cond-mat0410180/p.tex
1: %% ****** Start of file template.aps ****** %
2: %%
3: %%
4: %%   This file is part of the APS files in the REVTeX 4 distribution.
5: %%   Version 4.0 of REVTeX, August 2001
6: %%
7: %%
8: %%   Copyright (c) 2001 The American Physical Society.
9: %%
10: %%   See the REVTeX 4 README file for restrictions and more information.
11: %%
12: %
13: % This is a template for producing manuscripts for use with REVTEX 4.0
14: % Copy this file to another name and then work on that file.
15: % That way, you always have this original template file to use.
16: %
17: % Group addresses by affiliation; use superscriptaddress for long
18: % author lists, or if there are many overlapping affiliations.
19: % For Phys. Rev. appearance, change preprint to twocolumn.
20: % Choose pra, prb, prc, prd, pre, prl, prstab, or rmp for journal
21: %  Add 'draft' option to mark overfull boxes with black boxes
22: %  Add 'showpacs' option to make PACS codes appear
23: %  Add 'showkeys' option to make keywords appear
24: %\documentclass[aps,prl,preprint,groupedaddress]{revtex4}
25: %\documentclass[aps,prl,preprint,superscriptaddress]{revtex4}
26: 
27: %\documentclass[amsmath,amssymb,aps,prb,preprint,groupedaddress,showpacs,endfloats]{revtex4}
28: \documentclass[amsmath,amssymb,aps,prb,twocolumn,groupedaddress,showpacs,floatfix]{revtex4}
29: \usepackage{graphicx}
30: 
31: % You should use BibTeX and apsrev.bst for references
32: % Choosing a journal automatically selects the correct APS
33: % BibTeX style file (bst file), so only uncomment the line
34: % below if necessary.
35: %\bibliographystyle{apsrev}
36: 
37: \begin{document}
38: 
39: % Use the \preprint command to place your local institutional report
40: % number in the upper righthand corner of the title page in preprint mode.
41: % Multiple \preprint commands are allowed.
42: % Use the 'preprintnumbers' class option to override journal defaults
43: % to display numbers if necessary
44: %\preprint{NSF-KITP-04-51}
45: %Title of paper
46: \title{Quantum Fluctuations and Excitations in Antiferromagnetic Quasicrystals}
47: % repeat the \author .. \affiliation  etc. as needed
48: % \email, \thanks, \homepage, \altaffiliation all apply to the current
49: % author. Explanatory text should go in the []'s, actual e-mail
50: % address or url should go in the {}'s for \email and \homepage.
51: % Please use the appropriate macro foreach each type of information
52: % \affiliation command applies to all authors since the last
53: % \affiliation command. The \affiliation command should follow the
54: % other information
55: % \affiliation can be followed by \email, \homepage, \thanks as well.
56: %\email[Corresponding author : ]{wessel@phys.ethz.ch}
57: %\homepage[]{Your web page}
58: %\thanks{}
59: %\altaffiliation{}
60: \author{Stefan Wessel$^{(1)}$} 
61: \author{Igor Milat$^{(2)}$}
62: \affiliation{$^{(1)}$Institut f\"ur Theoretische Physik III, Universit\"at Stuttgart, 70550 Stuttgart, Germany}
63: \affiliation{$^{(2)}$Theoretische Physik, ETH Z\"urich, CH-8093 Z\"urich, Switzerland}
64: 
65: \date{\today}
66: \begin{abstract}
67: We study the effects of quantum fluctuations and the excitation spectrum for the antiferromagnetic Heisenberg model on 
68: a two-dimensional
69: quasicrystal, by numerically solving linear  spin-wave theory on finite approximants of the octagonal tiling. 
70: Previous quantum Monte Carlo results for the distribution of local staggered magnetic moments and the 
71: static spin structure factor are reproduced well within this approximate scheme.
72: Furthermore, the magnetic excitation spectrum consists of 
73: magnon-like low-energy modes, as well as dispersionless high-energy states of multifractal nature. 
74: The dynamical spin structure factor, accessible to inelastic neutron scattering, 
75: exhibits linear-soft modes at low energies, self-similar structures with bifurcations emerging at intermediate 
76: energies,
77: and
78: flat bands in high-energy regions.
79: We find that the distribution of local staggered moments stemming from  the inhomogeneity of the quasiperiodic structure
80: leads to a characteristic energy spread in the local dynamical spin susceptibility, 
81: implying distinct nuclear magnetic resonance spectra, specific for 
82: different local environments.
83: \end{abstract}
84: 
85: % insert suggested PACS numbers in braces on next line
86: \pacs{03.75.Hh,03.75.Lm,05.30.Jp}
87: % insert suggested keywords - APS authors don't need to do this
88: %\keywords{}
89: 
90: %\maketitle must follow title, authors, abstract, \pacs, and \keywords
91: \maketitle
92: 
93: % body of paper here - Use proper section commands
94: % References should be done using the \cite, \ref, and \label commands
95: 
96: \section{Introduction}
97: \label{sec:intr}
98: Quantum fluctuations are responsible for various degrees of disorder in low-dimensional quantum 
99: antiferromagnets. In particular two-dimensional systems show a variety of quantum
100: disordered phases, competing with conventional long-range magnetic order. 
101: For example, while the Heisenberg model on the square lattice 
102: exhibits true long-range order at zero 
103: temperature~\cite{manousakis}, spatial inhomogeneous magnetic exchange 
104: eventually leads to a complete suppression of magnetic order in structures 
105: such as the plaquette lattice, driven by  local singlet formation~\cite{plaquette,laeuchli}.
106: Other sources of magnetic disorder are frustration effects due to competing interactions~\cite{misguich}, and site/bond 
107: depletion~\cite{depletion}, where the reduction in the 
108: long-range magnetic 
109: order is accomplished  by  proliferation of localized low-energy excitations~\cite{mucciolo}.
110: All these systems share the translational invariance of the underlying lattice structure - assuming sufficient
111: self averaging in the case of quenched disorder. 
112: 
113: Quasiperiodic systems, lacking translational symmetry in addition to their inhomogeneous lattice structure were initially thought 
114: to not support sizeable correlations of localized magnetic moments.
115: However, recent neutron scattering 
116: experiments on  Zn-Hg-Ho~\cite{sato} as well as Cd-Mg-Tb~\cite{cdmgtb} icosahedral quasicrystals support the presence of 
117: significant magnetic 
118: correlations in these three-dimensional quasicrystalline compounds, with the absence of true long-range magnetic 
119: order~\cite{absence}  due to 
120: large frustrations in the 
121: antiferromagnetic exchange~\cite{wessel}.
122: 
123: \begin{figure}[t]
124: \begin{center}
125: \includegraphics[width=6cm]{inflat.eps}
126: \caption{
127: Finite approximant of the octagonal tiling with 239 sites (thin lines), along with a superimposed inflated 41 sites 
128: approximant (thick lines) of rescaled edge length by a factor of $\lambda=1+\sqrt{2}$. For the 41 sites approximant the vertices of one 
129: of the two sublattices are dressed with a disk to exhibit the bipartite nature of the octagonal tiling.
130: }
131: \label{fig:inflat}
132: \end{center}
133: \end{figure}
134: 
135: Bipartite, and thus unfrustrated, quasiperiodic crystal structures
136: were indeed shown to allow for sizeable two-sublattice antiferromagnetic order in a recent quantum Monte Carlo study of the spin-1/2 
137: Heisenberg 
138: model on the 
139: octagonal tiling~\cite{wessel}. Furthermore, the magnetic order in this two-dimensional system was found to exhibit
140: nontrivial patterns in 
141: the local staggered moment distribution, reflecting the self-similarity of the underlying quasiperiodic 
142: lattice structure~\cite{wessel}.
143: A  renormalization group approach based on this self-similarity indeed gives a gross account on the observed spread in the
144: staggered magnetization~\cite{anu}.
145: Furthermore, the static spin structure factor exhibits magnetic selection rules that impose a shift of reciprocal space 
146: indices~\cite{wessel}. 
147: The resulting neutron diffraction pattern~\cite{wessel} can be accounted for by analysis~\cite{lifshitz} of the quasicrystal 
148: spin 
149: group~\cite{ron}, also applicable to frustrated classical  models on the octagonal tiling with long-range exchange 
150: interactions~\cite{vedmedenko}.
151: 
152: While static properties of the Heisenberg antiferromagnet on the octagonal tiling are thus well studied, little is known about the 
153: spectral properties of these systems. On general grounds one would expect gapless Goldstone-modes to dominate 
154: at low-energies, even though the translational symmetry is absent. In addition, one would expect to find multifractional eigenstates,
155: as observed in tight-binding models on quasiperiodic lattices~\cite{grimm}.
156: Here, we investigate dynamical properties of quantum magnetic quasicrystals in order to 
157: identify the relevant energy scales of quantum fluctuations in such systems, determine the magnetic excitation spectrum, as 
158: well as to provide theoretical grounds for 
159: future experiments on magnetic quasicrystals, such as inelastic neutron scattering or magnetic resonance. In particular, 
160: we use linear spin-wave theory, which was successfully used in studies of periodic magnetically ordered systems, 
161: and apply it to the quasiperiodic case. 
162: 
163: The outline of the paper is as follows: Basic properties of the octagonal tiling are presented in the following section.  In 
164: Sec.~\ref{sec:line} we review linear spin wave theory in a real space formulation, and present
165: a numerical construction of the eigenmode expansion. An alternative scheme is given in the 
166: Appendix.
167: The results of applying this method to the octagonal tiling are discussed in Sec.~\ref{sec:resu}:
168: In Sec.~\ref{sec:stag} we discuss static 
169: properties of 
170: the magnetic correlations on the octagonal tiling, and compare our results to previous quantum Monte Carlo simulations. A detailed 
171: analysis of the excitation 
172: spectrum is presented in Sec.~\ref{sec:exci}, followed by a discussion of dynamical magnetic properties, such as the dynamical
173: spin structure factor in Sec.~\ref{sec:dyna}, and the local dynamical spin susceptibility (Sec.~\ref{sec:loca}). Finally, we conclude in 
174: Sec.~\ref{sec:conc} with a 
175: perspective on future investigations.
176: 
177:  
178: \section{Octagonal tiling}
179: \label{sec:octa}
180: In the following, we analyze the magnetic ground state properties and excitation spectrum of the nearest-neighbor 
181: antiferromagnetic spin$-1/2$ Heisenberg model,
182: \begin{equation}
183:  \label{eq:H-spin}
184:  H=J \sum_{\langle i,j \rangle} {\mathbf S}_i \cdot {\mathbf S}_j,\quad J>0, 
185: \end{equation}
186: on the most prominent example of a magnetic quasicrystal in two dimensions, the octagonal tiling.
187: The octagonal tiling is a bipartite quasiperiodic crystal system, and possesses an overall eightfold rotational symmetry, allowing for 
188: simple two-sublattice antiferromagnetism~\cite{lifshitz}. Sites in this tiling have coordination numbers $z$ ranging 
189: from 3 to 8, leading to a broad distribution of local staggered moments in the magnetically ordered ground state~\cite{wessel}.
190: A further important property of the octagonal tiling, in the absence of translational invariance, is its self-similarity under 
191: inflation 
192: transformations~\cite{levine}. This reversible operation refers to a well-defined decimation of a subset of vertices of the 
193: tiling, followed by a re-connection of the new vertices. Aside from a trivial rescaling of the length scale by a factor 
194: $\lambda=1+\sqrt{2}$, the infinite quasicrystal is left unchanged by this transformation. 
195: 
196: For our numerical study we consider finite square approximants of the octagonal tiling with $41$, $239$, and $1393$ sites.
197: These approximants can be obtained 
198: by the "cut-and-project" method
199: from a four-dimensional cubic lattice~\cite{duneau}, and 
200: are related by the inflation transformation, an example of which is shown in Fig.~\ref{fig:inflat}.
201: In order to avoid boundary-induced 
202: frustration effects, we 
203: apply toroidal boundary conditions~\cite{schulz}. 
204: Due to the lack of translational symmetry, we thus need to solve real-space linear spin-wave 
205: theory on lattices with up to 5572 sites. Before 
206: presenting our results, we provide details about the numerical scheme used in our calculations in the following section.
207: 
208: \section{Numerical spin-wave approximation}
209: \label{sec:line}
210: In this section, we review linear spin-wave theory, applied to the antiferromagnetic Heisenberg model on finite, bipartite 
211: lattices.
212: We consider a bipartite lattice with sublattices $A$ and $B$ consisting of $N_A$ and $N_B$ sites, respectively.
213: For the octagonal approximants considered in this work $N_A=N_B$, however
214: the following approach also applies if 
215: $N_A$ and $N_B$ are different. Following the standard Holstein-Primakoff approach\cite{spinwave}, we represent the spins in 
216: terms of 
217: bosonic operators $a_i$ and $b_i$. For $i\in A$, 
218: \begin{eqnarray}
219:   \label{eq:hpA}
220:   S^z_i & = &S - a^\dag_i a_i, \nonumber\\
221:   S^+_i &= &\sqrt{2S} \left(1 - \frac{a^\dag_i a_i}{2S}\right)^{1/2} a_i, \\
222:   S^-_i &= &\sqrt{2S} a^\dag_i \left(1 - \frac{a^\dag_i a_i}{2S}\right)^{1/2}, \nonumber
223: \end{eqnarray}
224: where $S$ denotes the spin magnitude, and
225: \begin{eqnarray}
226:   \label{eq:hpB}
227:   S^z_j & = &-S + b^\dag_j b_j, \nonumber\\
228:   S^+_j &= &\sqrt{2S} b^\dag_j \left( 1 - \frac{b^\dag_j b_j}{2S}\right)^{1/2},\\
229:   S^-_j &= &\sqrt{2S} \left(1 - \frac{b^\dag_j b_j}{2S}\right)^{1/2} b_j, \nonumber
230: \end{eqnarray}
231: for $j\in B$.
232: The linear spin-wave Hamiltonian is obtained by substituting the above identities in Eq.~(\ref{eq:H-spin}), 
233: expanding the square roots in of $1/S$, and keeping terms of lowest order $(1/S)^0$,
234: \begin{eqnarray}
235:   \label{eq:Hsw}
236:   H_{SW} &=& -J S(S + 1) N_b+ J S H_2,\\
237:   H_2&=& \sum_{\langle i,j \rangle}
238:   \left( a^\dag_i a_i + b_j b^\dag_j + a^\dag_i b^\dag_j + b_j a_i \right),
239: \end{eqnarray}
240: where $N_b$ denotes the number of bonds, and the sum in $H_2$ extends over all bonds of the bipartite lattice.
241: Introducing the $N_s=N_A+N_B$ component row vector
242: \begin{equation}
243:  \bar{a}^\dag=(a^\dag_1,...,a^\dag_{N_A},b_1,...,b_{N_B}),
244: \end{equation}
245: and the corresponding columnar conjugate, we can express the quadratic part of the Hamiltonian in matrix notation,
246: \begin{equation}
247:  H_2= \bar{a}^\dag M \bar{a} = \sum_{\langle i,j \rangle} \bar{a}^\dag M^{(i,j)} \bar{a},
248: \end{equation}
249: where
250: \begin{eqnarray}
251:  \label{eq:Mijkl}
252:  \left(M^{(i,j)}\right)_{k,l}&=&\delta_{i,k}\delta_{i,l}+\delta_{j+N_A,k}\delta_{j+N_A,l}\nonumber\\
253:                              & &+\delta_{i,k}\delta_{j+N_A,l}+\delta_{j+N_A,k}\delta_{i,l}
254: \end{eqnarray}
255: is the connectivity matrix of the lattice structure.
256: The bipartiteness of the lattice thus allows for a direct formulation in terms of the  $N_s \times N_s$ hermitian matrix $M$, 
257: instead of the 
258: general formulation
259: based on a $2 N_s \times 2 N_s$ matrix~\cite{blaizot}. 
260: We now seek a Bogoliubov transformation to the 
261: $N_n$ normal bosonic modes $\beta_k$, 
262: such that the spin-wave Hamiltonian $H_{SW}$ is diagonal when expressed in terms of the $\beta_k$,
263: \begin{equation}
264:   \label{eq:Hsw_diag}
265:   H_{SW} =N_s E_0+ \sum_{k=1}^{N_n} \omega_k \beta^\dag_k \beta_k,
266: \end{equation}
267: where $\omega_k>0$ denotes the eigenfrequency of the $k$-th mode, and $E_0$ the ground state energy per site.
268: To this end we make an Ansatz,
269: \begin{equation}
270:  \label{eq:T}
271:  \bar{a}=T\bar{\beta},
272: \end{equation}
273: with
274: \begin{equation}
275:  \bar{\beta}^\dag=(\beta^\dag_1,...,\beta^\dag_{N_+},\beta_{N_++1},...,\beta_{N_n}),
276: \end{equation}
277: so that we  divide the $N_n$ normal bosonic modes into two disjunct sets of
278: length $N_+$, and $N_-=N_n-N_+$, respectively, further determined
279: below.  Due to the bosonic commutation relations, the transformation
280: $T$ has to fulfill
281: \begin{equation}
282:  \label{eq:bcr}
283:  T \Gamma T^\dag = \Sigma,
284: \end{equation}
285: with matrices $\Gamma$ and $\Sigma$, defined by
286: \begin{equation} 
287: \label{eq:SL}
288:  \Gamma= \begin{pmatrix} 1_{N_+} & 0 \\ 0 & -1_{N_-} \end{pmatrix},
289:  \quad\Sigma= \begin{pmatrix} 1_{N_A} & 0 \\ 0 & -1_{N_B} \end{pmatrix}.
290: \end{equation}
291: Here, $1_{N}$ denotes the $N\times N$ identity matrix~\cite{white}.
292: Since $\Sigma^2 = 1$, as well as $\Gamma^2 = 1$, 
293: and because the left and right
294: inverse of a square matrix are identical, 
295: if $T$  satisfies
296: \begin{equation}
297: \label{eq:constr}
298: T^\dag \Sigma T = \Gamma,
299: \end{equation}
300: the bosonic commutation relations are fulfilled. Since $T$
301: diagonalizes $H_2$, we have
302: \begin{equation}
303:  T^\dag M T= \Omega, \quad \Omega={\rm diag}(\omega_1,...,\omega_{N_n}),
304: \end{equation}
305: from which we find upon multiplication from the left with $T \Gamma$,
306: that $T$ has to satisfy
307: \begin{equation}
308:  \Sigma M T=T\Gamma \Omega.
309: \end{equation}
310: The column vectors of $T$ are thus seen to be related to the right
311: eigenvectors of $\Sigma M$.  For semi-positive $M$, $\Sigma M$ has
312: real eigenvalues, and eigenvectors belonging to different eigenvalues
313: are orthogonal~\cite{blaizot}.  We denote the number of positive
314: (negative) eigenvalues by $N_+$ ($N_-$), the number of zero-modes by
315: $N_0$, and label the positive (negative) eigenvalues by $\lambda^+_i$,
316: $i=1,...,N_+$ ($\lambda^-_i$, $i=1,...,N_-$).  After numerically
317: solving the non-hermitian $N_s\times N_s$ eigenvalue problem for
318: $\Sigma M$~\cite{lapack}, we construct within the subspace of each
319: degenerate eigenvalue $\lambda^{\pm}_n>0 (<0)$ of dimension
320: $d^{\pm}_n$ eigenvectors $z^{\pm}_{n,1},...,z^{\pm}_{m,d^{\pm}_n}$,
321: obeying
322: \begin{equation}
323:  \label{eq:orthonormal}
324:  (z^{\pm}_{n,i})^\dag\Sigma z^{\pm}_{n,j}=\pm \delta_{i,j},\quad i,j=1,...,d^{\pm}_n\
325: \end{equation}
326: using Gram-Schmidt orthogonalization with respect to
327: $\Sigma$ (ref.~\onlinecite{lang}).  We then obtain $T$ from the orthonormal
328: eigenvectors as
329: \begin{equation}
330:  \label{eq:Tmatrix}
331:  T=(z^+_{1,1},...,z^+_{N_+,d^+_{N_+}},z^-_{1,d_1},...,z^-_{N_-,d^-_{N_-}}),
332: \end{equation}
333: where Eq.~(\ref{eq:orthonormal}) ensures that Eq.~(\ref{eq:constr}) is
334: satisfied.  The corresponding eigenfrequencies are given by
335: \begin{eqnarray}
336: \omega_k&=&J S \lambda^+_i,\quad k=1,...,N_+,\\
337: \omega_{N_++k}&=&- J S \lambda^-_k,\quad k=1,...,N_-.
338: \end{eqnarray}
339: The zero-modes of $\Sigma M$ correspond to collective modes due to the
340: broken continuous symmetry implied by the classical N\'eel
341: state~\cite{blaizot,mucciolo}.  Furthermore, the ground state energy becomes
342: \begin{equation}
343: E_0 N_s =-J S(S + 1) N_b+ J S \sum_{k=1}^{N_-} |\lambda^{-}_k|.
344: \end{equation}
345: An alternative means of numerically constructing the transformation matrix $T$ is presented in the appendix. We have 
346: verified, that both approaches indeed yield the same results.
347: 
348: \section{Results}
349: \label{sec:resu}
350: In this section, we present the results obtained by applying the linear spin-wave approximation to the octagonal tiling
351: introduced in Sec.~\ref{sec:octa}.
352: Within linear spin-wave theory, ground state expectation values of both static and dynamic magnetic correlations can 
353: be 
354: calculated from contractions of bosonic normal mode operators. It turns out convenient for this purpose, to
355: express the Bogoliubov transformation in terms of the row vectors
356: \begin{eqnarray}
357:  \tilde{a}^\dag&=&(a^\dag_1,...,a^\dag_{N_A},b^\dag_1,...,b^\dag_{N_B},a_1,...,a_{N_A},b_1,...,b_{N_B}),\nonumber\\
358:  \tilde{\beta}^\dag&=&(\beta^\dag_1,...,\beta^\dag_{N_n},\beta_1,...,\beta_{N_n}),\nonumber
359: \end{eqnarray}
360: as
361: \begin{equation}
362:  \label{eq:tildeT}
363:  \tilde{a}=\tilde{T}\tilde{\beta}, \quad \tilde{T}=\begin{pmatrix} U & V \\ V^* & U^* \end{pmatrix}.
364: \end{equation}
365: One obtains $U$ and $V$ from the transformation matrix $T$ of Eq.~(\ref{eq:T}), upon
366: defining 
367: $N_+\times N_+$ matrices $A$ and $B$, and $N_- \times N_-$ matrices $C$ and $D$ such that
368: \begin{equation}
369:  T=\begin{pmatrix} A & C \\ B & D \end{pmatrix},
370: \end{equation}
371: and gets
372: \begin{equation}
373:  U=\begin{pmatrix} A & 0 \\ 0 & D^* \end{pmatrix}, \quad V=\begin{pmatrix} 0 & C \\ B^* & 0 \end{pmatrix}. 
374: \end{equation}
375: 
376: 
377: \subsection{Staggered magnetization}
378: \label{sec:stag}
379: In this section we examine static properties of the magnetic ground state in the octagonal tiling.
380: For this purpose, we first calculate the staggered magnetization at each lattice site $i$
381: in linear spin-wave theory,
382: \begin{equation}
383:  \label{eq:lsm}
384:  m_s(i)=|\langle S^z_i \rangle |=S- \sum_k |V_{ik}|^2. 
385: \end{equation}
386: The spatially averaged staggered magnetization 
387: \begin{equation}
388: m_s=\sum_{i=1}^{N_s} m_s(i)
389: \end{equation}
390: is shown as a function of the system size $N_s$ of the approximant of the octagonal tiling
391: in Fig.~\ref{fig:fss}. The finite size values scale well as a function of $N_s^{-1/2}$, and indicate a sizeable staggered 
392: magnetization of the long ranged ordered ground state in the octagonal tiling.
393: The value of $m_s$ extrapolated to the thermodynamic limit is $m_s=0.34$,
394: and agrees well with the quantum Monte Carlo result, ${m}_s=0.337\pm 0.002$~\cite{wessel}. 
395: The  weak reduction of the order parameter by quantum fluctuations is indicative for the feasibility  of the linear 
396: spin-wave approach in this system. The ground state energy $E_0$, shown in the inset of Fig.~\ref{fig:fss}, is also found to 
397: scale well as a function of $N_s^{-3/2}$, as 
398: expected for an ordered state~\cite{einarsson}. The extrapolated value in the thermodynamic limit, $E_0=-0.646$, compares well 
399: with the quantum Monte Carlo result, $E_0=-0.6581(1)$~\cite{wessel_e0}.
400: 
401: \begin{figure}[t]
402: \begin{center}
403: \includegraphics[width=8cm]{fss.eps}
404: \caption{
405: Finite size scaling of the ground state staggered magnetization, $m_s$, and the lowest excitation gap, $\Delta$, for
406: the $S=1/2$ Heisenberg model on the octagonal tiling in linear spin-wave theory. The inset shows the finite size scaling of
407: the ground state energy, $E_0$, for the same system.
408: }
409: \label{fig:fss}
410: \end{center}
411: \end{figure}
412: 
413: In Ref.~\onlinecite{wessel}, the magnetic ground state on the octagonal tiling was found to exhibit a nontrivial 
414: local structure reflecting the self-similarity of the underlying quasiperiodic lattice structure. 
415: We now analyze to what extent spin-wave theory is able to reproduce this structure and thus to account for 
416: the specific nature of quantum fluctuations in an inhomogeneous connectivity.
417: For this purpose, we show in Fig.~\ref{fig:lsm} the linear spin-wave results of the local staggered magnetization, 
418: $m_s(i)$, for the 1393 sites approximant, and compare those with values grouped according to the coordination number $z$ of the various 
419: sites,
420: with the quantum Monte Carlo data of Ref.~\onlinecite{wessel}. 
421:   
422: We find linear spin-wave theory to qualitatively reproduce characteristic features of the local staggered moment distribution, 
423: such as (i) a wide 
424: spread of the moments, in particular for small values of $z$, (ii) a prominent bimodal splitting of the moments for $z=5$ and (iii) 
425: the hierarchical structure observed in the splitting of the moments for sites with $z=8$, shown in the inset of Fig.~\ref{fig:lsm}. 
426: These splittings in the local staggered moments can be accounted for by the 
427: properties of inequivalent sites under deflation transformations, reflecting their different local 
428: environments~\cite{wessel}.
429: In particular, fivefold sites with $z=5$ always occur in pairs, with two different types of  site. The first type is connected to 
430: four 
431: fourfold ($z=4$)
432: sites and the other fivefold site, while for the other type two neighbors are fourfold, two threefold ($z=3$) and one 
433: fivefold.
434: This difference in the local connectivity leads to the observed splitting~\cite{wessel}, and 
435: is also reflected in a  different behavior of the two types of fivefold sites under 
436: deflation transformations: while one type is decimated, the other remains as a threefold site in the deflated tiling.
437: The eightfold ($z=8$) sites exhibit an even richer, hierarchical structure shown in the inset of Fig.~\ref{fig:lsm}, with
438: moments grouped  according to the different deflation properties of the eightfold sites, 
439: namely their new coordination number $z'$. The high symmetry sites with $z'=8$ show a further hyperfine splitting, with 
440: moments grouped according to the value of  $z''$ under a second deflation, indicated by numbers next to the symbols in the inset 
441: of Fig.~\ref{fig:lsm}. As discussed 
442: in Ref.~\onlinecite{wessel}, these splittings eventually
443: lead to a multifractal distribution of local staggered moments in the infinite quasicrystal,  for this class of sites.
444: 
445: From a quantitative comparison of the local staggered magnetization between linear spin-wave theory and quantum Monte Carlo, 
446: we find 
447: characteristic limitations of the spin-wave 
448: approach to persist on a local level. Namely, while local staggered moments of high-connectivity sites are 
449: reproduced even quantitatively, deviations of 
450: about $8\%$ are observed for low-connectivity sites. In an inhomogeneous environment, linear spin-wave theory is thus 
451: more accurate at sites of large coordination, as might have been expected from its behavior in homogenous systems. 
452: 
453: The long-range antiferromagnetic order in the octagonal tiling
454: leads to characteristic neutron diffraction patterns,
455: due to selection rules imposed by the magnetic symmetry~\cite{wessel,lifshitz}.
456: These patterns can be obtained  from the 
457: static longitudinal structure factor 
458: \begin{equation}
459: S^{\parallel}({\mathbf k})=\sum_{i,j=1}^{N_s} e^{i{\mathbf k} \cdot ({\mathbf r}_i - {\mathbf r}_j)} \langle S^z_i S^z_j \rangle,
460: \end{equation}
461: which within linear spin-wave theory amounts to the Fourier transform of the real space distribution of $\langle S^z_i\rangle$.
462: Since linear spin-wave theory reproduces the spatial staggered moment distribution rather well, as seen from Fig.~\ref{fig:lsm}, 
463: the
464: resulting static longitudinal structure factor, shown in the left of Fig.~\ref{fig:Sxx_Szk} also compares well 
465: to the quantum Monte Carlo result~\cite{wessel}. In particular, it exhibits the extinction of 
466: nuclear Bragg peaks and the emergence of new, magnetic Bragg peaks. For a 
467: detailed theoretical derivation  of the various selection rules in the octagonal tiling, we refer to the explicit 
468: enumerations in Ref.~\onlinecite{lifshitz}.
469: 
470: 
471: \begin{figure}[t]
472: \begin{center}
473: \includegraphics[width=8cm]{ms.eps}
474: \caption{
475: Dependence of the local staggered magnetization on the coordination number $z$ for all sites in the 1393 sites 
476: approximant of the octagonal tiling within linear spin-wave theory of the $S=1/2$ Heisenberg model (LSWT). For reference,  results 
477: of quantum Monte Carlo simulations of the same system (QMC) are also shown~\cite{wessel}. The inset exhibits the hierarchy of 
478: the 
479: local 
480: staggered magnetization of the $z=8$ sites, grouped according to the value of $z'$ under a deflation transformation. Numbers next
481: to symbols give the value of $z''$ for $z'=8$ sites under a further deflation.
482: }
483: \label{fig:lsm}
484: \end{center}
485: \end{figure}
486: 
487: \begin{figure}[t]
488: \begin{center}
489: \includegraphics[width=4cm]{Szz_o_0.eps}
490: \includegraphics[width=4cm]{Sxx_o_sum.eps}
491: \caption{
492: Intensity plot of the static longitudinal magnetic structure factor $S^{\parallel}({\mathbf k})$ (left), and the 
493: integrated dynamical 
494: spin structure factor $S^{\perp}({\mathbf k})$ (right)
495: for the $S=1/2$ Heisenberg antiferromagnet on the 1393 sites approximant of the 
496: octagonal tiling.
497: }
498: \label{fig:Sxx_Szk}
499: \end{center}
500: \end{figure}
501: 
502: 
503: The structural distribution of local staggered moments observed in Fig.~\ref{fig:lsm} exhibits an inhomogeneous distribution of 
504: quantum
505: fluctuations. As seen from Eq.~(\ref{eq:lsm}), fluctuations in $m_s$ arise from distinct contributions from the various 
506: eigenstates of the
507: system. 
508: To quantify the relevance of the different eigenstates for the quantum fluctuations, 
509: we study the reduction of the staggered magnetization at each lattice site $i$, which can be parameterized as a function of 
510: energy,
511: \begin{equation}
512: \delta m_s(i,\omega)=\frac{\sum_k |V_{ik}|^2 \delta(\omega-\omega_k)}{\sum_k \delta(\omega-\omega_k)}.
513: \label{eq:dms}
514: \end{equation}
515: In Fig.~\ref{fig:dms} we show $\delta m_s(i,\omega)$, averaged separately over sites with $z=3$, and 
516: $z=8$.~\cite{footnote1}
517: 
518: In both cases, the dominant contribution to quantum fluctuations stem from the low-energy modes, with  $\omega/JS<2$. 
519: For threefold sites,
520: further contributions to $\delta m_s$  arise from higher energy modes, which are not relevant for the eightfold sites. 
521: In general, we find the  upper bound of the 
522: energy range that is relevant for quantum 
523: fluctuations  to
524: decrease with increasing coordination number $z$. 
525: Although sites with low coordination numbers thus receive quantum fluctuations over a larger range in energy space, their
526: staggered moment is typically larger than for high coordinated sites, as seen from Fig.~\ref{fig:lsm}.
527: A transfer of relevant quantum fluctuations
528: to low-energy modes is thus responsible for a decrease of the staggered moment at specific sites. 
529: This increased relevance of low-energy modes for quantum fluctuations 
530: is also observed for sites with the same coordination number. 
531: For example, the inset of 
532: Fig.~\ref{fig:dms} shows $\delta m_s$ for the fivefold sites, where the bimodal splitting was observed in Fig.~\ref{fig:lsm}, 
533: averaged 
534: separately over sites with a small and a large moment, respectively. 
535: The additional reduction of the local moment for one type of fivefold sites is clearly seen to be due to a proliferation of 
536: quantum fluctuations in the lower energy region.
537: 
538: This hints at a close  link between the structure of the inhomogeneous magnetic ground state and the
539: magnetic excitation spectrum.
540: In the following subsection, we proceed to analyze the magnetic excitations 
541: in more detail.
542: 
543: 
544: \begin{figure}[t]
545: \begin{center}
546: \includegraphics[width=8cm]{dms.eps}
547: \caption{
548: Frequency dependence of quantum fluctuations to the staggered magnetization, $\delta m_s(\omega)$, for the $S=1/2$ 
549: Heisenberg model on the octagonal tiling, averaged separately over sites with 
550: coordination numbers $z=3$ and $8$. The inset shows $\delta m_s(\omega)$ for $z=5$, averaged separately over sites with a
551: small and  large staggered magnetization, respectively.
552: }
553: \label{fig:dms}
554: \end{center}
555: \end{figure}
556: 
557: \subsection{Excitation Spectrum}
558: \label{sec:exci}
559: In the previous section we found that
560: spin-wave theory provides a qualitative, and even quantitative account on
561: the magnetic ground state properties of the Heisenberg model on the   
562: octagonal tiling. 
563: We now proceed to use linear spin-wave theory to gain insights into the 
564: spectral properties of this quasiperiodic antiferromagnet.
565: 
566: The spectra of spin-wave normal modes for the first three approximants of the octagonal tiling
567: are shown in Fig.~\ref{fig:dos}. For finite approximants the spectra consists of discrete sets of 
568: energy levels, spanning an extended range up to $\omega_{max}/JS\approx 7.3$. This implies a bandwidth almost a factor two
569: larger than for the Heisenberg model on the square lattice, where $\omega_{max}/JS=4$.
570: Upon increasing the system size, the energy spectrum appears to develop a
571: dense band, and two  isolated pockets at higher energies, near $\omega/JS\approx6.5$ and $7.3$, respectively. The presence of 
572: a third 
573: pocket near $\omega/JS\approx5.5$ cannot be excluded from the finite size data. However, from Fig.~\ref{fig:dos} the gap near 
574: $\omega/JS\approx 5$ appears to eventually close for higher approximants, whereas 
575: the gaps to the  higher energy pockets do not show 
576: any decrease for the approximants considered here.
577: 
578: For a more quantitative analysis of the distribution of normal modes, we calculate the density of states of the spin-wave excitation 
579: spectrum (DOS), 
580: \begin{equation}
581: \rho(\omega)=\sum_k \delta(\omega-\omega_k),
582: \end{equation}
583: shown for the largest approximant in the inset of Fig.~\ref{fig:dos}. The spin-wave DOS exhibits a characteristic spiky shape in the 
584: high-energy regime, in particular for energies larger than 
585: $\omega/JS\approx 3$, similar to shapes found for the tight-binding 
586: Hamiltonian on the 
587: same lattice structure~\cite{anu2000}.
588: In the low-energy region, the DOS appears more smooth, with a residual 
589: roughness due to the limited resolution of the  
590: energy spectra due to finite size effects. We indeed find similar resolution limited roughness also on finite square lattice systems.
591: 
592: On general grounds, the presence of a long-range ordered ground state is expected to characterise  the low-energy properties 
593: of 
594: the excitation spectrum.
595: In particular, due to the broken $SU(2)$ symmetry, we expect 
596: the gap to the lowest excitation, $\Delta$, to close
597: in the thermodynamic limit of the quasiperiodic tiling.
598: A finite size scaling analysis of $\Delta$, shown in 
599: Fig.~\ref{fig:fss}, is indeed consistent with $\Delta\propto N_s^{-\alpha}$, where $\alpha\approx 0.5$.
600: At low energies the DOS furthermore increases linearly, $\rho(\omega)\propto\omega$,
601: as seen from the quadratic low-energy behavior of the 
602: cumulative density of states,
603: \begin{equation}
604: N(\omega)=\int_0^{\omega} d\epsilon \: \rho(\epsilon),
605: \end{equation}
606: shown in Fig.~\ref{fig:cdos}. Here, we employ $N(\omega)$, since this quantity is less susceptible to resolution limited 
607: roughness than $\rho(\omega)$, due to the frequency integration.
608: The low-energy features of the spin-wave DOS in the octagonal tiling are also observed 
609: in  periodic antiferromagnets. For example, on the square lattice, antiferromagnetic 
610: spin-waves obey a linear dispersion relation at low energies
611: $\omega_{\mathbf k}\approx \sqrt{8}JS |{\mathbf k}|$,
612: resulting in a linear low-energy DOS in this system.
613: 
614: 
615: \begin{figure}[t]
616: \begin{center}
617: \includegraphics[width=8cm]{dos.eps}
618: \caption{
619: Excitation spectra of the Heisenberg antiferromagnet on finite approximants of the octagonal tiling of different sizes within 
620: linear spin-wave theory. The inset shows the density of state, $\rho(\omega)$, for the largest approximant. 
621: }
622: \label{fig:dos}
623: \end{center}
624: \end{figure}
625: \begin{figure}[t]
626: \begin{center}
627: \includegraphics[width=8cm]{cdos.eps}
628: \caption{
629: Cumulative density of states, $N(\omega)$, of the Heisenberg antiferromagnet on finite approximants of the octagonal tiling of 
630: different sizes within linear spin-wave theory. The inset exhibits the quadratic scaling of $N(\omega)$ at low energies 
631: for the largest approximant, indicative of a linear low-energy density of states.
632: }
633: \label{fig:cdos}
634: \end{center}
635: \end{figure}
636: 
637: The similarity of the low-energy DOS for the Heisenberg model on the quasiperiodic octagonal tiling to the periodic case suggest 
638: that the low-energy excitation could be delocalized, magnon-like modes also in the octagonal tiling. We are thus lead to 
639: analyze the spatial extent of the 
640: eigenstates found within linear spin-wave theory.
641: An appropriate method to characterize the localization properties
642: is the inverse participation ratio (IPR)~\cite{wegner} which expresses the spatial extent of the wavefunction of a given state.
643: For the bosonic  excitations of quantum magnets, the IPR of an eigenstate is given 
644: by its contribution to the quantum fluctuations of the staggered magnetization.
645: Following Ref.~\onlinecite{mucciolo},
646: we  define the energy dependent inverse participation ratio 
647: \begin{equation}
648:  I(\omega)=\frac{\sum_k\: I_k\: \delta(\omega-\omega_k)}{\sum_k \delta(\omega-\omega_k)},
649: \end{equation}
650: where
651: \begin{equation}
652:  I_k=\frac{\sum_{i} |V_{ik}|^4}{\left(\sum_{i} |V_{ik}|^2 \right)^2},
653: \end{equation}
654: and study its scaling behavior upon increasing the system size. 
655: For delocalized states of a $d$ dimensional quantum system, the IPR 
656: decreases with the system size, $N_s$, as $N_s^{-1}$.
657: Exponentially localized states should be very insensitive to the system size and one expects a size independent IPR.
658: 
659: Fig.~\ref{fig:ipr} shows the calculated IPR as a function of energy for the spin$-1/2$ Heisenberg model on finite approximants
660: of the octagonal tiling. For comparison, we show in the inset the corresponding quantity for the Heisenberg model 
661: on the square lattice. In  the square lattice case, all eigenstates show the characteristic $N_s^{-1}$ scaling, expected for
662: extended magnon states. For the octagonal tiling, we find such behavior only for the low-energy modes, but due to the limited 
663: resolution cannot exclude a reduced finite size scaling down to zero frequency.
664: The low-energy modes in the octagonal tiling thus appear as extended excitations out of the antiferromagnetic ground state, 
665: similar to coherent magnons in the periodic case. The higher energy states also do not appear exponential localized, but 
666: show a significantly reduced finite size scaling 
667: $I(\omega)\propto N_s^{-\beta}$, with a scaling exponent $\beta<1$, characteristic for multifractal states, as observed for 
668: critical states at the Anderson localization transition~\cite{schreiber}. 
669: The effective exponent $\beta$ decreases towards the upper edge of the spectrum, being lowest for energies near the
670: isolated pockets found in the DOS in Fig.~\ref{fig:dos}. 
671: While a reliable determination of $\beta$ would require the study of substantially larger approximants, we estimate a value of 
672: $\beta\approx 0.2$ for the high-energy modes.
673: The multifractality of eigenstates 
674: found in non-interacting models on quasiperiodic crystals~\cite{grimm} is thus clearly observed also for excitations of 
675: strongly correlated systems, such as the Heisenberg antiferromagnet.
676: 
677: \begin{figure}[t]
678: \begin{center}
679: \includegraphics[width=8cm]{ipr.eps}
680: \caption{
681: Frequency dependent inverse participation ratio, $I(\omega)$, for the $S=1/2$ Heisenberg antiferromagnet on finite approximants 
682: of the octagonal tiling of different sizes, obtained within linear spin-wave theory. Dashed lines indicate a finite size scaling 
683: as $N_s^{-1}$, expected for extended states.
684: The inset shows linear spin-wave results for $I(\omega)$ on square lattices of different sizes. 
685: }
686: \label{fig:ipr}
687: \end{center}
688: \end{figure}
689: 
690: \subsection{Dynamical spin structure factor}
691: \label{sec:dyna}
692: Having analyzed the spin-wave excitations of the Heisenberg antiferromagnet on the octagonal tiling,
693: we  proceed to study magnetic properties exhibiting the magnetic excitation spectrum, such as the dynamical spin structure 
694: factor, and  local dynamical spin susceptibility.
695: 
696: The dynamical spin structure factor, accessible experimentally by inelastic neutron scattering, reflects the time-dependent 
697: spin-spin 
698: correlation functions, transformed to  momentum space, 
699: \begin{equation}
700:  S^{\perp}({\mathbf k},\omega)=\frac{1}{N_s}\sum_{i,j=1}^{N_s} e^{i{\mathbf k} \cdot ({\mathbf r}_i - {\mathbf r}_j)} 
701: S^{\perp}(i,j,\omega),
702: \end{equation}
703: with 
704: \begin{equation}
705:  S^{\perp}(i,j,\omega)=\frac{1}{2}\int_{-\infty}^{\infty} dt\: e^{i\omega t} \langle S^+_i(t) S^-_j(0) + S^-_i(t) S^+_j(0) 
706: \rangle.
707: \end{equation}
708: Using the normal-mode expansion, Eq. (\ref{eq:tildeT}), we obtain the following expression for $S^{\perp}(i,j,\omega)$ within 
709: linear 
710: spin-wave theory,
711: \begin{equation}
712:  S^{\perp}(i,j,\omega)=S \sum_k \left( U_{ik} U^*_{jk} + V^*_{ik} V_{jk} \right) \delta(\omega-\omega_k),
713: \end{equation}
714: from which $S^{\perp}({\mathbf k},\omega)$ is readily obtained using a fast Fourier transformation. Most spectral 
715: weight in $S^{\perp}({\mathbf k},\omega)$ 
716: is located at momenta corresponding to magnetic Bragg peaks, as seen by comparing the integrated dynamical spin structure  
717: factor,
718: \begin{equation}
719:  S^{\perp}({\mathbf k})=\int \frac{d\omega}{2\pi} \: S^{\perp}({\mathbf k},\omega),
720: \end{equation}
721: shown in the right of Fig.~\ref{fig:Sxx_Szk} to the static longitudinal structure factor, $S^{\parallel}({\mathbf k})$, 
722: in the left of Fig.~\ref{fig:Sxx_Szk}.
723: For a detailed analysis of the dynamical spin structure factor, 
724: we consider both constant-frequency 
725: scans, shown in Figs.~\ref{fig:Sxx_o_1} to~\ref{fig:Sxx_o_3}, as well as scans along various momentum 
726: space 
727: directions, 
728: shown in Figs.~\ref{fig:Sxx_k_1} and~\ref{fig:Sxx_k_2}. To increase the contrast in these figures, we have rescaled 
729: the data to the maximum value in each plot, separately. 
730: 
731: \begin{figure}[t]
732: \begin{center}
733: \includegraphics[width=4cm]{Sxx_o_0.eps}
734: \includegraphics[width=4cm]{Sxx_o_1.8.eps}
735: \caption{
736: Intensity plot of the dynamical spin structure factor, $S({\mathbf k},\omega)$,
737: for fixed $\omega/JS=0$ (left) and $\omega/JS=1.8$ (right)
738: for the $S=1/2$ Heisenberg antiferromagnet on  the 1393 sites approximant of the 
739: octagonal tiling.
740: }
741: \label{fig:Sxx_o_1}
742: \end{center}
743: \end{figure}
744: 
745: In order to compare the relative spectral weight at different energies, we show in Fig.~\ref{fig:Sxx_o_all}
746: the momentum-integrated spectral function,
747: \begin{equation}
748: S^{\perp}(\omega)=\int\frac{d^2 k}{(2\pi)^2} \: S({\mathbf k},\omega).
749: \end{equation}
750: Compared to the DOS, 
751: we find that apart from the low-energy 
752: region below $\omega/JS\approx 2$, the shape of $S^{\perp}(\omega)$ closely reflects the structures in the DOS. In the 
753: low-energy region, we observe an disproportionately large contribution to $S^{\perp}(\omega)$, in contrast to the low DOS 
754: in this region. The difference between $S^{\perp}(\omega)$, and the DOS, $\rho(\omega)$, in linear spin-wave theory is obtained
755: using Eq. (\ref{eq:tildeT}) as 
756: \begin{equation}
757: S^{\perp}(\omega)-S\rho(\omega)=2 S \sum_{i,k} |V_{ik}|^2  \delta(\omega-\omega_k),
758: \end{equation}
759: and indicates, that the extra contributions to $S^{\perp}(\omega)$ at energies $\omega/JS<2$ are due to the spectral predominance
760: of low energy quantum fluctuations, $\delta m_s(\omega)$, Eq.~(\ref{eq:dms}), shown in Fig.~\ref{fig:dms}. 
761: In the following, we first consider the dynamical spin structure factor in this low-energy region, which 
762: is dominated by magnetic Bragg scattering. 
763: 
764: \begin{figure}[t]
765: \begin{center}
766: \includegraphics[width=8cm]{Sxx_o_all.eps}
767: \caption{
768: Frequency dependence of the momentum-integrated dynamical spin structure factor, $S^{\perp}(\omega)$, 
769: for the $S=1/2$ Heisenberg antiferromagnet on the octagonal tiling.
770: }
771: \label{fig:Sxx_o_all}
772: \end{center}
773: \end{figure}
774: 
775: In the elastic limit of $S^{\perp}({\mathbf k},\omega\rightarrow 0)$, shown in the left of Fig.~\ref{fig:Sxx_o_1}, we 
776: can indeed identify the
777: magnetic Bragg peak positions of the static longitudinal structure factor, shown in Fig.~\ref{fig:Sxx_Szk}.
778: Furthermore, for all energies up to $\omega/JS\approx 2$, similar patterns as in $S^{\perp}({\mathbf k},0)$ are 
779: observed, albeit with the 
780: width of the peaks increasing upon increasing the energy. Eventually, these peaks  evolve into ring-like structures, 
781: centered around the magnetic Bragg peaks,  such 
782: as shown for $\omega/JS=1.8$ in the right part of Fig.~\ref{fig:Sxx_o_1}.  This is a clear indication for
783: magnetic soft-modes at low energies, which dominate the magnetic response in this energy regime.
784: 
785: \begin{figure}[t]
786: \begin{center}
787: \includegraphics[width=4cm]{Sxx_o_2.9.eps}
788: \includegraphics[width=4cm]{Sxx_o_3.eps}
789: \caption{
790: Intensity plot of the dynamical spin structure factor, $S({\mathbf k},\omega)$,
791: for fixed $\omega/JS=2.9$ (left) and $\omega/JS=3.0$ (right)
792: for the $S=1/2$ Heisenberg antiferromagnet on  the 1393 sites approximant of the 
793: octagonal tiling.
794: }
795: \label{fig:Sxx_o_2}
796: \end{center}
797: \end{figure}
798: 
799: Typical examples of $S^{\perp}({\mathbf k})$ at higher frequencies are shown in 
800: Fig.~\ref{fig:Sxx_o_2},
801: and~\ref{fig:Sxx_o_3}. We find all plots to exhibit an eightfold overall 
802: symmetry, as expected for the octagonal tiling. However, the positions of the dominant peaks are different from the 
803: magnetic Bragg peaks 
804: found below $\omega/JS\approx2$. For example, for $\omega/JS=3.0$ (right of Fig.~\ref{fig:Sxx_o_2}), most spectral weight is 
805: located at ${\mathbf k}\approx(0, 1.3 \pi)$, and
806: ${\mathbf k}\approx(0, 4.3 \pi)$, as well as symmetry-related momenta. Peaks at these momenta are absent in both 
807: $S^{\parallel}({\mathbf k},0)$, and the nuclear Bragg scattering (Fig.~5 (a) of Ref.~\onlinecite{wessel}). In fact, we do 
808: not 
809: observe pronounced spectral weight at nuclear Bragg peak positions for any finite energy cut: to give a  further example
810: of this fact, we show 
811: $S^{\perp}({\mathbf k})$ at $\omega/JS=4.6$ in the left of Fig.~\ref{fig:Sxx_o_3}.
812: Upon changing the energy-level of the cut  only  slightly, the patterns found 
813: in the high-energy
814: region
815: change more drastically 
816: than those in the low-energy regime. As an example, in the left 
817: of Fig.~\ref{fig:Sxx_o_2}, we show $S^{\perp}({\mathbf k})$ for $\omega/JS=2.9$. 
818: Compared to $S^{\perp}({\mathbf k})$ at $\omega/JS=3.0$ (right of Fig.~\ref{fig:Sxx_o_2}), we indeed find a different set of 
819: dominant peaks. 
820: %% IMILAT: I don't understand what U want to say.
821: %%
822: %% This suggests
823: %% that a single mode approximation will not be adequate for describing the spin-wave spectrum in this energy region. 
824: %%
825: %%
826: 
827: \begin{figure}[t]
828: \begin{center}
829: \includegraphics[width=4cm]{Sxx_o_4.6.eps}
830: \includegraphics[width=4cm]{Sxx_o_6.4.eps}
831: \caption{
832: Intensity plot of the dynamical spin structure factor, $S({\mathbf k},\omega)$,
833: for fixed $\omega/JS=4.6$ (left) and $\omega/JS=6.4$ (right)
834: for the $S=1/2$ Heisenberg antiferromagnet on  the 1393 sites approximant of the 
835: octagonal tiling.
836: }
837: \label{fig:Sxx_o_3}
838: \end{center}
839: \end{figure}
840: 
841: The patterns in $S^{\perp}({\mathbf k})$ for energies corresponding to the high-energy pockets in the DOS 
842: (c.f.~Fig.~\ref{fig:dos})
843: consist of more diffusive structures than those at lower energies.
844: For example, we find broad ring-like structures, centered around the magnetic Bragg peaks
845: in $S^{\perp}({\mathbf k})$ for $\omega/JS=6.4$, shown in the right of Fig.~\ref{fig:Sxx_o_3}.
846: The appearance of such diffusive structures at high energies is expected from the results of 
847: Sec.~\ref{sec:exci}, where the 
848: the high-energy excitations were 
849: found 
850: to be spatially less extended than the low-energy states. 
851: This reduced spatial extent leads to smeared 
852: diffraction patterns observed in Fig.~\ref{fig:Sxx_o_3}.
853: 
854: \begin{figure}[t]
855: \begin{center}
856: \includegraphics[width=4cm]{Sxx_k_5_2.eps}
857: \includegraphics[width=4cm]{Sxx_k_9.15.eps}
858: \caption{
859: Intensity plot of the dynamical spin structure factor, $S({\mathbf k},\omega)$,
860: along the $(5,2)$ momentum space direction (left) and along the line $(2.9\pi,k_y)$ (right)
861: for the $S=1/2$ Heisenberg antiferromagnet on  the 1393 sites approximant of the 
862: octagonal tiling.
863: }
864: \label{fig:Sxx_k_1}
865: \end{center}
866: \end{figure}
867: 
868: In order to analyze the momentum dependence of the dominant peaks in the dynamical spin structure factor, we choose 
869: representative 
870: directions in momentum 
871: space, and plot $S^{\perp}({\mathbf k},\omega)$ along such cuts. We first consider momenta lying along the $(5\; 2)$ 
872: direction, 
873: shown in the left of 
874: Fig.~\ref{fig:Sxx_k_1}. 
875: This cut passes through two of the major magnetic Bragg peaks, namely at $|{\mathbf k}_B|/\pi\approx 1.3$,  
876: and $3.2$. In the low-energy region, below $\omega/JS\approx 2$, most spectral weight is located along straight lines, emerging 
877: from the magnetic Bragg peaks, and  with spectral weight that increases for decreasing energy, characteristic of magnetic 
878: soft-modes.
879: Near the magnetic Bragg peaks, ${\mathbf k}_B$, we thus observe linear dispersion relations
880: of the magnetic excitations, $\omega=c |{\mathbf 
881: k}-{\mathbf k}_B|$, with 
882: an estimated spin-wave velocity $c/JS \approx 2.1$, which is of similar order of magnitude than the linear spin-wave 
883: result for the square lattice 
884: ($c/JS=\sqrt{8}\approx 2.83$). Similar soft-modes are 
885: also observed along the cut of constant $k_x/\pi=2.9$, shown in the right of Fig.~\ref{fig:Sxx_k_1}, including a
886: further magnetic Bragg peak at ${\mathbf k}_B~\approx(2.9\pi,4.8\pi)$.
887: The emergence of these linear low-energy dispersion relations is consistent with the  linear 
888: low-energy DOS found in Sec.~\ref{sec:exci}, and furthermore explains the ring-like structures 
889: in $S^{\perp}({\mathbf k})$, as seen for $\omega/JS=1.8$ in the right of Fig.~\ref{fig:Sxx_o_2}. 
890: 
891: In spite of the absence of translational symmetry, the dynamical spin structure factor of 
892: the antiferromagnetic quasicrystal clearly exhibits 
893: the presence of soft-modes near the magnetic Bragg peak positions of the quasiperiodic 
894: crystal. We expect such a generic feature of magnetic long-range order to be present also in other magnetically ordered 
895: quasicrystals. 
896: 
897: In contrast, at high frequencies, $\omega/JS>5.5$, we do not observe any significant dispersion of the spectral weight 
898: distribution. Instead, we find two flat bands of only slightly modulated spectral weight located near $\omega/JS\approx 6.4$, 
899: and $7.3$ in Fig.~\ref{fig:Sxx_k_2}, which 
900: correspond
901: to the two isolated pockets of the spin-wave DOS in  Fig.~\ref{fig:dos}. We consider this  observation as further 
902: indication for the limited spatial extent of the corresponding eigenstates, concluded in Sec.~\ref{sec:exci} from the finite size 
903: scaling behavior of the inverse participation ratio.
904: 
905: \begin{figure}[t]
906: \begin{center}
907: \includegraphics[width=4cm]{Sxx_k_1_0.eps}
908: \includegraphics[width=4cm]{Sxx_k_3_2.eps}
909: \caption{
910: Intensity plot of the dynamical spin structure factor, $S({\mathbf k},\omega)$,
911: along the $(1,0)$  momentum space direction (left) and the $(3,2)$ direction (right)
912: for the $S=1/2$  Heisenberg antiferromagnet on  the 1393 sites approximant of the 
913: octagonal tiling.
914: }
915: \label{fig:Sxx_k_2}
916: \end{center}
917: \end{figure}
918: 
919: 
920: In the intermediate energy regime, between  $\omega/JS\approx 2$, and $\omega/JS\approx 5.5$, the distribution of 
921: spectral weight is more complex, and can be accounted for by  band-like segments, which recur at different energies 
922: and 
923: with varying bandwidths, as shown in  Fig.~\ref{fig:Sxx_k_2}. 
924: At points of increased spectral weight, such as for
925: $\omega/JS=3$ near $|{\mathbf k}|/\pi\approx 1.8$, and $4.2$, in the left of Fig.~\ref{fig:Sxx_k_2}, 
926: the corresponding gapped modes furthermore show a linear dispersion relation.
927: In addition, we find bifurcations emerging as branches of these band-segments
928: extending towards low-energies. This is seen for example in the right part of Fig.~\ref{fig:Sxx_k_2}, for 
929: momenta $|{\mathbf k}|<\pi$. 
930: Such self-similar structures might have been expected to dominate the dynamical spin structure factor,  due to the geometric 
931: properties of the octagonal tiling, reflecting its inflation symmetry. 
932: Nevertheless, we find them well separated in energy from more conventional low-energy features, that reflect the 
933: magnetic order in this system. 
934: 
935: \subsection{Local Dynamical Spin Susceptibility}
936: \label{sec:loca}
937: While the dynamical spin structure factor thus exhibits the peculiar nature of the excitations in the self-similar 
938: quasiperiodic system, the different local environments of the magnetic moments in the quasicrystal are accessible from the 
939: local dynamical spin 
940: susceptibility, the imaginary part of which  at 
941: each lattice site $i$ is given by
942: \begin{equation}
943:  \chi''_{\mathrm local}(i,\omega)=S^{\perp}(i,i,\omega)=S \sum_k \left( |U_{ik}|^2 + |V_{ik}|^2 \right) \delta(\omega-\omega_k),
944: \end{equation}
945: within linear spin-wave theory. The local dynamical spin
946: susceptibility
947: is accessible in nuclear magnetic resonance experiments,
948: in the form of Knight-shifts at nuclear sites in the vicinity of the magnetic sites. Here, we study the properties
949: of $\chi''_{\mathrm local}$ for the Heisenberg model on the octagonal tiling, as an example of a quasiperiodic lattice structure.
950: 
951: 
952: In Fig.~\ref{fig:sii}, $\chi''_{\mathrm local}$ is shown for the largest approximant, averaged separately for 
953: sites of different coordination. We observe a broad spread in the signal, with characteristic energies that
954: increase linearly from $\omega/JS\approx2.5$ for threefold sites to $\omega/JS\approx7.3$ for eightfold sites. 
955: Furthermore,
956: the energy-range over which there is a large signal narrows for sites of increasing coordination, reflecting a similar trend in the 
957: spread of the local staggered moments (c.f. Fig.~\ref{fig:lsm}). The inequivalent local environments of the various sites are responsible for this
958: extended range of signals. For example, the left insets of Fig.~\ref{fig:sii} exhibits that the two types of fivefold sites
959: show a different frequency dependence of the dynamical spin susceptibility. Namely, sites with a smaller moment have signals inside a narrow region 
960: near 
961: $\omega/JS\approx 4.5$, whereas sites with a larger moment produce signals over a more extended region, ranging from $\omega/JS\approx4.2$ to 
962: $\omega/JS\approx 5$. The high symmetry eightfold sites also exhibit a characteristic splitting in the local spin susceptibility, as seen in the 
963: right inset of Fig.~\ref{fig:lsm}. Here, the individual signals are labeled by the value of $z'$
964: for the site from which this signal results.
965: We observe from Fig.~\ref{fig:lsm}, that the widths of the signals narrow towards  $\omega/JS\approx 7.3$ for increasing values of $z'$.
966: The local dynamical spin susceptibility thus reflects the hierarchical structure of the local moment distribution of the eightfold sites. 
967: We expect such features to be generic properties of quasiperiodic magnets, eventually seen  
968: in nuclear magnetic resonance experiments on real quasiperiodic magnetic systems.
969: 
970: \begin{figure}[t]
971: \begin{center}
972: \includegraphics[width=8cm]{Sii.eps}
973: \caption{
974: Imaginary part of the local dynamical spin susceptibility, $\chi''_{\mathrm local}(i,\omega)$, 
975: for the $S=1/2$ Heisenberg antiferromagnet on the  octagonal tiling,
976: averaged separately over sites with
977: coordination numbers $z=3$ to $8$.
978: The left inset shows $\chi''_{\mathrm local}(i,\omega)$ for the fivefold sites, averaged separately over sites with a
979: small and  large staggered magnetization, respectively. The right inset shows $\chi''_{\mathrm local}(i,\omega)$ for the eightfold sites,
980: averaged separately over sites with different behavior under deflation transformation, grouped according to the value of $z'$.
981: }
982: \label{fig:sii}
983: \end{center}
984: \end{figure}
985: 
986: 
987: \section{Conclusions}
988: \label{sec:conc}
989: We studied the antiferromagnetic spin-1/2 Heisenberg model on the octagonal tiling, a two-dimensional 
990: quasiperiodic lattice structure, using linear spin-wave theory in a real space formulation. This approach was found to 
991: quantitatively reproduce previous quantum Monte Carlo results on static magnetic ground state properties of this system.
992: The spin-wave excitation spectrum was found to consist of  magnon-like low-energy soft-modes with a linear 
993: dispersion relation near the magnetic Bragg peaks, characteristic to long-range magnetic order.
994: It will be interesting to confirm the existence of  such linear soft-modes in the octagonal tiling
995: in future quantum Monte Carlo studies of the dynamical spin structure factor.
996: 
997: In addition, the dynamical spin structure factor shows
998: self-similar structures and bifurcations,
999: as well as flat bands
1000: at higher energies.
1001: We expect such features to be 
1002: generic to magnetic quasicrystals, which might
1003: eventually become observable in neutron scattering 
1004: experiments on magnetically ordered quasicrystals. 
1005: 
1006: Within the spin-wave approach, it is possible to include magnetic frustration, offering the 
1007: possibility of modeling
1008: more realistic quasiperiodic lattice structures, and their influence on magnetic properties. 
1009: Starting in 
1010: the unfrustrated limit, and increasing the magnetic frustration, 
1011: the evolution  of the classical N\'eel state can be examined, as well as the potential  relevance of the
1012: multifractal excitations for its breakdown in the case of strong frustration. 
1013: Another route to magnetic disorder, which can be taken at least theoretically, is by means of a quasiperiodic bilayer, which 
1014: is expected to 
1015: show a quantum phase transition upon increasing the interlayer coupling, due to local singlet formation. The presence of 
1016: multifractal excitations in the quasicrystal might be of possible relevance to quantum criticality in the transition 
1017: between the N\'eel ordered state and the disordered, gaped state. We leave such studies for future research.
1018: 
1019: 
1020: 
1021: 
1022: 
1023:  \section*{Acknowledgments}
1024: 
1025: We thank P. Frigeri, S. Haas, O. Nohadani, D. Rau, M. Sigrist, and
1026: S. Wehrli for fruitful discussions. The support of the MaNEP project of Swiss National Science Foundation is gratefully acknowledged.  Parts of the numerical calculations were done
1027: using the ALPS project library~\cite{alps}, and performed
1028: on the Asgard Beowulf cluster and Superdome at ETH Z\"urich.
1029: 
1030: \appendix
1031: 
1032: \section{Numerical Bogoliubov Transformation}
1033: 
1034: In this appendix we describe an alternative numerical scheme of finding the Bogoliubov
1035: transformation of the spin-wave Hamiltonian\cite{avery}. We need to construct a  matrix $T$, which diagonalizes the Hamiltonian
1036: matrix $M$ defined in Eq.~(\ref{eq:Mijkl}), and also satisfies the
1037: constraint in Eq.~(\ref{eq:constr}). We therefore need to simultaneously solve 
1038: \begin{equation}
1039:   T^\dag M T = \Omega,\quad \text{and} \quad   T^\dag \Sigma T = \Gamma.
1040: \end{equation}
1041: 
1042: The matrix $T$ can be constructed in two steps as follows:
1043: In a first step, an eigenvector matrix $Z$ and the set of eigenvalues
1044: $\lambda_i$ is obtained for the non-Hermitian eigenvalue problem~\cite{lapack},
1045: \begin{equation}
1046:   \Sigma M Z  = Z \Lambda,
1047: \end{equation}
1048: where $\Lambda = \text{diag}(\lambda_1, \ldots, \lambda_{N_n})$. 
1049: The columns of $Z$ are the right eigenvectors of $\Sigma M$.
1050: Hermitian conjugation of
1051: the above equation yields
1052: \begin{equation}
1053:   (\Sigma Z^\dag \Sigma) \Sigma M = \Lambda ( \Sigma Z^\dag \Sigma ),
1054: \end{equation}
1055: so that the rows of $\Sigma Z^\dag \Sigma$ form the left eigenvectors of $\Sigma M$.
1056: 
1057: In a second step, we diagonalize the Hermitian matrix,
1058: \begin{equation}
1059:   L = Z^\dag \Sigma Z,
1060: \end{equation} 
1061: obtaining a unitary matrix $U$, such that
1062: \begin{equation}
1063:   U^\dag L U = \text{diag}(l_1, \ldots, l_{N_n}).
1064: \end{equation}
1065: Furthermore, from the eigenvalues $l_i$ of $L$ we construct 
1066: the diagonal matrix $l^{-1/2}$, defined as
1067: \begin{equation}
1068:   l^{-1/2}_{ij} = 
1069:     \delta_{ij} |l_i|^{-1/2}.
1070: \end{equation}
1071: In case that the eigenvalues of $\Sigma M$ are non-degenerate,
1072: performing this second step is trivial: In this case $l_i$ is the
1073: $\Sigma$-norm of the corresponding eigenvector, and $U = 1$.
1074: We finally obtain the transformation $T$ as:
1075: \begin{equation}
1076:   T = Z U l^{-1/2}.
1077: \end{equation}
1078: Indeed, since 
1079: \begin{equation}
1080: \label{eq:app_L}
1081:  \bar \Gamma = T^\dag \Sigma T
1082: \end{equation}
1083: satisfies $\bar\Gamma^2 = 1$ by construction,
1084: $T$ satisfies the constraint Eq.~(\ref{eq:constr}). 
1085: Furthermore, right and left eigenvectors belonging to different
1086: eigenvalues are orthogonal. Thus $L$ is block diagonal with blocks
1087: corresponding to degenerate subspaces of $\Sigma M$. It follows that
1088: $U$ is block diagonal as well. Therefore $U$ only mixes columns of $Z$
1089: belonging to the same eigenvalue of $\Sigma M$, and consequently $T$
1090: satisfies
1091: \begin{equation}
1092:   \Sigma M T = T \bar \Lambda.
1093: \end{equation}
1094: Multiplying the above equation from the left by $T^{-1} = \Gamma T^\dag \Sigma$ 
1095: (which follows from Eq.~(\ref{eq:app_L})), yields
1096: \begin{equation}
1097:   T^\dag M T = \Lambda \bar \Omega.
1098: \end{equation}
1099: Hence, the matrix $T$  also diagonalizes the Hamiltonian matrix $M$.
1100: The energies of the bosonic eigenmodes are given as $\omega_i=JS\Gamma_{ii}\lambda_i$, respectively.
1101: 
1102: 
1103: 
1104: 
1105: \addcontentsline{toc}{section}{References}
1106: 
1107: 
1108: \begin{thebibliography}{8.}
1109: 
1110: \bibitem{manousakis} E. Manousakis, Rev. Mod. Phys. {\bf 63}, 1 (1991), and references therein.
1111: \bibitem{plaquette} A. Koga, S. Kumada, N. Kawakami, J. Phys. Soc. Jpn. {\bf 68} (7), 2373 (1999).
1112: \bibitem{laeuchli} A. L\"auchli, S. Wessel, and M. Sigrist,  Phys. Rev. B {\bf 66}, 014401 (2002).
1113: \bibitem{misguich} Gregoire Misguich, Claire Lhuillier, Report cond-mat/0310405, and references therein.
1114: \bibitem{depletion} A. Sandvik, Phys. Rev. B {\bf 66}, 024418 (2002).
1115: \bibitem{mucciolo} E. R. Mucciolo, A. H. Castro Neto, and C. Chamon, Report cond-mat/0402102.
1116: \bibitem{sato} T. J. Sato, H. Takakura, A. P. Tsai, K. Shibata, K. Ohoyama, and K. H. Andersen, Phys. Rev. B {\bf 61}, 476 
1117: (2000).
1118: \bibitem{cdmgtb} T. J. Sato, H. Takakura, G. Guo, A. P. Tsai, and K. Ohoyama, J. of Alloys and Compounds, {\bf 342}, 365 
1119: (2002).
1120: \bibitem{absence} T. J. Sato, H. Takakura, A. P. Tsai, and K. Shibata, Phys. Rev. Lett. {\bf 81}, 2364 (1998).
1121: \bibitem{wessel} S. Wessel, A. Jagannathan, S. Haas, Phys. Rev. Lett. {\bf 90}, 177205 (2003).
1122: \bibitem{anu} A. Jagannathan, Phys. Rev. Lett. {\bf 92}, 047202 (2004).
1123: \bibitem{lifshitz} R. Lifshitz and S. Even-Dar Mandel, Acta Cryst. A {\bf 60}, 167 (2004).
1124: \bibitem{ron} Ron Lifshitz, Material Science and Engineering A {\bf 294}, 508 (2000).
1125: \bibitem{vedmedenko} E. Y. Vedmedenko, U. Grimm, and R. Wiesendanger, Report cond-mat/0406373.
1126: \bibitem{grimm}U. Grimm, and M. Schreiber, in {\it  Quasicrystals - Structure and Physical Properties}, ed. H.-R. Trebin 
1127: (Wiley-VCH, Weinheim, 2003), and references therein.
1128: %\bibitem{oned1} J. Vidal, D. Mouhanna, and T. Giamarchi, Phys. Rev. Lett. {\bf 83}, 3908 (1999); Phys. Rev. B {\bf 65}, 014201 (2002).
1129: %\bibitem{oned2} M. Arlego, D. C. Cabra, and M. D. Grynberg, Phys. Rev. B {\bf 64}, 134419 (2001); K. Hida, J. Phys. Soc. Jpn. 68, 3177 (1999); Phys. Rev. Lett. {\bf 86}, 1331.(2001).
1130: %\bibitem{bulut} N. Bulut, D. Hone, D. J. Scalapino, and E. Y. Loh, Phys. Rev. Lett. {\bf 62}, 2192 (1989).
1131: \bibitem{levine} D. Levine and P. J. Steinhardt, Phys. Rev. B {\bf 34}, 596 (1986).
1132: \bibitem{white} R. M. White, M. Sparks, and I. Ortenburger, Phys. Rev. {\bf 139}, 450 (1965).
1133: \bibitem{duneau} M. Duneau {\it et al.}, J. Phys. A {\bf 22}, 4549 (1989).
1134: \bibitem{schulz} A. Jagannathan and H. J. Schulz, Phys. Rev. B {\bf 55}, 8045 (1997).
1135: \bibitem{spinwave} K. Yosida, {\it Theory of Magnetism} (Springer, 1996).
1136: \bibitem{blaizot} J.-P. Blaizot and G. Ripka, {\it Quantum Theory of Finite Systems} (MIT Press, Cambridge, MA, 1986).
1137: \bibitem{lapack} Using standard numerical routines, for example from LAPACK.
1138: \bibitem{lang} S. Lang, {\it Algebra}, (3rd Edition, Addison-Wesley 1993).
1139: \bibitem{einarsson} T. Einarsson and H. J. Schulz, Phys. Rev. B {\bf 51}, 6151 (1995).
1140: \bibitem{wessel_e0} S. Wessel, unpublished.
1141: \bibitem{footnote1} We verified, that the shown averages are indeed representative of the individual contributions.
1142: \bibitem{anu2000} A. Jagannathan, Phys. Rev. B. {\bf 61}, R834 (2000).
1143: \bibitem{wegner} F. Wegner, Z. Phys. B {\bf 36}, 209 (1980).
1144: \bibitem{schreiber} M. Schreiber and H. Gurssbach, Phys. Rev. Lett. {\bf 67}, 607 (1991).
1145: \bibitem{alps} M. Troyer {\it et al.}, Lecture Notes in Computer
1146:     Science {\bf 1505}, 191 (1998). Source codes of the library can
1147:     be obtained from \verb|http://alps.comp-phys.org/| .
1148: \bibitem{avery} Avery J. {\it Creation and annihilation operators} (McGraw-Hill, 1974), appendix 1 and references therein.
1149: \end{thebibliography}
1150: \end{document}
1151: 
1152: