1:  \section{Experimental results}\label{sec:twopoint}

2: 

3: \subsection{Two point statistics}

4: 

5: First of all, we characterize our data by two-point statistics (structure functions, pdf of increments etc.). In this way, we have a reference to results presented frequently in the literature and we get a first impression of the differences between longitudinal and transverse increments.

6:  

7: A central pre-assumption for many considerations of small scale turbulence is isotropy. To study isotropy we use the K\'arm\'an equation (isotropy relation) because it is the simplest exact relation for structure functions which rely on isotropy. The K\'arm\'an equation connects the second order longitudinal and transversal structure functions:

8:  \begin{eqnarray}\label{karman1}

9: \left<v_r^2\right>=\left<u_r^2\right>+\frac{r}{2}\frac{\partial}{\partial r}\left<u_r^2\right>\;\;\;\textrm{first K\'arm\'an equation}.

10: \end{eqnarray}

11: Fig. \ref{fig:karman} shows the left and right hand side of the K\'arm\'an equation, i.e. the second transverse structure function and the hypothetic transverse structure function for isotropic flows calculated from the longitudinal one. Taking the validity of the K\'arm\'an equation as indication of isotropy, we see that for small distances $r<L$ the isotropic relation is well fulfilled, the relative error is within 5\%. For large distances $r>L$ isotropy is violated (large scale anisotropy), the ratio $\langle v_r^2\rangle/\langle u_r^2\rangle$ is approximately 0.66, which is a typical value in the literature.

12: \begin{figure}[tbp]

13: \begin{center}

14: \includegraphics[width=3.5in]{Karman2.eps}

15: \end{center}

16: \caption{Second order transverse structure function estimated on the one hand directly from data (squares) and on the other hand from the longitudinal structure function via the K\'arm\'an equation (straight line); i.e. the left and right hand side of equation (\ref{karman1}) are plotted. The relative deviations are plotted by a dashed line. For smaller scales $r<L$ the K\'arm\'an equation is well fulfilled within 5\% and is isotropic in this sense. On larger scales the data are clear anisotropic. \label{fig:karman} }

17: \end{figure}

18: 

19: Next, we analyze the data with respect to scaling properties. Fig. \ref{fig:scaling}a) shows the energy spectrum with Kolmogorov's -5/3--law for comparison. In Fig. \ref{fig:scaling}b) the third order structure function is plotted in a compensated representation, i.e. $\langle |u_r^{3}|\rangle/r$ is plotted against $r$ to estimate the scaling range. The maximum lies at $10^{-2}\;$m, which defines according to Kolmogorov's 4/5-law the position and the width of the scaling range.

20: 

21: Because the scaling range is narrow, we use the extended self similarity (ESS) \cite{benzi93b,benzi95}, see also appendix \ref{appendixa},

22: \begin{equation}

23: \left<|u_r|^n\right>\propto\left<|u_r|^3\right>^{\xi_n^l}\label{ESSa1}

24: \end{equation}

25: \begin{equation}

26: \langle |v_r|^n\rangle\propto \langle|u_r|^3\rangle^{\xi_n^t}\label{ESSb1}

27: \end{equation}

28: to estimate the scaling exponents of the structure functions. In Fig. \ref{fig:ESS} we show the application of ESS to the third, fourth and sixth order structure functions. The third order structure function is of especially interest, because it serve as the reference in ESS. In all three figures it can be seen that the transverse exponent is smaller than the longitudinal one, $\xi_n^{t}< \xi_n^{l}$. This result is well accepted \cite{pearson01,dhruva97,antonia97a,zhou00,chen97a}. As a consequence, one scaling group is not enough to characterize the turbulent flow and the statistics is more complex than previous thought.

29: 

30: To make this more clear, Fig. \ref{fig:ESSb} shows the exponents $\xi_n^l$ and $\xi_n^t$ up to order 8. For the sixth order structure function we get $\xi_6^l= 1.76\;\pm 0.04$. Fitting Kolmogorov's log-normal model to the values of $\xi_n^l$ yields the intermittency parameter $\mu=$0.24. Both is in good accordance with the experimentally expected values, see for example \cite{arneodo96}. The transverse exponents are clearly smaller then the longitudinal exponents for $n>3$, i.e. the transverse structures are significantly more intermittent, see Fig. \ref{fig:ESSb}. This heavily discussed point will be discussed in detail at the end of this article.

31: 

32: \begin{figure}[tbp]

33: \begin{center}

34: \includegraphics[width=2.5in]{spectrum}% Here is how to import EPS art

35: \includegraphics[width=2.5in]{s3.eps}

36: \end{center}

37: \caption{a) Energy spectrum (dots) with a -5/3--power law and b) compensated plot of the third order structure function. The plateau of $\left<|u_r|^3\right>/r$ defines the scaling range SR. (Data for R$_\lambda$=180.)\label{fig:scaling}}

38: \end{figure}

39: 

40: \begin{figure}[tbp]

41: \begin{center}

42: \includegraphics[width=1.8in]{ESS3}% Here is how to import EPS art

43: \includegraphics[width=1.8in]{ESS4}

44: \includegraphics[width=1.8in]{ESS6}

45: \end{center}

46: \caption{The ESS representation of the longitudinal and transverse structure functions. a) The transverse third order structure function is plotted versus the longitudinal third order structure function. The exponent is $\xi^{t}_3=0.87$. The line has an exponent of 1, which is expected from Kolmogorov's 4/5-law for the third order longitudinal structure function.  b) The fourth order structure functions and c) the sixth order structure functions. Between the longitudinal and transverse structure functions clear differences can be seen. \label{fig:ESS} }

47: \end{figure}

48: 

49: \begin{figure}[tbp]

50: \begin{center}

51: \includegraphics[width=3.5in]{ESS.eps}% Here is how to import EPS art

52: \end{center}

53: \caption{\label{fig:ESSb} The scaling exponents estimated with ESS. Bold squares: the scaling exponents $\xi_n^l$ up to order 8. The dotted line is the forecast of K41, the straight line includes the intermittency corrections of K62 with the fitted intermittency parameter $\mu=0.24$. Open squares: transverse exponents.}

54: \end{figure}

55: 

56: An other quantity which enables to quantify intermittency is the flatness $F_\alpha\equiv\left<\alpha^4\right>/\left<\alpha^2\right>^2$, where $\alpha$ is $u_r$ or $v_r$, respectively. For a Gaussian distribution it is $F_\alpha\equiv 3$ and for an intermittent distribution it is $F_\alpha> 3$. As shown in Fig. \ref{fig:flatness}, for $r<L$ both components are intermittent. For small $r$ the transverse increments are considerably more intermittent than the longitudinal one.

57: \begin{figure}[tbp]

58: \begin{center}

59: \includegraphics[width=3.5in]{flatness.eps}% Here is how to import EPS art

60: \end{center}

61: \caption{\label{fig:flatness} Flatness of longitudinal and transverse increments. For scales $r<L$ the flatness lies for both components over the Gaussian value of 3. Furthermore the flatness is larger for the transverse component, i.e. it is more intermittent.}

62: \end{figure}

63: 

64: \begin{figure}[tbp]

65: \begin{center}

66: \includegraphics[width=5in]{pdf.eps}% Here is how to import EPS art

67: \end{center}

68: \caption{Probability density functions (pdf) for a) longitudinal and b) transverse increments on three different length scales $r=L,\;L/5,\;\lambda\approx \lambda$. The pdfs' width are normalized with respect to their standard deviations and are shifted along the ordinate for a better representation. On the largest scale $L$ a Gaussian distribution is fitted for comparison. Towards smaller scales the deviations from the Gaussian form become obvious. \label{fig:pdf} }

69: \end{figure}

70: 

71: \begin{figure}[tbp]

72: \begin{center}

73: \includegraphics[width=5in]{pdf_uv.eps}% Here is how to import EPS art

74: \end{center}

75: \caption{A direct comparison between the longitudinal and transverse pdfs, a) for $r=L/5$ and b) for $r=\lambda$. Longitudinal pdfs are represented by black squares, transverse pdfs by white squares. The width of the pdfs are normalized with respect to their standard deviation. For $r=L/5$ the deviations are mainly given by the skewness, for $r=\lambda$ also differences in intermittency, i.e. exponential tails, occurs. \label{fig:pdf_uv} }

76: \end{figure}

77: 

78: %einleitender Satz XX

79: In Fig. \ref{fig:pdf} the probability density of longitudinal and transversal velocity increments are plotted. The velocity distribution on each scale is normalized with the respective standard deviation, i.e. $u_{\sigma}:= u_r/\sigma_{u,r}$ and $v_{\sigma}:= v_r/\sigma_{v,r}$, to compare only the form of the curves. In Fig. \ref{fig:pdf_uv} the longitudinal and transversal probability densities for two different length scales ($r=L/5$ and $r=\lambda$) are shown. In both figures, the intermittent character of the statistics can be seen. Whereas for $L=L/5$ the main difference is seen only for positive increments, i.e. the main difference is the skewness, we find for smaller scales that the distributions for the transverse increments is for positive and negative increments more intermittent, see Fig. \ref{fig:pdf_uv}b).

80: 

81: Next we present briefly results from the high Reynolds number data set. The K\'arm\'an equation is well fulfilled, i.e. the data are isotropic in good approximation, see Fig. \ref{fig:KarmanHRe}. Fig. \ref{fig:scalingHRe}a) presents the energy spectrum of the longitudinal increments which shows a distinct scaling range with the exponent -5/3 in accordance to Kolmogorov's theory. The scaling range is more pronounced for this data set as it can be seen also from the third order structure function, compare Figs. \ref{fig:scalingHRe}b) and \ref{fig:scaling}b).  

82: \begin{figure}[tbp]

83: \begin{center}

84: \includegraphics[width=3.5in]{KarmanHRe}% Here is how to import EPS art

85: \end{center}

86: \caption{The second order transverse structure function calculated from data (squares) and calculated from the longitudinal structure function via the K\'arm\'an equation (line). Because the differences in the scaling range can not be seen, also the relative error between them are plotted (dashed line). (R$_\lambda$=550)\label{fig:KarmanHRe}}

87: \end{figure}

88: \begin{figure}[tbp]

89: \begin{center}

90: \includegraphics[width=5in]{scalingHRe}% Here is how to import EPS art

91: \end{center}

92: \caption{a) Energy spectrum (dots) with a -5/3-power law and b) the third order structure function for R$_\lambda$=550. The plateau of $\left<|u|^3\right>/r$ defines the scaling range SR.\label{fig:scalingHRe}}

93: \end{figure}

94: 

95: In this section, we have presented our data with respect to two-point statistics, i.e. regarding only the statistics of increments for one fixed scale separately. The results are comparable to them given in the literature for moderate Reynolds numbers. Differences between longitudinal and transverse increments are clearly visible in pdfs and structure functions indicating that the transverse increments are more intermittent as the longitudinal. In the following we apply a new analysis to study the dependence between different length scale and the interaction between both increments which was not studied so far.

96: 

97: 

98: 

99: