A GENERAL EXPRESSION FOR THE CONDUCTIVITY TENSOR1

Can. J. Phys. Downloaded from www.nrcresearchpress.com by UNIVERSITY OF ARIZONA LIBRARY on 12/14/12

I n connection with the papers of Professor Prigogine and Dr. Ziman and
also with the remark made by Professor Bardeen, I would like to discuss
briefly a general expression for the conductivity tensor. Without referring to
the Boltzmann-Bloch equation, which is a certain approximation valid only
under a rather restricted condition (as was discussed by Professor Prigogine),
we can write down an exact formula for the electrical conductivity. Let us
denote the Hamiltonian of a system by H . This includes everything in the
absence of the field: the kinetic energy, potential energy of the electrons and
of the lattice atoms, interaction between electrons and lattice, and so on. But
the interaction of the system with the applied electric field, which we call He,,
is not included in H . T h e conductivity tensor for an oscillating electric field
can be expressed in the form

[I] u,.(m) = ~'L-{T~ p~]j.(O)j,(t+ifiA) ePiu'dtdh,

For personal use only.

where po is the density matrix representing the equilibrium distribution of the

system in absence of the electric field:
po = e-B* /Ti-(,-a*] ,
/3 being equal to 1IkT. j, and j, are the current operators of the whole system
in the v and p directions respectively, and j,(t) gives the natural motion of the
current as determined by the Hamilto~lianH; thus
j,(t) = exp (iHt/fi) j, exp ( - iIIt/fi) .
Eq. [I] is derived by a simple perturbation calculation. We assume t h a t a t
t = -a the system was in the equilibrium represented by po. A sinusoidal
electric field was switched on a t t = - a , which however is assumed to be
sufficiently weak. From the equation of motion
= [I-I
ifi~ + Hex(t) PI
we can easily obtain the change of p to the first order of II,,:

p -po = r
1
zfi
S'
-a
exp ( -Htf /ifi)[He, ( t r )PO]exp (Htr/ifi)dtr+O(H.r).

Therefore the current is given by

where He,(tf) will be replaced by -exE(tf), ex being the total dipole moment
of the system. Using the relation

'Presented a t the International Conference on Electron Transport in Metals and Solids,

held in Ottawa a t the National Research Laboratories, September 10-14, 1956.
2Department of Physics, Uiliversity of Tokyo, Tokyo, Japan.
1274
 KUBO: CONDUCTIVITY TENSOR 1275

Eq. [2] is transformed into Eq. [I]. Either Eq. [ I ] or [2] can be used, but Eq. [I]
Can. J. Phys. Downloaded from www.nrcresearchpress.com by UNIVERSITY OF ARIZONA LIBRARY on 12/14/12

is more syinmetrical.
This formula ti or^ was actively discussed last year in Japan. First Nakano (1)
discussed a particular case of Eq. [ l ] as an application of the method used by
Kubo and Tomita in their general theory of the magnetic resonance problem.
I t was further discussed by several people including m)~self (3). Professor
Feynman also discussed the same problem while he was visiting Japan last
summer. I was informed t h a t it was discussed in the United States by Lax,
Luttinger, Kohn, and maybe by some others. I would like to discuss here
briefly the implications of this sort of expression, which is quite general but
admittedly quite abstract.
(1) We know, for example, t h a t the Debye formula for the magnetic
susceptibility is generalized to
x = (M2)/kT,
where 144 is the magnetic moment of the whole system. This equation can be
further generalized to the non-equilibrium case (2) in which the complex
For personal use only.

susceptibility can be expressed generally iri terms of spontaneous fluctuation

of the magnetization of the system. This is to show that our formula is to be
regarded as a straightforward extension of the familiar formulae of suscepti-
bility (response in general) to non-equilibrium situations. Such an extension is
always possible so long as we confine ourselves to "linear dissipative" systems.
(2) I t can be easily seen t h a t Eq. [l]or [2] is actually equivalent to the well-
known Heisenberg-Kramers dispersion formula plus the familiar expressioit
of absorption given by the usual application of time-dependent perturbation,
provided that one allows the use of the integral expression of delta (plus or
minus) functions. We must admit a certain mathematical difficulty for the
convergence of the integral, but any convergence factor which may be intro-
duced is enough to give the right answer. Physically, we have to exclude the
long Poincari. cycle. This is esseiltially the very difficult question as to how
irreversibility arises from reversible motion; but we shall not discuss this here,
(3) T h e correlation of the spontaneous currents may be defined by
*,.(t) = 1)
((jv(o)jP(t) = ((jv(71jp(f+71) .
T h e average sign meails the average over the density matrix po. T h e con-
ductivity can be also written in terms of these correlation functions. For
example,

where the superscript s means the symmetric part of the tensor, and EB(w)is
the average energy of an oscillator with the frequency w a t the temperature
T = I/kP. Eq. [3] is the so-called fluctuation-dissipation theorem, a particular
case of which is the well-l<nown Nyquist theorem for the thermal noise in a
 1276 CANADIAN JOURNAL O F PHYSICS. VOL. 34. 195G

resistive circuit. Derivation of [3] by our argument is equivalent to t h a t of

Callen and Welton (4). We can further show that the fluctuation-dissipation
theorem applies also to Im u & ( w ) (antisymmetric part), but the other parts
Can. J. Phys. Downloaded from www.nrcresearchpress.com by UNIVERSITY OF ARIZONA LIBRARY on 12/14/12

of the tensor, Im a,S,(w) and Re a,",( w ) , are more complicated. They call also
be written in terms of correlation functions, but it is not so simple.
(4) This is related to the following fact. In the classical case, we call show
that the correlatioil fuilctions defined above are related to the relaxation
functions in a simple manner. The relaxation functioil @,,(t) meails the decay
of the current j,(t) when an electric field of unit strength is applied in the
v-direction from t = - up to t = 0. The classical relation is simply
@,Y(t) = *,"(t)/kT.
In quantum mechanical systems we have instead

This complication for the quantum mechanical system should be noticed.

( 5 ) T h e Onsager relations can easily be obtained from [I]:
For personal use only.

Re u,,(w, (HI) = Re a,,(-w, (HI),

Im u,,(w, (HI) = -1m u,,(-w, IHI),
where H is the magnetic field.
(6) For the static case ( w = 0 ) , Eq. [3] goes illto

a;. =
2kT
J-,
*;.(t) cos wt dt.

I t may be worth while to notice that this is a generalized form of Einstein's

relation connecting the mobility to the diffusion constant. For some simple
cases, one can easily obtain Einstein's relation from this formula. This con-
sideration is often very useful.
(7) We must admit that Eq. [I] or [2] does not solve the problem of calcu-
lating the conductivity for a given system, unless one has the prescription how
to handle it. T h e method of calculation, moreover, depends on the situation.
First of all, we can show that the usual Bloch equation is related to this general
formula. When one applies the damping theory technique with a certain
perturbation term in H one can easily go back to the Bloch equation a s an
approximation, which may of course be sufficient in the circumstances. For
metallic conduction, the well-known Griineisen formula is directly derived
from Eq. [I] if one assumes that the integrand decays in time a s a simple
exponential. I t is sometimes essential to go beyond the Bloch equation, and
here the general equatioil can give us the starting point; the approximation
needed depends on the physical situation. We tried t o apply the method to
 KUBO: CONDUCTIVITY TENSOR 1277

conductioil in a strong magnetic field where the quantization effect becomes

serious. This study will be reported elsewhere.
(8) Similar expressioils call be derived for other lriilds of transport quantities
Can. J. Phys. Downloaded from www.nrcresearchpress.com by UNIVERSITY OF ARIZONA LIBRARY on 12/14/12

such as heat conductivity and thermoelectric power. There is a certain diffi-

culty which I have not yet completely cleared up as to the quantum mechanical
representation of the presence of gradients of temperature or chemicaI po-
tential. But I believe that our final answers are right; these give expressions
involving the energy currents instead of mass currents.
(9) I do not know whether it is possible for there to exist a variational
theorem for the expression [I]. We already lalow the existence of variational
theorems for the partition function and also for the Bloch equation, but if a
variational principle could be found giving equation [I] it would be more pro-
found and more widely applicable than the known ones.
REFERENCES
1. NAKANO.
~
H. Prom.
- 7 -a~

2. KWBO,R. and TOMITA,

- ~.,.
Theoret. Phvs. (Taoan1.,, 15., 77 (19561.
I(.-J. ~h;s. Soc. Japan, 9, 888 (1954).
3. NAKAJIMA. Proc. Phys. Soc. A, August 1956. KUBO.Busseiron Kenkyu, 89, 79 (1955).
KUBOand YOKOTA. Busseiron Kenkyu, 89, 99 (1955). KUBO,HASHIZUME,
and YOKOTA.
Busseiron Icenkyu, 91 (1955). (These are preliminary reports.)
4. CALLEN,H. B. and WELTON,T. A. Phys. Rev. 83, 34 (1951).
For personal use only.