1: \documentstyle[prl,twocolumn,epsf,floats,aps]{revtex}
2: \begin{document}
3: \draft
4:
5: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname @twocolumnfalse\endcsname
6:
7: \title{High-energy magnon dispersion and multi-magnon continuum in the
8: two-dimensional Heisenberg antiferromagnet}
9:
10: \author{Anders W. Sandvik$^1$ and Rajiv R. P. Singh$^{2}$}
11:
12: \address{$^{1}$Department of Physics, University of Illinois at
13: Urbana-Champaign, 1110 West Green Street, Urbana, Illinois 61801 \\
14: $^{2}$Department of Physics, University of California, Davis,
15: California 95616}
16:
17: \date{\today}
18:
19: \maketitle
20:
21: \begin{abstract}
22: We use quantum Monte Carlo simulations and numerical analytic continuation
23: to study high-energy spin excitations in the two-dimensional $S=1/2$
24: Heisenberg antiferromagnet at low temperature. We present results
25: for both the transverse $(x)$ and longitudinal $(z)$ dynamic spin structure
26: factor $S_{x,z}({\bf q},\omega)$ at ${\bf q}=(\pi,0)$ and $(\pi/2,\pi/2)$.
27: Linear spin-wave theory predicts no dispersion on the line connecting these
28: momenta. Our calculations show that in fact the magnon energy at $(\pi,0)$
29: is 10\% lower than at $(\pi/2,\pi/2)$. We also discuss the transverse and
30: longitudinal multi-magnon continua and their relevance to neutron scattering
31: experiments.
32: \end{abstract}
33:
34: \pacs{PACS: 75.40.Gb, 75.40.Mg, 75.10.Jm, 75.30.Ds}
35: \vskip2mm]
36:
37: It is now well established that the spin-wave theory of the two-dimensional
38: Heisenberg model \cite{anderson} correctly describes the low-energy dynamics
39: of layered antiferromagnets such as La$_{\rm 2}$CuO$_{\rm 4}$
40: \cite{neutrons1,chn}. Several methods have been used to calculate the quantum
41: renormalization factor $Z$ of the spin wave velocity, $c=Zc_0$, with the
42: result $Z \approx 1.18$ for $S=1/2$\cite{oguchi,singh,spinwave2,wiese,sandvik}.
43: Early neutron scattering experiments were consistent with this renormalization
44: over the entire Brilloin zone \cite{neutrons2}. However, recently more
45: accurate measurements have shown clear deviations at high energies in
46: La$_{\rm 2}$CuO$_{\rm 4}$ \cite{neutrons3} and other materials
47: \cite{neutrons4}. In particular, spin-wave theory predicts no dispersion on
48: the magnetic zone boundary, i.e., along the line of momenta ${\bf q}=
49: (\pi-x,x)$, whereas
50: the experimental data show a significant variation in the excitation energy.
51: This could be caused by interactions other than the nearest-neighbor
52: super-exchange $J$ normally used to model the materials \cite{neutrons3,morr}.
53: The deviations could also be at least partially due to a failure of low-order
54: spin-wave theory to account for the high-energy dynamics of the Heisenberg
55: model. A series expansion calculation \cite{singh2} indeed indicates
56: a momentum dependent renormalization factor, with the magnon energy at
57: ${\bf q}=(\pi,0)$ approximately $7$\% lower than at $(\pi/2,\pi/2)$. This,
58: however, is opposite to the trend found experimentally for
59: La$_{\rm 2}$CuO$_{\rm 4}$ \cite{neutrons3}.
60:
61: Another important issue, that has not yet been confronted experimentally,
62: is the multi-magnon continuum expected to be present in the dynamic
63: structure factor $S({\bf q},\omega)$ above the single-magnon mode. The
64: neutron scattering cross-section has been analyzed assuming only a
65: delta-function representing the spin waves with dispersion $\omega_{\bf q}$
66: \cite{neutrons1,neutrons2,neutrons3,neutrons4}.
67: The resolution of present experiments is not sufficient for detecting the
68: presence of additional spectral weight, much less to determine its
69: distribution. Theoretical calculations have so far also not been
70: successful in accurately determining the full dynamic structure factor.
71: In order to provide guidance for improved fitting procedures it is imperative
72: to obtain quantitative estimates of the multi-magnon spectral features, in
73: particular considering that experiments may soon become sufficiently accurate
74: for detecting the continuum. Within spin-wave theory, correctly accounting
75: for the interactions that give rise to the continuum is extremely complicated
76: and has led to contradictory results \cite{spinwave2,igarashi,canali}.
77: Numerical calculations of $S({\bf q},\omega)$ are challenging because large
78: lattices have to be used to converge to the thermodynamic limit, and the
79: extraction of real-time dynamics from quantum Monte Carlo (QMC) data is
80: difficult. Such calculations have therefore so far only given limited insights
81: \cite{makivic,syljuasen}. The series work \cite{singh2} suggests a significant
82: continuum but gives no information on its shape.
83:
84: Here we address the issues of high-energy magnon dispersion and multi-magnon
85: continuum using QMC calculations in a way that explicitly separates the
86: transverse and longitudinal components of the dynamic structure factor. This
87: allows us to more accurately determine both the magnon energy and the
88: continuum part of the spectrum. We consider the Heisenberg model on a square
89: lattice, defined in standard notation by the Hamiltonian
90: \begin{equation}
91: H = J \sum\limits_{\langle i,j\rangle} {\bf S}_i \cdot {\bf S}_j .
92: \label{ham}
93: \end{equation}
94: The coupling $J \approx 1500$ K in typical planar cuprates. On an infinite
95: lattice the spin-rotational symmetry of this model is spontaneously broken at
96: $T=0$ and long-range order develops along an axis that we take as the
97: $z$-direction. The two-point correlations then have different longitudinal
98: ($z$) and transverse $(x,y)$ components. In a finite system the symmetry
99: is not broken and all correlation functions are equal to the same
100: rotational average. We want to access separately the transverse dynamic
101: structure factor, which contains the single-magnon mode as well as a
102: continuum, and the longitudinal component comprising only a multi-magnon
103: continuum. To this end, we explicitly break the symmetry by applying
104: a staggered magnetic field;
105: \begin{equation}
106: H \to H - h\sum_{i} (-1)^{x_i+y_i} S^z_i.
107: \end{equation}
108: We adjust the field strength $h$ so that the induced staggered
109: magnetization equals the known value in the thermodynamic limit;
110: $m = |\langle S^z_i \rangle | = 0.307$
111: \cite{oitmaa,reger,singh,spinwave2,wiese,sandvik}.
112: In the limit of large system sizes $h \to 0$ (at $T=0$) and the correlation
113: functions become equal to their values in a system with spontaneously broken
114: symmetry.
115:
116: In real layered cuprates there is a small
117: inter-layer coupling and some degree of anisotropy, resulting in a
118: finite-$T$ transition to an ordered state, in La$_{\rm 2}$CuO$_{\rm 4}$ at
119: $T_{\rm N} \approx 300$ K. We here consider a low temperature, $\beta=J/T=32$,
120: corresponding to approximately $50$ K, where the staggered magnetization in
121: real materials is very close to the saturated $T=0$ value. Our staggered
122: field $h$ can then also be viewed as a mean-field treatment of the
123: inter-layer coupling \cite{scalapino}.
124:
125: Working with
126: lattices with $N= L\times L$ spins and periodic boundary conditions, we
127: have used the stochastic series expansion method \cite{sse} to calculate
128: the imaginary-time dependent spin-spin correlation function
129: \begin{equation}
130: G_\alpha ({\bf q},\tau) =
131: \langle S^\alpha (-{\bf q},\tau) S^\alpha ({\bf q},0) \rangle ,
132: \end{equation}
133: where $\alpha = x,z$ and
134: \begin{equation}
135: S^a({\bf q},\tau) = {1\over \sqrt{N}}
136: \sum_i {\rm e}^{-{\bf q} \cdot {\bf r}_i}
137: {\rm e}^{-\tau H} S^\alpha_i {\rm e}^{\tau H} .
138: \end{equation}
139: The correlation function is related to the dynamic structure factor
140: according to
141: \begin{equation}
142: G_\alpha ({\bf q},\tau) = {1\over \pi}\int_{-\infty}^\infty
143: d\omega S_\alpha({\bf q},\omega){\rm e}^{-\tau \omega},
144: \label{gtos}
145: \end{equation}
146: which in principle can be inverted to yield the real-frequency dynamics from
147: the correlation function computed in the simulations. With simulation data
148: affected by statistical fluctuations there are well known difficulties in
149: carrying out this analytic continuation in practice, and one can only expect
150: limited frequency resolution using, e.g., the maximum entropy method
151: \cite{maxent}. Here we will instead assume reasonable functional forms for
152: the transverse ($x$) and longitudinal ($z$) dynamic structure factors, with
153: parameters that are adjusted to satisfy the equality (\ref{gtos}). With the
154: QMC method used, derivatives of $G_\alpha ({\bf q},\tau)$ can also be directly
155: calculated \cite{ramanpaper} and impose additional constraints on
156: $S_\alpha({\bf q},\omega)$;
157: from Eq.~(\ref{gtos})
158: \begin{equation}
159: G^{(n)}_\alpha ({\bf q},\tau) = {(-1)^n\over \pi} \int_{-\infty}^\infty
160: d\omega S_\alpha({\bf q},\omega)\omega^n {\rm e}^{-\tau \omega}.
161: \label{gtder}
162: \end{equation}
163: We use the first $(n=1)$ and $(n=2)$ second derivatives, which at $\tau =0$
164: correspond to the first two frequency moments of the spectrum. We also use
165: the sum rule
166: \begin{equation}
167: \chi_\alpha ({\bf q})= {2\over \pi}
168: \int_{-\infty}^\infty {d\omega \over \omega}S_\alpha ({\bf q},\omega),
169: \end{equation}
170: where $\chi_\alpha ({\bf q})$ is the static susceptibility
171: \begin{equation}
172: \chi_\alpha ({\bf q}) = \int_0^\beta d\tau G_\alpha ({\bf q},\tau).
173: \end{equation}
174: The spectrum also obeys the bosonic relation $S_\alpha ({\bf q},-\omega) =
175: {\rm e}^{-\beta\omega}S_\alpha ({\bf q},\omega)$.
176:
177: For the model forms of the transverse and longitudinal structure factors we
178: take simple functions that reflect the expected gross spectral features.
179: The transverse component should include a delta-function at an energy
180: $\omega_{\bf q}$, representing the magnon, and a continuum that does not
181: extend below $\omega_{\bf q}$ (temperature broadening of the high-energy
182: magnons should be insignificant at $\beta=32$). The longitudinal
183: component is entirely in the continuum and can also not extend below
184: $\omega_{\bf q}$. Both continua must decay to zero rapidly as
185: $\omega \to \infty$, to ensure convergence of all frequency moments.
186: We use
187: \begin{eqnarray}
188: S_x({\bf q},\omega) & = & A_1({\bf q}) \delta (\omega - \omega_q) +
189: A_2({\bf q}) f_x({\bf q},\omega) \label{sxform}, \\
190: S_z({\bf q},\omega) & = & B({\bf q}) f_z({\bf q},\omega), \label{szform}
191: \end{eqnarray}
192: where the continua $f_{x,z}({\bf q},\omega)$ are given by
193: \begin{eqnarray}
194: f_x({\bf q},\omega)
195: & = & r_x {\rm e}^{-(\omega -\nu)^2/2\sigma^2},
196: ~~ (0~{\rm for}~ \omega < \omega_q) , \label{tcont} \\
197: f_z({\bf q},\omega)
198: & = & r_z (\omega-\omega_q)^p {\rm e}^{-a(\omega-\omega_q)^b},
199: ~~ (0~{\rm for}~ \omega < \omega_q) \label{lcont},
200: \end{eqnarray}
201: where $r_x$ and $r_z$ are factors normalizing the frequency integrals
202: to unity. For clarity, we have suppressed the dependence of the parameters
203: $\nu,\sigma,p,a$, and $b$ on ${\bf q}$. The static structure factors
204: (the integrated spectral weights) are given by
205: \begin{eqnarray}
206: S_x({\bf q}) & = & \langle S^x(-{\bf q})S^x({\bf q})\rangle =
207: A_1({\bf q}) + A_2({\bf q}), \\
208: S_z({\bf q}) & = & \langle S^z(-{\bf q})S^z({\bf q})\rangle =
209: B({\bf q}).
210: \end{eqnarray}
211: In scattering experiments with unpolarized neutrons, a rotational average of
212: the transverse and longitudinal dynamic structure factors is measured.
213: The rotationally averaged static structure factor $S({\bf q})=2S_x({\bf q})+
214: S_z({\bf q})= 2A_1+2A_2+B$.
215:
216: The values we find for the ``self-consistent'' staggered field (demanding
217: $m=0.3070$ to an accuracy of better than $0.00005$) are $h/J=0.10652$ for
218: $L=4$, $0.022575$ for $L=8$, $0.005450$ for $L=16$, and $0.001615$ for $L=32$.
219: With the QMC algorithm \cite{sse} implemented in the $z$-basis, and with the
220: field also in the $z$-direction, the longitudinal correlations can be easily
221: calculated. In order to compute the transverse correlations it is more
222: efficient to apply the field in the $x$-direction, so that the measurements
223: can be carried out with diagonal operators \cite{henelius}. Hence, we perform
224: two independent simulations for each lattice size.
225:
226: \begin{figure}
227: \centering
228: \epsfxsize=7.2cm
229: \leavevmode
230: \epsffile{fig1.eps}
231: \vskip1mm
232: \caption{Ratio of the longitudinal and transverse static structure factors
233: (integrated spectral weights) along a path in the Brilloin zone. Open
234: circles are for $L=16$ and solid ones for $L=32$.}
235: \label{fig1}
236: \end{figure}
237:
238: Fig.~\ref{fig1} shows our results for the ratio of the longitudinal and
239: transverse spectral weights versus the momentum, along a standard path in
240: the Brilloin zone. At ${\bf q}=(\pi,\pi)$ the ratio diverges (in the
241: thermodynamic limit), reflecting the divergence of $S_z(\pi,\pi)$ due to
242: the long-range order. For ${\bf q} \to (0,0)$ and ${\bf q} \to (\pi,\pi)$
243: it approaches zero and hence the magnon completely exhausts the total
244: spectral weight in these limits (other treatments have shown that the magnon
245: exhausts the low-energy transverse spectral weight \cite{stringari}). The
246: ratio is the largest at ${\bf q}=(\pi/2,\pi/2)$ where it is above $0.7$ ---
247: it is between $0.6$ and $0.7$ over much of the Brilloin zone. Hence, in
248: unpolarized neutron scattering experiments, $30-35$\% of the cross-section
249: at the zone boundary is due to the longitudinal continuum.
250:
251: Next we study the full dynamic structure factor, focusing on the zone-boundary
252: momenta ${\bf q} = (\pi,0)$ and $(\pi/2,\pi/2)$. The analytic continuation
253: using fits to the functions (\ref{sxform}) and (\ref{szform}) was carried out
254: using QMC imaginary-time data with a spacing $\Delta\tau=0.5$ up to $\tau = 4$
255: (the data become too noisy for higher $\tau$). For $L=4$, we have
256: compared with exact diagonalization results and find that the magnon energy
257: is very accurately reproduced (better than 0.5\%) but the weight in the magnon
258: is underestimated by about $5\%$ because the form of the continuum,
259: Eq.~(\ref{tcont}), is not appropriate for a very small system where only
260: a small number of significant delta-functions represent the continuum.
261: Our results for larger systems should have smaller systematic errors.
262:
263: Fig.~\ref{fig2} shows the size dependence of the magnon weight and
264: energy (the scatter of the data gives an indication of the statistical
265: errors). The $L=4$ data is from exact diagonalization; in this case the
266: two momenta are degenerate due to equivalence of this lattice
267: and a $2^4$ hyper-cube. Rough extrapolations in $1/L$ give the infinite-size
268: energy $\omega_{\bf q}/J \approx 2.15$ for ${\bf q}=(\pi,0)$ and $2.39$ for
269: $(\pi/2,\pi/2)$. The relative weight of the magnon in $S_x({\bf q},\omega)$
270: is $\approx 60$\% at $(\pi,0)$ and $85$\% at $(\pi/2,\pi/2)$. Within
271: linear spin-wave theory, including the quantum-renormalization $Z=1.18$,
272: the energy $\omega_q/J = 2.36$ for both momenta. Hence, the results show that
273: the interactions neglected in spin-wave theory have strong effects at
274: ${\bf q}=(\pi,0)$, lowereing the energy by almost $10$\% and transferring
275: $\approx$ 40\% of the magnon weight into the continuum. At $(\pi/2,\pi/2)$
276: the excitation energy is instead slightly increased by the interactions and
277: a much smaller fraction of the weight is in the continuum.
278:
279: \begin{figure}
280: \centering
281: \epsfxsize=7.7cm
282: \leavevmode
283: \epsffile{fig2.eps}
284: \vskip1mm
285: \caption{Properties of the single-magnon part of the transverse dynamic
286: structure factor vs inverse system size ($L=4,8,12,16,20,24$, and $32$).
287: (a) single-magnon fraction of the weight, (b) magnon energy.}
288: \label{fig2}
289: \end{figure}
290:
291: Fig.~\ref{fig3} shows the magnon as well as the transverse and longitudinal
292: continua for $L=32$. The rotationally averaged structure factor measured in
293: neutron scattering experiments with unpolarized neutrons is also shown. The
294: exponent $p$ in the longitudinal continuum, Eq.~(\ref{lcont}), is small and
295: positive, $p \alt 0.05$, giving a spectral weight concentrated close to the
296: magnon. The longitudinal continuum is significantly narrower at $(\pi/2,\pi/2)$
297: than at $(\pi,0)$. For both momenta, the transverse continuum is broad and
298: centered further above the magnon. All these spectral features are consistently
299: present for system sizes $L \ge 8$. As seen in the insets of Fig.~\ref{fig3},
300: for $L=32$ only $\approx 45$\% of the rotationally averaged spectral weight
301: is in the magnon mode at $(\pi,0)$, increasing to $\approx 65$\% at
302: $(\pi/2,\pi/2)$. The magnon weight decreases only slightly for even
303: larger systems, as can be inferred from Fig.~\ref{fig2}.
304:
305: In previous work using QMC and analytic continuation \cite{makivic,syljuasen},
306: the magnon energy was extracted as the first moment of the rotationally
307: averaged relaxation function, $F({\bf q},\omega) \sim S({\bf q},\omega)/
308: \omega$. Using this procedure, we find the moment $2.53$ at $(\pi,0)$ and
309: $2.58$ at $(\pi/2,\pi/2)$, i.e., both significantly above the actual magnon
310: energies (the first moment of $S({\bf q},\omega)$ is higher yet; approximately
311: $2.66$ for both momenta). This is clearly due to the large spectral weight
312: in the continuum. There is potentially a similar problem in the analysis of
313: neutron scattering data. Since the present-day energy resolution is such that
314: a non-negligible part of the continuum that we have found here would be
315: included in an experimental fit to a single resolution-broadened peak, the
316: extracted magnon energy may be too high. This effect would be
317: particularly important at and close to ${\bf q}=(\pi,0)$, where our results
318: show that less than $50$\% of the total weight is in the magnon peak and the
319: continuum is relatively broad.
320:
321: Experimental results do appear to show the presence of some weight consistent
322: with a continuum above the resolutioned broadened magnon peak \cite{neutrons3},
323: but the statistics is not sufficient for extracting its size and shape.
324: Our calculations should be useful for analyzing future experiments with higher
325: resolution. Such experiments will be very important for determining the
326: significance of other interactions in the antiferromagnetic cuprates beyond
327: the nearest-neighbor exchange $J$. The significant effects of magnon
328: interactions that we have found at ${\bf q}=(\pi,0)$ may also have important
329: consequences for the broadening of the Raman spectrum \cite{canali} and the
330: momentum space anisotropy of the photoemission spectrum \cite{wells}.
331:
332: \vskip1mm
333: We would like to thank G. Aeppli, L. Balents, G. Sawatzky, and D. Scalapino
334: for stimulating discussions. This research was supported by the NSF under
335: grants No.~DMR-97-12765 and DMR-99-86948. The work was started at the
336: Institute of Theoretical Physics at UC Santa Barbara, under support of NSF
337: grant No.~PHY-94-07194. Most of the numerical calculations were carried out
338: on the SGI Origin2000 system the NCSA.
339: \null\vskip-4mm
340: \begin{references}
341: \null\vskip-20mm\null
342: \bibitem{anderson}
343: P. W. Anderson, Phys. Rev. {\bf 86}, 694 (1952).
344:
345: \bibitem{neutrons1}
346: D. Vaknin {\it et al.}, Phys. Rev. Lett. {\bf 58}, 2802 (1987);
347: G. Shirane {\it et al.}, {\it ibid.} {\bf 59}, 1613 (1987);
348: G. Aeppli {\it et al.}, {\it ibid.} {\bf 62}, 2052 (1989).
349:
350: \bibitem{chn}
351: S. Chakravarty, B. I. Halperin, and D. R. Nelson, Phys. Rev. Lett. {\bf 60},
352: 1057 (1988); Phys. Rev. B {\bf 39}, 2344 (1989).
353:
354: \bibitem{oguchi}
355: T. Oguchi, Phys. Rev. {\bf 117}, 117 (1960).
356:
357: \bibitem{singh}
358: R. R. P. Singh, Phys. Rev. B {\bf 39}, 9760 (1989).
359:
360: \bibitem{spinwave2}
361: C. J. Hamer, Z. Weihong, and P. Arndt, Phys. Rev. B {\bf 46}, 6276 (1992);
362: J. Igarashi, Phys. Rev. B {\bf 46}, 10763 (1992);
363: C. M. Canali and M. Wallin, Phys. Rev. B {\bf 48}, 3264 (1993).
364:
365: \bibitem{wiese}
366: U.-J. Wiese and H.-P. Ying, Z. Phys. B {\bf 93}, 147 (1994);
367: B. B. Beard and U.-J. Wiese, Phys. Rev. Lett. {\bf 77}, 5130 (1996).
368:
369: \bibitem{sandvik}
370: A. W. Sandvik, Phys. Rev. B {\bf 56}, 11678 (1997).
371:
372: \bibitem{neutrons2}
373: S. Hayden {\it et al.}, Phys. Rev. Lett. {\bf 67}, 3622 (1991).
374:
375: \bibitem{neutrons3}
376: R. Coldea {\it et al.} (unpublished).
377:
378: \bibitem{neutrons4}
379: Y. J. Kim et al Phys. Rev. Lett. {\bf 83}, 852 (1999);
380: H. M. R{\o}nnow {\it et al.} (unpublished).
381:
382: \bibitem{morr}
383: D. K. Morr, Phys. Rev. B {\bf 58}, R587 (1998).
384:
385: \bibitem{singh2}
386: R. R. P.Singh and M. P. Gelfand, Phys. Rev. B {\bf 52}, 15695 (1995).
387:
388: \bibitem{igarashi}
389: J. Igarashi and A. Watabe, Phys. Rev. B {\bf 43}, 13456 (1991).
390:
391: \bibitem{canali}
392: C. M. Canali and S. M. Girvin, Phys. Rev. B {\bf 45}, 7127 (1992).
393:
394: \bibitem{makivic}
395: M. S. Makivic and M. Jarrell, Phys. Rev. Lett. {\bf 68}, 1770 (1992).
396:
397: \bibitem{syljuasen}
398: O. F. Sylju{\aa}sen and H. M. R{\o}nnow, cond-mat/0003350.
399:
400: \bibitem{oitmaa}
401: J. Oitmaa and D. D. Betts, Can. J. Phys. {\bf 56}, 897 (1978).
402:
403: \bibitem{reger}
404: J. D. Reger and A. P. Young, Phys. Rev. B {\bf 37}, 5978 (1988).
405:
406: \bibitem{scalapino}
407: D. J. Scalapino, Y. Imry, and P. Pincus, Phys. Rev. B {\bf 11},
408: 2042 (1975).
409:
410: \bibitem{sse}
411: A. W. Sandvik, Phys. Rev. B {\bf 59}, R14157 (1999).
412:
413: \bibitem{maxent}
414: M. Jarrell and J. E. Gubernatis, Phys. Rep. {\bf 269}, 133 (1996).
415:
416: \bibitem{ramanpaper}
417: A. W. Sandvik, S. Capponi, D. Poilblanc, and E. Dagotto,
418: Phys. Rev. B {\bf 57}, 8478 (1998).
419:
420: \bibitem{henelius}
421: P. Henelius, A. W. Sandvik, and S. M. Girvin,
422: Phys. Rev. B {\bf 61}, 364 (2000).
423:
424: \bibitem{stringari}
425: S. Stringari, Phys. Rev. B {\bf 49}, 6710 (1994).
426:
427: \bibitem{wells}
428: B. O. Wells {\it et al.}, Phys. Rev. Lett. {\bf 74}, 964 (1995);
429: C. Kim {\it et al.}, {\it ibid.}, {\bf 80}, 4245 (1998).
430:
431:
432: \end{references}
433:
434: \begin{figure}
435: \centering
436: \epsfxsize=7.8cm
437: \leavevmode
438: \epsffile{fig3.eps}
439: \vskip1mm
440: \caption{The transverse (solid curves) and longitudinal (dashed curves)
441: dynamic structure factor at ${\bf q}=(\pi,0)$ (upper panel)
442: and $(\pi/2,\pi/2)$ (lower panel). $S_x({\bf q},\omega)$ and
443: $S_z({\bf q},\omega)$ are normalized to $1$ and $S_z({\bf q})/S_x({\bf q})$
444: [see Fig.~\protect{\ref{fig1}}], respectively. The delta-function piece of
445: $S_x({\bf q},\omega)$ is represented by a vertical line with height
446: (indicated by a star) equal to its relative weight $A_1({\bf q})$. The
447: insets show the rotationally averaged spectra, $S({\bf q},\omega)=
448: 2S_x({\bf q},\omega) + S_z({\bf q},\omega)$.}
449: \label{fig3}
450: \end{figure}
451:
452:
453: \end{document}