise
=
=
=
=
h~ (~k)i
1 1
~ ~ )|ih|~ |+i)
~
h+|(k B
(h+|~ |ih|(k B)|+i
2|B|2 i
1
~ ~ )]|+i
Imh+|[~ , (k B)
2|B|2
1
~
b (k B)
|B|2
1
b (k~b).
|B|
10
~ =
In going from the second last to last line we made use of the facts that k B
~ b) and that b (k |B|)b = 0.
k (|B|
(I havent been very careful about factors of 2 and 1s in this. ) Further,
noting that (na nb )Im(ha|A|bi|biBha|) = (nb na )Im(hb|A|ai|aiBhb|) we can
take 1/2 of the value evaluated for h~ (~k)i above and denote that as the contribution to the non equilibrium spin distribution due to the electric field due to
each spinor branch. This then matches the expressions obtained by the other
methods mentioned.
As a side comment also we could have, in place of the ~ we have been working
with above, replaced it with ji = ki H since the parts of this which have non~ ~ only without the kinetic energy T term, and
zero spin contribution are ki B
we would have ended up with the familiar expression that the non equilibrium
~ form probably
~ (kj B)
Hall conductivity is a Berry curvature of the (ki B)
made famous by the XL Qi and SC Zhang PRB74, 085308.
11