1: \documentstyle[mprocl,psfig,epsf,subfigure]{article}
2:
3: \newcommand{\D}{\displaystyle}
4: \newcommand{\T}{\textstyle}
5: \newcommand{\conj}[2]{\begin{samepage}
6: {\vspace{1ex}
7: \noindent
8: {\large \sc #1}. {\em #2} \\[1ex]}
9: \end{samepage}}
10: \renewcommand{\thesubfigure}{\thefigure.\arabic{subfigure}}
11: \makeatletter
12: \renewcommand{\@thesubfigure}{\thesubfigure:\space}
13: \renewcommand{\p@subfigure}{}
14: \makeatother
15:
16: %\setlength{\textwidth}{14cm}
17: %\setlength{\oddsidemargin}{1cm}
18: %\setlength{\evensidemargin}{3cm}
19: %\renewcommand{\baselinestretch}{1.2}
20:
21: \title{UNIVERSAL MULTIFRACTAL PROPERTIES OF THE SMALL SCALE INTERMITTENCY IN
22: ANISOTROPIC AND INHOMOGENEOUS TURBULENCE}
23:
24: \author{M. ALBER}
25: \address{Dept. of Medical Physics,\\ Radiologische Uniklinik, Universit\"at T\"ubingen,
26: D-72076 T\"ubingen, Germany \\ email msalber@med.uni-tuebingen.de}
27:
28: \author{S. L\"UCK, C. RENNER, J. PEINKE}
29: \address{FB 8 Physik, Universit\"at Oldenburg,
30: D-26111 Oldenburg, Germany \\ email Peinke@uni-oldenburg.de}
31:
32: \begin{document}
33: \maketitle
34:
35: \abstracts{The notion of self-similar energy cascades and multifractality has long
36: since been connected with fully developed, homogeneous and isotropic turbulence.
37: We introduce a number of amendments to the standard methods for analysing the
38: multifractal properties of the energy dissipation field of a turbulent flow. We
39: conjecture that the scaling assumption for the moments of the energy dissipation
40: rate is valid within the transition range to dissipation introduced by Castaing
41: {\it et al.}( Physica D {\bf 46}, 177 (1990)). The multifractal spectral functions
42: appear to be universal well within the error margins and exhibit some as yet
43: undiscussed features. Furthermore, this universality is also present in the neither
44: homogeneous nor isotropic flows in the wake very close to a cylinder or the
45: off-centre region of a free jet.}
46:
47: \section{Introduction}
48:
49: The concept of an iterated distributive process behind turbulence, commonly
50: called the ``energy cascade'', is much older than fractal geometry. It might be
51: said that multifractal theory has its roots, partially, in turbulence research
52: \cite{Mandelbrot74}. Of course, by being finite the statistics of the energy cascade
53: do not have to be multifractal, and even if it may be conceived to be so, the
54: expedience of a multifractal description is far from obvious. For several decades
55: one incentive for turbulence research has been the idea of a universal behaviour in
56: small scale intermittency \cite{Frisch95}, but it has not been possible to bring
57: this universality to light in the regime of length scales which is accessible to
58: a multifractal model. These questions were adressed in pioneering works by Meneveau
59: and Sreenivasan \cite{Meneveau87,Sreenivasan91,SreenivasanAntonia97}, and it became
60: clear then that multifractal geometry is an issue in the long process of understanding
61: turbulence.
62:
63: This paper is devoted to an extensive examination of experimental data in a multifractal
64: framework. We shall not venture to relate our experimental findings to multifractal
65: models of turbulence here, however, we vouch for the multifractality of turbulence.
66:
67: The central quantity of our investigation, as introduced by Obukhov \cite{Obukhov62}
68: and Kolmogorov \cite{Kolmogorov62} in 1962, is the energy dissipation averaged over
69: a volume of size $r$. In the case of locally homogeneous and isotropic turbulence this
70: is approximated by
71: \begin{equation}
72: \epsilon_r(x) \quad \propto \quad \frac{1}{r} \int_{x-r/2}^{x+r/2} \left(\frac{\partial u}
73: {\partial x'} \right)^2 \mbox{d}x' \quad,
74: \end{equation}
75: where $u(x)$ denotes the longitudinal velocity component at point $x$ along a 1-D cut
76: through the energy dissipation field (Figure \ref{ftshom}).
77:
78: We define the multifractal measure $\mu_i$ on a grid of size $r$ to be
79: $\mu_i = r \epsilon_r(ir)$.
80: The $q$-th order moment of $\mu$ is defined as :
81: \begin{equation} \label{partsum}
82: S_q(r) = \sum_i \mu_i^q \quad \propto \quad r^{\tau(q)} \quad,
83: \end{equation}
84: where the sum is taken over all boxes of some $r$-grid and $\tau(q)$
85: is the multifractal scaling function.
86:
87: To derive these quantities from the data, we employ an improved multifractal box counting
88: algorithm recently suggested \cite{Alber98}. We sketch this algorithm in section two.
89: Section three concerns the details of the experimental data and the particular aspects
90: of the selection of the proper scaling range. Section four presents the results obtained
91: for locally isotropic flow conditions indicating a universal multifractal behaviour in
92: the range defined by the transition from the inertial range to the dissipative limit.
93: These universal features are also recovered in inhomogeneous and not fully developed
94: turbulence, as shown in the concluding section.
95:
96: \begin{figure}[htb]%
97: \begin{center}%
98: \hspace{-2.0cm}\psfig{file=fig1.eps,width=14cm,height=6cm}%
99: \caption{\label{ftshom} \em The energy dissipation time series (equivalent to a 1-D cut along the mean
100: flow) for experiment J1. Units on the x-axis are in terms of the Kolmogorov microscale
101: $\eta$.}
102: \end{center}
103: \end{figure}
104:
105: \section{An Improved Multifractal Box-Counting Algorithm}
106:
107: The necessity of taking great care when utilising fixed-size box counting algorithms
108: has frequently been expressed. However, here we deem fixed-size algorithms to be the
109: method of choice for two reasons: their shortcomings are well understood
110: \cite{Caswell87,Provenzale93} and can in some cases be balanced \cite{Alber98,Riedi96},
111: and they are closer to the theoretical concepts of turbulence than their fixed-mass
112: counterparts \cite{Badii88} in that they do not mix length scales . We also refrain from
113: the direct evaluation of the $f(\alpha)$ curve \cite{Chhabra89}, because we find
114: artefacts are easier to trace in the standard approach \cite{Meneveau87,Meneveau91}.
115: The information content of both methods is equivalent since they are based on
116: essentially the same counting procedure.
117:
118: We want to outline in brief some of the improvements introduced to the method of moments
119: \cite{Alber98}. The prime source of errors was identified as the finite size artefacts
120: due to ill-fitting grids. To reduce the influence of these ``clipping errors'' we replace
121: in eq. (\ref{partsum}) the local measure $\mu_i$ by a local average $<\mu>_i$ and define
122: \begin{eqnarray}
123: <\mu>_i = \mu_i \prod_{j=1}^n \Theta (\mu(x_i+\delta_j)) \quad & \quad \mbox{for} \quad & q \geq 0 \\
124: <\mu>_i = \left( \prod_{j=1}^n \mu(x_i+\delta_j) \right)^{1/n} & \mbox{for} & q < 0 \quad,
125: \end{eqnarray}
126: where $\{ \delta_j \}$ is a set of displacements smaller than $r$.
127:
128: Secondly, we introduce the concept of wandering intercepts which describe the deviations
129: from linear scaling of the partition function (\ref{partsum}) \cite{Alber98,Cutler91,Cutler93}.
130: For a given $q$ we define
131: \begin{equation} \label{wnd}
132: C(q,r) = \log S_q(r) - \hat{\tau}(q) \log r \quad ,
133: \end{equation}
134: where $\hat{\tau}(q)$ is an estimate of $\tau(q)$, e.g. the result of a least squares
135: line fit to $\log S_q(\log r)$. The $C(q,r)$ can be thought of as a ``lacunarity function''.
136: Note that the concept of Extended Self-Similarity introduced by Benzi \cite{Benzi95}
137: is a special case of eq. (\ref{wnd}). Our development relies on the assumption that the
138: $C(q,r)$ does not behave too erratically in $q$ and can ideally be factorised into functions
139: of $r$ and $q$ only. Although we shall not make detailed assumptions about the $C(q,r)$
140: in general, a lack of coherence with respect to $q$ at a point $q_0$ betrays the presence
141: of a phase transitional behaviour of the multifractal \cite{Alber98}.
142: For these purposes we employ plots of $C(q_1,r_i)$ against $C(q_2,r_i)$ for
143: $q_1 < q_0 < q_2$ and expect that a lack of correlation between the fluctuations
144: in $r$ of the wandering intercepts appears as a scattered plot.
145:
146: The greatest obstacle in the multifractal analysis of turbulence data (as with any other
147: physical data) is the determination of the proper scaling range. For our findings it is
148: characteristic that the multifractal scaling assumption applies only
149: to the transitional range to dissipation, which can loosely be defined by the
150: lengthscale $\eta$ where the local Reynolds number becomes of the order of unity:
151: $ \frac{\bar{u} \eta}{\nu} = 1$.
152: In most cases, the lower bound is dictated by the detector resolution, which usually
153: exceeds $\eta$ quite considerably for turbulent flows of a reasonable Reynolds number.
154: The upper bound on the scaling range is much more indistinct, since typically one can
155: only discern a sweeping transition to a power law behaviour of $S_q(r)$, which in
156: turn sometimes extends over less than one order of magnitude. Although it is tempting to
157: expand the scaling range both to acquire a better statistical basis and to extend the
158: validity of the analysis, we handle the problem quite restrictively. Since the essence
159: of numerical multifractal analysis is extracting to a high precision minor differences
160: in interrelated scaling laws, we deem our approach justified.
161:
162: \section{The Statistical Properties of Experimental Data}
163:
164: The turbulence velocity data were obtained in three series of experiments of flows
165: in air. Series J was a free jet through a nozzle of $d=8$ mm diameter and an output
166: velocity of 40 m/s. The estimated $R_{\lambda}$ in the centre was 210. The jet was
167: scanned axially and radially at an axial distance of $60d = 48$ cm. The time series
168: had a size of $1.25 \times 10^7$ data points. Series CF was a wind tunnel experiment
169: with a cylinder of diameter $D=5$ cm. The turbulent wake was examined at a distance
170: of $32d = 160$ cm for a number of velocities, yielding $R_{\lambda}$ from 200 to 520.
171: The size of the time series was $1.25 \times 10^6$. Series CV consisted of a
172: variation of the location of the probe in the wake field of the cylinder at 21m/s
173: velocity, ranging from $4d = 20 \mbox{cm}$ to $40d = 200 \mbox{cm}$. Again, $1.25 \times 10^7$
174: points were sampled. The equipment was a DANTEC anemometer StreamLine 90CN10 with
175: single wire probes 55P01 (sensitive length 1.25mm) for the jet measurements and
176: x-wire probes 55P61 (sensitive length 1mm) for the cylinder measurements. The use
177: of the latter probe facilitated a more accurate measurement of the longitudinal
178: velocity component.
179:
180: An extensive analysis was performed for all time series. This included the
181: calculation of characteristic mean-field quantities (see table 1) and the generation
182: of probability density functions (PDFs) of velocity increments. The changing form
183: of the PDFs as a function of the spatial resolution $r$ was quantified by a parameter
184: $\Lambda^2(r)$ according to Castaing \cite{Castaing90}. The power law behaviour of
185: the resulting function $\Lambda^2(r)$ locates the transitional range between the
186: classical inertial range and the dissipative limit \cite{Chabaud94} (see figure
187: \ref{flam}). In the following we restrict our investigations of the scaling
188: behaviour of $<\epsilon_r^q>$, respectively $S_q(r)$, to this transitional range.
189: Critical evidence for the coincidence of these scaling ranges was deduced from our
190: experimental data. The chosen ranges for each example are given in tables 1 and 2.
191:
192: \begin{figure}[htb]%
193: \hspace{1.0cm} \psfig{file=fig4.eps,width=10cm,height=6cm}%
194: \caption{\label{flam} \em A plot of $\log \Lambda^2$ against $\log (r /\eta)$ for experiment J1.
195: The transitional range corresponds to the linear part of the curve abutting the
196: maximum.}
197: \end{figure}
198:
199: It is one of the traits of box-counting, that it delivers a result under almost all
200: circumstances. However, it is sometimes open to arbitrariness to find out which results
201: comprise significant information about the multifractal. As it turned out, statistical
202: resolution is a severe issue for moments as small as 3 for about 1 million data points,
203: and 10 million points improve the situation only marginally. It has been shown
204: \cite{Meneveau91}, that $f(\alpha)$ will assume negative values due to the 1-D
205: nature of the data set as opposed to the three dimensional energy field,
206: which infers that the main contribution to a moment that corresponds to a negative
207: $f(\alpha)$ comes from very few singularities. This being the case, it comes as no
208: surprise to find that the correlation plots for $q_1, q_2 >5$ are extremely well
209: correlated. In general, we attribute little credibility to any moment larger than 4.
210: For negative $q$, noise and digitalisation artefacts put an end to the reliability
211: of the results at sometimes quite moderate values of $q \approx -1$. A special case are
212: the inhomogeneous examples, where due to the background of the intertwined laminar
213: flow no multifractal fine structure can be resolved with negative moments.
214:
215: \begin{figure}[hbt]%
216: \label{tqfa}%
217: \centering%
218: \subfigure[\em The Experiment J1 serves as a standard against which the others can be compared
219: since the detector resolution is close to $\eta$. As it turns out, resolution has an impact
220: both on positive and negative moments. This figure shows the $\tau(q)$ curve of this experiment]
221: {\label{figreftq} \psfig{file=fig2.eps,width=6.3cm}}%
222: \subfigure[\em The $f(\alpha)$ curve corresponding to experiment J1, see also figure
223: \ref{figreftq}. This curve was obtained by a Legendre transform of $\tau(q)$. Errors
224: are amplified due to the great leverage of $q$.]
225: {\label{figreffa} \psfig{file=fig3.eps,width=6.3cm}}%
226: %\caption{}%
227: \end{figure}
228: \addtocounter{figure}{1}
229:
230: \section{Multifractal Spectra in the Locally Isotropic Case}
231:
232: In the following, four examples of classical fully developed turbulence are
233: investigated in the centre of the flows, where local isotropy is
234: known to hold for small scales. The experiments were chosen to cover a range
235: of $R_{\lambda}$ from 190 to 520.
236: \mbox{Table 1} gives the relevant details. An example of an energy time series
237: is given in \mbox{figure \ref{ftshom}.}
238:
239: \begin{table}[h]
240: \begin{center}
241: \begin{tabular}{|l|c|c|c|c|} \hline
242: flow & free jet & free jet & wake & wake \\
243: & & & of a cylinder & of a cylinder \\ \hline \hline
244: reference & J1 & J2 & CF1 & CF2 \\ \hline
245: position of probe & 125 diam. & 60 diam. & 32 diam. & 32 diam. \\ \hline
246: number of points & $1.25\times 10^7$ & $1.25\times 10^7$ & $1.25\times 10^6$ & $1.25 \times 10^6$ \\
247: \hline
248: mean velocity & 2.26 m/s & 4.2 m/s & 22.2 m/s & 4.6 m/s \\ \hline
249: sampling rate & 8 kHz & 50 kHz & 120 kHz & 20 kHz \\ \hline
250: Kolmogorov & 0.25 mm & 0.1 mm & 0.07 mm & 0.22 mm \\
251: microscale $\eta$ & & & & \\ \hline
252: Taylor microscale $\lambda$ & 27 $\eta$ & 28 $\eta$ & 44 $\eta$ & 29 $\eta$ \\
253: $\lambda = \frac{\langle(u-\bar{u})^2\rangle}{\langle (\partial_x(u-\bar{u}))^2 \rangle}$ & & & & \\ \hline
254: Taylor-Reynolds number& 190 & 210 & 520 & 220 \\
255: $R_{\lambda} = \frac{\lambda \sqrt{\langle (u - \bar{u})^2 \rangle}}{\nu}$ & & & & \\ \hline
256: transitional range & 4 .. 113 $\eta$ & 8 .. 83 $\eta$ & 26 .. 79 $\eta$ & 8 .. 85 $\eta$ \\ \hline
257: detector resolution & 4 $\eta$ & 10 $\eta$ & 14 $\eta$ & 5 $\eta$ \\ \hline
258: permissible moments & -2 .. 5 & -0.2 .. 4 & -0.6 .. 4 & -4 .. 4 \\ \hline
259: \end{tabular}
260: \end{center}
261: \vspace{.3cm}
262: Table 1 {\em Summary of experimental conditions and evaluation parameters
263: for the locally isotropic flows.}
264: \end{table}
265:
266: In the context of this paper, we want to establish experiment J1 as the standard for a
267: simple reason: the detector resolution and $\eta$ are quite close, so that one can
268: assume that the fine structure of the original velocity signal is well resolved. In fact,
269: a direct comparison with experiment J2 shows that this point is crucial for the
270: calculation of extreme moments. Figures \ref{figreftq} and \ref{figreffa}
271: show the $\tau(q)$ and $f(\alpha)$ curve for the transitional range.
272:
273: Figure \ref{fhtb} shows
274: a superposition of the respective $f(\alpha)$ spectra for all experiments of table 1.
275: The errorbars are significantly larger than the deviations between the curves. Given
276: these results, we have to assume universality of the spectra of singularities.
277:
278:
279: \begin{figure}[htp]%
280: \hspace{1cm} \psfig{file=fig5.eps,width=11cm,height=6.8cm}%
281: \caption{\label{fhtb} \em The superposition of the $f(\alpha)$ curves of all locally isotropic
282: experiments (J1, J2, CF1, CF2) are in reasonable agreement. The solid line is the standard J1,
283: the errorbars given for CF1 are representative for all experiments with 'poor' statistics.}
284: \end{figure}
285:
286: Although it is notoriously difficult to establish negative values of $f$ reliably,
287: we have confidence in the results we have given. A self consistent analysis \cite{Alber98}
288: of the $\tau(q)$ curve showed that statistics are still sound for $q \approx 3$,
289: whereas the intersection with the $\alpha$-axis occurs for $q=3.2$ c.f. figure \ref{tqfa}.
290:
291: There is a couple of interesting, but less obvious results. Firstly, from correlation
292: plots of $C(q,r)$ one is led to deduce the presence of a phase transition at $q_0 = 1$, which
293: does not express itself in a kink in $f(\alpha)$ since both tangents necessarily
294: have slope 1. We interpret this as the presence of two phases: the highly intermittent
295: or turbulent phase which dominates the moments for $q>1$, and the ''laminar'' phase
296: which does so by the less pronounced singularities for moments with $q<1$.
297:
298: Secondly, if one sets a threshold for the analysis of the measure, one finds that
299: for all (very small) thresholds the laminar branch of the multifractal spectrum
300: disappears. The conclusion is, that all energy dissipation must be concentrated
301: on the set which corresponds to the H\"older exponent $\alpha(q=1)$. Whilst this
302: is necessarily so for mathematical multifractals, it is remarkable for a physical,
303: hence finite one. In other words, the steepest singularities feed on most of the
304: energy and dissipation attracts further energy to dissipate.
305:
306: \pagebreak
307: \section{Multifractal Spectra in the Inhomogeneous Case}
308:
309: Inherent to most theories pertinent to small scale turbulence is the assumption of isotropy
310: and homogeneity of the turbulent velocity field. A question of practical importance
311: is if similar properties can be found where these conditions are violated.
312: To approach this we investigated a free jet with an radial offset of 9 cm in 48 cm
313: axial distance
314: from the nozzle, i.e. in the outer layers where laminar phases are mixed in.
315: As a second inhomogeneous situation the wake of a cylinder was examined in 4 diameters
316: distance, i.e. under conditions of not fully developed turbulence.
317: \mbox{Figure \ref{finhom}} shows the energy dissipation field for these cases.
318: In comparison to figure \ref{ftshom} we clearly see intermittent
319: burst of turbulent activity in an otherwise smooth flow.
320:
321: \begin{figure}[htbp]%
322: \label{finhom}%
323: \centering%
324: \subfigure[\em The energy time series of experiment J3 in the off-centre region of a free jet.
325: The width was chosen to match figure \ref{ftshom} in terms of $\eta$. Laminar phases
326: are mixed with turbulent bursts which were transported out of the central region of the jet.]
327: {\psfig{file=fig6.eps,width=6.2cm}}%
328: \subfigure[\em The energy time series of experiment CV1 in the wake close behind the cylinder.]
329: {\psfig{file=fig7.eps,width=6.2cm}}%
330: \end{figure}
331: \addtocounter{figure}{1}
332:
333: \begin{table}[hp]
334: \begin{center}
335: \begin{tabular}{|l|c|c|} \hline
336: flow & free jet & wake \\
337: & off-centre & near a cylinder \\ \hline \hline
338: reference & J3 & CV1 \\ \hline
339: position of probe & 60 diam. & 4 diam. \\
340: & 11.25 diam. off-centre & \\ \hline
341: number of points & $1.25\times 10^7$ & $1.25\times 10^7$ \\ \hline
342: mean velocity & 0.43 m/s & 21.2 m/s \\ \hline
343: sampling rate & 6 kHz & 200 kHz \\ \hline
344: Kolmogorov & 0.17 mm & 0.035 mm \\
345: microscale $\eta$ & & \\ \hline
346: Taylor microscale $\lambda$ & 11 $\eta$ & 53 $\eta$ \\ \hline
347: Taylor-Reynolds & 34 & 720 \\
348: number $R_{\lambda}$ & &\\ \hline
349: transitional range & 4 .. 25 $\eta$ & 18 .. 129 $\eta$ \\ \hline
350: detector resolution & 6 $\eta$ & 28 $\eta$ \\ \hline
351: permissible moments & 0 .. 4 & 0 .. 4 \\ \hline
352: \end{tabular}
353: \end{center}
354: \vspace{.3cm}
355: Table 2:{\em Summary of experimental conditions and evaluation parameters
356: for the inhomogeneous flows.}
357: \end{table}
358:
359: The analysis is to some extent different from the homogeneous case. Firstly, most
360: quantities like $R_{\lambda}$ and $\eta$ loose their physical meaning.
361: Secondly, the $\Lambda^2$ plots become distorted by the laminar
362: phases, so that the analysis is open to some greater degree of arbitrariness than in
363: the homogeneous case. Table 2 gives the approximate values.
364:
365:
366: \begin{figure}[htnp]
367: \hspace{1.5cm} \psfig{file=fig8.eps,width=10cm}
368: \caption{\label{fiht} \em In the inhomogeneous case, the right branch of $f(\alpha)$ is not accessible.
369: The solid line again is the standard J1, errorbars are given for J3. Statistcs are quite
370: poor for the jet measurement due to the great dilatation of the turbulent bursts.}
371: \end{figure}
372:
373: From the spectra in Figure \ref{fiht} it could be deduced that the small scale
374: properties of the energy dissipation field are to some degree indepedent of the
375: prerequisites of global uniformity and isotropy. We cannot state yet to which extent
376: these conditions do not hold within a turbulent burst, but we accept the possibility
377: that the multifractal aspects of the energy dissipaion field are in fact universal.
378:
379: Notice that the `laminar' branch of the spectrum is not accessible due to the presence
380: of a laminar background. Thus the right hand part of the spectrum collapses to a point
381: at the maximum. The peak at unity indicates that the energy dissipation is still space
382: filling locally, despite its sparse appearance.
383:
384: \section{Conclusion}
385:
386: We have reported on a correspondence in small scale energy dissipation spectra
387: in both locally isotropic and inhomogeneous turbulence.
388: The results are plausible if one takes into consideration
389: that firstly the analysis is restricted to small scales of less than
390: 100$\eta$, and secondly, by the observation that the inhomogeneous turbulence in the
391: boundary layer of the free jet can be thought of as a composition of zones of turbulence
392: transported out of the centre and inserts from the surrounding laminar flow
393: \cite{Sreenivasan91}.
394: Less accessible to intuition are our findings for the non-developed turbulent wake.
395:
396: To set our findings in context with previous work on small scale turbulence,
397: we want to stress once more that our work focuses on the transitional regime,
398: a range of length scales between the inertial range and the dissipative limit.
399: Here we expect that the dissipation plays a
400: role of increasing importance. Traditionally, scaling behaviour is investigated
401: on larger scales in the inertial range, where the dissipation should be of minor
402: influence.
403: Some recent work has shed new light on the turbulent cascade in the inertial range.
404: \cite{Frisch95,Greiner97,Nelkin96,Naert97}. Our results suggest that in the neighbouring
405: regime the statistics of the energy dissipation field $\epsilon_r$ become universally
406: multifractal.
407: The challenge remains to match the various regimes in which turbulence is described
408: to finally arrive at a truly comprehensive view.
409:
410:
411: \section{References}
412: \bibliographystyle{unsrt}
413: \bibliography{diplu}
414:
415: \end{document}
416:
417:
418:
419:
420:
421:
422: