Power-law spin correlations in pyrochlore antiferromagnets
Extracted source text · arXiv cond-mat/0407005
%%% Edited from "pyrice-orig.tex" edited July 2003 from
%%% \OMIT{Note that $a\approx 1.02$ nm in the pyrochlore lattice.)}
%%% the version I originally wrote in April 10-17, 1995,
\documentclass[pre,aps,twocolumn,showpacs,floatfix,eqsecnum]{revtex4}
\usepackage{graphicx}
\begin{document}
%%% \def\OMIT#1 {{\sl #1}}
\def\OMIT#1 {}
\def\LATER#1 {}
\newcommand{\eqr}[1]{(\ref{#1})}
\newcommand{\half}{{\textstyle{1\over2}}}
\newcommand{\fourth}{{\textstyle{1\over4}}}
\def \kag {kagom\'e }
\def \pb {{\overline {\bf p}}}
\def \n {{\hat {\bf n}}}
\def \p {{\bf p}}
\def \hh {{\bf h}}
\def \qq {{\bf q}}
\def \ss {{\bf s}}
\def \G {{\bf G}}
\def \uu {{\bf u}}
\def \pp {{\bf p}}
\def \Pol {{\bf P}}
\def \QQ {{\bf Q}}
\def \RR {{\bf R}}
\def \rr {{\bf r}}
\def\JJ{{\bf J}}
%% \def \k {{\bf k}}
\def \kk {{\bf k}}
\def \KK {{\bf K}}
\def\LL{{\bf L}}
%%% \def\Fcal{{\cal F}}
\def \HH {{\cal H}}
\def \F {{\bf F}}
\def\Hfield{{\bf H}}
\def\Jspin{{J_{\rm spin}}}
\newcommand{\la}{\langle}
\newcommand{\ra}{\rangle}
\def\Struc#1{{\la |\tt(#1) |^2\ra}}
%%%%%%%
%%%%%%%% NEW
\def \tt {{\tilde t}}
\def \tPol {{\tilde {\Pol}}}
\def\Tcmf{{T_c^{\rm MF}}}
\def\betaeff{{\beta_{\rm eff}}}
\title{POWER-LAW SPIN CORRELATIONS
IN PYROCHLORE ANTIFERROMAGNETS}
\author{C.~L.~Henley}
\affiliation{Department of Physics,
Cornell University, Ithaca, New York 14853-2501}
\begin{abstract}
The ground state ensemble of
the highly frustrated pyrochlore-lattice antiferromagnet
can be mapped to a coarse-grained ``polarization''
%%% configurations of arrows on the edges of a diamond lattice,
field satisfying a zero-divergence condition
%% Furthermore, the (entropic) free energy is maximized at $\Pol(\rr)=0$;
From this it follows that the correlations of this field, as
well as the actual spin correlations, decay with separation
like a dipole-dipole interaction ($1/|R|^3$).
Furthermore, a lattice version of the derivation gives an
approximate formula for spin correlations, with several
features that agree well with
simulations and neutron-diffraction measurements of diffuse scattering,
in particular the pinch-point (pseudo-dipolar) singularities
at reciprocal lattice vectors.
This system is compared to others in which constraints also imply
diffraction singularities, and other possible applications of the
coarse-grained polarization are discussed.
\end{abstract}
\maketitle
\section{introduction}
\label{sec:intro}
Highly frustrated antiferromagnets are characterized by
a very large number of essentially degenerate states,
such that in a range of temperatures much smaller than the
spin interaction scale
they have strong local correlations, yet fail to order~\cite{ramirez}:
diffuse scattering is the obvious probe of such a state.
In this paper, I make the point that the classical ground-state
ensemble can entail constraints that ensure a generic
power-law decay of correlations: a picture of theses state
as ``liquid-like'' (i.e. featureless) is thus imcomplete.
%% in spinel, pyrochlore, or Laves crystal structuress
%% (the last category are alloys with semi-itinerant magnetism).
We consider specifically the pyrochlore lattice,
consisting of corner-sharing tetrahedra, partly
because of its simplicity as a model
[high (cubic) symmetry and nearest-neighbor interactions],
but most importantly because it is the magnetic
lattice in many real systems:
the B-site spinels~\cite{lee00,lee02},
two families of pyrochlores represented by
%%% defect pyrochlores $\rm CsnFeF_6$~\cite{CsMnFeF6}
$\rm CsNiCrF_6$~\cite{harris94,harris97a},
and the oxides $\rm Y_2Mn_2O_7$~\cite{Y2Mo2O7}, and
even the transition-metal lattice in ``frustrated
itinerant'' antiferro- or ferrimagnetic Laves phases, such as
$\rm Y_{1-x}Sc_xMn_2$~\cite{shi93,ballou}.
Ferroelectric degrees of freedom (water ice~\cite{bernal-pauling,youngblood})
and charge order (in magnetite~\cite{an56}) are also equivalent
to pyrochlore systems.
Different pyrochlores show a wide variety of magnetic behaviors
at low temperatures:
apart from long-range order, these include
lattice distortions~\cite{lee00,tcherny},
``spin-ice'' behavior~\cite{harris97b,ramirez-spinice,Si99,Bra01a,Bra01b},
paramagnetism (in $\rm Tb_2Ti_2O_7$)
down to low temperatures,~\cite{Tb2Ti2O7}
or spin glass freezing (in $\rm Y_2Mo_2O_7$
even when structural disorder is quite small~\cite{Y2Mo2O7}.
However, most of these have a cooperative paramagnet regime
at higher temperatures -- but still quite low compared to
the interaction scale $J$ -- demonstrating that they have
a large set of nearly degenerate (and accessible) states.
The present paper addresses this phase, as it would be
extrapolated to zero temperature.
In the remainder of this section, I review the lattice and
the ground states of a pyrochlore antiferromagnet,
and survey three situations in which it can
be modeled by the ensembles used in this paper.
Then (Sec.~\ref{sec:coarse}
the mapping of the low-temperature states
of the Ising model to the ``diamond ice'' model
is to construct a coarse-grained polarization field,
which functions somewhat like an order parameter
for this model, and this is used to
predict some unusual features of the magnetic diffuse scattering.
Sec.~\ref{sec:lattice} presents another version of the
derivation, which not only gets the long-wavelength behavior
(corresponding to the neighborhoods of special points in the
Brillouin zone) but provides an approximation for the entire zone.
I also discuss (Sec.~\ref{sec:disc}) other situations in
which local constraints in a ground state produce diffraction
singularities, and speculate on other questions that can
be addressed using the insight that the polarization field
contains the important long-wavelength degrees of freedom.
A parallel paper~\cite{moessner-dipolar} contains many of the same results.
\subsection{Pyrochlore lattice and Hamiltonian}
The pyrochlore lattice consists of corner-sharing tetrahedra
arranged in cubic (fcc) symmetry.
Its key property is that
the sites are the bond midpoints of a diamond lattice.
[See Fig.~\ref{fig:lattice};
other figures emphasizing the tetrahedra are in
Refs.~\onlinecite{tcherny} or~\onlinecite{moessner98a,moessner98b}.]
For future use, we define vectors pointing towards the four corners
of a tetrahedron,
\LATER{Use ``cases''? or eqalign?}
\begin{eqnarray}
\ \uu_{1,2,3} \equiv [\fourth,-\fourth,-\fourth] \quad
\nonumber \\
\uu_4 \equiv [\fourth, \fourth, \fourth] \qquad
+ \text{cyclic perms}.
\label{eq:uu}
\end{eqnarray}
In the diamond lattice, every even-to-odd bond vector is a $\uu_m$.
The basic results derived in this paper are transferable to other
lattices in which the spins sit on the bonds of a bipartite
lattice, and with antiferromagnetic interactions among
all spins on bonds sharing a
common endpoint,such that their sum is constrained
to a fixed value.
Besides the pyrochlore lattice,
this class includes the kagom\'e lattice,
the two-dimensional
``checkerboard'' lattice,~\cite{moessner98a,moessner98b},
the ``sandwich'' lattice modeling SCGO and consisting
of two kagom\'e layers linked by an additional
triangular layer of
spins,~\cite{broholm-SCGO,henley-HFM00,kawamura}
and the garnet lattice~\cite{petrenko}.
The spin Hamiltonian,
having only nearest-neighbor antiferromagnetic exchange,
can be cast as a sum of squares:
\begin{equation}
\HH_{\rm spin}= \Jspin
\sum _{\la ij \ra} \ss_i \cdot \ss_j +{\rm const}
= \frac{\Jspin}{2} \sum _\alpha \LL_\alpha^2,
\label{eq:Hspin}
\end{equation}
where the total spin of four sites in a tetrahedron is
\begin{equation}
\LL_\alpha \equiv \sum _{i\in \alpha} \ss_i
\label{eq:L}
\end{equation}
%% (\RR_m+\half \uu_m).
where $\alpha$ labels each diamond site
and ``$i\in\alpha$'' runs over the surrounding four spins.
For simplicity, I will mostly treat the case with Ising spins $t_i$.
(There is little loss of generality, in view of Sec.~\ref{sec:Escales}, below.)
\begin{equation}
\HH = \half J \sum _\alpha L_\alpha^2
= J \sum _{\la ij \ra} t_i t_j +{\rm const} .
\label{eq:Ham}
\end{equation}
From \eqr{eq:Ham} it is evident that any state with
\begin{equation}
L_\alpha = 0.
\label{eq:ground}
\end{equation}
is a classical ground state; these are massively degenerate.
The frustration of the pyrochlore Ising model,
and the extensive ground state entropy implied by \eqr{eq:ground},
were recognized very early~\cite{an56}.
\subsection{Temperature regimes and validity of Ising model}
\label{sec:Escales}
Here I discuss the three different situations in which realistic
models can be treated under this paper's scheme.
[Readers who are content with a treatment of an Ising toy model,
and do not immediately demand any connection to experiment,
may skip to Sec.~\ref{sec:coarse} where the central result
is derived.]
First, it is possible that spin anisotropies
(additional to \eqr{eq:Hspin}) reduce the ground state
manifold to the Ising states.
If the lattice is to remain cubic, the preferred axis cannot
be the same for each spin, but instead is $\hat \uu_{m(i)}$,
where $\hat \uu_m = 4 \uu_m /\sqrt{3}$ is the unit vector
of \eqr{eq:uu},
and $m(i)$ labels the direction of the diamond-lattice
bond on which spin $i$ sits.
Assuming the spins are strongly aligned, one puts
$\ss_i = t_i \hat \uu_{m(i)}$ and the Hamiltonian
reduces to \eqr{eq:Ham} with $J=- \Jspin/3$,
since $\hat \uu_m \cdot \hat \uu_{m'} = -1/3$ for
$m\neq m'$, and $m(i)\neq m(j)$ for nearest neighbors.
This is the ``spin-ice'' model~\cite{harris97b,ramirez-spinice}.
The other two situations correspond to isotropic Heisenberg
models in the large-$S$ semiclassical regime (where $S$ is the
spin length). This is necessary to guarantee that \eqr{eq:ground} is
a good starting point for specifying the ground states.
One can identify a succession of energy scales in
the pure nearest-neighbor Heisenberg antiferromagnet,
corresonding to successive breakings of the grond state
degeneracy as more effects are considered.
The largest scale is $E_J \sim zJS^2$ where $z$ is the
coordination number; this is the scale of the mean-field
(classcial) part of the exchange energy, and of the Curie-Weiss
constant. At temperatures $T \ll E_J$, it is reasonable
to approximate the spin ensemble by a subset of the
ground-state ensemble.
The spin-wave spectrum $\{\omega_j \}$ depends
upon which state we linearized around.
Thus, the total spin-wave zero-point energy (to harmonic order,
$\sum_j \half \hbar \omega_j$)
partially breaks the ground-state degeneracy, favoring {\it collinear} states,
~\cite{shen82,hen89}. This is expressed quantitatively by an
effective Hamiltonian~\cite{oja,larhen90,henley-HFM00},
with coupling constant $\sim E_{\rm coll}$, where
the next largest scale is $E_{\rm coll} \sim JS \ll E_J$.
(Even when $S$ is not so large, e.g. $S=5/2$ as with real spins,
the coefficient of $JS$ is commonly less than $1/10$
so the inequality is valid.)
In lattices other than pyrochlore, one finds a similar regime in
which a discrete subset of the classical ground states
(e.g. coplanar states,~\cite{ritchey,chubukov} in the \kag case) gets
selected by the harmonic zero-point energy.
Finer treatment of the spin-wave fluctuations produces an
even smaller energy scale $E_{\rm disc}$ for
selection among the discrete states.
In the \kag case,~\cite{chubukov,chan};
this is due to anharmonic terms
in the Holstein-Primakoff expansion,
so that $E_{\rm disc} \propto S^{2/3} \ll S$.
In the pyrochlore case, besides the anharmonic terms,
$E_{\rm disc}$ does include contributions of $O(JS)$
from the harmonic terms.~\cite{pyro-Heff}.
However, these terms only partly
break the degeneracy, and the coefficients are small, so
we can still assume $E_{\rm disc} \ll E_{\rm coll}$.
Thus, the second regime in which the Ising theory
models the correlations is
$E_{\rm disc} < T < E_{\rm coll}$.
One expects a symmetry breaking to
long-range collinear order, in which
a global axis $\n$ is spontaneously adopted
and each spin is given by $\ss_i = t_i \n$, with $t_i=\pm1$,
plus small fluctuations.
This phase is a kind of ``spin nematic''~\cite{chancole91}.
The energy terms distinguishing different Ising states
are unimportant since $T> E_{\rm disc}$, hence
the spin ensemble is roughly the ground states of \eqr{eq:Ham},
as claimed.
\OMIT{The phase of the XY (2-component) model, which
develops long-range collinear order,
would be analogous.~\cite{moessner-dipolar}}
Of course, in the real world there is another energy
scale $E_{\rm pert}$ for perturbations of the pure Heisenberg
Hamiltonian, representing the magnitude further-neighbor exchange and
dipole couplings, as well as lattice distortions, and
disorder, which cause a transition or freezing into
some other state at sufficiently low temperature, so
the regime of the spin nematic-like phase is really
${\rm max}(E_{\rm disc},E_{\rm pert}) < T < E_{\rm coll}$.
\OMIT {Such a collinear ordering was apparently not observed in
Monte Carlo simulations of the
{\it classical} Heisenberg system~\cite{reimers92a}}
Finally, the third situation is when $E_{\rm coll} \ll T \ll E_J$.
This ensemble is well modeled by the low-$T$ limit
of the classical Heisenberg model.
%%%%%%%%
\OMIT{Note that, although thermal fluctuations
tend to favor local collinearity, in the Heisenberg spin
case this does not diverge~\cite{moessner98a,moessner98b}
as $T\to 0$,and is insufficient to induce long-range order
of the collinear axis.}
%%%%%%%%
It turns out (Sec.~\ref{sec:Heis-pol}, below)
that the key notion of ``polarization'' does extend to the
Heisenberg spin case. I have mostly neglected that case in this
paper because, at my level of approximation, the results
look identical; it would merely obscure the notation by
adding spin-component indices everywhere.
\section{Coarse-graining, Fourier mode fluctuations,
and long-range correlations}
\label{sec:coarse}
In this section, I present the steps leading to power-law
correlations using the framework of a continuum theory,
in the ideas appear more transparently.
\subsection{Ice mapping and local polarization}
\label{sec:icemap}
In fact the pyrochlore (Ising) ground states
map 1-to-1 onto those of the well-known diamond-lattice
ice model~\cite{an56,liebmann86}, in which the degrees of freedom are arrows
along the lattice bonds.
In the map, every tetrahedron center becomes a vertex of
the diamond lattice, while the spin sites map to bond centers
of the diamond lattice. Each spin $t_i =+ 1(-1)$ maps to an arrow
pointing along the corresponding diamond lattice edge,
in the positive (negative) sense from the even to the odd vertex.
This well-known mapping ~\cite{an56,liebmann86,Hen92}
is also used to model the ``spin ice'' system Dy$_2$Ti$_2$O$_7$
(and also Ho$_2$Ti$_2$O$_7$), wherein
local $\la 111 \ra$ anisotropy plus
{\it ferromagnetism} makes a highly frustrated
Ising model~\cite{harris97b,ramirez-spinice,Si99,Bra01a}.)
The ground state condition -- net spin of every tetrahedron is zero
-- maps to the ``ice rule''
the numbers of incoming and outgoing arrows are equal at every vertex
~\cite{bernal-pauling}.
The key object in this paper is the ice polarization field.
On a diamond-lattice bond a polarization
$t_i \uu_{m(i)}$ can be defined, aligned from the even to odd diamond-lattice
vertex if the spin is up, oppositely if it is down;
here $m(i)$ is the local 3-fold axis of site $i$.
On every diamond vertex, we define the mean of this polarization
over the surrounding tetrahedron of spins:
\begin{equation}
\Pol(\RR_\alpha) \equiv \sum _{i \in \alpha}
t_i \uu_{m(i)}.
\label{eq:Pol-alpha}
\end{equation}
The six possible ground-state configurations
of that tetrahedron correspond
to $\Pol(\RR_\alpha)= (\pm 1,0,0)$, $(0,\pm 1, 0),$ or $(0,0,\pm 1)$.
Finally, the coarse-grained arrow field $\Pol(\rr)$ is the
coarse-grained polarization averaged over some larger neighborhood and
assumed to vary smoothly.
\subsection{Effective free energy and correlations}
\label{eqc:FtotFT}
The ground-state entropy density is a function of the
average polarization.
A subensemble of states in which $\Pol$ is large
(which can be forced by boundary conditions)
has relatively little freedom for rearrangements of the spins
or arrows; indeed, for a saturated polarization
such as $\Pol = (1,0,0)$ the ensemble consists
of a single microstate.
Thus it is very plausible that the entropy density has a
maximum for zero polarization.
Therefore, to lowest order,
the total free energy (arising entirely from entropy), as
a function of coarse-grained $\Pol(\rr)$, has the form
\begin{equation}
F_{\rm tot} (\{ \Pol (\rr) \})/T = v_{\rm cell}^{-1}
\int d^3\rr \half \kappa |\Pol(\rr)|^2,
\label{eq:Fcoarse}
\end{equation}
where $v_{\rm cell}=a^3/4$ is the volume of a primitive unit cell.
The ``stiffness'' $\kappa$ is dimensionless,
as appropriate since it is purely entropic in origin.
Corresponding to the condition \eqr{eq:ground}, i.e. the ice rule,
$\Pol(\rr)$ satisfies a divergence constraint
\begin{equation}
\nabla \cdot \Pol(\rr)=0
\label{eq:divergence}
\end{equation}
like a magnetic field without monopoles.
Eqs.~\eqr{eq:Fcoarse} and \eqr{eq:divergence} look, respectively,
like the field energy of a magnetic (or electric) field,
and its divergence constraint, in the absence of monopoles (or charges).
These equations, together,
signify that that the probability distribution
of the (long-wavelength portion of) the polarization field is
the (constrained) Gaussian distribution
\begin{equation}
{\rm Prob}(\{ \Pol(\rr) \} ) \propto
e^{- F(\{ \Pol (\rr) \})/T}
\prod _\rr \delta (\nabla \cdot \Pol (\rr))
\label{eq:Poldist}
\end{equation}
Fourier transforming \eqr{eq:Fcoarse} simply gives
$F_{\rm tot}= \sum _\kk \half \kappa |\Pol(\kk)|^2$,
so a naive use of equipartition would give
$\langle P_\mu (-\kk) P_\nu(\kk) \rangle = (1/{\kappa}) \delta_{\mu\nu}$.
But the divergence constraint \eqr{eq:divergence} imposes
\begin{equation}
\kk\cdot \Pol(\kk)=0
\label{eq:div-Fourier}
\end{equation}
in Fourier space. Thus the correct result has
the longitudinal fluctuations projected out:
\begin{equation}
\la P_\mu(-\kk) P _\nu(\kk)\ra
= \frac{1}{\kappa}\Bigl( \delta _{\mu\nu} -
\frac {k_\mu k_\nu}{|\kk|^2} \Bigr),
\label{eq:flucts}
\end{equation}
Fourier transforming \eqr{eq:flucts} back to direct space gives
\begin{equation}
\langle P_\mu (0) P_\nu(\rr)\rangle \cong
\frac {4\pi}{\kappa} \Bigl[ \delta(\rr) +
\frac{1}{r^3} ( \delta _{\mu\nu} -3 \hat r_\mu\hat r_\nu) \Bigr]
\label{eq:ice-dipolar}
\end{equation}
at large separations $\rr$ (where $\hat \rr \equiv \rr/|\rr|.$)
{\it Correlations have the
spatial dependence of a dipole-dipole interaction,}
which is a power law.
Models that exhibit such correlations
-- including the pyrochlore system,
so often described as ``liquid-like'',
-- are thus, in a sense, in a {\it critical} state.
[The generalization of \eqr{eq:ice-dipolar}
for $d$-dimensional real space.
would be a $1/ r^{d}$ decay.]
This criticality was first appreciated in the ice model itself,
being detected originally in a simulation~\cite{stillinger}.
A functional form with a dipolar singularity
like \eqr{eq:flucts} was produced by
a clever random-walk approximation to a series expansion~\cite{villain}.
The universal explanation, made here, that dipolar correlations
arise from \eqr{eq:Fcoarse} with the divergence condition,
was first put forward to explain experiments on
two-dimensional ice-like systems~\cite{youngblood}.
Ref.~\onlinecite{huse-dimer}
have also presented an ansatz equivalent to \eqr{eq:Poldist},
derived the dipolar correlations,
and confirmed them by simulations, for
the dimer covering of a simple cubic lattice.
[See their eq.~(3).]
The $1/|\rr|^3$ decay has also been obtained analytically and
numerically in Ref.~\onlinecite{hermele} , for the
pyrochlore model of this present paper.
Ref.~\onlinecite{moessner98b}, Sec.~II~D 1, recognized
that the ground-state constraint in the Heisenberg
pyrochlore antiferromagnet \eqr{eq:ground} entails
long-range correlations, but in the absence of
the polarization concept, the argument's form is diffuse,
and it was not possible to predict an explicit functional
form. An interesting heuristic argument was made there to justify the
empirical fact of the ``bowtie'' shape (see Fig.~\ref{fig:strucf})
taken by the diffuse scattering: in other words,
that it takes a scaling form in terms of $q_x/q_\perp$.
\OMIT{Although a planar section of the diffuse
scattering does look like a ``bowtie'', in three-dimensional
reciprocal space the contours are surfaces of revolution.}
\subsection{Diffraction consequences}
\LATER{NEED introduxtory sentence}
\subsubsection{Spin structure factor}
\label{sec:spinstrucf-1}
There is a linear relationship
between the actual Ising spins and the ice arrows:
specifically, the map~\eqr{eq:Pol-alpha},
from spins to the the diamond-vertex polarization is actually
invertible:
\begin{equation}
t(\RR \pm \half \uu_m) = 4\uu_m \cdot \Pol(\RR),
\label{eq:Pol-spin}
\end{equation}
where we take the $+$ or $-$ signs for even and odd
vertices $\RR$, respectively.
Thus it is not surprising that
the pseudodipolar correlations \eqr{eq:ice-dipolar}
of the latter imply similar $1/r^3$ correlations for the former.
However, the coefficients relating these (namely,
the $\{ \uu _m \}$ vectors) are staggered in sign.
To see this, first Fourier transform \eqr{eq:Pol-spin}, using
the fact that
\begin{equation}
\uu_{m,x}= \fourth e^{\half i\KK_{200}\cdot (\uu_m-\uu_4)},
\label{eq:uuKK}
\end{equation}
similarly $uu_{m,y}$ with $\KK_{020}$, etc. When this is inserted
into the formula for $\tt(\kk)$ [similar to \eqr{eq:t-Fourier}],
the staggering contained in the $\uu_m$
factors shifts the argument by the reciprocal space vector
$\KK_{200}$ so that $\tt(\KK_{200}+\qq) \propto P_x(\qq)$, etc.
Consequently, the singularities
of the spin structure factor in reciprocal space
are pseudo-dipolar in form, just like \eqr{eq:spindipolar}, but
occur at nonzero reciprocal lattice vectors $\KK$
rather than at $\qq=0$.
%%% (This is worked out precisely in Sec.~\ref{sec:spinstrucf-2}.)
Let $\tt(\kk)$ be the Fourier transform of the spins.
The structure factor (i.e. the neutron diffraction intensity.
modulo polarization factors) has pseudo-dipolar singularities
\begin{equation}
\Struc{\KK_{200}+\qq} \propto
\frac{q_\perp^2}{q_\parallel^2 + q_\perp^2}
\label{eq:spindipolar}
\end{equation}
Here $\KK_{200}\equiv [2\pi/a] (2,0,0)$
is a reciprocal lattice vector;
and $\qq_\parallel \equiv q_x$ and $q_\perp^2 \equiv q_y^2 + q_z^2$.
Near $\KK_{111}\equiv 2\pi(1,1,1)$, the same form holds with
$\qq_\parallel$ and $\qq_\perp$ being the components of $\qq$
parallel and perpendicular to the $(111)$ direction.
The elastic constant $\kappa$ must be determined
from Monte Carlo simulations.
Yet, without knowing it, we can still make the nontrivial prediction that the
diffuse scattering has quantitative the same strength near $\KK_{111}$
as it does near $\KK_{200}$.
In an isotropic Heisenberg model,
these pseudodipolar singularities can be distinguished
from true dipolar singularities
in neutron scattering experiments,
since they are independent of spin direction.
They can be seen in only one spatial direction around each reciprocal
lattice point.
The functional form \eqr{eq:spindipolar}
has nodes which we expect (correctly) are a consequence
of symmetry and therefore extend throughout reciprocal
space, beyond the small-$\qq$ limit in which
\eqr{eq:spindipolar} was derived.
For example, \eqr{eq:spindipolar} indicates that
\begin{equation}
\Struc{k_x,0,0} \equiv 0
\label{eq:100zero}
\end{equation}
for wavevectors anywhere along the entire (100) axis.
Similarly the structure factor is
zero along the (111) axis
near $\KK_{111}$ from \eqr{eq:spindipolar}
Since $\kk=(000)$ is the intersection of seven distinct axes,
along each of which the structure factor must vanish,
we deduce that the diffuse scattering vanishes near the origin as
\begin{equation}
\langle |t(\kk)|^2\rangle \propto
(k_x^2 k_y^4 + {\text{5 permutations}})-6 k_x^2 k_y^2 k_z^2.
\end{equation}
Indeed, experiments and all simulations on pyrochlore
antiferromagnets find quite small diffraction
throughout the first Brillouin zone.
\LATER{Citing Nagle~\cite{nagle}.}
\subsubsection{Derivation of nodal lines}
\def\tplane{{t_{\rm plane}}}
It is possible to confirm \eqr{eq:100zero}
without using coarse-graining, by sharpening
an argument made in Sec.~II D of Ref.~\onlinecite{moessner98b}.
(I have written the steps in detail, so it will be clear whether
they do or do not carry through for other lattices.)
Partition the pyrochlore lattice sites into (100) planes labeled by $x$,
and define $\tplane(x) \equiv \sum _{x_i=x} t_i$.
Then consider, for example, the tetrahedra with
centers at $x=0$: each such tetrahedron includes a pair of spins
with $x_i=-\fourth$ and another pair with $x_i=+\fourth$.
By the ground-state constraint \eqr{eq:ground} the sum of
one pair is the negative of the sum of the other pair.
Since every spin with $x_i=\pm \fourth$
belongs to exactly one such tetrahedron,
this implies $\tplane(-\fourth)=-\tplane(\fourth)$;
and by Bravais lattice periodicity,
$\tplane(\fourth+\half n)=(-1)^2 \tplane(\fourth)$.
Since all spins have $x_i=\fourth+\half n$ for some $n$,
\begin{equation}
\tt(k_x,0,0) \propto \sum _n e^{-ik_x(\fourth+\half n)}
(-1)^n \tplane(\fourth)
\end{equation}
which cancels, except for a possible Bragg peak when $k_x$ is a
multiple of $2\pi$.
However, in the maximum entropy state that we
assumed in Sec.~\ref{eqc:FtotFT},
the average spin in one of these planes
%%% $\tplane(\fourth)$
-- which is proportional to the mean polarization $P_x$
-- is zero, so $\tt(k_x,0,0)=0$ for all $k_x$.
A similar, but more complicated, argument works for a $\{111\}$ direction.
Let $\xi_i\equiv \rr_i \cdot \frac {1}{\sqrt 3}(1,1,1)$ be the projection
of site $i$ on the $(111)$ axis, and redefine $\tplane(\xi)$ as the
spin sum over the spin plane with $\xi_i=\xi$.
Each tetrahedron with $\xi=0$ includes
three spins in the $\xi=-1/4\sqrt{3}$ plane and one spin
in the $\xi=\sqrt{3}/4$ plane, and every spin in either plane
belongs to a unique tetrahedron; consequently
$\tplane(-1/4\sqrt{3})=-
\tplane(+\sqrt{3}/4)$,
even though one plane has three times as many spins as the other.
Furthermore these unequal planes are equally spaced, so
$\tplane(\sqrt{3}/4 + n/\sqrt{3}) = (-1)^n \tplane(\sqrt{3}/4)$
and the rest of the argument follows as before.
\subsubsection{Alternate derivation of nodal lines}
There is an alternate way to see that diffuse intensity must
be small near the zone center. Let $\Pol^+(\kk)$
be Fourier transform of the discrete polarization
on (say) even diamond sites, defined like \eqr{eq:L-Fourier}.
On the one hand, the Fourier transform of the
constraint \eqr{eq:ground} at {\it odd} vertices
\begin{equation}
4 \G(-\kk) \cdot \Pol(\kk) =0
\label{eq:lattice-con-FT}
\end{equation}
where
\begin{equation}
\G(\kk) = \sum _{m=1}^4 e^{ i\kk\cdot \uu_m} \uu_m \approx
\fourth i \kk
\end{equation}
at small wavevectors.
On the other hand, we can start from \eqr{eq:Pol-spin}
and take its Fourier transform -- this time, not taking
advantage of \eqr{eq:uuKK}; the result is
\begin{equation}
\tt(\kk) = 4 \G(\half \kk)\cdot \tPol(\kk).
%%%% = \sum _{\R,m} \exp(i\kk\cdot(\RR+\uu_m/2)) s(\RR+\uu_m/2))
\label{eq:Pol-spin-FT}
\end{equation}
Near $\kk =0$,
comparing \eqr{eq:Pol-spin-FT} and \eqr{eq:lattice-con-FT}
and recalling that $\G(-\half \kk) \approx -\half \G(\kk)$,
we conclude that $\tt(\kk)\approx 0$, as I asserted earlier.
Furthermore, whenever $\kk$ is along a $\{100\}$ or a $\{111\}$ symmetry axis,
$\G(\kk)$ is parallel to that same axis by symmetry (for all $\kk$),
hence structure factor \eqr{eq:100zero} is null all along those axes.
\subsection{Generalization to Heisenberg spins}
\label{sec:Heis-pol}
In the case of isotropic $n=3$ component spins, we can
reiterate all the coarse-graining arguments of Sec.~\ref{sec:coarse}
in terms of the Heisenberg spin components.
Polarization components are defined as in \eqr{eq:Pol-alpha}
but now for each Cartesian spin component, so the polarization field is now
a {\it tensor} carrying not only space but also spin indices:
$(\Pol_\alpha)_{\beta\mu} \equiv \sum _{i\in \alpha}
(\ss_i)_\beta (\uu_{m(i)})_\mu$.
As in the Ising case, it is easy to convince oneself that
the entropy density is maximum when $\Pol=0$. Hence we expect
that \eqr{eq:Fcoarse} remains valid, except that each $|\Pol(\rr)|^2$
is now interpreted as a tensor norm (sum of squares of all
tensor elements).
Finally, \eqr{eq:divergence} now becomes three equations, one for
each spin-space component.
The basis of this paper
-- and the signatures implied in the diffuse diffraction --
extends even to Heisenberg pyrochlore antiferromagnets
in a magnetic field $\Hfield$. The replacement
$\HH \to \HH - \Hfield \cdot \sum_i \ss_i$
is equivalent to substituting
\begin{equation}
\LL_\alpha \to \LL_\alpha - \Hfield/2\Jspin
\label{eq:L-shifted}
\end{equation}
in \eqr{eq:Ham}.
In the case of vector spins (but not Ising spins!),
one achieves a ground state by
satisfying $\LL_\alpha=\Hfield/2\Jspin$ on every tetrahedron.
Simply replacing $\delta \ss_i\equiv \ss_i-\Hfield/8$
in the definition of $\Pol(\rr)$
(i.e subtracting off the mean spin expectation),
provides a polarization appropriate
to this ground state, which should exhibit the
same sort of long-range correlations
(but with an $|\Hfield|$ dependence.)
\section{Lattice approximation}
\label{sec:lattice}
The continuum theory presented in Sec.~\ref{sec:coarse}
can only predict the shape
of singularities at special points in reciprocal space.
For a better comparison to experiments,
a theory of the diffuse scattering
throughout reciprocal space is desirable.
Of course, whereas the result \eqr{eq:ice-dipolar}
is universal across a class of models
as enumerated in Sec.~\ref{sec:icemap}
and Sec.~\ref{sec:more-lattices},
the detailed shape of the scattering calculated here
is specific to the pyrochlore lattice.
\OMIT{To handle the case of an external field, one
would replace the $\delta$-functions
in \eqr{eq:maxlike}
by $\delta(\LL(\RR)-\Hfield)$.
But I don't trust the validity of the lattice derivation
with the Heisenberg model, so this is cut.}
The derivation depends on lattice Fourier transforms.
This unfortunately forces the introduction, for this section,
a new indexing $t(\RR+\half\uu_m)$ and $L(\RR)$
for the same objects $t_i$ and $L_\alpha$
defined previously, where $\RR$ designate the diamond sites;
the even sites $\RR$ form an fcc Bravais lattice and
we let $N$ be the number of primitive unit cells.
Thus each pyrochlore site $\rr_i$ is written
$\RR+\half \uu_m$; the surrounding odd
diamond vertices are at $\RR+\uu_m$.
\begin{equation}
\tt_m(\kk)= \frac{1}{\sqrt N}
\sum _\RR e^{-i\kk\cdot (\RR+\half \uu_m)}
t(\RR+\half \uu_m) .
\label{eq:t-Fourier}
\end{equation}
The definition \eqr{eq:L} is rewritten
$L(\RR)=\sum _{m=1}^4 t(\RR\pm\half \uu_m))$
taking $+$ and $-$ when $\RR$ is an even or odd diamond vertex,
respectively. Thus the Fourier transform of \eqr{eq:L},
restricted to even or odd vertices as indicated by the ``$\pm$'', is
\begin{equation}
\tilde{L}^{\pm} (\kk) =
\frac{1}{\sqrt N}
\sum _{m=1}^4 e^{\pm \half i \kk\cdot \uu_m} \tt_m
\label{eq:L-Fourier}
\end{equation}
\OMIT{The phase factor used belongs to the actual
position of the spin, rather than to the center of the unit cell
associated with $\RR$.}
\subsection{Derivation of fluctuations}
\label{sec:lattice-deriv}
The simplest way to describe the approximation is
first to imagine a distribution of spins
\begin{equation}
{\rm Prob}(\{ t_i \}) \propto
e^{-\sum _i \frac{t_i^2}{2 t_0^2}} \exp (-\betaeff \HH),
\label{eq:gaussianT}
\end{equation}
where $\HH$ is given by \eqr{eq:Ham},
and $\betaeff$ ought to be the inverse temperature.
The Gaussian factor in \eqr{eq:maxlike}
may be viewed, in the Bayesian spirit of the maximum-likelihood
approach, as a trivial {\it a priori} independent distribution before we
account for any spin interaction. The second factor
enforces spin correlations; here $\HH(\{ t_i \})$ is the Hamiltonian
\eqr{eq:Ham} for Ising spins, except that now
$t_i$ is allowed to take any real value.
Adopting the limit $\betaeff \to \infty$, \eqr{eq:gaussianT} reduces to
\begin{equation}
{\rm Prob}(\{ t_i \}) \propto
e^{-\sum _i \frac{t_i^2}{2 t_0^2}}
\prod_\alpha \delta(L_\alpha)
\label{eq:maxlike}
\end{equation}
The second factor in \eqr{eq:maxlike} imposes the ground state
constraints \eqr{eq:ground} around every tetrahedron, both even and odd.
\subsubsection{Ising spin correlations}
Our goal is to evaluate $\langle t_i t_j \rangle$, and this is not hard
to do using the Fourier transform \eqr{eq:t-Fourier}.
We must first rewrite
$\prod_\alpha \delta(L_\alpha)
\propto \prod_\kk \delta(L^+(\kk)) \delta(L^-(\kk))$
in \eqr{eq:maxlike}, and re-express this in terms of $\{ \tt_m(\kk) \}$.
Since different wavevectors decouple, it will be convenient to
view $\tt_m (\kk)$ as a complex 4-component vector.
Then the even and odd vertex constraints take the form
of orthogonality conditions:
%%% $(E^+(\kk), \tt(\kk))= (E^+(\kk), \tt(\kk))= 0$,
\begin{equation}
{\tilde L}^\pm(\kk) \equiv
\sum _{m=1}^4 e^{\pm \half i \kk\cdot \uu_m} \tt_m(\kk)
\equiv (E^\pm(\kk,\tt(\kk))=0
\label{eq:orthog}
\end{equation}
The coefficients are
\begin{equation}
E^\pm_m(\kk) \equiv e^{\mp\half i \kk \cdot \uu_m}
\label{eq:E}
\end{equation}
so $E^-_m \equiv (E^+_m)^*.$
The inner product is defined by
\begin{equation}
(A,B) \equiv \sum _{m=1}^4 A_m^* B_m.
\end{equation}
So, the distribution \eqr{eq:maxlike}
can be rewritten in Fourier space as
\begin{equation}
\prod_\kk \left[ e^{-\frac{1}{2 t_0^2} \sum_m^4 |\tt_m(\kk)|^2 }
\delta((E^+(\kk),\tt) \delta((E^-(\kk),\tt) \right].
\label{eq:spin-con-FT}
\end{equation}
Thus, for each $\kk$ in \eqr{eq:spin-con-FT},
we now have a Gaussian distribution over a four-dimensional space,
with two constraints reducing it to a two-dimensional subspace.
\OMIT{[Technically, the space is eight
dimensional since the components $\tt_m$ are complex, but
the other four dimensions are ascribed to $\tt_m(-\kk) =
\tt_m(\kk)^*$, as is customary.]}
Our object is to obtain correlations of the form
$\la \tt_l(-\kk) \tt_m (\kk) \ra$.
The result will essentiallly be
the projector $\Pi$ into the subspace satisfying \eqr{eq:orthog}.
\LATER{Use slant font for $\Pi$?}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
If we define the $4\times 2$ matrix $E$, the columns of which
are $E^+$ and $E^-$,
then $4\times 4$ projection matrix for the (desired)
space orthogonal to $E^+$ and $E^-$ is
$\Pi \equiv I-E(E^\dagger E)^{-1} E^\dagger$.
In fact
\begin{eqnarray}
(E^\dagger E) = 4 \pmatrix{ 1 & H\cr
H^* & 1} , \\
(E^\dagger E)^{-1} = \frac{1}{4} (1-|H|^2)^{-1}
\pmatrix{ 1 & -H^*\cr
-H & 1}
\label{eq:EEinverse}
\end{eqnarray}
where the $\kk$ argument in $E$ and $H$ was suppressed; here
\begin{eqnarray}
H(\kk) \equiv \frac{1}{4} \sum _m e^{i\kk\cdot \uu_m} =
\cos \frac{k_x}{4} \cos \frac{k_y}{4} \cos \frac{k_z}{4}
\nonumber \\
-i \sin \frac{k_x}{4} \sin \frac{k_y}{4} \sin \frac{k_z}{4} .
\end{eqnarray}
[$H(-\kk)\equiv H(\kk)^*$,
$|H(\kk)|^2 = \frac{1}{4}(1+
\cos \frac{k_y}{2} \cos \frac{k_z}{2}
+ \cos \frac{k_x}{2} \cos \frac{k_z}{2}
+ \cos \frac{k_x}{2} \cos \frac{k_y}{2} )$.
The final result is
\begin{equation}
\la \tt_l(-\kk) \tt_m (\kk) \ra =
t_0^2 \left\{\delta_{lm}- [E(E^\dagger E)^{-1} E^\dagger]_{lm}\right\}
\label{eq:tt-final}
\end{equation}
where $E$ and $(E^\dagger E)^{-1}$
are implicitly functions of $\kk$ and are defined by \eqr{eq:E}
and \eqr{eq:EEinverse}.
Since ${\rm Tr} \Pi =2$, we must have $t_0^2=2$ to satisfy
the normalization condition $\la t_i^2 \ra=1$.
\OMIT{If we want to be more formal, one can alternatively reach the
projection matrix by writing the two $\delta$ functions (for each $\kk$) as
$\delta(u_+) \delta(u_-)
=\int d\mu_+ d\mu_- e^{i(\mu_+ u_+ + \mu_- u_-)}$;
Switching the order of
integration over $\tt_m$ and over $\mu_+$, $\mu_-$ in the integrals,
we can do the four-dimensional integrals over $\tt_m$ exactly using
a shift of variable, and what remains is a two-dimensional
Gaussian integral over the $\mu$ variables, in which the
$2\times 2$ matrix of coefficients is the same one as appears
in \eqr{eq:matrix}.}
\subsubsection{Spin structure factor}
\label{sec:spinstrucf-2}
The structure factor, as measured in neutron diffraction,
combines the four sublattice contributions.
Write the Fourier transform of all the spins as
$\tt(\kk) \equiv
\frac{1}{\sqrt N}
\sum_m \sum _\RR e^{-i\kk\cdot (\RR+\half \uu_m)}
t(\RR+\half \uu_m) = (M,\tilde t(\kk))$,
where we used the 4-vector inner product and
\begin{equation}
M\equiv (1,1,1,1)
\label{eq:Mcoeff}
\end{equation}
The structure factor is then a projection of \eqr{eq:tt-final}
onto the $M$ vector, thus
\begin{equation}
\la |\tt(\kk)|^2\ra
= t_0^2 M^T [I- E(E^\dagger E)^{-1} E^\dagger] M .
\label{eq:struc-result}
\end{equation}
In the case of spin ice, the mapping of the Ising spins $\{ t_i \}$
to the real spins is different (it is staggered).
In this case, $M_{\beta m} = (\hat{\uu} _m)_\beta $.
The pinch-point singularities will appear, but at
different places (including the origin), because
in this case the spins {\it are} the local polarizations.
In this case, the scattering around
the origin is no longer suppressed.
In either the antiferromagnet or the spin-ice case,
there will be extinctions of the expected singularity
at points where $(E^{\pm}, M)$ happens to cancel.
\LATER{Work out the form of $|\Pol(\kk)|^2$ using a
similar technlogy, to verify the claims,
connect to Sec. II, and extract the stiffness $\kappa$.}i
Considering the form of \eqr{eq:tt-final},
singularities in the spin fluctuations
may (but do not necessarily) occur when the
$2\times 2$ matrix $(E^\dagger E)$ is singular.
In view of \eqr{eq:EEinverse}, this occurs when
$|H(\kk)|^2=1$. But that is precisely the
defining condition for the fcc
reciprocal lattice vectors $\KK$.
[This was derived by a different route in
Sec.~\ref{sec:spinstrucf-1}.]
Indeed, $E^+(\KK_{200})=i(-1,-1,1,1)$ and
$E^+(\KK_{111})=\frac{1+i}{\sqrt 2} (-1,1,1,1)$.
We see $E^+(\KK)$ is proportional
to a {\it real} four-vector.
[That generically happens at
a point in three-dimensional $\kk$-space, since there are
three independent phase relationships to be satisfied
among the components $\{ E^+_m \}$.]
Then $E^+$ is proportional to $E^-$,
and the rank of $E(\KK)$ is reduced from two to one
at these points, confirming that $E(\KK)$ is singular.
\LATER{WORK OUT EXPLICITLY THE EXPANSION NEAR THESE POINTS.}
The special lines in reciprocal space on
which diffraction is zero, are those where
the 4-vector of coefficients \eqr{eq:Mcoeff},
relating the physical spin
to the four sublattice
spins $\tt_m(\kk)$, happens to be orthogonal
to both of the vectors in the null space of the
$4\times 4$ matrix \eqr{eq:tt-final}.
%%% \old-section{Other analytic calculations}
\subsection{Finite temperature}
In the approximations of Sec.~\ref{sec:coarse}
and Sec.~\ref{sec:lattice-deriv},
where the tetrahedron constraint was imposed rigorously,
we were forced to project out the corresponding fluctuation modes.
It is possible to extend the approximation so as to
permit small fluctuations of the tetrahedron magnetizations,
as must be excited at $T>0$, starting from
\eqr{eq:gaussianT}. Now, no projection
is required, since a large coefficient $\betaeff$
tends to suppress those fluctuations, and
so this $\betaeff<\infty$ case is
actuallly more straightforward.
%%% which is needed to describe a
[This subsection is essentially
a streamlining of Ref.~\onlinecite{yoshida},
as converted to my notations.]
The Hamiltonian in \eqr{eq:gaussianT} is a quadratic
form in the $t_i$'s. Fourier transforming\eqr{eq:Ham},
and putting it into the 4-vector notation with \eqr{eq:orthog}, we get
\begin{equation}
\HH = \half J \sum _\kk \left[ |L^+(\kk)|^2 + |L^-(\kk)|^2 \right]
= \half J (\tt, E^\dagger E \tt) .
\end{equation}
Thus, \eqr{eq:gaussianT} becomes ${\rm Prob}(\{\tt_m\})
\propto \exp(-\half (\tt,\lambda \tt)$, where
\begin{equation}
\Lambda \equiv \frac{1}{t_0^2} I + \betaeff E^\dagger E.
\end{equation}
and \eqr{eq:tt-final} gets replaced by
\begin{equation}
\la t_l(-\kk) t_m (\kk) \ra = (\Lambda^{-1})_{lm}.
\label{eq:tt-thermal}
\end{equation}
The matrix $E^\dagger E$ has rank two.
It can be seen that, as $\betaeff \to \infty$,
\eqr{eq:tt-thermal} indeed reduces to the projector \eqr{eq:tt-final}.
The consequence of $\betaeff < \infty$ for correlations is that,
in reciprocal space the pseudodipolar singularities
\eqr{eq:spindipolar}, get rounded by the substitution
\begin{equation}
\qq^2 \to \qq^2+\xi^{-2}
\label{eq:screened}
\end{equation}
in the denominator,
where $\xi \sim \betaeff ^{1/2}$ is a a correlation length.
In real space, the power-law decays
\eqr{eq:ice-dipolar} acquire an
$\exp(-r/\xi)$ factor that cuts them off,
as noted in Ref.~\onlinecite{garanin}.
\subsection{Other analytic approximations}
Three prior treatments of the diffuse scattering in the
pyrochlore arrived
at a mathematical form more or less identical to \eqr{eq:gaussianT},
but with different formulas for the coefficients in these equations
as a function if temperature.
Thus, of course, their result is \eqr{eq:tt-thermal};
however, they did not note the pseudodipolar
(or, in the spin-ice case, literally dipolar) correlations
which are implicit in these formulas at the $T=0$ limit.
%% For the limiting case
%% $\betaeff=\infty$, as noted already, this reduces to exactly the
%% approximation I described here.
%%% \subsubsection{Reimers}
The diffuse scattering problem was first addressed by
Reimers~\cite{reimers92b} for Heisenberg, or in general
$n$-component vector spins.) That derivation is based on
mean field theory (Ornstein-Zernike correlations),
which ought to be valid in the critical regime $T\to \Tcmf$,
where $\Tcmf=J/n$.
However, that is invalid in the pyrochlore case, since
the real critical temperature is driven to zero.
Even so, the $\kk$ dependence of the result
(eq.~(15) of Ref.~\onlinecite{reimers92b}), for the
limit $T\to \Tcmf$, is exactly of the form \eqr{eq:gaussianT},
with $1/t_0^2 = 3 (T/\Tcmf-1)$ and $\betaeff=1/T$.
%%% \subsubsection{Canals and Garanin}
Canals and Garanin~\cite{garanin,canals-garanin}
considered the classical pyrochlore antiferromagnet with
$n$-component unit spins, in the large-$n$ limit which is
tractable analytically. This is, in effect,
like the ``spherical model'' approximation for finite-$n$,
in that a constraint $|\ss_i|=1$ on every spin
is replaced by one on all $N$ spins, $\sum_i |\ss_i|^2 = N$.
As they noted in Ref.~\onlinecite{garanin}, Sec. III,
the $n=\infty$ limit is completely described by Gaussian
approximations.
Their result for small temperatures can be reduced to
\eqr{eq:gaussianT} with $1/t_0^2 = n/3$
and $\betaeff=1/T$; this reduces to \eqr{eq:maxlike}
as $T\to 0$.
\LATER{CHECK Canals and G.}
%%% \subsubsection{Yoshida et al}
Finally, Yoshida {\it et al}~\cite{yoshida} use an elaborate
cluster-variational method to derive a sensible formula for the
temperature-dependent diffuse scattering
for the {\it Ising} ($n=1$) pyrochlore antiferromagnet
(specifically, spin ice).
Their result, taking the lowest-order finite temperature correction,
amounts to \eqr{eq:gaussianT}
with $t_0^2=2$
and $\betaeff J = \frac{3}{16} e^{2 J/T}$,
which in the zero-temperature limit reduces to \eqr{eq:maxlike}.
[Note that $J_{\rm eff}/3$ of Ref.~\onlinecite{yoshida} is my $J$.
My $E^{\pm m}$ are essentially linear combinations of the
4-vectors $c$ and $s$ of their eq.~(B-3).]
Until now, I have only mentioned approximations which
involve the connectivity of the sites and which can incorporate
the long-range nature of the constraint.
Some other approximations, intended mainly to model the bulk
susceptibility, are based on a single isolated tetrahedron
~\cite{harris95,moessner-berlinsky,garcia-huber}.
Apart from the restriction that this tetrahedron has
zero net spin,
these approaches necessarily miss the long-range constraint
and hence give a poor picture of the long-range correlations.
\subsection {Comparison to diffraction in
experiment and simulation}
Several pyrochlore systems show the same characteristic
diffuse scattering features:
(i) the entire first Brillouin zone has a very low intensity
(ii) intensity is zero along $\{100\}$ abd $\{111\}$
axes; (iii) there is a pinch-point singularity
of form \eqr{eq:spindipolar}
at $\KK_{200}$ and also $\KK_{111}$ reciprocal
lattice vectors.
It should be noted that, since there are four spin sites per
unit cell, the diffuse scattering
is not periodic with the Brillouin zone.
Experimentally, such features were seen in an
itinerant Laves phase (Ref.~\onlinecite{ballou}, Fig.~3);
in the pyrochlore CsNiCrF$_6$''
(Ref.~\onlinecite{harris97a}, Fig.~4);
and most recently in the spinel
ZnCrO$_4$ (Ref.~\onlinecite{lee02}, Fig.~3(a,b)),
in the higher-temperature regime above
a structural transition.
In simulations, such patterns appeared
in Fig.~2 of Ref.~\onlinecite{zinhar95},
Fig.~4 of ~Ref.~\onlinecite{harris97a},
%%% {\sl CITE FIG. 4, hhl plane}
and in Ref.~\onlinecite{moessner98b}.
For comparison, spin-1/2 results
from exact diaagonalization are
shown in Ref.~\onlinecite{canals98}, Fig.~4;
they are qualitatively similar, but less sharp.
Images of analytic large-$n$ calculations
of Garanin and Canals can also be compared:
Ref.~\onlinecite{canals-garanin}, Fig.~4
(which is the \kag system)
and Ref.~\onlinecite{Can00}, Fig.~6.
\LATER{Add Yoshida to this list.}
Ref.~\onlinecite{moessner-kagice}, in their
Fig.~5, plot diffuse scattering from
simulation of a two-dimensional spin problem
equivalent to the honeycomb dimer covering.
(This is a plane of spins in the
``\kag-ice'' phase, whereby an external field applied to
a ``spin-ice'' pyrochlore system causes $\{111\}$ planes to decouple).
The pinch-points (called ``bowties'' by those authors) are
prominently visible, which are diagnostic of pseudodipolar correlations
on real space.
The structure factor \eqr{eq:spindipolar}
has a local maximum at $q+x=0$, if
we vary $q_x$ along a line $\qq_\perp={\rm const}$,
offset slightly from the $q_x$-axis.
The same behavior is found around $\KK_{111}$ and,
of course, all other symmery-equivalent reciprocal lattice vectors:
the structure factor has maxima in the plane perpendicular to
the radial direction in reciprocal space.
The union of these planar facets forms the same shape (a cuboctahedron)
as the boundary of the FCC lattice's first Brillouin boundary,
but doubled in all three directions.
Indeed, in a plane of reciprocal space from a single crystal
of Y$_{1-x}$Sc$_x$Mn$_2$, the diffuse scattering
is concentrated near the {\it lines} where this
plane cuts the Brillouin zone faces \cite{ballou}.
Powder diffraction data from pyrochlores showed
a characteristic maximum
~\cite{reimers92b}
which was seen experimentally~\cite{harris94},
and in simulation~\cite{reimers92a}.
\OMIT{Note that $a\approx 1.02$ nm in the pyrochlore lattice.)}
This is consistent with the fact that the (doubled)
Brillouin zone boundary (where diffraction is
maximum along any ray in reciprocal space) is roughly a sphere.
(Note that powder averaging, in the vicinity of $\KK_{200}$, amounts to
integrating \eqr{eq:spindipolar} over $q_y$ and $q_z$, which yields
a weakly cusped function ${\rm Const}- 2 \pi q_x^2 \ln q_x$.)
\section{Discussion}
\label{sec:disc}
In summary, it was found that a polarization can be
defined in pyrochlore antiferromagnets which
-- in a ground state -- exactly satisfies a
divergence condition. The coarse-grained version
of this is analogous to an order parameter,
in being the natural variable to describe
large-scale properties.
The correlations were found (Sec.~\ref{sec:coarse})
to have a pseudo-dipolar form which -- since this is a
pure power law -- implies an infinite correlation length.
These behaviors were repeated in a lattice-based derivation
(Sec.~\ref{sec:lattice}.
In the rest of this section, I discuss other problems
to which these findings or approaches could be extended
or related.
\subsection{Implications for diffraction experiments}
\label{sec:neutrons}
The argument of this paper suggests that
the analytis of diffraction experiments
ought to focus more on the characteristic features,
such as pinch points, identified in
Sec.~\ref{sec:spinstrucf-1} and Sec.~\ref{sec:spinstrucf-2}.
Deviations from the predictions at those places
are sensitive measures of the extent to
which the tetrahedron constraint is violated
in the actual ensemble. Thus, it is suggested
to analyze these experiments so as to extract
the correlation length $\xi$, and to check how
well the diffraction is suppressed along the
predicted nodal lines.
[A correlation length
was extracted in Ref.~\onlinecite{zinhar95}
from simulations, however in this case it was
actually the finite size cutoff.]
Deviations in the overall pattern from the shape
predicted in Sec.~\ref{sec:spinstrucf-2}
(which are consistent with simulations~\cite{moessner-dipolar})
are likely to reflect additional
terms in the Hamiltonian, as in Ref.~\onlinecite{lee02}.
Ref.~\onlinecite{lee02} have fitted the diffuse intensity as
the Fourier transform, not of a single tetrahedron, but a single loop
of 6 spins.
However, contrary to the speculation in that paper, this does
not necessarily indicate a physical state built from such hexagons.
To explain this, I will outline an alternative, equally
systematic way of fitting the diffuse diffraction data.
The constraint \eqr{eq:Ham} implies a similar constraint on
the matrix of correlation functions. Then one can express any
valid correlation function using a
basis of linearly independent, orthogonal functions in real space satisfying
this constraint, in the spirit of e.g. spherical harmonics.
The first of these terms is the same correlation that derives
from the ring of six. The form observed~\cite{lee02} is,
from this viewpoint, the simplest possible shape, as one
might expect at higher temperatures when all the other
terms are damped out. To produce a pinch point, an infinite
number of such functions would be needed, corresponding to
a large spatial extent.
\subsection{Dynamics }
It is well known that violations of the tetrahedron
constraint (due to disorder, or thermal excitation)
map to electric charges (in the language where $\Pol(\rr)$
is an ``electric field.''
A defect in which three arrows point
outwards has a ``charge'' of $Q=+1$, or $Q=-1$ if three point
inwards~\cite{FN-Gausslaw}; in general, $Q=L_\alpha/2$.
These ``charges'' can only be created in opposite
pairs; such a pair feels an (entropic)
effective interaction which behaves (at coarse-grained
distances) exactly like a Coulomb interaction.
At $T>0$, the interaction would be screened in the usual
fashion, and the screening length is the correlation
length mentioned in Subsec.~\ref{sec:neutrons}.
Apart from these defects, the total (classical) spin is constrained to
be exactly zero. [The total magnetization is
$\half \sum_\alpha{\LL_\alpha} =0$ by \eqr{eq:ground}.]
Hence, the defects are central to any theory of the
magnetic relaxation, as observed by inelastic neutron diffraction,
or in AC susceptibility
measurements~\cite{snyder03,snyder-dilution}.
To sharpen this point, note that within the ground states,
there is {\it no} local move that produces another valid state:
an entire loop must be updated at once~\cite{stillinger}.
But the movement of a charge along a loop leaves behind the
same change and thus implements this nonlocal ``flip'' operation.
The interpretation of relaxation experiments ought to be cast in
terms of the diffusion and recombination of pseudo ``electric charges".
The relaxation rate of the real magnetization is proportional
to the drift mobility of the ``charges''. The theory of their
behavior is isomorphic to an intrinsic semiconductor;
the cost a $Q=\pm1$ defect (in an Ising model) is $2J$, so
$4J$ plays the role of the bandgap.
Nonmagnetic impurity sites~\cite{snyder-dilution}
act like impurity levels in a semiconductor
(exactly at midgap). In the Ising model, a tetrahedron with
one missing site has a ground state with $L_\alpha=\pm 1$,
corresponding to $Q=\pm 1/2$: it is as if a quenched charge
of $-1/2$ has been placed there, with the possibility of
binding a $+1$ charge to it.
In a Heisenberg model, such a tetrahedron often
still satisfies $\LL_\alpha=0$ and the behavior is
more subtle~\cite{henley-HFM00}.
We could also make predictions for the dynamic neutron
structure factor. The polarization $\Pol(\rr)$ is a conserved
quantity, so (in a classical model) it must diffuse.
Thus
\begin{equation}
\frac{\partial \Pol}{\partial t} = - \nabla \cdot \JJ_\Pol
= \Gamma \nabla^2 \Pol
\label{eq:diffusion}
\end{equation}
Hence, near a pinch point $\KK$, this [classical] dynamics
implies vanishing widths in the dynamic structure factor
${\cal S}(\KK+\qq,\omega) \sim \Gamma \qq^2$.
Dynamical conclusions were derived by Moessner and Chalker
directly from an equation of motion in terms of $\LL_\alpha$.
[The consequence in a quantum model appears to be
a gapless excitation with linear dispersion~\cite{hermele}.]
\subsection{Other models}
\subsubsection{Other three-dimensional lattices}
\label{sec:more-lattices}
The polarization construction can be generalized to any model
in which (i) ``spins'' sit on the edges of a bipartite
graph, and (ii) around every vertex, the sums of the ``spins''
is the same; here ``spin'' degree might be structural
as well as magnetic.
In three dimensions, the best
examples (apart from the pyrochlore lattice) are
antiferromagnets on the garnet lattice~\cite{petrenko}
or dimer coverings on the simple cubic, bcc, and diamond
lattices.
Dimer coverings of the diamond lattice
might realized by the ice model
in which ions of difference valences sit on the even
and odd diamond sites; they also correspond
to the ground states of an Ising pyrochlore in
an external field.
This last system is most
plausibly realized by a 1/4-occupancy lattice gas,
with nearest-neighbor repulsion,
representing a charge order problem.
Alongside the dimer models are vertex models
on the simple cubic or triangular lattice,
in which each site has three inwards and three
outwards arrows~\cite{balents,hermele}.
\subsubsection{Two-dimensional models}
In two dimensions,
constraints such as \eqr{eq:ground} are encountered in several models,
notably in two-dimensional ice (= six-vertex model), the
triangular Ising antiferromagnet ground state,
the square lattice dimer covering, and especially the
\kag Heisenberg antiferromagnet~\cite{kagome,huse,kondev}.
To compare the last of these to the pyrochlore Ising ground states, it is
fairest to consider the ground states of the three-state Potts model
antiferromagnet on the \kag lattice.
In that system, as in the pyrochlore, one can predict the structure factor
$\Struc{\kk}=0$ along the lines through the origin and
its first star of reciprocal lattice vectors; and here too
the scattering tends
to concentrate along a surface which is the Brillouin zone boundary,
scaled up by a factor of 2.
Following the analogy to the pyrochlore, one would guess
the scattering has a smooth maximum at the zone corner
$\QQ\equiv \{ {2\over 3}, {2\over 3}\}$
type points, but in this respect point the behavior differs from $d=3$.
To understand that, consider that \eqr{eq:divergence} is
satisfied formally by writing
\begin{equation}
\Pol(\rr) = \nabla \times \hh(\rr)
\end{equation}
where $\hh(\rr)$ is a ``vector potential''.
In $d=2$, there is no gauge freedom:
for a given configuration $\{\Pol(\rr)\}$,
$\hh(\rr)$ is {\it uniquely} determined (apart from a constant),
and can be visualized as
parametrizing a (rough) interface in a $2+1$ dimensional
space~\cite{blote}.
Following standard ``Kosterlitz-Thouless'' (also known as
``Coulomb-gas'') prescriptions, spin operators
have a component which is a periodic function of the
local $\hh(\rr)$. This implies correlations
with a power-law proportional to $1/\kappa$,
and the structure factor
must have a power-law cusp at $\QQ$
(which I will call a ``zone corner singularity'')
reflecting this quasi-long-range order~\cite{huse}.
Ref.~\onlinecite{moessner-kagice} have noticed
these ``zone-corner singularities'' in simulations;
they also appear when one
directly measures the fluctuations $\la |h(\qq)|^2 \ra$
of a discretely defined height field~\cite{henley-height,zeng}.
In such ``height models'' it is also possible that
the free energy favors a state with bounded fluctuations
of $\hh(\rr)$, corresponding to long-range order of the
spins.~\cite{huse,zeng}.
In the rare cases of a height
model in $d=3$ (e.g. three-state Potts antiferromagnet
on the simple cubic lattice), or in the ground state
of a $d=2$ quantum system,
this ``locking'' behavior is always expected, except in
some quantum models which contain nontrivial Berry phases.
Ref.~\onlinecite{moessner-kagice} have calculated
the asymptotic correlation function for
the dimer coverings of a triangular lattice,
a system which is realized in the ``spin-ice''
class of pyrochlore in a magnetic field
oriented along $\la 111 \ra$: see their
eqs.~(4.3)-(4.4).
It should be noted that two-dimensional dimer models
(which are solvable by the methods used for the
Ising model, i.e. free fermions), have
the peculiarity that
the correlations arising from the
``height'' field $\hh(\rr)$ have
exactly the same decay ($1/r^2)$ as
the pseudodipolar terms;
the contributions can be distinguished
because the first kind of correlation does not
depend on the orientation of the vector $\rr$
between sites, while the pseudo-dipolar kind does.
\LATER{CHECK THAT Ref.~\onlinecite{henley-height}
do both mention zone-corner singularities.}
\subsubsection{Quantum models}
For the $S=1/2$ case, there is
believed to be no spin order~\cite{canals98}.
The theory of this paper is not literally applicable
to small-$S$ auantum systems, since no wavefunction is
possible in which every tetrahedron is simultaneously
a singlet. Nevertheless, it is claimed~\cite{canals-garanin}
that the diffraction from exact diagonalizations
of the spin-1/2 case agrees well with formulas such as
\eqr{eq:struc-result}.
\OMIT{Critique Fisher et al for using 6-loop update:
is it really ergodic? Also, cite Newman et al
algorithm for ice.}
A pyrochlore Ising antiferromagnet, made into a
quantum model by a small transverse ring exchange,
has also defined the same coarse-grained field
$\Pol(\rr)$, which in their theory is called
an ``electric field''~\cite{hermele}.
Out of their quantum-mechanical
variables conjugate to $\Pol(\rr)$ they construct
a ``magnetic field''; the resulting $U(1)$ gauge
theory has gapless modes with $|\kk|$ dispersion
analogous to light waves, and correspondingly
there are power law correlations (though with
a different power law.) Ref.~\onlinecite{huse-dimer}
also studied a model with a polarization
(they call $\Pol(\rr)$ the ``magnetic'' field)
as a means to constructing quantum models with
no long range order and fractionalized excitations.
Recently, interesting phenomena have been observed in
certain (electron) {\it conductors} containing a
pyrochlore sublattice~\cite{fulde}, which are
speculated to be related to the frustration of
this lattice.
In particular,
heavy fermion behavior is seen~\cite{fulde}
in the spinel $\rm LiV_2 O_4$,
and unsual ferromagnetic behavior in pyrochlore $\rm Nd_2Mo_2O_7$
is ascribed to a Berry phase acquired by
the fermions in a spin background~\cite{nagaosa}.
Perhaps the coarse-grained polarization
field can help in modeling the long-wavelength behavior
of these systems.
There is also a speculation~\cite{ihm}
that the low-temperature state
of ice itself (neglecting the dipole couplings beyond the nearest
neighbor!) is dominated by proton tunneling with nontrivial
Berry phases.
Even though this model is (probably) not relevant to real ice,
its exotic ground state is of interest in its own right
and the $\Pol(\rr)$ field is likely to enter its decription.
\subsection{Diffraction singularities due to constraints}
Local constraints produce diffraction singularities
in other systems,
specifically in the so-called ``transition state"
of certain metal alloys~\cite{deridder}.
In an FCC lattice, let $ (\RR) =\pm 1$ represent two
different chemical species.
To model a state which has strong short-range order,
assume a constraint resembling \eqr{eq:ground},
\begin{equation}
\sum _m s(\RR+\uu_m) =0,
\label{eq:tr-st-con}
\end{equation}
where $\{ \RR+ \uu_m \}$ are the 12 nearest-neighbor sites.
After Fourier transforming, we obtain
\begin{equation}
F(\kk) {\tilde s}(\kk) =0,
\label{eq:tr-states}
\end{equation}
where $F(\kk) \equiv \sum _m \exp (i\kk\cdot \uu_m)$.
It follows from \eqr{eq:tr-states} this that the diffuse scattering
$\langle |{\tilde s}(\kk)|^2 \rangle$
is {\it zero everywhere} in reciprocal space,
except that it {\it diverges}~\cite{deridder}
along the two-dimensional surfaces defined by $F(\kk)=0$.
When these surfaces intersect the Ewald plane of an
electron diffraction experiment, they produce
well-known arcs observed in fcc metal alloys near ordering transitions
~\cite{deridder}.
Similar behavior is seen in the $\rm Na_{22}Ba_{14}CaN_6$ structure,
a triangular arrangement of rods each having an Ising degree of
freedom with ``antiferromagnetic'' correlations~\cite{tri-rods}.
Arcs are also seen in quasicrystals, where they are
ascribed to the constraints of tiling space~\cite{Ne95},
Eq.~\eqr{eq:tr-states} is a sharper singularity
than is found for the pyrochlore lattice in this paper.
The fundamental reason is that the ``spins'' in
\eqr{eq:tr-st-con} are on a
primitive Bravais (FCC) lattice, so in a lattice
of $N$ cells there are $N$ constraints and the same
number of variables.
In contrast, in the pyrochlore problem analyzed in the present paper,
there are four sites per primitive cell
and only two constraints.
Consequently \eqr{eq:tr-states} gets replaced by
the $4\times 2$ matrix equation \eqr{eq:orthog}.
Just as \eqr{eq:tr-states} has its singular surfaces at $\kk$
values where the number of constraints, so \eqr{eq:orthog}
has its pseudodipolar singularities
at the points (reciprocal lattice vectors) at which the
two equations are linearly dependent and reduce to one equation.
The above constraint-counting arguments have been
phrased so as to make clear how they might be
adapted to other systems. For example, the triangular
Ising antiferromagnet (in zero field) is a highly frustrated system
on a Bravais lattice, so one might naively expect
stronger kinds of singularity in its diffuse scattering.
However, in that case there is no equality constraint but
instead $\sum _{i\in \alpha}t_i =\pm 1$, so the whole
approach breaks down.
\acknowledgments
\OMIT {M. Kvale for unpublished collaboration and discussions on
the pyrochlore antiferromagnet.}
I thank R. Ballou and C. Broholm for sharing data before
publication,
J. F. Nagle and F. Stillinger for information on the ice model,
as well as O. Tchernyshyov, R. Moessner, and S. Sondhi for discussions.
%%% M. Kvale, unpublished (1993).
%%% Spin-wave calculation, including Hartree-Fock for the 4th order term,
%%% for some pyrochlore lattice states, along the lines of Chubukov's
%%% calculation for the large-$S$ \kag antiferromagnet.
This work was supported by the National Science Foundation
(NSF) grant no. DMR-0240953.
Part of it was completed at the Kavli
Institute for Theoretical Physics, supported
by NSF grant PHY99-0794.
\OMIT{
I really won't make a point of powder data.
It all looks the same -- a hump at about the right place, followed
by a valley. (Indeed, when Reimers back-transformed the scattering to
obtain radial correlations, there were some big differences
among his various papers of 1991 -- in one case he concluded nearest
neighbor ferromagnetic interactions, which seems bizarre.
That's just the data which most closely resembles the
expected hump, Fig. 9 on p. 5686 of [greeet91].
It is for $\rm Tb_2 Mo_2 O_7$.
Furthermore both Tb and Mo have moments, and he claims that
the dominant coupling is AFM Tb-Mo; therefore, this isn't
a clean example!}
\OMIT{The powder data from $\rm CsMnFeF_6$ in W. Kurtz et al,
in Sol. State. Comm. 18, 1479 (1976), seems to show such a hump.
But it is suspicious: Laurent L\'evy told me (1987) that
$\rm CsNiFeF_6$ and $\rm CsMnFeF_6$ have strong FM tendencies.
Fig. 4.9 in Liebmann quotes {\it single-crystal} data
from the same group
(M. Steiner et al, J. Phys. Soc. Jpn. Suppl. 52, 173 (1983) --
I have this somewhere]. It clearly doesn't look much like
the expectation from nearest-neighbors (e.g. Zinkin \& Harris;
a cruder version from MC is shown in
Liebmann's Fig. 4.12; including weakish nnn interactions was enough
to radically change it to look like the data (his Fig. 4.13).
}
\begin{thebibliography}{99}
\bibitem{ramirez}
%% Phase Transitions, c. 1996. -- must be a false memory
%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%
%% or p. 423 in {\it Handbook of Magnetic Materials}, vol. 13
%% (ed. K.J. H. Buschow), Elsevier Science, Amsterdam, 2001.
%% pp 423-520
%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%
%% also: Geometrical frustration in magnetism
%% Ramirez AP
%% CZECHOSLOVAK JOURNAL OF PHYSICS 46: 3247-3254 Suppl. 6 1996
%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%
A. P. Ramirez, Annu. Rev. Mater. Sci 24, 453 (1994).
%% 453-480 1994
%% "STRONGLY GEOMETRICALLY FRUSTRATED MAGNETS"
% Greedan JE
% Geometrically frustrated magnetic materials
% J MATER CHEM 11 (1): 37-53 2001
%%%% PYROCHLORE EXAMPLES
%%% do spinels ZnFe2O4??
%%% spinel ZnCr2O4
\bibitem{lee00}
S. H. Lee, C. Broholm, T. H. Kim, W. Ratcliff II, and S. W. Cheong,
Phys. Rev. Lett. 84, 3718 (2000).
%% ``Local Spin Resonance and Spin-Peierls-like Phase Transition in a
%% Geometrically Frustrated Antiferromagnet''
\bibitem{lee02}
S-H Lee, C. Broholm, W. Ratcliff, G. Gasparovi\'c,
Q. Huang, T. H. Kim, and S.-W. Cheong, Nature 418, 856 (2002).
%%% ``Emergent excitations in a geometrically frustrated magnet''
%% (ZnCrO$_4$).
%%% Fig 3(a,b) is expt, cd is their sim.
\bibitem{harris94}
%%% ``Magnetic structure of the spin-liquid state in a frustrated pyrochlore''
M. J. Harris, M. P. Zinkin, Z. Tun, B. M. Wanklyn, and I. P. Swainson
%% CsNiCrF6
Phys. Rev. Lett. 73, 189 (1994).
%%% powder diffraction: no LRO
\bibitem{harris97a}
M. J. Harris, M. P. Zinkin, and T. Zeiske, Phys. Rev. B 56, 11786 (1997)
%%% ``Short-range magnetic order in the frustrated pyrochlore antiferromagnet
%%% CsNiCrF$_6$''
%% 11786-90
%%% Did MC simulation, single xtal: Heisenberg, Metropolis:
%% ALSO Simulation with pyrochlore diffraction.
\bibitem{Y2Mo2O7}
J. N. Reimers, J. E. Greedan, R. K. Kremer, E. Gmelin,
and M. A. Subramanian,
Phys. Rev. B 43, 3387 (1991);
%%% ``Short-range magnetic ordering in the highly frustrated '
%%% pyrochlore Y2Mn2O7'' 3387-3394
%%% spin-glass w/o struc.disorder
%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-d-%-%-%-%-%-%-%-%-%-%-%
%%% M. J. P. Gingras, C. V. Stager, N. P. Raju, B. D. Gaulin,
%%% and J. E. Greedan, Phys. Rev. Lett. 78, 947 (1997);
%%% Y2Mo2O7: spin glass expts, found chi-NL exponents
%% ``Glassy Statics and Dynamics in the Chemically Ordered Pyrochlore
%% Antiferromagnet Y2Mo2O''
%%% Y2Mo2O7: chemically ordered but spin glass
%%% Greedan, c. 1991, asserted spin-glass freezing w/o disorder;
%%% see Can J Phys more recent.
%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-d-%-%-%-%-%-%-%-%-%-%-%
J. S. Gardner, B. D. Gaulin, S.-H. Lee, C. Broholm, N. P. Raju,
and J. E. Greedan, Phys. Rev. Lett. 83, 211 (1999);
%%% Y2Mo2O7/Tb2Mo2O7: Heisenberg type
%%% neutron diff.
C. H. Booth, J. S. Gardner, G. H. Kwei, R. H. Heffner, F. Bridges, and
M. A. Subramanian, Phys. Rev. B 62, R755 (2000)
%%% EXAFS, there is bond disorder, after all.
%-%-%-%-%-
% [greeet91] J. Greedan, Reimers, Stager, and Penny,
% PRB 43, 5682 (1991)
%%% Not pertinent (?)
\bibitem {shi93}
M. Shiga, K. Fujisawa, and H. Wada,
J. Phys. Soc. Jpn. 62, 1329 (1993).
%%% ``SPIN LIQUID BEHAVIOR OF HIGHLY FRUSTRATED Y(SC)MN2 AND
%%% EFFECTS OF NONMAGNETIC IMPURITY''
\bibitem{ballou}
R. Ballou, E. Leli\`evre-Berna, and B. F{\aa}k, Phys. Rev. Lett.
76, 2125-8 (1996).
%%%$\rm Gd3Ga_5O_{12}$.
%%% ``Spin Fluctuations in (Y$_{0.97}$Sc$_{0.03}$)Mn$_2$: %%% A Geometrically
%%% Frustrated, Nearly Antiferromagnetic, Itinerant Electron System''.
%%% See Fig. 3
%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-
%% NOT APPROPRIATE
%% \bibitem{balet92}
%% R. Ballou, B. Ouladdiaf, P. J. Brown,
%% M. D. Nunez Regueiro, and C. Lacroix,
%% Phys. Rev. B45, 3158 (1992)
%% 3158-60.
%%``Unusual field-induced transition in a frustrated itinerant antiferromagnet''
\bibitem {bernal-pauling}
J. D. Bernal and R. H. Fowler, J. Chem. Phys.
1, 515 (1933);
%%% 515-548. Ice rules!
L. Pauling, J. Am. Chem. Soc. 57, 2680 (1935).
%%% approximation for ice model entropy.
\bibitem{youngblood}
R. W. Youngblood and J. D. Axe, Phys. Rev. B 23, 232 (1981).
%%% 232-8 (1981),
%%% ``Polarization fluctuations in ferroelectric models''.
\bibitem {an56}
P. W. Anderson,
Phys. Rev. 102, 1008 (1956).
%%% 1008-13 (1956),
%%% ``Ordering and antiferromagnetism in ferrites''. %%% pyrochlore
\bibitem{tcherny}
%%% structural distortion papers.
%%% ``Order by Distortion and String Modes in Pyrochlore Antiferromagnets
O. Tchernyshyov, R. Moessner, and S. L. Sondhi,
Phys. Rev. Lett. 88, 067203 (2002)
%% 4 pp.
%%% I omitted PRB,Spin-Peierls phases in pyrochlore antiferromagnets.
%%%%%%%%%%%%%%%%% SPIN ICE
\bibitem{harris97b}
M. J. Harris, S. T. Bramwell, D. F. McMorrow, T. Zeiske, and K. W. Godfrey
Phys. Rev. Lett. 79, 2554-7 (1997),
%%% 2554-7 (1997),
%%% ``Geometrical frustration in the ferromagnetic pyrochlore Ho$_2$Ti$_2$O$_7$
%%% coin the term ``spin ice''.
\bibitem{ramirez-spinice}
A. P. Ramirez, A. Hayashi, R. J. Cava, R. Siddharthan,
and B. S. Shastry, Nature 399, 333-5 (1999).
%%% ``Zero-point entropy in `spin ice' ''
%%% expts and measurements
%%% 333-5 (1999).
\bibitem{Si99}
R. Siddharthan, B. S. Shastry, A. P. Ramirez, A. Hayashi,
R. J. Cava, and S. Rosenkranz,
Phys. Rev. Lett. 83, 1854 (1999),
%%% 1854-7 (1999),
%%% ``Ising pyrochlore magnets: Low-temperature properties, `ice rules,'
%%% and beyond''.
\bibitem{Bra01a}
S. T. Bramwell and M. J. P. Gingras,
Science 294, 1495 (2001).
%% 1495-1501 (2001).
%%% ``Spin ice state in frustrated magnetic pyrochlore materials''
\bibitem{Bra01b}
S. T. Bramwell et al.
M. J. Harris, B. C. den Hertog, M. J. P. Gingras,
J. S. Gardner, D. F. McMorrow, A. R. Wildes, A. L. Cornelius,
J. D. M. Champion, R. G. Melko, and T. Fennell,
Phys. Rev. Lett. 87, 047205 (2001).
%%%4pp;
%%% ``Spin Correlations in Ho$_2$Ti$_2$O$_7$: A Dipolar Spin Ice System''
see their Fig. 1.
%%% R. G. Melko, B. C. den Hertog, and M. J. P. Gingras,
%%% Phys. Rev. Lett. 87, 067203 (2001), 4 pp.
%%% ``Long-range order at low temperatures in dipolar spin ice''
%%% Dy$_2$Ti$_2$O$_7$ and Ho2Ti2O7
%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%
%%% \bibitem {Den00}
%%% B. C. den Hertog and M. J. P. Gingras,
%%% Phys. Rev. Lett. 84, 3430-3 (2000).
%%% ``Dipolar interactions and origin of spin ice in Ising pyrochlore magnets''
\OMIT{Cite CsMnFeF6: Old, old diffraction papers about $\rm CsMnFeF_6$.}
\bibitem{Tb2Ti2O7}
J. S. Gardner, B. D. Gaulin, A. J. Berlinsky, P. Waldron,
S. R. Dunsiger, N. P. Raju,and J. E. Greedan,
Phys. Rev. B64, 224416 (2001)
%%% Tb2Ti2O7: paramagnet down to 0.1K (but SRO below 50K)
%% ``Neutron scattering studies of the cooperative paramagnet
%%% pyrochlore Tb2Ti2O7'' %% 10pp. %%% Ising, 111-like.
%%% they present, and fit the diffuse by nearest-nbr correls
%%% they think dipole enters the act?
\bibitem{moessner98a}
R. Moessner and J.~T.~Chalker,
%% ``Properties of a classical spin liquid: The Heisenberg pyrochlore antiferromagnet''
Phys. Rev. Lett. 80, 2929 (1998).
%% 2929-2932 (1998).
\bibitem{moessner98b}
R. Moessner and J. T. Chalker,
Phys. Rev. B 58, 12049-12062 (1998)
%% ``Low-temperature properties of classical geometrically frustrated antiferromagnets''
%% II D 1 argues about the diffuse scatt.
\bibitem{moessner-dipolar}
S.~V.~Isakov, K.~Gregor, R.~Moessner, ad S.~L.~Sondhi,
preprint, ``Dipolar spin correlations in classical pyrochlore
antiferromagnets.''
% \bibitem{Kao02}
% preprint (cond-mat/0207270),
% Y.-J. Kao, M. Enjalran, A. Del Maestro, H. R. Molavian, and M. J. P. Gingras
% Phys. Rev. B 68, 172407 (2003).
% {\sl SO FAR, CITED ONLY ABOUT FIG'S.}
%%% ``Understanding Paramagnetic Spin Correlations in
%%% the Spin-Liquid Pyrochlore Tb$_2$Ti$_2$O$_7$''.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%% "By including single-ion excitations from the ground state doublet to
%%% higher crystal field levels, we successfully describe the elastic neutron
%%% scattering pattern and dispersion relations in Tb2Ti2O7, quantitatively
%%% consistent with experimental observations.
\bibitem {henley-HFM00}
C.~L.~Henley, Can. J. Phys. 79, 1307 (2001).
%% 1307-21 (Dec. 21, 2001).
%%% Effective Hamiltonian
%% ``Effective Hamiltonians and dilution effects in
%% Kagom\'e and related antiferromagnets'',
\bibitem{broholm-SCGO}
%%%`` Isolated Spin Pairs and Two-Dimensional Magnetism in SrCr9pGa12-9pO19''
S.-H. Lee, C. Broholm, G. Aeppli, T. G. Perring, B. Hessen, and A. Taylor
Phys. Rev. Lett. 76, 4424 (1996).
\bibitem{kawamura}
%% ``Ordering of the Antiferromagnetic Heisenberg Model on a Pyrochlore Slab
T. Arimori and H. Kawamura,
J. Phys. Soc. Jpn. 70, 3695 (2001).
%% \bibitem {haset86}
%% J. M. Hastings, L. M. Corliss, W. Kunnmann, and D. Mukamel,
%% Phys. Rev. B 33, 6326 (1986). %%% $MnTe_2$
\bibitem{petrenko}
O. A. Petrenko and D. McK. Paul,
Phys. Rev. B 63, 024409 (2001)
%%% ``Classical Heisenberg antiferromagnet on a garnet lattice:
%%% A Monte Carlo simulation'' %%% [8 pages]
\bibitem {hen87}
C. L. Henley,
J. Appl. Phys. 61, 3962 (1987).
%%% ``Ordering by disorder: ground state selection in FCC antiferromagnets'',
\bibitem {larhen90}
B. E. Larson and C. L. Henley, unpublished, 1990.
{\sl ``Ground State Selection in Type III FCC Vector Antiferromagnets''}
\bibitem {shen82}
E. F. Shender, Sov. Phys. JETP {56}, 178 (1982).
%%% collinearity.
\bibitem {hen89}
C. L. Henley, Phys. Rev. Lett. 62, 2056 (1989).
\bibitem {oja}
M. T. Heinil\"a and A. S. Oja,
Phys. Rev. B 48, 16514 (1993).
%%% "Long-range order produced by the interaction between spin waves
%%% in classical fcc Heisenberg models" 10 pp.
%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-
%% H. E. Viertio and A. S. Oja,
%% Phys. Rev. B 36, 3805 (1987);
%%% GROUND-STATE OF NUCLEAR SPINS IN FCC METALS
%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-
%% {\sl After I drafted this paper, there was a review by
%% A. S. Oja, and O. V. Lounasmaa, Rev. Mod. Phys. 69, 1 (1997).}
\bibitem{ritchey}
I. Ritchey, P. Coleman and P. Chandra, Phys. Rev. B 47, 15342 (1993).
\bibitem{pyro-Heff}
C.L. Henley and U. Hizi, unpublished.
\bibitem{chan}
%%% ``Ground state selection in Kagom\'e antiferromagnet''
C. L. Henley and E. P. Chan,
J. Mag. Mag. Mater. 140-144, 1693 (1995);
E. P. Chan, Ph. D. Thesis, Cornell University, 1994.
\bibitem{chubukov}
A. Chubukov, {Phys. Rev. Lett.}, 69, 832 (1992).
%%%% See also A. Chubukov, {\it J. Appl. Phys.}, {\bf 73}, 5639(1993).
\bibitem {liebmann86}
R. Liebmann,
{\it Statistical Physics of Periodically Frustrated Ising Systems}
(Springer Lecture Notes on Physics No. 251, 1986).
\bibitem
{Hen92} C. L. Henley, Bull. Amer. Phys. Soc.
37, 441 (1992, APS March Meeting).
%%% ``Algebraic correlations in the 3D ice model''.
%%% 441-2
\bibitem{chancole91}
``Spin nematic'' is defined as a state
with a symmetry-breaking to preferred subspace of spin space
while $\la \ss_i \ra =0$ for all spins.
[P. Chandra and P. Coleman
Phys. Rev. Lett. 66, 100 (1991).]
%% Spin nematic state...
\bibitem{stillinger}
F. H. Stillinger and M. A. Cotter,
J. Chem. Phys. 58, 2532 (1973).
%%% ``Local orientational order in ice.''
\bibitem{villain}
J. Villain and J. Schneider,
in {\it Physics and Chemistry of ice}, ed. E. Whalley,
S. J. Jones, and L. W. Gold
(Royal Society of Canada, Ottawa, 1973);
see also J. Villain, Sol. St. Comm. 10, 967 (1972).
%%% OTHERS ABOUT THE CORRELATIONS
\bibitem {nagle}
Since the arrows in real ice are associated with real dipole moments,
this entropic effect appears there as a modification of the dielectric constant:
see J. F. Nagle, J. Chem. Phys. 61, 883 (1974);
%%% \OMIT {(``Dielectric constant of ice:'' arguments about $G$ vs. $g$);}
Chem. Phys. 43, 317 (1979).
%%% ``Theory of the dielectric contant of ice.''}
\bibitem{moessner-kagice}
%%% ``Theory of the [111] magnetization plateau in spin ice''
R. Moessner and S. L. Sondhi, Phys. Rev. B68, 064411 (2003).
%%% kagome-ice; DyTiO7 also.
%%% ...The application of a magnetic field along the [111] direction ...
\bibitem{zeng}
%% ``Zero-temperature phase transitions of an antiferromagnetic Ising
%% model of general spin on a triangular lattice''
C. Zeng and C. L. Henley,
Phys. Rev. B 55, 14935 (1997).
%%% 14935-947 (1997).
\bibitem{henley-height}
R. Raghavan, C. L. Henley, and S. L. Arouh,
J. Stat. Phys. 86, 517 (1997);
%%% 517-550 (Feb. 1997);
%%% ``New two-color dimer models with critical ground states''
%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-%-
J. Kondev and C. L. Henley,
Phys. Rev. B 52, 6628 (1995).
%% 6628-39 (1995).
%%% `` Four coloring model on the square lattice: a critical ground state
\OMIT{Did Burton and Kondev actually mention these?}
\bibitem {zinhar95}
M. P. Zinkin and M. J. Harris, J. Mag. Mag. Mater.
140-144, 1803 (1995).
%% 1803-1804. ``Local mean field calculation of the
%% frustrated pyrochlore spin structure''
%% SIMULATION: quench to a classical gr.state
%% shows nice single xtal results w/pinch points
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\bibitem{huse-dimer}
%% ``Coulomb and Liquid Dimer Models in Three Dimensions''
D. A. Huse, W. Krauth, R. Moessner, and S. L. Sondhi,
Phys. Rev. Lett. 91, 167004 (2003).
\bibitem{reimers92a}
J. N. Reimers, Phys. Rev. B 45, 7287 (1992).
%%% 7287-94 (1992).
%% 7287-94 (1992).
%%% `Absence of long-range order in a three-dimensional geometrically
%%% frustrated antiferromagnet"
% %% ``Absence of long-range order in a three-dimensional geometrically
% %% frustrated antiferromagnet'' (MC resuults)
% (the data I need is quoted in [reimers92b])
%%% Cite for lack of LRO.
\bibitem{garanin} D. A. Garanin and B. Canals,
Phys. Rev. B59, 443(1999).
%%% 443-459
%%% ``Classical spin liquid: exact solution for
%%% the infinite-component antiferromagnetic model on the kagomé lattice
\bibitem{canals-garanin}
B. Canals and D. A. Garanin,
Can. J. Phys. 79, 1323 (2001),
%%% 1323-1331 (2001),
%%% ``Spin-liquid phase in the pyrochlore antiferromagnet.''
%%% See their Fig. 4.
\bibitem {reimers92b}
J. N. Reimers, Phys. Rev. B 46, 193 (1992).
%% 193-202
%% ``Diffuse-magnetic-scattering calculations for frustrated
%% antiferromagnets''
%%% The bastard powder-averaged everything, so I can't test it.(!)
\bibitem{yoshida}
S. Yoshida, K. Nemoto, and K. Wada,
J. Phys Soc. Jpn. 71, 948 (2002)
%% 948-954 MAR 2002
%% ``Application of the cluster variation method to spin ice systems on the
%% pyrochlore lattice''
\bibitem{harris95}
M.~J.~Harris, M.~P.~Zinkin, Z.~Tun, B.~Wanklyn,
and I.~Swainson,
J. Mag. Mag. Mater. 140-144, 1763 (1995)
%% 1763-64. [ICM'94]
%% A single-crystal neutron-scattering study of a
%% highly frustrated pyrochlore
%% fit to a single tetrahedron
\bibitem{moessner-berlinsky}
R. Moessner and A. J. Berlinsky,
Phys. Rev. Lett. 83, 3293 (1999)
%% ``Magnetic susceptibility of diluted pyrochlore and
%% SrCr9-9xGa3+9xO19 antiferromagnets''
%%% 3293-7 (1999)
%%% toy model for diffraction.
\bibitem{garcia-huber}
A. J. Garcia-Adeva and D.~L.~Huber,
Phys. Rev. Lett. 85, 4598 (2000).
%% 85 (21): 4598-4601 NOV 20 2000
%% ``Quantum tetrahedral mean field theory of the magnetic susceptibility
%% for the pyrochlore lattice''
\bibitem{canals98}
B. Canals and C. Lacroix, Phys. Rev. Lett. 80, 2933 (1998).
%%% 2933-6 (1998).
%%% ``Pyrochlore Antiferromagnet: A Three-Dimensional Quantum Spin Liquid''.
\bibitem{Can00}
B. Canals and C. Lacroix, Phys. Rev. B 61, 1149 (2000).
%%% ``Quantum spin liquid: The Heisenberg antiferromagnet on the
%%% three-dimensional pyrochlore lattice''
%% \bibitem {haret91}
%% A. B. Harris, A. J. Berlinsky, and C. Bruder,
%% ``ORDERING BY QUANTUM FLUCTUATIONS IN A STRONGLY FRUSTRATED
%% HEISENBERG-ANTIFERROMAGNET
%% J. Appl. Phys. 69, 5200 (1991).
%% \bibitem{reiet91b}
%% J. N. Reimers, A. J. Berlinsky, and A.-C. Shi,
%% Phys. Rev. B 43, 865 (1991).
%% {\sl Mean-field approach; ignore this.}
\bibitem{FN-Gausslaw}
With this normalization of
charge, each ice-arrow must carry a discrete flux of $1/2$.
By the same rule,the maximum-polarization state
(which by \eqr{eq:Pol-alpha} is $\Pol=(1,0,0)$) carries a flux
per unit area of $2/a^2$ across a $(100)$ plane.
Hence, to satisfy Gauss's law, we must use a coarse-grained
polarization $2 P(\rr)/a^2$.
\bibitem{snyder03}
J. Snyder, B. G. Ueland, J. S. Slusky, H. Karunadasa, R. J. Cava, A. Mizel,
and P. Schiffer, Phys Rev. Lett. 91, 107201 (2003).
%%% ``spin ice'' $\rm Dy_2Ti_2O_7$
%%% ``Quantum-Classical Reentrant Relaxation Crossover in Dy2Ti2O7 Spin Ice''
\bibitem{snyder-dilution}
J. Snyder, J. S. Slusky, R. J. Cava, and P. Schiffer,
Phys. Rev. B66, 064432 (2002).
%%% preprint (cond-mat/0205020),
%%% ``Dirty Spin Ice: The Effect of Dilution on Spin Freezing in Dy$_2$Ti$_2$O$_7$''
%%% ...We have studied spin freezing in the diluted spin ice compound
%%% Dy2-xYxTi2O7 where the nonmagnetic Y ions replace the magnetic Dy ions
%%% on the frustrated pyrochlore lattice...
\bibitem {konhen95}
J. Kondev and C. L. Henley,
%%% ``Kac-Moody symmetries of critical ground states''
Nuc. Phys. B 464, 540-575 (1996).
\bibitem{hermele}
%%% ``Pyrochlore photons: The U(1) spin liquid in a S = (1/2)
%%% three-dimensional frustrated magnet
M. Hermele, M. P. A. Fisher, and L. Balents,
Phys. Rev. B 69, 064404 (2004)..
\bibitem{balents}
%%% ``Fractionalization in an easy-axis Kagome antiferromagnet
L. Balents, M. P. A. Fisher, and S. M. Girvin,
Phys. Rev. B 65, 224412 (2002).
\bibitem{deridder}
R. De Ridder, G. van Tendeloo, and S. Amelinckx,
Acta Crystallogr. A 32, 216 (1976).
%%% ``Cluster model for transition from short-range order
%%% to long-range order in FCC based binary-systems and its
%%% study by means of electron-diffraction'';
R. De Ridder, G. Van Tendeloo, D. van Dyck, and S. Amelinckx,
Phys. Stat. Sol. 38, 663 (1976); Phys. Stat. Sol. 40, 669 (1977).
%%% R. De Ridder, G. van Tendeloo, D. van Dyck, and S. Amelinckx,
%%% Phys. Stat. Sol. 40, 669 (1977),
%% 669-683 (1977),
%%% ``Cluster model for transition-state
%%% and its study by means of electron-diffraction.
%%% 2. Application to some particular systems''
%%% ``transition states'' model
%%% {De76} -> {De76b} R. De Ridder, G. van Tendeloo, D. van Dyck, and S. Amelinckx,
%%% Phys. Stat. Sol. 38, 663-674 (1976),
%%% ``Cluster model for transition-state
%%% and its study by means of electron-diffraction'';
\bibitem
{Ne95} M. E. J. Newman, C. L. Henley, and M. Oxborrow,
Philos. Mag. 71, 991 (1995), and references therein.
%%% only for diffuse
\OMIT{\bibitem{spintunnel}
%%% CLIPPED references in old PYRICE
J.~von~Delft and C.~L.~Henley,
Phys. Rev. Lett. 69, 3236 (1992);
%%% ``Destructive quantum interference in spin tunneling problems''
D. Loss, D. P. DiVincenzo, and G. Grinstein,
Phys. Rev. Lett. 69, 3232 (1992).
%%%``Suppression of Tunneling by interferenmce in half-integer-spin particles''.
}
%%% \bibitem {vil79}
%%% J. Villain. Z. Phys. B 33, 31 (1979).
%% 31-42
%%% ``Insulating spin glasses''
%%% {\sl Does this really have the ice mapping?!!}
\bibitem{zhang02}
%% ``Effective Hamiltonian and low-lying eigenstates
%% of four-sublattice antiferromagnets'', by
N.-G.~Zhang, C.~L.~Henley, C.~Rischel, and K.~Lefmann,
Phys. Rev. B 65, 064427 (2002).
\bibitem{huse} D.A. Huse and A.D. Rutenberg,
Phys. Rev. B 45, 7536 (1992).
\bibitem{kondev}
%% ``Kac-Moody symmetries of critical ground states''
J. Kondev and C. L. Henley,
Nuc. Phys. B 464, 540 (1996).
%%% 540-575 (1996).
\bibitem{blote} H.~W.~J.~Bl\"{o}te and H.~J.~Hilhorst,
J.~Phys. A 15, L631 (1982);
B. Nienhuis, H.J. Hilhorst, and H.W. Bl\"{o}te,
J. Phys. A 17, 3559 (1984).
\bibitem{kagome}
J. Chalker, P. C. W. Holdsworth and E. F. Shender,
%%% HIDDEN ORDER IN A FRUSTRATED SYSTEM - PROPERTIES OF THE HEISENBERG KAGOME
%%% ANTIFERROMAGNET
Phys. Rev. Lett., 68, 855 (1992).
\bibitem{fulde}
P. Fulde, J. Phys.: Condens. Matter 16, S591 (2004).
%%(24 March 2004) S591-S598
%% HFM2004 proceedings
%%% ``Itinerant d electrons in spinels''
\bibitem{ihm}
J. Ihm, J. Phys. A 29, L1 (1996).
%%% Residual entropy of ice: a manifestation of the fractional exclusion
%%% statistics in real three-dimensional space
%%% 1 (7 January 1996) L1-L7
\bibitem{nagaosa}
%% ``Spin chirality, Berry phase, and anomalous Hall effect in
%% a frustrated ferromagnet
Y. Taguchi, Y. Oohara, H. Yoshizawa, N. Nagaosa, and Y. Tokura,
Science 291, 2573 (2001).
%%% (5513): 2573-2576 MAR 30 2001
%-%-%-%-%-%-%
%%% ``Spin chirality fluctuations and anomalous Hall
%%% effect in itinerant ferromagnets
S. Onoda and N. Nagaosa, Phys. Rev. Lett.
90, 196602 (2003).
%%% Nd2Mo2O7
\bibitem{tri-rods}
U. Steinbrenner and A. Simon, Z. Kristallogr. 212, 428 (1997).
%%% 428-438.
%%% ``Structural frustration in a rod packing -- An analogy
%%% to the disordered triangular Ising net
\end{thebibliography}
\begin{figure}[ht]
\includegraphics[width=1.0\linewidth]{pyro-ice.eps}
\caption{
\LATER{NEED TO REDRAFT FIGURE TO MAKE IT VISUALLY CLEAR.}
(a) Ising ground state on fragment of pyrochlore lattice;
spins are shown on edges of a diamond lattice.
(b). The corrsponding ice-model arrows between the diamond vertices.}
%%% (c). polarizations on the even diamond vertices.
\label{fig:lattice}
\end{figure}
\begin{figure}[ht]
\includegraphics[width=1.6\linewidth]{pyrn-conxyy.ps}
\caption{
\LATER{CHECK I USED THE NEW, NOT 1995 FORMULA.}
Calculated structure factor in the plane normal to $[1 \bar 1 0]$.
(The figure is the same slice as shown in
Refs.~\protect\onlinecite{zinhar95} and
\protect\onlinecite{harris95}.)}
\label{fig:strucf}
\end{figure}
\end{document}