Thermal currents in highly correlated systems
Extracted source text ยท arXiv cond-mat/9603079
\tolerance = 10000
% This is for two column
\documentstyle[floats,multicol,aps,epsf,prb]{revtex}
% This is for one column
%\documentstyle[preprint,floats,aps,epsf,prb]{revtex}
\renewcommand\floatpagefraction{.7}
\renewcommand\dblfloatpagefraction{.7}
\begin{document}
%%%% the next line was added by xxx admin to ensure that
%%%% this paper doesn't use hypertex because of problems with references
\ifx\href\undefined\else\errmessage{hyperTeX disabled by xxx admin}\fi
\draft
% This is for two column
\twocolumn[\hsize\textwidth\columnwidth\hsize\csname @twocolumnfalse\endcsname
%\twocolumn
\title{Thermal currents in highly correlated systems}
\author{J. Moreno and P. Coleman}
\address{Department of Physics and Astronomy, Rutgers University,
Piscataway, NJ 08855}
\date{\today}
\maketitle
\begin{abstract}
Conventional approaches to thermal conductivity in itinerant systems
neglect the contribution to thermal current due to
interactions. We derive this contribution
to the thermal current and show how it produces
important corrections to the thermal conductivity in
anisotropic superconductors. We discuss the possible
relevance of these corrections
for the interpretation of the thermal conductivity of
anisotropic superconductors.
\end{abstract}
%\narrowtext
%\newpage
%\vskip 0.2 truein
%\pacs{Pacs numbers: 74.70.Tx, 74.25.Ld, 74.72.-h}
\vskip2pc]
\def\vx{\vec x}
\def\vj{\vec j}
\def\vr{\vec r}
\def\vk{\vec k}
\def\vq{\vec q}
\def\vp{\vec p}
\input psfig.tex
%\vskip 0.15 truein
Historically, understanding thermal conductivity
has played an important role in the development of
materials theory.
There has recently been a growth
of interest in studying this basic property
in strongly correlated materials.\cite{Taillefer,recent}
One of the recurrent consequences of strong interactions
is the development of anisotropic gaps in the
excitation spectrum: possible examples include
heavy fermion superconductors, Kondo insulators and the chevrel
superconductor $V_3Si$.
\cite{hfermion,v3si} In such
systems, the thermal conductivity has an important role
to play in the elucidation of the gap symmetry.
Pioneering work on the theory of thermal conductivity was carried out
in the sixties\cite{Chester,Langer,super}; by and large, the theoretical
approaches taken today
are a direct application of this early work \cite{phonon,Tmatrix,Boltzman}.
In this paper, we show how these
classic approaches require modification
to take account of new thermal conduction channels
that are introduced by interactions. As part of this
discussion, we shall show how thermal conductivity, like
particle or charge conductivity, can be regarded as a boundary condition
response which is most directly computed as a response to a
fictitious gauge field. We illustrate this approach in the context of
anisotropic superconductivity, showing how
interaction contributions to the thermal current
significantly modify the flow of heat produced by excitations
near a gap node.
Thermal conductivity
is directly related, via the Kubo formula, to
thermal current fluctuations
\begin{equation}
\kappa^{ab} = - \beta \hspace{0.05in} lim_{\omega \rightarrow 0}
\frac{\partial} {\partial \omega}
Re \{ \int _0^{\infty} dt e^{i \omega t}
\langle [ j^{a}(t), j^{b}(0)]\rangle \}.
\end{equation}
There is an important distinction between heat and charge
conductivity. Unlike charge density, the energy density does not
commute with the interactions: this means that heat is directly
transmitted by interactions, introducing new interaction contributions
to the microscopic thermal current operator. The effect of these new
thermal conduction channels is important when interactions severely
modify the dispersion of the electrons, as in the case of a
Mott insulator, or a superconductor with gap nodes.
\cite{mahan,polaron,castellani}
To derive the thermal current
we appeal to Noether's theorem, which relates
continuity of energy flow to the covariance
of the action under co-ordinate transformations in time
$
%\begin{equation}
t\rightarrow t' = t + \phi[\vec x, t'].
$
Consider an
electronic system described by the Lagrangian
\begin{equation}
{\cal L} = \int d^3x(i\psi^{\dagger}
\!\stackrel{\leftrightarrow}{\partial_t}\!
\psi) -H,
\end{equation}
where $H$ is the Hamiltonian
and
$\! \stackrel{\leftrightarrow}
{\partial_t} = {1 \over 2}
(\stackrel{\rightarrow}{\partial_t} -
\stackrel{\leftarrow}{\partial_t})\!$ is the antisymmetrized
time-derivative.
By Noether's theorem, the change in the action
$S= \int {\cal L}(t) dt$ is the energy continuity equation
\begin{equation}
{\delta S/ \delta \phi[\vec x, t] } =
-\bigl[\partial _t \epsilon({\vec x},t) + {\vec \nabla}\cdot
{\vec j}({\vec x},t)
\bigr],
\end{equation}
where $\epsilon$ and $\vec j$ are the energy
and thermal current density respectively.
Continuity of energy flow follows from the invariance
of the action under co-ordinate transformations
${\delta S/ \delta \phi[\vec x, t] } = 0$.
Writing $\delta S
= \int [\delta S/ \delta \phi] \delta \phi(x,t)$ and integrating
by parts, we find
\begin{equation}
\delta S = \int dt d\vec x \biggl[
\dot \phi(\vec x,t) \epsilon(\vec x, t)
+
\vec \nabla \phi(\vec x,t)\cdot \vec j(\vec x, t)
\biggr]
\end{equation}
so that
$\epsilon(x) = {\delta S / \delta {\dot \phi}}$, while the
heat current is ${\vec j} = {\delta S / \delta {\vec \nabla} \phi}
$.
The
calculation of the thermal current requires special care
because interactions are non-local
and all higher derivatives of the
fields must be taken into account when taking the
functional derivative.
This important point was overlooked in the classic treatment by Langer
\cite{Langer}.
We can take these effects into account by
noting that the derivative
operators inside the action are evaluated at constant time $t$,
so that under the transformation $t\rightarrow t'$, they
acquire a covariant form
\begin{equation}
- i \vec \nabla \rightarrow
- i {\vec \nabla} - \vec A \hat w,
\end{equation}
where we denote $\hat \omega= i\!\stackrel{\leftrightarrow}{\partial_{t'}}$.
The field
$\vec A = \vec \nabla \phi$
adjusts for the fact that spatial derivative $\nabla \equiv\nabla\vert_{t'}$
is
taken at constant $t'$. It follows that
the energy current is
\begin{equation}
\vec j = {\delta S
/\delta \vec \nabla \phi } = -{\partial H[ \vec A] / \partial \vec A},
\end{equation}
where $H[\vec A]$ is the ``gauged'' Hamiltonian with the
replacement $\vec k \rightarrow \vec k - \vec A \hat \omega$
in both the kinetic {\em and} interaction terms.
We see that $\vec A$ plays the role of
a ``fictitious'' gauge
field conjugate to the thermal current.
To illustrate this procedure, consider
a fluid of electrons with kinetic energy
$\epsilon_{\vec k}$ and an exchange interaction $J_{\vec k}$, where
\def\dg{^\dagger}
\begin{equation}
H[\vec A] =
\sum_{\vec k}\bigl[
\psi\dg_{\vec k}
\epsilon_{(\vec k - \vec A \hat \omega)}\psi_{\vec k}
+\frac {1}{2}
\vec \sigma_{-\vec k}
\cdot
J_{(\vec k - \vec A \hat \omega)}
\vec \sigma_{\vec k}\bigr],
\end{equation}
where $\vec \sigma_{\vec k }$ is the Fourier transform of the spin density
at wavevector
$\vec k$ .
Differentiating with respect to $\vec A$ gives
\begin{equation}
\vec j =i \sum_{\vec k}
\biggl[
\bigl(\vec\nabla_{\vec k}\epsilon_{\vec k})
\psi\dg_{\vec k}
\!\stackrel{\leftrightarrow}{\partial_t}\!
\psi_{\vec k}
+
\frac{1}{2}\bigl(\vec \nabla_{\vec k} J_{\vec k}\bigr) \vec
\sigma _{-\vec k}\cdot
\!\stackrel{\leftrightarrow}{\partial_t}\!
\vec \sigma_{\vec k}
\biggr].\end{equation}
The second term reflects the additional
heat flow created by the exchange interaction. In an antiferromagnetic
Mott insulator, it is this term which is responsible for the
conduction of heat by spin-waves. The approach illustrated
here can be used on any Hamiltonian,
without the need to develop the equations of motion for that
particular model.
Past efforts to compute the thermal current contribution from
interactions have adopted an equation of motion approach on a case-by-case
basis.
This yields long expressions for the thermal current, where the time derivatives
are expanded in their full glory.\cite{mahan,polaron}
The usefulness of the gauge-theoretic derivation of the thermal current
lies partly in its invariance
under the renormalization group. This means that it can
be directly applied to the effective Lagrangian that describes
the low energy physics of an interacting system. In any case where the
low energy physics is described by weakly interacting quasiparticles
with energy $E_{\vec k}$, application of the same procedure yields
\begin{equation}
\vec j = i\sum_{\vec k} \bigl(\nabla_{\vec k}E_{\vec k}\bigr)a\dg_{\vec k}
\stackrel{\leftrightarrow}{\partial_t}
a_{\vec k} = \sum_{\vec k}\bigl( E_{\vec k}\nabla_{\vec k}E_{\vec k}\bigr) n_{\vec k},
\end{equation}
where $n_{\vec k}= a\dg_{\vec k}a_{\vec k}$ is the quasiparticle number
operator and we have used the equation of motion to make the last substitution.
When the mass renormalization of the electrons is highly anisotropic,
the quasiparticle thermal current contains a large interaction component.
Microscopic calculations that ignore these
contributions fail to recover the correct quasiparticle description
of the heat current at low temperatures.
A particularly important illustration
of this effect occurs in an
anisotropic superconductor. In this case, the BCS Hamiltonian
takes the form
\begin{eqnarray}
H[\vec A ] &=& {1/2}
\sum_{\vec k}\Psi\dg_{\vec k}{\cal E}_{{\vec k} - \vec A \hat \omega }
\Psi_{\vec k}, \nonumber \\
{\cal E}_{\vec k} &=&
\left(
\begin{array}{cc}
\epsilon_{\vk}\hat{1} & \hat{\Delta}_{\vk} \\
\hat{\Delta}^{\dagger}_{\vk} & -\epsilon_{\vk}\hat{1},
\end{array} \right)
\end{eqnarray}
where $\hat{\Delta}_{\vec k}$
is the anisotropic gap function,
$\Psi\dg_{\vec k}$
%= ( \psi\dg_{\vec k} , \psi_{-\vec k})$
is the four-component Nambu spinor for the electrons.
The thermal current operator is then
\begin{equation}
\vec j =
{ 1\over 2}\sum_{\vec k}\Psi\dg_{\vec k}
\nabla_{\vec k}{\cal E}_{\vec k}\hat \omega
\Psi_{\vec k}.
\end{equation}
The off-diagonal terms in this expression derive
from the interaction contribution to the thermal current.
In the presence of weak scattering, we can use the equation of motion
to replace
\begin{equation}
\nabla_{\vec k}{\cal E}_{\vec k}\hat \omega \rightarrow
\frac{1}{2}\{
{\cal E}_{\vec k},
\nabla_{\vec k}{\cal E}_{\vec k}\} = E_{\vec k}
\nabla_{\vec k} E_{\vec k}\underline{1},
\end{equation}
where $E_{\vec k}
= (\epsilon_{\vec k}^2 + \Delta_{\vec k} ^2)^{1/2}$. This
recovers the quasiparticle form (9). Note however, that the
Fermi velocity is replaced by the
the quasiparticle group velocity
$\vec {\cal V}_{\vec k}=\nabla_{\vec k} E_{\vec k}$, which
has important components {\em parallel} to the Fermi surface.
These can significantly
affect the anisotropy of the thermal conductivity.
\begin{figure}[btp]
\centerline{\psfig{file=fig1.ps,width=3.0in}}
\vskip -0.4truein
\caption{Constant energy contours surrounding a line
node in a ``polar'' superconductor. The thermal current operator is normal to the energy contours and points radially outwards from the gap node.
The dense line indicates position of
Fermi surface and the dark region represents the
line node, which lies perpendicular to the plane of the paper.
}
\label{fig1}
\end{figure}
We now use (11) and apply the Kubo formula, to obtain
\begin{eqnarray}
\kappa^{ij}= \frac{\pi}{2 T} \sum_{\vk}
\int^{\infty}_{-\infty} {d\omega} \omega^2
\left( -\frac{\partial f}{\partial \omega}\right)
{\Lambda^{ij}(\vec k, \omega)},
\label{thcond}
\end{eqnarray}
where \begin{equation}
\Lambda^{ij}(\vec k, \omega) =
Re \left\{ Tr\! \left[
({\nabla_{k}}\hat{{\cal E}_{\vec k}})^i
{\rm A}_{\vec k}(\omega)
( {\nabla_{k}}\hat{{\cal E}_{\vec k}})^j {\rm A}_{\vec k}(\omega)
\right]\right\}\ ,
\end{equation}
${\rm A}_{\vec k}(\omega)$ is the matrix spectral function
and $f$ is the Fermi function.
Carrying out the trace we find
\begin{equation}
\kappa^{ij} = { 1 \over 2T} \int _{-\infty}^{\infty}
d \omega
\omega^2\biggl(
-{d f \over d\omega}
\biggr){ N(\omega) \over \Gamma(\omega)}
\langle \vec {\cal V}^i\vec {\cal V}^j \rangle_{\omega}
,
\end{equation}
where $\Gamma(\omega)$ is the quasiparticle scattering rate
and
\begin{equation}
N(\omega)\langle \vec {\cal V}^i\vec {\cal V}^j\rangle_{\omega}
=\sum_{\vec k}
\vec {\cal V}_{\vec k}^i
\vec {\cal V}_{\vec k}^j \delta(\omega- \vec E_{\vec k})
\end{equation}
describes the quasiparticle velocity distribution.
To show how the interaction contribution to the thermal current
modifies the thermal conductivity,
consider the case of a ``polar'' superconductor, with a spherical
Fermi surface and gap function
$|\Delta_{\vk}|^{2}=\Delta_{0}^{2} \hat{k}_{z}^{2}$, which leads to
a line of gap nodes in the basal plane $k_z=0$.
The kinetic contribution to the thermal current operator always
lies normal to the electron Fermi surface, and if we omit the
effects of interaction, the low temperature thermal conductivity
lies predominantly in the basal plane. In the correct calculation,
the current carried by each excitation lies in the direction of the
quasiparticle velocity $\vec v^{QP}_{\vec k}$, which
lies normal to the surfaces of constant energy (Fig. 1).
The line-node in the basal plane is surrounded by elliptical surfaces
of constant energy and consequently, in the vicinity of the gap node
there are quasiparticles which propagate perpendicular to the basal
plane, giving rise to a increased thermal conductivity in this direction.
At low temperatures the
thermal conductivity in both directions
is proportional to $T^2$, with a finite ratio at $T=0$
\begin{equation}
{\kappa^{zz} \over \kappa^{xx}} = \frac {1}{2}\left( {\Delta_o
\over \epsilon_F}\right)^2.
\end{equation}
Without interaction effects, this ratio vanishes at $T=0$.
\begin{figure}[btp]
\psfig{file=fig2.ps,width=3.375in}
\caption{Ratio between the thermal conductivity perpendicular and
parallel to the basal plane as a function of the normalized
temperature for different values of the ratio
$r=\Delta_{0}/\epsilon_{F}$.}
\label{fig2}
\end{figure}
In Figure 2 we show how the anisotropy in the
thermal conductivity varies as the ratio between
the maximum gap and the Fermi energy
is increased from zero. Two effects of interest are observed:
(i) the anisotropy is reduced by these interaction effects;
(ii) the thermal conductivity is enhanced in the region close to the
transition temperature. These effects become important
when the transition temperature
is a substantial fraction of the Fermi energy.
Such considerations may be particularly relevant to heavy fermion
superconductors, of which two cases are particularly worthy of mention:
$UBe_{13}$ and $UPt_{3}$. In both cases,
NMR measurements suggest the presence of line nodes.
In the former case,
superconductivity develops before a Fermi liquid regime
has been established, so it is very likely that $\Delta/\epsilon_F$ is
large. In the latter case, the conventional wisdom is that
$\Delta/\epsilon_F\sim 0.05-0.1$. However, measurements of the thermal conductivity
on very pure samples of this material\cite{Taillefer} show that
$\kappa^{zz}/\kappa^{xx}\sim {\ \rm constant}$ at low temperatures;
they also display a marked peak in the ratio $\kappa^{sc}/\kappa^n$ near
$T_c$. Existing approaches \cite{recent} attribute these effects to a
combination of
both elastic and inelastic scattering. While such an explanation is
surely feasible, an alternative explanation might be obtained
by assuming a larger
ratio $\Delta/\epsilon_F$, and attributing these features to
the interaction contribution to the thermal current that has been
hitherto ignored.
The main point of this paper has been to emphasize that interactions
actually modify the thermal current, which can have important effects
on the thermal conductivity. As an illustration of these ideas,
we have considered the effect
of gap anisotropy in polar superconductors. Our basic approach
can be applied to
other gap anisotropic systems. One very interesting case
in this respect may be the small gap Kondo insulators,
where the insulating gap appears to exhibit a point node\cite{Miyake}.
Similar considerations can also be applied
to the momentum current, a point which is important in the interpretation
of measurements of the
ultrasound attenuation in strongly correlated systems.
This could explain why early ultrasound measurements in $UBe_{13}$
\cite{Batlogg}
found no anisotropy, despite the clear suggestion of line nodes from
NMR measurements.\cite{maclaughlin}
A secondary aspect of our work concerns the
observation that thermal currents can be treated using a gauge theoretic
approach. Just as momentum currents, or stress ($\sigma$)
in a material result
from the gauge field we call strain ($u$), $\sigma = R u$,
energy currents result from
a ``temporal strain'' which we represent by the gauge field $\vec A$.
If we calculate the heat current response to this fictitious field
\begin{equation}
\vec j_T(\omega) = - Q(\omega) \vec A(\omega),
\end{equation}
we find that the ``London Kernel'' is directly
related to the thermal conductivity by the
relation $Q(\omega) = i \omega T\kappa (\omega)$, from which we deduce that
\begin{equation}
{\partial \vec A \over \partial t} = -{ \vec \nabla T \over T}.
\end{equation}
In other words, $\vec A$ is the gauge field conjugate to thermal
gradients.
By analogy with the case of momentum currents,
the quantity $\vec A$ may be regarded as a sheer
in time. Just as broken spatial translation symmetry enables
a material to support a persistent stress
or ``momentum superflow'', we are led to speculate that if it were
possible to produce a state of condensed matter with spontaneously
broken time translation symmetry, such a system would
exhibit ``heat superflow'', manifested by an infinite thermal
conductivity.
We are grateful to Micha Berkooz, Denis Golosov, Andres Jerez,
Gabriel Kotliar and Andrew Schofield for useful
discussions. This research was supported by NSF grant DMR-93-12138.
\begin{references}
\bibitem{Taillefer} B. Lussier, Brett Ellman and Louis Taillefer,
Phys. Rev. Lett. {\bf73}, 3294 (1994), B. Lussier et al.
(electronic archive xxx.lanl.gov/cond-mat/9504072)
\bibitem{recent} A. Fledderjohann and P. J. Hirschfeld, Solid
St. Comm. {\bf94}, 163 (1995); M.R. Norman and P.J. Hirschfeld,
(electronic archive xxx.lanl.gov/cond-mat/9509045);
M.J. Graf, S.-K. Yip and J.A. Sauls ,
(electronic archive xxx.lanl.gov/cond-mat/9512054)
\bibitem{hfermion} For a review of heavy fermion materials,
see e.g. N. Grewe and F. Steglich, Handbook on the Physics and
Chemistry of Rare Earths, vol. 14, (1991), R. H. Heffner and
M. R. Norman, Comments on Condensed Matter Physics (1995)
\bibitem{v3si} For an overview of strong coupling
superconductors, see e.g. T. Ohno et al., Physica B {\bf186-188}, 1034 (1993)
%\bibitem{cuprates} For a review of cuprate superconductors, see e.g.
%E. Dagotto, Rev. Mod. Phys. {\bf66}, 763 (1994), W.E. Pickett, Rev. Mod.
%Phys. {\bf61},433 (1989)
\bibitem{Chester} G. V. Chester, Reports on Progress in Physics
vol. XXVI, 411 (1963)
\bibitem{Langer} J. S. Langer, Physical Rev. {\bf128}, 110 (1962)
\bibitem{super} L. Kadanoff and P. C. Martin, Physical Rev.
{\bf124}, 670 (1961); V. Ambegaokar and L. Tewordt, Physical Rev.
{\bf134}, A805 (1964); V. Ambegaokar and A. Griffin, Physical Rev.
{\bf137}, A1151 (1965)
\bibitem{phonon} Y. Ono, Progress of Th. Phys. {\bf46}, 757 (1971),
S. K. Lyo, Phys. Rev. B {\bf17}, 2545 (1978), A. Vilenkin et al.,
Phys. Rev. B {\bf18} 5280 (1978), D. V. Livanov et al., Sov. Phys.
JETP {\bf72} 760, (1991)
\bibitem{Tmatrix} H. Monien et al., Solid St. Comm. {\bf61},
581 (1987), S. Schmitt-Rink et al., Phys. Rev. Lett. {\bf57}, 2575
(1986); P.J. Hirschfeld et al., Phy. Rev. B {\bf37} 83 (1988)
\bibitem{Boltzman} B. Arfi et al., Phys. Rev. B {\bf39}, 8959
(1989)
\bibitem{mahan}A brief discussion of electron-electron interaction effects on
thermal current, cast the language of equation of motion methods,
is to be found in
G. Mahan, ``Many Particle Physics'', pp 665, Plenum (1981).
\bibitem{polaron}Though we are not aware of any literature
on electron-electron interaction effects on thermal currents,
a small literature on interaction effects on thermal
current in the small polaron problem does exist, see, e.g
P.E. Nielson and P.L. Taylor, Phys. Rev. Letters, {\bf 21}, 893 (1968).
\bibitem{castellani}In the special case of isotropic systems,
it is possible to correctly calculate $\kappa^{ab}= \kappa \delta^{ab}$
from the momentum dependent part of the energy fluctuations,
without direct knowledge of the effect of interactions on $j^a$.
This method was applied to interacting disordered systems by
G. Strinati et al., Phys. Rev. B {\bf44}, 6078 (1991).
\bibitem{Miyake} H. Ikeda and K. Miyake, to be published in J. Phys.
Soc. Jpn.
\bibitem{Batlogg} B. Batlogg et al., Phys. Rev. Lett. {\bf55}, 1319 (1985);
B. Goding et al., Phys. Rev. Lett. {\bf55}, 2479 (1985)
\bibitem{maclaughlin} D.E. MacLaughlin et al., Phys. Rev. Lett. {\bf53},
1833 (1984)
\end{references}
\end{document}