Spectrum of elementary and collective excitations in the dimerized S=1/2 Heisenberg chain with frustration
Extracted source text ยท arXiv cond-mat/9901302
%\documentstyle[prl,aps]{revtex}
\documentstyle[preprint,aps]{revtex}
\begin{document}
\draft
\tightenlines
\title{{Spectrum of elementary and
collective excitations in the dimerized S=1/2 Heisenberg chain with frustration}}
\author{P.V. Shevchenko$^{a}$, V.N. Kotov$^{b}$, and O.P. Sushkov$^{c}$}
\address
{School of Physics, The University of New South Wales,
Sydney 2052, Australia}
\date{11 December 1998}
\maketitle
\begin{abstract}
We have studied the low-energy excitation spectrum of a dimerized and frustrated
antiferromagnetic Heisenberg chain. We use
an analytic approach,
based on a description of the excitations as triplets above a strong-coupling
singlet ground state.
The quasiparticle spectrum is calculated by treating the excitations as a dilute Bose gas with infinite on-site repulsion.
Additional singlet (S=0) and triplet (S=1) modes are found as two-particle bound states of the elementary triplets. We have also calculated the
contributions of the elementary and collective excitations into the spin structure
factor. Our
results are in excellent agreement with exact diagonalizations and dimer
series expansions data as long as the dimerization parameter
$\delta$ is not too small ($\delta>0.1$), i.e. while the
elementary triplets can be treated as localized objects.
\end{abstract}
%Corresponding author: Pavel Shevchenko, School of Physics, The University of
% New South Wales, Sydney, 2052, Australia ph.(612)9385 61 32, fax(612)9385 60 60, e-mail: pavel@newt.phys.unsw.edu.au
\pacs{PACS: 75.10.Jm, 75.30.Ds, 75.40.Gb}
%keywords: Heisenberg antiferromagnet, spin liquid}
\baselineskip 0.6cm
\section{Introduction}
The properties of a variety of recently discovered quasi one-dimensional materials can
be described by the Alternating Heisenberg Chain (AHC) model.
The Hamiltonian of the model reads:
\begin{equation}
\label{Ham}
H=J\sum_i [(1+\delta(-1)^i){\bf S}_i .{\bf S}_{i+1}+\alpha{\bf S}_i .{\bf S}_{i+2}]
,\end{equation}
where $i$ denotes the sites of a chain with length $N$ and ${\bf S}_i$ are $S=1/2$
spin operators. The parameter $\alpha$ is the next-nearest neighbor coupling,
leading to frustration, and $\delta$ is the dimerization. We assume $J>0$.
The AHC is realized in nature in many materials that have two important but structurally inequivalent superexchange paths that are spatially linked, so that an alternating spin-spin interaction results. Representative examples of materials of this type are $(VO)_2P_2O_7$ \cite{Garret} and various aromatic free-radical compounds \cite{Nordio}. Alternating chains may also arise as a result of spontaneous dimerization, due to dynamical spin-phonon coupling.
Examples include the recently discovered inorganic spin-Peierls compounds
$CuGeO_3$ \cite{Hase} and $\alpha'-NaV_2O_5$ \cite{Isobe}.
The AHC model (\ref{Ham}) is a straightforward generalization of the uniform Heisenberg chain, which is the most widely studied quantum spin system. The uniform $S=1/2$ chain has a gapless excitation spectrum with a known dispersion relation and a rather complicated ground state which is characterized by strong quantum fluctuations, making it highly unstable to perturbations. The AHC generalizes the uniform chain by alternating the spin-spin interaction between two values, $J(1+\delta)$ and $J(1-\delta)$, and including next-nearest neighbor coupling (frustration) $\alpha J$. Since the Hamiltonian (\ref{Ham}) is rotationally invariant with respect to spin, the total spin is a good quantum number, and the ground state is a spin singlet. The translational symmetry however is broken by the dimerization, and the resulting system has a gap to the first excited state with $S=1$.
% in accordance to the Lieb-Schultz-Mattis theorem \cite{Lieb}.
The model has a rather complicated spectrum of states at higher energies, including multimagnon continua and bound states.
Recent neutron scattering studies of $(VO)_2P_2O_7$ carried out by Garrett {\it et al.}
\cite{Garret,Garret1,Garret2} show that this material can be described well by the
alternating spin chain model. In addition, these measurements
provide evidence for the existence of a
two-magnon triplet ($S=1$) bound state. Similar studies of the spin-Peierls
material $CuGeO_3$ by Ain {\it et al.}\cite{Ain}, in combination with
Raman spectroscopy \cite{Paul}, suggest that a singlet $S=0$ bound
state is present in this material.
These experiments have strongly motivated the theoretical studies of the
low-energy excitations of the AHC model, since the existence of magnetic bound states
is one of its characteristic features.
More generally, bound states seem to appear in all
low-dimensional dimerized quantum antiferromagnets which have a
gapped triplet spectrum above a singlet ground state \cite{Sushkov1}.
Besides the relevance to real compounds, the model (\ref{Ham}) is interesting from theoretical point of view since it contains two independent mechanisms for spin gap formation. Subject of a special interest is the nature of excitations. At $\delta=0$ the model
exhibits two phases, separated by a critical value of frustration $\alpha_c=0.2411$, known from numerous studies \cite{Okamoto}. For $\alpha<\alpha_c$ the ground state
is similar to that of the uniform Heisenberg chain and frustration is irrelevant in
this parameter regime.
The elementary excitations are massless
unconfined spinons with spin S=1/2 \cite{ft}. At $\alpha=\alpha_c$ there is a transition into a
phase with
spontaneously dimerized two-fold degenerate ground state. The spectrum acquires a gap and the elementary excitations are massive spinons \cite{Haldane,Shastry,Chitra}. On the other hand Haldane \cite{Haldane} has shown that for any $\delta\neq 0$ the spinons become confined into
triplet $S=1$ excitations with a gap in the spectrum.
Interactions between the triplets can lead to additional massive singlet, $S=0$, and triplet, $S=1$, excitations below the continuum \cite{Uhrig,Shastry} and even a sequence of further massive excitations \cite{Affleck1}. As $\delta\rightarrow 0$ the size of the elementary triplet
(or, equivalently, the spinon radius of confinement), $r_0$, is known to scale as \cite{Uhrig1,Affleck2}: $r_0\sim \delta^{-2/3}$ for $\alpha<\alpha_c$ and $r_0\sim \delta^{-1/3}$ for $\alpha>\alpha_c$. Therefore
the triplets become nonlocal objects for small dimerization, ultimately giving way to
completely unconfined spinons in the strict limit $\delta=0$.
The excitation spectrum of the AHC has been extensively studied by
a variety of numerical methods.
Most recent work includes: Exact Diagonalizations (ED) carried out
by Bouzerar, Kampf, and Japaridze \cite{Bouzezar} and Barnes, Riera and Tennant \cite{Barnes},
dimer series expansions by Singh and Weihong \cite{Singh}, and density matrix renormalization
group (DMRG) studies by Sorensen {\it et al.} \cite{Affleck}.
However there are only few analytical works studying the AHC model.
Uhrig and Schulz\cite{Uhrig} have used the continuum limit field theory approach
and predicted the possibility of singlet (S=0) and triplet (S=1) two-magnon bound states below
the two-particle continuum in the dimerized phase ($\delta>0$). Affleck \cite{Affleck1} and Uhrig {\it et al.}\cite{Uhrig1} developed a description, based on the soliton picture in which the elementary excitations are treated as
confined $S=1/2$ solitons (spinons). It was demonstrated that $S=1,0$ soliton anti-soliton bound states are formed and their number increases
as $\delta\rightarrow 0$.
Bouzerar and Sil \cite{Bouzezar1} have studied the excitation spectrum of the dimerized phase using the Bond Operator Technique (BOT) introduced by Chubukov\cite{Chubukov} and Sachdev and Bhatt\cite{Sachdev}, in combination with the Brueckner approach developed by
Kotov, Sushkov, Weihong, and Oitmaa \cite{Sushkov}. The results of Ref.\cite{Bouzezar1} are mainly inconsistent with the ED data. They are also inconsistent with our results in spite of the fact that we use the same approach. The reason for this discrepancy, as will become clear from
our presentation below, is that the treatment of Ref.\cite{Bouzezar1}
is quite incomplete and misses several important
technical aspects of the problem.
The aim of the present work is to investigate theoretically
the excited states of the dimerized phase ($\delta>0$) of the AHC by using the
analytic Brueckner approach. Previous applications of the Brueckner approach include the two-layer
Heisenberg model \cite{Sushkov,Shevchenko}, the quantum spin-ladder \cite{Sushkov1} as well as the spin ladder with frustration \cite {Kotov}. In all
cases excellent agreement was achieved between the theoretical
and numerical spectra.
The Brueckner approach is based on the description of the excitations as triplets above a strong coupling singlet ground state.
In essence it represents an effective way of taking into account
the hard-core constraint, which has to be imposed on the triplets,
by treating the excitations as a dilute Bose gas with
infinite on-site repulsion.
The method is valid while the on-site density of excitations (triplets)
$n_{i}$ is small
and under the assumption that the elementary triplets are well localized
objects. In spite of the fact that the
on-site density in the AHC is small even
close to the point
$\delta=0$ ($n_i\approx 0.08$ for $\delta=0.01,\alpha=0$), we find that
our approach fails at $\delta\sim 0.1$ because the triplets become nonlocal objects. In the region $\delta > 0.1$ the present method provides an excellent
quantitative description of the excitation spectrum.
The rest of the paper is organized as follows.
In Sec.II
we find the system of self-consistent equations describing the AHC with frustration and calculate the one-particle (elementary triplet) spectrum for
a wide range of parameters.
Sec.III describes the
additional singlet and triplet modes which appear below the
two-magnon continuum. These excitations are bound states of two
elementary triplets. In Sec.IV we calculate
the contributions of the elementary and the additional triplet into
the neutron scattering structure factor. Sec.V contains our conclusions.
\section{Formulation of the problem and quasiparticle spectrum.}
In order to analyze the excitation spectrum of (\ref{Ham}) it is
convenient to adopt the strong-coupling viewpoint. In the
limit $\delta=1, \alpha=0$ the ground state consists of non-overlapping spin singlets $|GS>=|1,0>|2,0>|3,0>...$, where $|i,0>=\frac{1}{\sqrt{2}}[
|\uparrow>_{2i}|\downarrow>_{2i+1}-|\downarrow>_{i}|\uparrow>_{2i+1}]$.
Here, and in all formulas below which involve triplet operators,
the "generalized" site index $i$ represents the bonds with
spin exchange $J(1+\delta)$.
Each singlet can be excited into a triplet state and therefore it is natural to introduce a creation operator $t_{\alpha i}^{\dagger}$ for this excitation: $|i,\alpha>=t_{\alpha i}^{\dagger}|i,0>$, $\alpha=x,y,z$. The representation of the spin operators in terms of the triplets $t_{\alpha i}^{\dagger}$ was introduced in \cite{Chubukov,Sachdev} and reads:
\begin{equation}
\label{bond}
S_{\alpha,2i}=\frac{1}{2}(t_{\alpha i}+t^{\dagger}_{\alpha i}-i\epsilon_{\alpha \beta
\gamma}t^{\dagger}_{\beta i}t_{\gamma i})
,
\end{equation}
$$
S_{\alpha,2i+1}=\frac{1}{2}(-t_{\alpha i}-t^{\dagger}_{\alpha i}-i\epsilon_{\alpha \beta
\gamma}t^{\dagger}_{\beta i}t_{\gamma i})
,$$
where $\epsilon_{\alpha \beta\gamma}$ is the fully antisymmetric tensor.
After application of this transformation to (\ref{Ham}), or, equivalently, after calculating the matrix elements of the ``hopping'' terms, we find
the effective Hamiltonian
\begin{equation}
\label{ham1}
H=H_2+H_3+H_4+H_U,
\end{equation}
\begin{equation}
\label{h22}
H_2=J_\perp\sum^{N/2}_{i=1,\alpha}t^{\dagger}_{\alpha i}t_{\alpha i}+\frac{\lambda}{2}\sum_{\alpha i}^{N/2}(t^{\dagger}_{\alpha i}t_{\alpha i+1}+t_{\alpha i}^{\dagger}
t^{\dagger}_{\alpha i+1}+h.c.),
\end{equation}
\begin{equation}
\label{h33}
H_3=\frac{\nu}{4}\sum^{N/2}_{i=1,\alpha \beta \gamma} (i\epsilon_{\alpha \beta \gamma}[t^{\dagger}_{\alpha i}t^{\dagger}_{\beta i+1}t_{\gamma i+1}-t^{\dagger}_{\alpha i+1}t^{\dagger}_{\beta i} t_{\gamma i}]+h.c.),
\end{equation}
\begin{equation}
\label{h44}
H_4=\frac{\mu}{2}\sum^{N/2}_{i=1,\beta \gamma}[t^{\dagger}_{\beta i}t^{\dagger}_{\gamma i+1}t_{\gamma i}t_{\beta i+1}-t^{\dagger}_{\beta i}t^{\dagger}_{\beta i+1}t_{\gamma i}t_{\gamma i+1}],
\end{equation}
\begin{equation}
\label{hu}
H_U=\frac{U}{2}\sum^{N/2}_{i=1,\alpha\beta} t^{\dagger}_{\alpha i}t^{\dagger}_{\beta i}t_{\beta i}t_{\alpha i},\hspace{0.3cm} U\rightarrow \infty,
\end{equation}
where we have adopted the notation:
\begin{equation}
J_\perp=J(1+\delta), \lambda=-\frac{1}{2}J(1-\delta-2\alpha), \mu=\frac{1}{2}J(1-\delta+2\alpha), \nu=J(1-\delta).
\end{equation}
The present notation for the coefficients is chosen following Ref.\cite{Kotov}.
For the problem studied in Ref.\cite{Kotov}
the Hamiltonian represents a spin ladder with frustration and the term $H_3$
cancels out due to the symmetry of the ladder with respect to interchange of
its two legs. The coupling $J(1+\delta)$ corresponds to the
rung exchange $J_{\perp}$ in the spin ladder model, and gives the
on-site energy of the triplets (first term in Eq.(\ref{h22})).
The rest of the terms in the Hamiltonian represent
an effective hopping of the triplets,
spontaneous creation of two triplets on nearby sites (second term in
Eq.(\ref{h22})), as well as interactions between the triplets (Eq.(\ref{h33}) and Eq.(\ref{h44})).
In addition, an infinite on-site repulsion $H_U$ is introduced to take into account the hard-core constraint $t^+_{\alpha i}t^+_{\beta i}=0$, which
has to be satisfied on every site.
This condition (only one triplet can be excited
on a site) follows from quantization of spin and guarantees the
one to one correspondence between the spin model Eq.(\ref{Ham}) and
its effective triplet description Eq.(\ref{ham1}).
Let us first consider the Hamiltonian (\ref{ham1}) without the $H_3$
term. A detailed analysis of the spectrum was presented in
Refs.\cite{Sushkov,Sushkov1,Kotov} an we will only summarize the results here.
The interaction $H_U$ gives the dominant contribution
to the renormalization of the triplet spectrum. It was demonstrated in \cite{Sushkov} that in the Brueckner approximation this renormalization is given by the normal self-energy operator
\begin{equation}
\label{en}
\Sigma^{Br}_n(k,\omega)=4\sum_q Z^{Br}_{q}v_q^2\Gamma^{Br}(k+q,\omega-\omega_q)
,\end{equation}
where the scattering amplitude is
\begin{equation}
\label{ver}
\Gamma^{Br}(q,\omega)=-\left(\sum_p\frac{Z^{Br}_{p}Z^{Br}_{q-p}u_p^2u_{q-p}^2}{\omega-\omega_p-\omega_{q-p}}\right)^{-1}.
\end{equation}
The renormalized spin-wave spectrum $\omega_k$, the quasiparticle residue $Z_k$ and the Bogoliubov parameters $u_k$, $v_k$ can be found from the solution of the coupled Dyson equations for the normal $G_n(k,t)=-i<T(t_{k\alpha}(t)t^{\dagger}_{k\alpha}(0))>$ and anomalous $G_a(k,t)=-i<T(t^{\dagger}_{-k\alpha}(t)t^{\dagger}_{k\alpha}(0))>$ Green's functions (see Ref.\cite{Sushkov}):
\begin{equation}
\label{sp}
\omega_{ k}=Z^{Br}_{ k}\sqrt{\tilde A_{k}^2-\tilde B_{k}^2},
\end{equation}
\begin{eqnarray}
\label{ab}
&&\tilde A_{k}=J_{\perp}+\lambda\mbox{cos}{2k}+\Sigma^{Br}({ k},0)+
2\mu\mbox{cos}{2k}\sum_{q}Z^{Br}_{q}v_{q}^2\mbox{cos}{2q},\\
&&\tilde B_{k}=\lambda\mbox{cos}{2k}-2\mu\mbox{cos}{2k}
\sum_{q}Z^{Br}_{q} u_{q}v_{q}\mbox{cos}{2q},\nonumber
\end{eqnarray}
\begin{eqnarray}
\label{zuv}
&&Z^{Br}_{k}=\left(1-\left.\frac{\partial \Sigma^{Br}}{\partial \omega}
\right|_{\omega=0}\right)^{-1},\\
&&u^2_{ k},v^2_{k}=\left(\frac{Z^{Br}_{k}\tilde A_{k}}
{2\omega_{k}}\pm\frac{1}{2}\right).\nonumber
\end{eqnarray}
These equations also take into account the
quartic interaction, $H_4$ (\ref{h44}),
in the lowest order, one-loop approximation.
For the sake of simplicity, in all of the above formulas, and everywhere from now on, we
omit the number of lattice sites.
All momenta take values inside the Brillouin zone of the original
(non-dimerized) lattice $(-\pi/2<k\le \pi/2)$.
The discrete lattice sums can be converted into integrals by
$\sum_{k} \rightarrow \pi^{-1}\int_{-\pi/2}^{\pi/2}dk$.
The basic approximation made in the derivation of $\Gamma^{Br}(q,\omega)$,
as discussed in Ref.\cite{Sushkov}, is the neglect of all anomalous scattering vertices, which are present in the theory due to the existence of anomalous Green's functions (i.e. quantum fluctuations).
All anomalous contributions however are suppressed by the small parameter of the Brueckner approach: the density of triplet excitations $n_i
=\sum_{\alpha}<t_{i\alpha}^{\dagger} t_{i\alpha}>
=3\sum_q Z^{Br}_{q}v_q^2$. As follows from Eq.(\ref{en}) the
Brueckner self-energy is first order in $n_{i}$, while
it is easy to prove that all anomalous contributions lead to
higher powers of the density \cite{remark}.
This observation justifies the Brueckner approximation (i.e. keeping
only Eq.(\ref{en})), since the triplet
density
is small throughout the disordered phase, even close to $\delta=0$.
We find that
$n_{i} \approx 0.0066$ for $\delta=0.6, \alpha=0$, increasing at decreasing dimerizetion
$n_{i} \approx 0.066$ for $\delta=0.1,\alpha=0$.
Now consider the effect of the cubic interaction $H_3$, Eq.(\ref{h33}).
In order to incorporate the effect of $H_3$
self-consistently, the normal and anomalous self-energies have to be found
to a given order and subsequently inserted into the system of coupled Dyson's equations for the anomalous and normal Green's functions. The poles of these Green's functions
determine the renormalized one-particle spectrum \cite{Kotov1}.
This procedure however is rather involved and in the present work
we will use only the lowest order perturbation theory result (i.e.
we will keep one diagram only).
This is justified, since we find that the lowest order contribution
is relatively small and therefore $H_3$
does not require a fully self-consistent
treatment.
The cubic interaction Eq.(\ref{h33}), after Fourier: $t_{\alpha j}=\sum_{k}t_{\alpha k}e^{2ikj}$,
and Bogoliubov: $t_{k\alpha}=u_ka_{k\alpha}+v_ka_{-k\alpha}^{\dagger}$ transformations can be rewritten as
\begin{equation}
\label{h3}
H_3=\frac{1}{2}\sum_{k_1,k_2,k_3=k_1+k_2}\epsilon_{\alpha \beta \gamma}\left(\Gamma_1(1,2,3)[a^{\dagger}_{k_1\alpha}a^{\dagger}_{k_2\beta}a_{k_3\gamma}+h.c]
\right. +
\end{equation}
\begin{equation}
+\left.\frac{1}{3}\Gamma_2(1,2,3)[a^{\dagger}_{-k_1\alpha}a^{\dagger}_{-k_2\beta}a^{\dagger}_{k_3\gamma}+h.c.]\right).
\end{equation}
We have defined
$$
\Gamma_1(1,2,3)=(u_{k_3}D(1,2,3)-v_{k_3}C(1,2,3))\sqrt{Z^{Br}_{k1}}\sqrt{Z^{Br}_{k2}}\sqrt{Z^{Br}_{k3}}
,$$
$$
C(1,2,3)=\Gamma_0(1,2)v_{k_1}v_{k_2}+\Gamma_0(-3,2)u_{k_2}v_{k_1}+\Gamma_0(1,-3)u_{k_1}v_{k_2},$$
where $\Gamma_0(1,2)=-\nu\mbox{sin}(k_1-k_2)\mbox{cos}(k_1+k_2)$ and $D(1,2,3)=C(1,2,3)$ with replacing all $u_{\bf k}$ by $v_{\bf k}$ and vice versa $\{u \leftrightarrow v\}$,
and $\Gamma_2(1,2,3)=-\Gamma_1(1,2,3)$ with replacing $\{u_{k_3}\leftrightarrow v_{k_3}\}
$.
Only a normal Green's function $G_n=-i<T(a_{k\alpha}(t)a_{k\alpha}^{\dagger}(0))>$ exists for the physical operators $a_{k\alpha}$, and therefore the
self-energy induced by $H_3$ is given, to lowest order, by the two
diagrams shown in Fig.1:
\begin{equation}
\label{se3}
\Sigma_3(k,\omega)=\sum_{q}\left(\frac{\Gamma_1^2(q,k-q,k)}{\omega-\omega_q-\omega_{k-q}}-\frac{\Gamma_2^2(q,k-q,k)}{\omega+\omega_q+\omega_{k-q}}
\right).\end{equation}
The correction $\delta\omega_k$ to the one-particle spectrum is: $\delta\omega_k=\Sigma_3(k,\omega_k)$.
The self-energy (\ref{se3}) also gives a
correction to the quasiparticle residue:
$
Z^{(3)}_k\approx 1+\left.\frac{\partial \Sigma_3(k,\omega)}{\partial \omega}\right|_{\omega=\omega_k}
$.
In order to find the spectrum, equations (\ref{en},\ref{ver},\ref{sp},\ref{ab},\ref{zuv})
have to be solved self-consistently for $\Sigma^{Br}(k,0)$ and $Z^{Br}_{k}$,
which corresponds to an infinite re-summation of diagrams (solution of Dyson's equation), and leads to the renormalized spectrum $\omega_{k}$.
Then the correction due to $H_3$ has to be added resulting in the
spectrum $\Omega_k=\omega_k+\delta\omega_k$.
In this approximation the quasiparticle residue is $Z_k=Z^{Br}_kZ_{k}^{(3)}$.
In the limit $\nu,\lambda\ll J_\perp$ all expressions can be easily
evaluated
analytically to leading order, and for example the corrections, due to
$H_{3}$ are:
$\delta\omega_k=-\frac{\nu^2}{2J_\perp}\mbox{cos}^2k$, and $Z_{k}^{(3)}=1-\frac{\nu^2}{2J_\perp^2}\mbox{cos}^2k$. Finally, combining the three-particle corrections with the results of Ref.\cite{Kotov} for the one-particle
spectrum without $H_{3}$, we obtain in the limit $\lambda,\nu\ll J_\perp$:
\begin{equation}
\label{s0}
\frac{\Omega_k}{J_\perp}=1+\frac{\lambda}{J_\perp}\mbox{cos}2k+\frac{3}{4}\frac{\lambda^2}{J^2_\perp}-\frac{1}{4}\frac{\lambda^2}{J^2_\perp}\mbox{cos}4k-\frac{1}{2}\frac{\nu^2}{J^2_\perp}\mbox{cos}^2k
.\end{equation}
Higher orders also can be evaluated, similarly to Ref.\cite{Kotov},
but we shall not present the lengthy expressions here.
Next we present results obtained by numerical self-consistent solution of
the Dyson's equations.
Plots of the triplet gap $\Delta=\Omega_{k=0}/J$ as a function of $\alpha$ at $\delta=0.4,0.2$ are presented in Fig.2. Our results agree quite well with the
ED data \cite{Bouzezar}.
We also note that our results disagree with those presented in
Ref.\cite{Bouzezar1}. The main reason for this is that the
three-leg vertices $H_3$, Eq.(\ref{ham1}) were not taken into account in
Ref.\cite{Bouzezar1}. In figures 4a,b,c,d we also
present spectra of the elementary triplet $\Omega_k$ for different parameter values. The agreement with the numerical data (when available) is very good.
The approach presented here relies heavily on the effective
description of the spin problem in terms of localized triplets.
As mentioned in the Introduction, in the limit $\delta\rightarrow 0$,
for any $\alpha$, the spinons, confined to form the triplets,
become unconfined. Our calculation clearly breaks down at this point.
We would like to emphasize however, that due to the power law
dependence of the radius of confinement $r_{0}$ on $\delta$,
namely $r_0\sim \delta^{-2/3}$ for $\alpha<\alpha_c=0.2411$ and
$r_0\sim \delta^{-1/3}$ for $\alpha>\alpha_c$ \cite{Uhrig1,Singh},
the triplets are well localized for $\delta > 0.1$. For
$\delta \sim 0.1$ the triplet size is about 2-3 lattice spacings.
Therefore the strong-coupling approach in fact is expected
to describe
quantitatively well a very wide range of parameter space.
Let us also mention that the effective Hamiltonian Eq.(\ref{ham1})
can be used even for $\delta < 0.1$, when $r_{0}$ increases
strongly. However technically the problem becomes much more
difficult and truly non-perturbative in nature.
One way to deal with it is to develop a perturbation theory to high
order, e.g.
via the dimer series expansion \cite{Singh}. Physically
a finite $r_{0}$ leads to the creation of a "string" (with length
$r_{0}$) of excited triplets. This string
leads to the formation of low-energy many-particle bound states
which mix strongly with the one-particle excitations, causing
ultimately (for $\delta \rightarrow 0$)
the vanishing of the one-particle spectral weight.
These effects can be taken into account
by diagonalizing the Hamiltonian in Fock space (spanned by
the one-magnon, two-magnon, etc. states).
A detailed analysis however
is beyond the scope of the present work \cite{Kotov2}.
\section{ Spectrum of collective excitations: two-magnon bound states.}
The quartic term Eq.(\ref{h44}) in the Hamiltonian Eq.(\ref{ham1}) leads to attraction between two elementary triplets. It has been already demonstrated in
the case of a
spin ladder \cite{Sushkov1,Kotov} that this attraction is strong enough to form additional singlet (S=0) and triplet (S=1) bound states below the two-particle continuum $E^{C}(Q)=\mbox{ min}_q(\Omega_{\frac{Q}{2}-q}+\Omega_{\frac{Q}{2}+q})$. Here we will calculate the energies of
these additional singlet and triplet states
in the AHC model. Such bound states have already been
observed in the ED calculations \cite{Bouzezar,Barnes} of the AHC.
Consider the scattering of the two triplets: $q_1\alpha+q_2\beta\rightarrow q_3\gamma+q_4\delta$ and introduce the total, $Q$, and relative, $q$, momenta of the pair.
The two-particle singlet and triplet wave functions are:
\begin{equation}
\label{singlet}
|\psi^{S}(Q)>=\frac{1}{\sqrt{6}}\sum_{q \alpha}\psi^S(Q,q)a^{\dagger}_{\alpha \frac{Q}{2}-q}a^{\dagger}_{\alpha \frac{Q}{2}+q}|0>
,\end{equation}
\begin{equation}
\label{triplet}
|\psi_{\alpha}^T(Q)>=\frac{1}{2}\sum_{q,\beta\gamma}\psi^T(Q,q)\epsilon_{\alpha \beta \gamma}a^{\dagger}_{
\frac{Q}{2}+q \beta}a^{\dagger}_{\frac{Q}{2}-q \gamma}|0>,
\end{equation}
where $\psi^{S,T}(Q,q)$ are determined from the Schr{\"{o}}dinger equation,
$H|\psi^{S,T}(Q)>=E^{S,T}(Q)|\psi^{S,T}(Q)>$ with bound state energy
$E^{S,T}(Q)$. From this equation one can readily derive the
integral Bethe-Salpeter equation satisfied by the bound state
wave-functions:
\begin{equation}
\label{bs}
[E^{S,T}(Q)-\Omega_{\frac{Q}{2}+q}-\Omega_{\frac{Q}{2}-q}]\psi^{S,T}(Q,q)=
\sum_p M^{S,T}(Q,q,p)\psi^{S,T}(Q,p)
.\end{equation}
Let us at first
neglect the cubic interaction $H_3$ in the Hamiltonian (\ref{ham1}).
Then
the scattering amplitudes $M^{S,T}(Q,q,p)$ in the singlet and triplet channels
are (see Ref.\cite{Kotov}):
$
M^S(Q,q,p)=(U-2\mu\mbox{cos}2q\mbox{cos}2p)$,
and $
M^{T}(Q,q,p)=-\mu\mbox{sin}2q\mbox{sin}2p
$, respectively. Eq.(\ref{bs}) should be solved with the substitution
\begin{equation}
M^{S,T}(Q,q,p)\rightarrow \sqrt{Z^{Br}_{\frac{Q}{2}-p}}\sqrt{ Z^{Br}_{\frac{Q}{2}+p}}\sqrt{ Z^{Br}_{\frac{Q}{2}-q}}\sqrt{ Z^{Br}_{\frac{Q}{2}+q}}
u_{\frac{Q}{2}-q}u_{\frac{Q}{2}+q}u_{\frac{Q}{2}-p}u_{\frac{Q}{2}+p}M^{S,T}(Q,q,p)
\end{equation}
in order to take into account that the triplet excitation $a^{\dagger}_{k\alpha}$
differs from
the bare one $t^{\dagger}_{k\alpha}$, due to the Bogoliubov transformation and the quasiparticle residue.
We stress that the solution of the integral equation (\ref{bs}) in the singlet channel
has to satisfy the condition
$\sum_{q}\psi^S(Q,q)=0$, due to the
infinite on-site repulsion. A Lagrange multiplier has to be introduced to enforce
this condition.
The normalization constants of the wave functions (\ref{singlet},\ref{triplet})
are chosen to satisfy the conditions: $\sum_{Q} |\psi(Q)|^2=
\sum_q|\psi(Q,q)|^2=1.$
The solution of Eq.(\ref{bs}) for the triplet and singlet bound states (\ref{singlet},\ref{triplet}) has already been found in \cite{Kotov} for the
Hamiltonian (\ref{ham1}) without the $H_3$ term. In the leading order
in $\lambda/J_\perp,\mu/J_\perp$ the energies and the
wave functions of the bound states are:
\begin{equation}
\label{es}
E_Q^S=2J_\perp-\mu(1+C_Q^2)+\frac{3}{2}\frac{\lambda^2}{J_\perp}-\frac{\lambda^2}{4J_\perp}\mbox{cos}2Q,
\end{equation}
$$\psi^S(Q,q)= \sqrt{2(1-C_Q^2)}\frac{\mbox{cos}2q+C_Q}{1+2C_Q\mbox{cos}2q+C_Q^2}
,
$$
\begin{equation}
\label{et}
E_Q^T=2J_\perp-\frac{\mu}{2}(1+4C_Q^2)+\frac{3\lambda^2}{2J_\perp}+\frac{\lambda^2}{4J_\perp}\mbox{cos}2Q,
\end{equation}
$$\psi^T(Q,q)=\sqrt{1/2-2C_Q^2}\frac{\mbox{sin}2q}{\frac{1}{2}+2C_Q\mbox{cos}2q+2C_Q^2}
,$$
where we have defined $C_Q=\frac{\lambda}{\mu}\mbox{cos}Q$.
Next, we analyze how $H_{3}$ affects the two-particle spectra.
The contribution of the three-particle scattering into binding should change the form
of Eq.(\ref{bs})
since it leads to retardation and thus to a non-trivial frequency dependence in the Bethe-Salpeter equation.
In the present work we treat
$H_3$ as a perturbation and find the second order corrections, $\delta E_Q^{S,T}$, to the two-particle bound state energies $E^{S,T}_Q$.
The correction in the singlet channel
is given as a sum of the diagrams, shown in Fig.3a,b,c,
convoluted with the singlet wave function
\begin{equation}
\label{dsin}
\delta E_1^S(Q)=\sum_{p,q}\delta M_a^S(Q,q,p) \psi^S(Q,q)\psi^S(Q,p)
+
\delta E^S_b(Q)+\delta E^S_c(Q),
\end{equation}
with the vertex
\begin{equation}
\delta M_a^S(Q,q,p)=-\frac{4\Gamma_1(\frac{Q}{2}-q,q-p,\frac{Q}{2}-p)\Gamma_1(\frac{Q}{2}+p,q-p,\frac{Q}{2}+q)}{E^S_Q-\omega_{q-p}-\omega_{\frac{Q}{2}+p}-\omega_{\frac{Q}{2}-q}}
\end{equation} and
\begin{eqnarray}
\label{e0}
&&\delta E^S_b(Q)=
2\sum_{p,q}\psi_S^2(Q,p)\frac{\Gamma_1^2(q,p-q+\frac{Q}{2},p+\frac{Q}{2})}{E^S_Q-\omega_{p-q+\frac{Q}{2}}-\omega_{\frac{Q}{2}-p}-\omega_q}
,\\
&&\delta E_c^S(Q)=2\sum_{p,q}\psi_S^2(Q,p)
\frac{\Gamma_2^2(q,p-q+\frac{Q}{2},p+\frac{Q}{2})}
{E^S_Q-\omega_{p-q+\frac{Q}{2}}-\omega_q-2\omega_{\frac{Q}{2}+p}-\omega_{\frac{
Q}{2}-p}}.
\end{eqnarray}
The correction to the triplet energy $E^T(Q)$ is given by the diagrams a,b,c,d,e in Fig.3,
convoluted with the triplet wave function
\begin{equation}
\label{dtrip}\delta E^T_1(Q)=\sum_{p,q}(\delta M^T_a+\delta M^T_d+\delta M^T_e)\psi^T(Q,q)\psi^T(Q,p)
+\delta E_b^T(Q)+\delta E_c^T(Q)
\end{equation}
with the vertices: $\delta M_{a}^T(Q,q,p)=\frac{1}{2}\delta M^S_a(Q,q,p)\{E^S(Q)\rightarrow E^T(Q)\}$,
$$
\delta M_{d}^T(Q,q,p)=\frac{\Gamma_1(\frac{Q}{2}+p,\frac{Q}{2}-p,Q)\Gamma_1(\frac{Q}{2}+q,\frac{Q}{2}-q,Q)}{E^T_Q-\omega_Q}
,$$
$$
\delta M_{e}^T(Q,q,p)=\frac{\Gamma_2(\frac{Q}{2}+q,\frac{Q}{2}-q,Q)\Gamma_2(\frac{Q}{2}+p,\frac{Q}{2}-p,Q)}{E^T_Q-\omega_Q-\omega_{\frac{Q}{2}-q}-\omega_{\frac{Q}{2}+q}-\omega_{\frac{Q}{2}-p}-\omega_{\frac{Q}{2}+p}}
,
$$
and the terms: $\delta E_{b,c}^T(Q)=\delta E_{b,c}^S(Q)\{\psi^S,E^S\rightarrow
\psi^T,E^T\}$.
In the leading order in $\nu,\lambda\ll J_\perp$ the corrections (\ref{dsin},\ref{dtrip}) can be easily found analytically:
\begin{equation}
\label{mom}
\delta E_1^S(Q)=-\frac{\nu^2}{2J_\perp}\mbox{cos}^2Q
,\hspace{0.2cm}\delta E_1^T(Q)=-\frac{\nu^2}{2J_\perp}(1-\frac{1}{2}\mbox{cos}^2Q)
.\end{equation}
We observe that the contributions (\ref{dsin},\ref{dtrip})
have an appreciable effect on
the two-particle energies and therefore the cubic interaction (\ref{h33})
can not be neglected.
It is worth noting that Eqs.(\ref{bs},\ref{dsin},\ref{dtrip}) are not quite correct because they do not take fully into account the hard-core constraint $t^{\dagger}
_{i\alpha}t^{\dagger}_{i\beta}=0$.
Let us consider the limit $\lambda,\nu\ll \mu\ll J_\perp$. A direct expansion in powers of
$1/J_\perp$ applied to the Hamiltonian (\ref{ham1}) in coordinate space, gives the
following energies for the singlet and triplet bound states:
\begin{eqnarray}
\label{real1}
&&E^S_{real}=2J_\perp-\mu-\frac{\lambda^2}{2\mu}(1+\mbox{cos}2Q)+\frac{9\lambda^2}{8J_\perp}-\frac{\nu^2}{4J_\perp}(1+\mbox{cos}2Q),\\
\label{real2}
&&E^T_{real}=2J_\perp-\frac{\mu}{2}-\frac{\lambda^2}{\mu}(1+\mbox{cos}2Q)+\frac{9\lambda^2}{8J_\perp}+\frac{\nu^2}{8J_\perp}\mbox{cos}2Q
.
\end{eqnarray}
The hard-core constraint is completely taken into account in these formulas by
not allowing hopping processes leading to two triplets occupying the same site.
On the other hand the momentum space diagrammatic calculation
Eqs.(\ref{es},\ref{et}) and the three-particle correction
(\ref{mom}) result in
\begin{eqnarray}
\label{mom1}
&&E^S_{mom}=2J_\perp-\mu-\frac{\lambda^2}{2\mu}(1+\mbox{cos}2Q)+\frac{\lambda^2}{4J_\perp}(6-\mbox{cos}2Q)-\frac{\nu^2}{2J_\perp}\mbox{cos}^2Q,\\
\label{mom2}
&&E^T_{mom}=2J_\perp-\frac{\mu}{2}-\frac{\lambda^2}{\mu}(1+\mbox{cos}2Q)+\frac{\lambda^2}{4J_\perp}(6+\mbox{cos}2Q)-\frac{\nu^2}{2J_\perp}(1-\frac{1}{2}\mbox{cos}^2Q)
.
\end{eqnarray}
The physical reason for the difference between the
momentum space calculation and the exact real space result is in the
additional blocking of virtual excitations which is not included in Eqs.(\ref{bs},\ref{dsin},\ref{dtrip}). Let us consider one example of such additional blocking. When we have
a state with one triplet, the quasiparticle blocks virtual excitations due to quantum fluctuations, $t^{\dagger}_{i\alpha}t^{\dagger}_{i+1\alpha}$, on two links. This is the physical origin of the third term in the one-particle dispersion (\ref{s0}). When we have two
excited
triplets and they are separated by more than one lattice
spacing they block four links (this
corresponds to the third terms in (\ref{es},\ref{et})).
However when the two triplets are on nearest-neighbor sites they block only three links,
which is energetically more favorable and leads to the increase of the binding energy.
This effect represents a non-potential contribution and is not taken into
account by the integral equation (\ref{bs}).
%Another process which is not taken into account in (\ref{bs}) is that one spac$e hopping of two triplets on the nearest sites is forbidden by constraint, th$us the term $\sim \mbox{cos}2Q$ should disappear in spectrum.
The additional contribution of the quantum fluctuations into binding is analogous to the
calculation of the Lamb shift in atomic physics.
It is clear from the above discussion that binding due to blocking of
quantum fluctuations exists only
when the triplets are on nearest-neighbor sites.
We have also identified several additional processes which have to be described more
accurately by inserting the vertex (\ref{ver}), responsible for the infinite one-site
repulsion, into the
diagrams presented in Fig.3. These diagrams should account correctly for
the hard-core repulsion
between the particles in the initial and intermediate states.
However it is very difficult to identify and calculate in momentum space
all diagrams of this type.
We will follow a simpler, effective way to take into account these additional
contributions.
Comparing the coordinate space results (\ref{real1},\ref{real2}) with the momentum space calculations (\ref{mom1},\ref{mom2}) one can see that the latter
can be corrected by adding the following expressions to Eqs.(\ref{dsin},\ref{dtrip}):
\begin{eqnarray}
\label{con}
&&\delta E^S_2(Q)=\frac{\lambda^2}{4J_\perp }\left[-\frac{3}{2}+\mbox{cos}2Q\right]\left|
\sum_{k}\sqrt{2}\mbox{cos}2k\psi^S(Q,k)\right|^2
,\\
\label{con1}
&&\delta E^T_2(Q)=\frac{3\lambda^2}{8J_\perp}\left[-1-\frac{2}
{3}\mbox{cos}2Q+\frac{\nu^2}{\lambda^2}\right]\left|\sum_{k}
\sqrt{2}\mbox{sin}2k\psi^T(Q,k)\right|^2
.\end{eqnarray}
In the above equations the expressions in the absolute value signs
give the probability amplitudes for two quasiparticles to be on nearest-neighbor sites in the appropriate channel.
Finally, to obtain the energies of the triplet and singlet bound states, the integral equation (\ref{bs}) should be solved first, and then the
corrections $\delta E^{S,T}_1(Q)$ (\ref{dsin},\ref{dtrip}) and $\delta E^{S,T}_2(Q)$ (\ref{con},\ref{con1}) should be added.
Results obtained by a self-consistent numerical
solution for the spectrum of the singlet and triplet bound states are
presented in Fig.4 for different values
of the frustration, $\alpha$, and dimerization, $\delta$.
We find that there are no bound states in the vicinity of $q=0$ for zero frustration ($\alpha=0$). Singlet and triplet bound states
are always present in the vicinity of $q=\pi/2$, however the singlet typically
exists within a much wider momentum range.
At $q=0$ the singlet splits off
from the continuum at any non-zero $\alpha$, while the triplet
exists only above a certain value of frustration.
One can see also that, quite generally, the singlet is below the triplet, i.e.
the binding in the singlet channel is stronger.
We have also calculated
the binding energies, $\varepsilon^{S,T}(Q)=E^C(Q)-E^{S,T}(Q)$,
of the singlet and triplet bound states as a function of $\delta$ at
$Q=\pi/2$ for $\alpha=0$.
The results are presented in Fig.5.
There is excellent agreement with the ED data \cite{Bouzezar,Barnes} in
the singlet channel even for $\delta=0.1$,
whereas the agreement for the triplet is not so good for $\delta<0.5$.
We attribute this disagreement to the fact that
the triplet binding energy is relatively small in comparison with the singlet one for
the whole range of parameters ($\alpha,\delta$).
Therefore taking into account the three-particle scattering in simple perturbation theory is not as good approximation for the triplet as it is for the singlet.
We have performed a more
accurate calculation for the cases when the binding energy
is very small. An
improvement can be achieved by
adding the corrections $\delta M^{S,T}(Q,q,p)$ to the scattering amplitude $M^{S,T}(Q,q,p)$ into the Bethe-Salpeter integral equation (\ref{bs}) and further solving
the equation to find the binding energies.
Such "self-consistent" calculations are presented by the dashed lines in Fig.5 and one can see that the agreement with the ED data indeed becomes better for small energies.
There are several important points which have been overlooked in Ref.\cite{Bouzezar1},
ultimately leading to incorrect results for the energies of the bound states. The contribution of the constraint (\ref{hu}) has not been taken into account in the integral equation (\ref{bs})
resulting in non-zero binding energy of the singlet at $q=0$ even for large $\delta$
and $\alpha=0$ (notice that our calculation always gives zero binding at this point).
The contribution of the three-particle scattering (\ref{h33})
into binding has not been taken into account in Ref.\cite{Bouzezar1}
and the effect given by the corrections (\ref{con},\ref{con1}) (Lamb shift) has not been included.
In addition, the error in the calculation of the one-particle spectrum (see the previous section)
has propagated into the two-particle energies as well.
Finally we would like to compare our results for the bound state energies
with the lowest order perturbation theory results, presented in Ref.\cite{Barnes}.
The binding energy $\varepsilon^{S,T}(Q)=E^C(Q)-E^{S,T}(Q)$ at $Q=\pi/2,\alpha=0$ resulting from (\ref{real1},\ref{real2}) and (\ref{s0}) in the
singlet channel, $\varepsilon^S(\frac{\pi}{2})/J=\frac{1}{2}(1-\delta)-\frac{17}{32}\frac{(1-\delta)^2}{1+\delta}$, is different from
the result of Ref.\cite{Barnes}: $\varepsilon^S(\frac{\pi}{2})/J=\frac{1}{2}(1-\delta)-\frac{14}{32}\frac{(1-\delta)^2}{1+\delta}$.
In the triplet channel our result is
$\delta\varepsilon^T(\frac{\pi}{2})/J=\frac{1}{4}(1-\delta)-\frac{13}{32}\frac{(1-\delta)^2}{1+\delta}$, which coincides with the formula presented in Ref.\cite{Barnes}.
\section{Structure factor.}
The one-particle triplet and the triplet bound state
can be observed in neutron scattering
experiments. The inelastic neutron scattering cross-section is directly proportional to the dynamic structure factor:
\begin{equation}
S(q,\omega)=\frac{1}{2}\mbox{Im}\int e^{i\omega t}<\mbox{T}({\bf S}(q,t).{\bf S}(-q,0))>.
\end{equation}
Here we calculate the spectral weights of the elementary and the additional triplets.
After the bond operator (\ref{bond}) and Bogoliubov transformations, the on-site spin operator ${\bf S}_{i}$ can be written in momentum space as:
\begin{equation}
S_\alpha(-q,0)=-ie^{iq/2}\left[T_1(q)a_{q\alpha}^{\dagger}+\epsilon_{\alpha\beta\gamma}\sum_{k}T_2(k,q)a^{\dagger}_{k+q\beta}a^{\dagger}_{k\gamma}\right],
\end{equation}
where we have defined the
vertices $T_1(q)=(u_q+v_q)\mbox{sin}\frac{q}{2}$ and
$T_2(k,q)=u_{k+q/2}v_{k-q/2} \mbox{cos}\frac{q}{2}$.
Then, after averaging, the structure factor can be rewritten in the form $S(q,\omega)=S_1(q)\delta(\omega-\Omega_q)+S_2(q)\delta(\omega-E^T_q)$ where
\begin{equation}
\label{str1}
S_1(q)=\frac{1}{2}\left[\sqrt{Z^{(3)}}T_1(q)+2
\sum_{k}\frac{T_2(k,q)\Gamma_1(k+\frac{q}{2},-k+
\frac{q}{2},q)}{\omega_q-\omega_{k+\frac{q}{2}}-\omega_{k-\frac{q}{2}}}
\right]^2
\end{equation}
is the spectral weight of the elementary triplet, formally represented by the diagram in Fig.6a, and
\begin{equation}
\label{str2}
S_2(q)=\frac{1}{2}\left|2\sum_{k}T_2(k,q)
\psi^T(q,k)+
\right.\end{equation}
$$\left.
\frac{T_1(q)}{E^T_q-\omega_q-\Sigma_3(E^T_q,q)}
\sum_{k}\Gamma_1(k+\frac{q}{2},-k+\frac{q}{2},q)\psi^T(q,k)\right|^2
$$
is the contribution of the two-particle triplet bound state, represented by the diagrams in Fig.6b. Note that in calculations of the spectral weights $S_1(q)$ and $S_2(q)$ the total Green's function, including the three particle scattering, should be used, i.e
the corrected quasiparticle residue $Z=Z^{Br}Z^{(3)}$ and the total elementary spectrum $\Omega_k=\omega_k+\delta\omega_k$, should be substituted.
Along the disordered line $\lambda=0$ there are no quantum fluctuations
(i.e. $v_k=0,u_k=1,Z^{Br}=1$), and
the dependence of $S_1(q)$ and $S_2(q)$ upon $\delta$ is given by the three particle scattering contribution only.
In the leading order in $\lambda\ll \mu,J_\perp$, we have calculated the spectral weights analytically:
\begin{eqnarray}
&&S_1(q)\approx \frac{1}{2}\mbox{sin}^2\frac{q}{2}(1-\frac{\lambda}{2J_\perp}\mbox{cos}2q-\frac{\nu^2}{4J^2_\perp}\mbox{cos}^2q-\frac{\lambda\nu}{J^2_\perp}\mbox{cos}q\hspace{0.1cm}\mbox{cos}^2\frac{q}{2})^2, \\
&& S_2(q)\approx \frac{1}{4}(\frac{\nu}{J_\perp}\mbox{cos}q\hspace{0.1cm}\mbox{sin}\frac{q}{2}+\frac{\lambda}{J_\perp-\mu/2}\mbox{cos}\frac{q}{2}\hspace{0.1cm}\mbox{sin}q)^2.
\end{eqnarray}
We have also performed a self-consistent numerical evaluation of $S_1(q)$ and $S_2(q)$ using (\ref{str1},\ref{str2}), and the results are presented in Fig.7a and Fig.7b.
For comparison we have also plotted $S_1(q)$
obtained by the dimer series expansion technique \cite{Singh}.
The structure factor of the elementary triplet is in excellent agreement with the
dimer series results even for $\delta=0.2$. From Fig.7a, which shows the
case $\lambda=0= 1-\delta-2\alpha$,
it is clear that for small $\alpha$ (and large $\delta$),
the structure factor of the bound state
triplet is much smaller than the one for the elementary triplet.
However as $\delta$ decreases the spectral weight of the bound state triplet
increases and eventually can even become equal to
the one-particle contribution.
\section{Summary and conclusions.}
In summary, by using self-consistent diagrammatic analysis,
we have calculated the one-particle and the two-particle excitation
spectra of the dimerized and frustrated antiferromagnetic Heisenberg chain.
The quasiparticle excitations are described as a dilute, strongly-correlated
Bose gas of triplets. The important parameter, which controls the validity
of the diagrammatic expansion and allows us to re-sum effectively the
most important diagrams is the density of triplets. The latter quantity
is quite small for a wide range of parameters, making the technique, presented
in this paper, effective for dimerization $\delta>0.1$.
For smaller dimerization the triplets become non-local objects and
a more adequate picture would be one of loosely bound spinons.
We emphasize however that the region of applicability of the presented
technique is, perhaps even surprisingly, quite large. The reason for this
can be traced to the fact that the quasiparticle weight decreases rather
slowly as $\delta$ decreases. Indeed, from Fig.7a one can see that
even for the quite small value $\delta=0.1$ the residue is around $0.3$
(the maximum is chosen to be 0.5).
Even though this is true only in the special case without
quantum fluctuations, it appears to be qualitatively correct even in the
general case. Therefore the triplets are well localized objects with large
quasiparticle residue even for rather small values of $\delta$.
We have found that our results for the spectrum are in excellent agreement with
the numerical data, obtained by exact diagonalizations and dimer series
expansions, for $\delta \approx 0.2$ and higher.
We have presented a detailed diagrammatic analysis of the
collective excitations and have found that singlet and triplet bound states of two
elementary triplets
generally exist below the two-particle continuum.
In the absence of frustration there are no bound states in the
vicinity of $q=0$, while both singlet and triplet ones exist at $q=\pi/2$.
Finite frustration increases the binding energies of both excitations,
making the singlet split off from the continuum at $q=0$ for any $\alpha\neq0$,
while a finite $\alpha$ is required for the triplet bound state to appear.
We have also calculated the contribution of the elementary triplet and the triplet bound state into
the structure factor, which is directly measurable in inelastic neutron
scattering experiments. We have found that for small $\alpha$ and large
$\delta$ the bound state contribution is quite small, however it grows
with decreasing $\delta$ and can become equal to the
quasiparticle contribution.
The technique used in this work is quite similar in spirit to
the dimer series expansion, the major difference being that while
we re-sum only the most important classes of diagrams, the dimer
series contains all of them, but only to a certain (finite) order.
Let us mention however
that at present the dimer series expansion technique
has not yet been
extended to calculate two-particle properties, such as bound state
energies and their contribution into the structure factor \cite{raj}.
This fact makes it worthwhile, in our view, to develop and further
refine the technique, presented in this work, especially concerning
the two-particle properties of quantum spin models with dimerization, of
which the AHC is a particular example.
\section{acknowledgements}
We are very grateful to Zheng Weihong, Rajiv Singh, and Anders Sandvik for
numerous stimulating discussions. This work was supported by a grant from the Australian Research Council.
\begin{thebibliography}{99}
\bibitem[a]{refa}{e-mail: pavel@newt.phys.unsw.edu.au}
\bibitem[b]{refb}{Present address: Department of Physics,
University of Florida, Gainesville, FL 32611-8440; e-mail:
valeri@phys.ufl.edu}
\bibitem[c]{refc}{e-mail: sushkov@newt.phys.unsw.edu.au}
\bibitem{Garret} A.W. Garrett, S.E. Nagler, D.A. Tennant, B.C. Sales, and T. Barnes,
Phys. Rev. Lett. {\bf 79}, 745 (1997).
\bibitem{Nordio} For early work in this area see P.L. Nordio, Z.G. Soos, and H.M. McConnell,
Ann. Rev. Phys. Chem. {\bf 17}, 237 (1966).
\bibitem{Hase} M. Hase, I. Terasaki, and K. Uchinokura, Phys. Rev. Lett. {\bf 70}, 3651
(1993).
\bibitem{Isobe} M. Isobe and Y. Ueda, J. Phys. Soc. Jpn. {\bf 65}, 1178 (1996); D. Augier,
D. Poilblanc, S. Haas, A. Delia, and E. Dagotto, Phys. Rev. B {\bf 56}, R5732 (1997).
\bibitem{Garret1} A.W. Garrett, S.E. Nagler, T. Barnes, and B.C Sales, Phys. Rev. B
{\bf 55}, 3631 (1997).
%\bibitem{Lieb} E. Lieb, T. Schultz, and D.C. Mattis, Ann. Phys. (N.Y.) {\bf 16}, 407 (1961);
%E. Lieb and D. Mattis, J. Math. Phys. {\bf 3}, 749 (1962).
% D.C.Mattis, The Theory of Magnetism (Harper{\&}Row, New York, 1965)
\bibitem{Garret2} D.A. Tennant, S.E. Nagler, T. Barnes, A.W. Garrett,
J. Riera, and B.C. Sales, Physica B {\bf 241}, 501 (1997).
\bibitem{Ain} M. Ain, J.E. Lorenzo, L.P. Regnault, G. Dhalenne,
A. Revcolevschi, B. Hennion, and T. Jolicoeur, Phys. Rev. Lett {\bf 78}, 1560 (1997).
\bibitem{Paul} For a review of optical spectroscopy studies, see P.H.M. Van Loosdrecht,
cond-mat/9711091, and P. Lemmens {\it et al.}, cond-mat/9809289.
\bibitem{Sushkov1} O.P. Sushkov and V.N. Kotov, Phys. Rev. Lett. {\bf 81}, 1941 (1998).
\bibitem{Okamoto} K. Okamoto and K. Nomura, Phys. Lett. A {\bf 169}, 433 (1992); G. Castilla,
S. Chakravarty, and V.J.Emery, Phys. Rev. Lett. {\bf 75}, 1823 (1995).
\bibitem{ft} J. des Cloizeaux and J.J. Pearson, Phys. Rev. {\bf 128},
2131 (1962); L.D. Faddeev and L.A. Takhtajan, Phys. Lett. A {\bf 85}, 375 (1981).
\bibitem{Haldane} F.D.M. Haldane, Phys. Rev. B {\bf 25}, 4925 (1982).
\bibitem{Shastry} B.S. Shastry and B. Sutherland, Phys. Rev. Lett. {\bf 47}, 964 (1981).
\bibitem{Chitra} R. Chitra,
S. Pati, H.R. Krishnamurthy, D. Sen, and S. Ramasesha, Phys. Rev. B {\bf 52}, 6581 (1995);
S. Mori, I. Harada, and T. Tonegawa, J. Phys. Soc. Jpn. {\bf 67}, 1409 (1998).
\bibitem{Uhrig} G.S. Uhrig and H.J. Schulz, Phys. Rev. B {\bf 54}, R9624 (1996).
\bibitem{Affleck1} I. Affleck, in the Proceedings of the NATO ASI Workshop on Dynamical properties of unconventional magnetic systems, Geilo, Norway, April 1997; cond-mat/9705127 and references therein.
\bibitem{Uhrig1} G.S. Uhrig, F. Sch{\"{o}}nfeld, M. Laukamp, and E. Dagotto, cond-mat/9805245.
\bibitem{Affleck2} I. Affleck, D. Gepner, H.J. Schulz, and T. Ziman, J. Phys. A {\bf 22}, 511 (1989); R.R.P. Singh, M.E. Fisher, and R.Shankar, Phys. Rev. B {\bf 39}, 2562 (1989).
\bibitem{Bouzezar} G. Bouzerar, A.P. Kampf, and G.I. Japaridze,
Phys. Rev. B {\bf 58}, 3117 (1998).
\bibitem{Barnes} T. Barnes, J. Riera, and D.A. Tennant, cond-mat/9801224.
\bibitem{Singh} R.R.P. Singh and Z. Weihong, cond-mat/9811028.
\bibitem{Affleck} E. Sorensen, I. Affleck, D. Augier, and D. Poilblanc,
Phys. Rev. B {\bf 58}, 14701 (1998).
\bibitem{Bouzezar1} G. Bouzerar and S. Sil, cond-mat/9805042.
\bibitem{Chubukov} A.V. Chubukov, JETP Lett. {\bf 49}, 129 (1989).
\bibitem{Sachdev} S. Sachdev and R. Bhatt, Phys. Rev. B {\bf 41}, 9323 (1990).
\bibitem{Sushkov} V.N. Kotov, O.P. Sushkov, Z. Weihong, and J. Oitmaa, Phys. Rev. Lett.
{\bf 80}, 5790 (1998).
\bibitem{Shevchenko} P.V. Shevchenko and O.P. Sushkov, to appear in
Phys. Rev. B (1999),
cond-mat/9809368.
\bibitem{Kotov} V.N. Kotov, O.P. Sushkov, and R. Eder, to appear in
Phys. Rev. B, 1 March (1999), cond-mat/9808169.
\bibitem{remark} Strictly speaking there are other diagrams,
such as the "rainbow" contribution to the anomalous self-energy,
which are of the same order in the density. These additional
contributions however are quite small, as discussed in
Ref.\cite{Kotov}, and we neglect them here.
\bibitem{Kotov1} See, e.g. V.N. Kotov, J. Oitmaa, and Z. Weihong,
cond-mat/9812330, where the Green's functions incorporating
similar three-particle diagrams to lowest order are explicitly written.
\bibitem{Kotov2} V.N. Kotov and O.P. Sushkov, unpublished.
\bibitem{raj} R.R.P. Singh and Zheng Weihong, private communication.
\end{thebibliography}
\vspace{0.5cm}
\begin{figure}
\caption{
Diagrams representing (second-order) corrections to the one-particle
normal self-energy due to the three-particle scattering $H_{3}$.}
\label{fig1}
\end{figure}
\begin{figure}
\caption{
The elementary triplet gap,
$\Delta=\Omega_{k=0}/J$ versus $\alpha$ for
$\delta=0.2, 0.4$. The squares represent Exact Diagonalization data \protect\cite{Bouzezar,Barnes}.}
\label{fig2}
\end{figure}
\begin{figure}
\caption{
Diagrams representing the three-particle
scattering contributions to the two-particle bound states. a),b),c) for the singlet bound state, and a),b),c),d),e) for the triplet bound state.}
\label{fig3}
\end{figure}
\begin{figure}
\caption{
Energy spectrum, $\Omega_q/J$, for a) $\delta=0.2,\alpha=0.3$
(squares are Exact Diagonalization data \protect\cite{Bouzezar,Barnes}),
b) $\delta=0.6,\alpha=0.0$, c) $\delta=0.6,\alpha=0.15$,
d) $\delta=0.6,\alpha=0.2$, e) $\delta=0.8, \alpha=0.0$, f) $\delta=0.8, \alpha=0.3$. The solid lines are the elementary triplet
spectrum (lower curve) and the lower edge of the two-triplet
continuum (upper curve). The dashed and dot-dashed lines represent
respectively the triplet, $S=1$, and singlet, $S=0$, bound states.}
\label{fig4}
\end{figure}
\begin{figure}
\caption{
Binding energies, $\varepsilon^{S,T}/J$, of the two-particle singlet, $S=0$, and triplet, $S=1$,
bound states
at $Q=\pi/2$, $\alpha=0$ versus $\delta$, compared with the
numerical Exact Diagonalization results (filled circles) \protect\cite{Barnes}. The solid lines represent the calculations
where
the three particle scattering $H_3$ is treated in
simple perturbation theory, and the dashed lines represent the self-consistent
inclusion of $H_3$ into the integral equation \protect(\ref{bs}). }
\label{fig5}
\end{figure}
\begin{figure}
\caption{
Diagrams for the structure factors of a)
the elementary triplet, and b) the two-particle triplet. }
\label{fig6}
\end{figure}
\begin{figure}
\caption{
Structure factors $S_1(q)$ and $S_2(q)$ of the elementary and
the two-particle triplet:
a) along the disordered line $1=\delta+2\alpha$ ($\lambda=0$),\\
b) for $\delta=0.4,\alpha=0.2,0.4$. The solid lines are the results of the
calculations using Eqs.(\ref{str1},\ref{str2}).
The symbols connected by dotted lines are the dimer series expansions results \protect\cite{Singh} for $S_1(q)$. }
\label{fig7}
\end{figure}
\end{document}