1:
2:
3: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
4: % Manuscrito sobre la Dimension Fractal de la Distribucion
5: % de Regiones HII en Galaxias (espirales e irregulares).
6: % Autores: Sanchez y Alfaro.
7: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
8: % Version 01 (enviada): 2008, Febrero 14.
9: % Version 02 (referee): 2008, Abril 09.
10: % Version 03 (referee2): 2008, Abril 17.
11: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
12:
13: \documentclass[12pt,preprint]{aastex}
14:
15: \begin{document}
16:
17: \title{The Fractal Distribution of HII Regions in Disk Galaxies}
18:
19: \author{N\'estor S\'anchez\altaffilmark{1} and
20: Emilio J. Alfaro\altaffilmark{1}}
21:
22: \altaffiltext{1}{Instituto de Astrof\'{\i}sica de Andaluc\'{\i}a,
23: CSIC, Apdo. 3004, E-18080, Granada, Spain.}
24:
25: \email{nestor@iaa.es}
26:
27: \begin{abstract}
28: It is known that the gas has a fractal structure in a wide
29: range of spatial scales with a fractal dimension that seems
30: to be a constant around $D_f \simeq 2.7$. It is expected
31: that stars forming from this fractal medium exhibit similar
32: fractal patterns. Here we address this issue by quantifying
33: the degree to which star-forming events are clumped. We develop,
34: test, and apply a precise and accurate technique to calculate
35: the correlation dimension $D_c$ of the distribution of HII
36: regions in a sample of disk galaxies. We find that the
37: determination of $D_c$ is limited by the number of HII
38: regions, since if there are $\lesssim 100$ regions available
39: then a bias tending to underestimate the dimension is produced.
40: The reliable results are distributed in the range
41: $1.5 \lesssim D_c \lesssim 2.0$ with an average value $\langle
42: D_c \rangle = 1.81$. This corresponds to a three-dimensional
43: dimension of $\langle D_f \rangle = 2.73$, very similar to
44: the value measured in the interstellar clouds. However,
45: we get significant variations in the fractal dimension
46: among galaxies, contrary to a universal picture sometimes
47: claimed in literature. The fractal dimension exhibits a weak
48: but significant correlation with the absolute magnitude and,
49: to a lesser extent, with the galactic radius. The faintest
50: galaxies tend to distribute their HII regions in more
51: clustered (less uniform) patterns.
52: The fractal dimension for the brightest HII regions within
53: the same galaxy seems to be smaller than for the faintest
54: ones suggesting some kind of evolutionary efffect, but the
55: obtained correlation remains unchanged if only the brightest
56: regions are taken into account.
57: \end{abstract}
58:
59: \keywords{HII regions ---
60: catalogs ---
61: galaxies: structure ---
62: stars: formation}
63:
64: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
65: % AQUI EMPIEZA LA INTRODUCCION
66: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
67:
68: \section{Introduction}
69:
70: One important issue in Astronomy is how the structure and
71: distribution of interstellar gas at different spatial scales
72: is connected to the distribution of newborn stars in the host
73: galaxy. It is known that the gas follows a hierarchical and
74: self-similar structure in a wide range of scales from $\sim
75: 0.1$ pc to $\sim 1$ kpc \citep{Elm04,Ber07}. The fractal
76: dimension characterizing this self-similar structure seems
77: to have a nearly universal value around $D_f \simeq 2.7$
78: \citep{San05,San07Df}. This value is in perfect agreement with
79: recent simulations of compressively driven turbulence in the
80: Interstellar Medium done by \citet{Fed07}, who obtained $D_f
81: \simeq 2.6-2.7$ in their standard simulations. In principle one
82: would expect that newborn stars forming from the high density
83: peaks in these fractal regions should exhibit similar fractal
84: patterns \citep{Elm01}. The analysis of the young stellar
85: population belonging to the Gould Belt yielded a fractal
86: dimension $D_f = 2.68 \pm 0.04$ that is consistent with
87: this picture \citep{San07GB}. However, late-type stars in
88: the Gould Belt have a significantly larger fractal dimension
89: ($D_f = 2.85 \pm 0.04$). There exist several possible
90: causes for this difference
91: \citep[see the discussion in][]{San07GB} but one possibility
92: is that these two stellar populations simply are reflecting
93: two different gas distributions of the parental clouds at
94: different spatial scales according to a multifractal scenario
95: \citep{Cha01,Fue06}.
96:
97: To test this possibility detailed studies of fractal properties
98: at larger spatial scales are required. There is clear evidence
99: that gas also follows fractal patterns at galactic scales (i.e.
100: larger than the kpc). For example, the HI distribution displays
101: a scale-free nature in SMC \citep{Sta99}, LMC \citep{Kim03}, and
102: other external galaxies (both irregulars and spirals) with very
103: different intrinsic properties \citep{Wes99,Wil05,Beg06,Dut08}.
104: A fractal (or multifractal) topology can explain some galactic
105: properties such as flat rotation curves \citep{Pfe94} or the
106: Kennicutt-Schmidt star formation law \citep{Tas07}. What about
107: the spatial distribution of newborn stars at galactic scales?
108: Again, a hierarchical and self-similar picture is consistent
109: with the distribution of star fields and star-forming sites
110: on galaxy-wide scales
111: \citep{Fei87,Elm01,Elm03,Par03,Elm06,Fue06,Ode06,Bas07}.
112: However, it is not clear whether this kind of fractal
113: distributions are connected/related or not to some
114: properties of the host galaxies, such as radius, rotation,
115: brightness, morphology, etc. In spite of the great variety
116: of $D_f$ values reported in the literature for different
117: galaxies (for both the gas and the distribution of star
118: forming sites) most of the authors argue in favor of a
119: more or less universal picture (see the references already
120: mentioned). In this universal description, the constancy of
121: the fractal dimension is a natural consequence of the fact
122: that the same physical processes are structuring these systems.
123: However, there are some indications that the situation could
124: be different and the mechanism that arrange the gas at spatial
125: scales larger than $\sim 1$ kpc could modify the final
126: distribution of the star formation at such scales. The
127: fractal dimension of the distribution of HII regions could
128: be different in grand design and flocculent galaxies
129: \citep[as suggested by][]{Hod85}, and the brightest
130: galaxies could have fractal dimensions higher than
131: faintest ones \citep{Par03,Ode06}. On the contrary,
132: \citet{Fei87} do not find any correlation between the
133: fractal dimension and the galactic properties, but their
134: uncertainties are so large that the robustness of this
135: conclusion is questionable.
136:
137: Part of the problem that prevents achieving unequivocal
138: conclusions lies in the great diversity of analysis
139: techniques used in the literature and/or the application
140: to not large enough samples of galaxies. This paper aims
141: to contribute to fill this gap by providing, firstly,
142: a carefully designed method that has been tested on
143: simulated data and that clearly establishs its accuracy
144: and applicability depending on the sample itself. Therefore,
145: this method can be used in a reliable way to investigate the
146: distribution of HII regions in disk galaxies. We apply this
147: method to the most complete sample of galaxies that we have
148: found in literature expecting to draw significant conclusions
149: regarding this matter. This paper is organized as follows.
150: Section~\ref{sec:teoria} explains the method used to calculate
151: the fractal dimension. It also analyzes the problems arising
152: from projection and sample size effects. Section~\ref{sec:sample}
153: describes the data collected from different sources on which we
154: apply the method in Section~\ref{sec:resultados}. The possible
155: correlations between the fractal dimension of the distribution
156: of HII regions and other galactic properties are discussed in
157: Section\ref{sec:correlaciones}. Finally, the main results are
158: summarized in Section~\ref{sec:conclusiones}.
159:
160: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
161: % AQUI EMPIEZA EL METODO
162: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
163:
164: \section{The fractal dimension of disk-like point distributions}
165: \label{sec:teoria}
166:
167: The primary goal of this work is to calculate the fractal
168: dimension of the distribution of HII regions in galaxies.
169: One way to do this is by using the so-called correlation
170: dimension \citep{Gra83} which gives robust results when
171: dealing with distributions of points in space. Let us
172: consider $N$ points in space with positions ${\bf x}$.
173: The number of other points within a sphere of radius $r$
174: centered on the ${\it i}$-th point is given by
175: \begin{equation}
176: \label{eq:n}
177: n_i(r) = \sum_{j=1,\ j \neq i}^{N}
178: H \left( r- | {\bf x}_i-{\bf x}_j | \right) \ \ ,
179: \end{equation}
180: where $H(x)$ is the Heaviside step function. We can choose
181: $M$ different points as centers and then calculate the
182: commonly called correlation integral in the form
183: \begin{equation}
184: \label{eq:c}
185: a(r) = \frac{1}{M(N-1)} \sum_{i=1}^{M} n_i(r) \ \ .
186: \end{equation}
187: Thus, this quantity represents the probability of finding a
188: point within a sphere of radius $r$ centered on another point.
189: For a fractal set $C(r)$ scales at small $r$ as
190: \begin{equation}
191: \label{eq:dc}
192: C(r) \sim r^{D_c} \ \ ,
193: \end{equation}
194: being $D_c$ the correlation dimension. For a homogeneous
195: distribution of points in a plane we expect $D_c = 2$,
196: whereas if the points are distributed obeying a fractal
197: geometry then $D_c < 2$. When evaluating $C(r)$ for real
198: data the power-law behavior (eq.~\ref{eq:dc}) is valid
199: only within a limited range of $r$ values, even if the
200: distribution follows an underlying fractal law. If $r$
201: is of the order of the mean distance between nearest
202: neighbors then the distribution looks like a set of
203: isolated points. Furthermore, if $r$ tends to the full
204: data set size then boundary effects become increasingly
205: important. In both cases $C(r)$ deviates from the expected
206: power-law behavior and $D_c$ tends to be underestimated
207: \citep[see][and references mentioned therein]{San07GB}.
208: These effects are magnified when the number of available
209: data decrease. We have developed an algorithm to
210: calculate $D_c$ in a reliable way for two-dimensional
211: distributions of points \citep{San07GB}. The novelty of
212: the algorithm lies in the implementation of objective
213: and suitable criteria to avoid both boundary effects and
214: finite-data problems at small scales. First, the algorithm
215: finds the boundary of the set of points by using the
216: procedure proposed by \citet{Edd77}, which determines the
217: vertices of the minimum-area convex polygon containing the
218: whole set of data points. Then we place circles of different
219: radii and evaluate $C(r)$ according to equations~(\ref{eq:n})
220: and (\ref{eq:c}). For this we impose the condition that all
221: circles must be kept inside the sample, this means that circles
222: are not allowed to cross the previously defined boundary. The
223: calculation is done for $r$ values ranging from the minimal
224: distance between two points to the maximum allowable value.
225: The number of possible samplings $M$ decreases as $r$ increases.
226: In each case we calculate both $C(r)$ and the corresponding
227: standard deviation $\sigma_C$. The correlation dimension $D_c$
228: is given by the slope of the best linear fit in a $\log C -
229: \log r$ plot. We establish a lower limit for this fit given
230: by the $r$ value for which $\sigma_C = C(r)$. This simple
231: criterion removes poorly estimated $C(r)$ values occurring
232: mainly at small $r$ values. An upper limit is automatically
233: set by the largest circle fitting into the sample. Finally,
234: the uncertainty associated to $D_c$ is calculated using
235: bootstrap techniques: we repeat the calculation on a series
236: of random resamplings of the data and the standard deviation
237: of the obtained set of $D_c$ values is taken as the error in
238: our estimation ($\sigma_{boot}$). This algorithm yields very
239: good results for distributions of points with a well-defined
240: fractal dimension $D_f$ (monofractals). For simulated
241: two-dimensional fractals with $1 < D_f < 2$ we verified that
242: the relation $D_c-\sigma_{boot} \lesssim D_f \lesssim D_c+
243: \sigma_{boot}$ is always fulfilled \citep{San07GB}. That is,
244: the algorithm gives unbiased fractal dimensions for sufficiently
245: representative samples.
246:
247: What we observe on the image of an external galaxy is the
248: distribution of HII regions projected onto the celestial
249: sphere. From the inclination and position angles we can
250: obtain the image as projected on the mean plane of the
251: galaxy. One important question is the relationship between
252: the correlation dimension measured on this plane (let us call
253: it $D_c$) and the ``real" dimension of the three-dimensional
254: distribution of HII regions within the volume occupied by the
255: galactic disk ($D_f$). On the one hand, if the total thickness
256: of the galactic disk $Z_{disk}$ is much more smaller than the
257: galactic diameter $D_{gal}$, i.e. if $f \equiv Z_{disk}/D_{gal}
258: \ll 1$, then the system can be considered as a very thin
259: slice of the three-dimensional distribution. In this case,
260: the dimension of the three-dimensional distribution and the
261: two-dimensional dimension are related through the expression
262: \citep{Fal90}
263: \begin{equation}
264: \label{eq:slice}
265: D_c = D_f - 1 \ \ .
266: \end{equation}
267: On the other hand, if we have the extreme case $f \simeq 1$,
268: then the image should be treated as a projection of the
269: three-dimensional set and the expected result is \citep{Fal90}
270: \begin{equation}
271: \label{eq:pro}
272: D_c = \min\{2,D_f\}\ .
273: \end{equation}
274: In a previous paper \citep{San05} we showed that for real
275: fractals\footnote{By ``real" fractals we mean distributions of
276: points following an underlying fractal law but not infinite in
277: spatial scale or in number of points and, moreover, having some
278: random component.} the dimension measured on the projected image tends
279: to be lower than the theoretical value given by eq.~(\ref{eq:pro}).
280:
281: \subsection{Projection effects}
282:
283: In a disk galaxy the value of $f$ for the distribution of HII
284: regions would be some intermediate value between a perfect
285: slice ($f=0$) and a projection ($f = 1$). For a typical
286: galaxy like the Milky Way this value should be around $f \sim
287: 100\ {\rm pc} / 10\ {\rm kpc} = 0.01$ \citep[e.g.][]{Pal04}.
288: In order to assess how the slice thickness $f$ would alter
289: the results we have done some numerical tests. First, we
290: have generated three-dimensional fractals following a
291: simple recipe in order to ensure a perfectly defined
292: fractal dimension. Within a cube of size $R$ (half of
293: cube height) we placed
294: $8$ smaller cubes (one in each octant) with size $R/L$
295: ($L \geq 2$). Each cube is placed randomly but always within
296: the volume of its corresponding octant. This procedure is
297: repeated successively $H$ times such as at the end we get
298: a distribution of $8^H$ points (we have used $H=8$ which
299: yields $\sim 16$ millions of points). The fractal dimension
300: of this distribution is given by $D_f = \log 8 / \log L$.
301: For the extreme case $L=2$ we get a homogeneous distribution
302: with $D_f=3$, but if $L>2$ then $D_f < 3$. Then we took a
303: slice along a random plane in space and projected it onto
304: its mean plane. The ratio $f$ between the slice thickness
305: and the size of the fractal is set by hand and is kept as
306: a free parameter. Finally, we randomly removed points from
307: the fractal until reaching the desired sample size $N$ which
308: was also kept as a free parameter.
309:
310: What we have done is to study in detail how the calculated
311: dimension $D_c$ depends on the original dimension $D_f$.
312: Figure~\ref{fig:dcn1000}
313: shows the results for three different values of $f$ but the
314: same sample size ($N=1000$ points). The error bars in this
315: figure come from the random component in the fractal
316: generation process. In general, these error bars tend
317: to be higher at low $D_f$ values because as $D_f$
318: decreases the filling factor decreases and, therefore,
319: the random component increases when each child cube is
320: placed. In other words, these error bars are related to
321: the random variations when generating the fractals and
322: not to the uncertainties in the determination of $D_c$
323: ($\sigma_{boot}$), the latter ones always being smaller
324: than the former ones. The result for the case $f=1$ is
325: quite similar to the previously reported \citep{San05}:
326: the projected dimension tends to be smaller than the
327: theoretical value given by equation~(\ref{eq:pro}).
328: This is because the projection on random planes may
329: produce random groupings and this implies a decrease in
330: $D_c$. This behavior is less pronounced or even non-existent
331: for the cases $D_f \simeq 3$ (the projection of a homogeneous
332: distribution of points is again homogeneous) and $D_f \simeq 1$
333: (the probability of occurrence of chance groupings is low
334: because the filling factor is also low). In general, the
335: projection on the mean plane does not affect the results for
336: the very thin slice case ($f=10^{-4}$) with the calculated
337: values being very close to the theoretical values expected
338: for a zero-thickness slice (eq.~\ref{eq:slice}). The random
339: fluctuations involved in the simulations tend to ``hide" the
340: fractal structure, randomizing any underlying fractal pattern.
341: This may yield $D_c$ values slightly higher than expected for
342: a perfect slice. An interesting point is that when $D_f$
343: decreases too much, structures belonging to lower levels
344: of the hierarchy are so small in comparison with the slice
345: thickness that the resulting distribution is essentially
346: equivalent to a projection on the mean plane of the slice.
347: This change in behavior from thin slice to projection
348: can be clearly seen in Figure~\ref{fig:dcn1000}. The
349: slice-projection transition zone depends, obviously, on
350: both $D_f$ and $f$: as the slice thickness increases the
351: projection behavior appears at higher $D_f$ values. In
352: any case, we have obtained a nearly linear $D_c - D_f$
353: relationship when $D_c \gtrsim 1.5$. Moreover, this
354: relationship is almost independent of the exact $f$
355: value as long as it is around or less than the typical
356: value $f \simeq 0.01$ (compare, for example, cases
357: $f=10^{-2}$ and $f=10^{-4}$ in Figure~\ref{fig:dcn1000}).
358: Thus, any uncertainty in $f$ would translate into a large
359: uncertainty when estimating $D_f$ from $D_c$ only if $D_c
360: \ll 1.5$ and/or $f \gg 10^{-2}$. Otherwise the fractal
361: dimension would be well-determined. We will see later
362: that our ``reliable" results for the distribution of
363: HII regions in galaxies lie in the range $1.5 \leq D_c
364: \leq 2.0$. Table~\ref{table1} presents the resulting
365: values for $D_c \geq 1.5$. The results shown correspond
366: to the case $f=10^{-2}$ but, as mentioned before, these
367: results do not significantly depend on the exact $f$
368: value.
369:
370: \subsection{Sampling size effects}
371:
372: How does the measured dimension depend on the sample size?
373: This is an important question because when working with
374: observational data many times the number of available data
375: points is rather small. There can be both extrinsic causes
376: for this (biased subsamples due to selection effects or
377: criteria in the detection of HII regions) and intrinsic
378: ones (for example, very low star formation rates would
379: produce a small number of clearly distinguishable HII
380: regions). Previously \citep{San07GB} we showed that
381: $D_c$ tended to be underestimated in two-dimensional
382: fractals having a relatively small number of points.
383: This effect could yield unrealistic results and/or
384: trends. In order to quantify this effect we repeated
385: the previous calculations with the same fractals and
386: slices but randomly removing more points until reaching
387: smaller sample sizes $N$. Figure~\ref{fig:dcfslice}
388: shows the results for $f=10^{-4}$ and two different fractal
389: dimensions $D_f=3$ and $D_f=2$, for which the expected values
390: for thin slices are $D_c=2$ and $D_c=1$, respectively
391: (indicated by horizontal lines). Since we are dealing
392: exactly with the same original data, the observed
393: decreasing in the average $D_c$ value with decreasing $N$
394: have to be attributed exclusively to sample size itself.
395: Thus, in general, the algorithm is able to estimate the
396: fractal dimension in a reliable way when applied to random
397: subsamples of fractal distributions of points. However, if
398: the sample size is too small ($N \lesssim 100$) a bias
399: tending to underestimate the dimension is produced.
400:
401: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
402: % AQUI EMPIEZA LA MUESTRA USADA
403: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
404:
405: \section{The sample of galaxies}
406: \label{sec:sample}
407:
408: We have used VizieR\footnote{http://vizier.u-strasbg.fr}
409: \citep{Och00} and ADS\footnote{http://adswww.harvard.edu}
410: databases, in conjunction with the papers of \citet{Gar91}
411: and \citet{Gar02}, to search for catalogs of external
412: galaxies containing positions of HII regions available
413: in machine readable format. We found a total of 93
414: spiral galaxies with positions for at least 50 HII
415: regions. If several different catalogs were available
416: for the same galaxy we choose the one having the largest
417: amount of data. We have also included data for 8 irregular
418: galaxies with HII positions kindly provided by D. Hunter
419: \citep{Roy00}. The properties of the selected galaxies
420: are listed in Table~\ref{table2}. The columns contain the
421: following information: (1) galaxy name, (2) morphological
422: Hubble type, (3) and (4) position $\phi$ and inclination $i$
423: angles respectively (in degrees), (5) morphological type
424: encoded in the de Vaucouleurs scale $T$, (6) spiral arm
425: class $A$, (7) distance to the galaxy $D$ (in Mpc), (8)
426: B-band absolute magnitude $M_B$, (9) radius corresponding
427: to the isophotal level 25 mag/arcsec$^2$ in the B-band
428: $R_{25}$ (in kpc), (10) maximum rotation velocity corrected
429: for inclination $V_{rot}$ (in km s$^{-1}$), and (11) the
430: reference from which positions of HII regions were obtained
431: (Ref). Most of the data for spiral galaxies was taken from
432: HyperLeda\footnote{http://leda.univ-lyon1.fr} database
433: \citep{Pat03}. The arm class was taken from \citet{Elm87}.
434: The position and inclination angles were taken from
435: \citet{Gar91} and \citet{Gar02}, if available, or from
436: HyperLeda otherwise. All the data for irregular galaxies
437: come from \citet{Roy00}. Figure~\ref{fig:galaxias} shows
438: the deprojected distributions of HII regions for each
439: galaxy considered here according to the references
440: given in the last column of Table~\ref{table2}.
441:
442: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
443: % AQUI EMPIEZAN LOS RESULTADOS
444: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
445:
446: \section{The fractal dimension of the HII region distributions}
447: \label{sec:resultados}
448:
449: It can be the case that clustering properties of HII regions
450: differ from region to region in the same galaxy. These
451: differences may be due to some bias (as, for example, if
452: a smaller amount of data exists in a more poorly observed
453: region of the galaxy) or they may be real differences (for
454: example, the organizing role of spiral waves will be smaller
455: close to the central region). Obviously, the algorithm
456: returns only an average estimation of $D_c$ for the
457: entire distribution of points. Given the sufficiently
458: representative sample of galaxies we expect that this
459: effect does not affect our main results and conclusions.
460: One important point that has to be taken into consideration
461: while estimating $D_c$ is the possible presence of outliers.
462: If a relatively small number of points are clearly separated
463: from the rest of the data then the resulting $D_c$ value
464: could be artificially low because of the presence of large
465: empty spaces. The algorithm takes this possibility into
466: account in a first approximation by defining the boundary.
467: The points that determine the boundary are not considered
468: in the calculation because any circle including any of
469: these points is automatically outside the boundary. Thus,
470: it does not matter if any of these points is an outlier
471: because it is not used in the calculation. However, we
472: have added an additional criterion to minimize possible
473: problems arising from this effect: if a small fraction
474: of points (we have chosen 10\% of the total sample) is
475: located outside $R_{25}$ then those points are removed
476: from the calculation. This criterion does not affect
477: the results if those points are not actually outliers.
478: The points that have been excluded from the calculation
479: are surrounded by circles in Figure~\ref{fig:galaxias}.
480:
481: The result of calculating the correlation dimension of
482: the distribution of HII regions in the sample of galaxies
483: is shown in Table~\ref{table3}. The columns are the galaxy
484: name, the number of cataloged HII regions $N$, the calculated
485: dimension $D_c$, the uncertainty resulting from bootstrapping
486: $\sigma_{boot}$, and the range of spatial scales $R_{min}-R_{max}$
487: (in kpc) over which the HII regions are distributed.
488: The calculated $D_c$ values are also shown in parenthesis in
489: Figure~\ref{fig:galaxias}. It can be seen that $D_c$ is consistent
490: with the appearance of the HII region distributions. $D_c$ values
491: close to 2 correspond to nearly homogeneous distributions. Smaller
492: $D_c$ values correspond to more irregular distributions having
493: clumps and/or filaments separated by low density (or empty) regions.
494: As an example, we can compare the smallest and the largest $D_c$
495: values in first page of Figure~\ref{fig:galaxias}: NGC~337A
496: ($D_c=1.34$) exhibits a much more clumpy distribution than
497: the relatively homogeneous galaxy NGC~1068 ($D_c=1.99$).
498:
499: Figure~\ref{fig:dcnhii}
500: shows the calculated correlation dimension for all the
501: galaxies in the sample as a function of the number of
502: available data points. The first thing we note is that
503: there exist significant differences among the galaxies:
504: in general, the uncertainties in the determination of
505: the dimension ($\sigma_{boot}$) are smaller than the
506: dispersion of the obtained values ($\sigma \simeq 0.3$).
507: This result differs considerably from the conclusion
508: outlined by \citet{Fei87} for their sample of 19 spiral
509: galaxies who argued in favor of a constant $D_c$ around
510: the mean value $\langle D_c \rangle = 1.68$. Two trends
511: are clearly visible in Figure~\ref{fig:dcnhii}. First,
512: the uncertainties increase as $N(\mathrm{HII})$ decreases,
513: consistent with the results obtained in the previous
514: simulations. But moreover, at small number of regions
515: the obtained values are substantially more spread out
516: toward lower values. The results were averaged in bins
517: of the number of HII regions with each bin having
518: approximately the same number of galaxies ($\sim 25$).
519: Table~\ref{table4} summarizes the results, where for
520: each bin we show mean values (also shown in
521: Figure~\ref{fig:dcnhii}), medians, and standard
522: deviations both for the fractal dimensions and for the
523: corresponding uncertainties. The overall trend, i.e.
524: the convergence of the mean value to a certain value
525: (in this case $\sim 1.8$) as the number of data points
526: increases, is very similar to the behavior described for
527: the simulated fractals (Figure~\ref{fig:dcfslice}). It is
528: clear that at least part of this trend is due to a bias in
529: the estimated value of $D_c$ for galaxies with small number
530: of HII regions. It is only when $N(\mathrm{HII}) \gtrsim 200$
531: that we obtain uncertainties below $\sigma_{boot} = 0.1$.
532: Therefore, to overcome this bias, we focus the detailed
533: analysis (next section) on galaxies having more than 200
534: HII regions (46 spiral galaxies). The average fractal
535: dimension in this case is $\langle D_c \rangle = 1.81$
536: with a standard deviations of $\sigma = 0.14$. According to
537: the results of Section~\ref{sec:teoria} (Table~\ref{table1}) this
538: corresponds to an average three-dimensional dimension of $\langle
539: D_f \rangle = 2.73$. Our average result is very similar
540: (within the uncertainties) to the result
541: of \citet{Fei87} ($\langle D_c \rangle = 1.68$).
542: However, the average uncertainty obtained by \citet{Fei87}
543: is $\sim 0.3$ whereas we get an average $\sigma_{boot}$ of
544: $\sim 0.03$. By considering these associated uncertainties
545: we can say that, contrary to our results, their $D_c$ values
546: do not differ among themselves. Regarding the irregular
547: galaxies, we have only a very small number of galaxies (8)
548: and they have no more than 200 HII regions (the value we
549: are considering as the limit for an unbiased determination
550: of $D_c$). Thus, any conclusion based on these data should
551: be treated with great caution. The average fractal dimension
552: for irregulars is $\langle D_c \rangle = 1.79$ (similar to
553: that found for spirals) with a standard deviation of
554: $\sigma = 0.20$. The average value for $\sigma_{boot}$
555: is $0.14$ so that there are not significant variations
556: among these galaxies. The results obtained by \citet{Par03}
557: are spread over the range $1.2 \lesssim D_c \lesssim 2.0$,
558: but if we only consider their five galaxies having $\sim 200$
559: or more HII regions then this range narrows to $1.55 \lesssim
560: D_c \lesssim 1.62$. This is below our average result mostly
561: because our result represents a lower limit in the case of
562: bias due to small sample size.
563:
564: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
565: % AQUI EMPIEZA LA DISCUSION
566: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
567:
568: \section{Dependence on galactic properties}
569: \label{sec:correlaciones}
570:
571: At this point, we have a set of fractal dimension values
572: determined in a precise and accurate way and a set of
573: variables describing some galactic properties, such
574: as the position angle $\phi$, inclination $i$,
575: morphological type $T$, arm class $A$, distance
576: $D$, absolute magnitude $M_B$, radius of the isophotal
577: 25 $R_{25}$, rotation velocity $V_{rot}$, and the number
578: of HII regions $N(\mathrm{HII})$. Now we proceed to examine
579: possible correlations between the fractal dimension and these
580: variables but taken into account that they represent properties
581: of different nature. Some of these quantities are in someway
582: involved in the determination of the fractal dimension and
583: may, therefore, introduce some bias or uncertainties into the
584: calculated $D_c$ values. These biases can be introduced either
585: by taking part directly in the calculation or through the
586: observation or deprojection processes. If this is the
587: case then spurious correlations can arise. We refer to
588: these as {\sl intrinsic} properties; they are $\phi$, $i$,
589: $D$, and $N(\mathrm{HII})$. On the other hand, the rest of
590: the observed variables are related to galactic properties
591: but they do not participate directly in the determination
592: of $D_c$; these {\sl extrinsic} variables are $T$, $A$,
593: $M_B$, $R_{25}$, and $V_{rot}$. We also analyze a new
594: variable that we have called the average surface density
595: of star forming regions, given by the number of star
596: forming regions divided by the square of the radius
597: ($N(\mathrm{HII})/R_{25}^2$).
598:
599: \subsection{Analysis method}
600:
601: The main goal is to identify possible relationships between
602: the degree of clumpiness of HII regions in galaxies and their
603: {\sl intrinsic} and/or {\sl extrinsic} properties, in the above
604: sense. So we have to suggest a model of dependence and evaluate
605: its ``goodness" via some statistical test defined for such
606: purpose. We propose, in a first approximation, a linear model
607: linking the variables and we choose as statistical criterion
608: the $AIC$ \citep[Akaike's Information Criterion,][]{Aka74},
609: defined as $AIC=2k-2\ln(L)$ where $L$ is the maximized likelihood
610: value and $k$ is the number of estimable parameters in the model.
611: Let $RSS$ be the residual sum of squares and $n$ the number of
612: observations, then the $AIC$ becomes $AIC=2k+n\left[\ln(2\pi
613: RSS/n)+1\right]$. Increasing the number of free parameters to
614: be estimated improves the goodness of fit, regardless of the
615: number of free parameters in the data generating process. Hence
616: $AIC$ not only rewards goodness of fit, but also includes a
617: penalty that is an increasing function of the number of
618: estimated parameters. This penalty discourages overfitting.
619: The preferred model is the one with the lowest $AIC$ value.
620: Once the best model has been selected according to this
621: criterion we perform the $t$ and $F$ tests to infer the
622: significance of the coefficients of the model and of the
623: overall fit. This analysis is performed in the {\bf R}
624: environment for statistical computing \citep{R08}.
625:
626: \subsection{Intrinsic variables}
627:
628: We begin with the intrinsic variables ($\phi$, $i$, $D$, and
629: $N(\mathrm{HII})$) and considering first the whole sample of
630: galaxies. Starting from a linear regression model based on $n$
631: variables we apply a stepwise regression process to select the
632: subset of variables providing the best model. The analysis
633: yields the position angle of the major axis and the number
634: of HII regions as the most significant variables. Thus, the
635: best model to explain $D_c$ variations from these variables
636: has the form $D_c = a\, \phi + b\, N(\mathrm{HII}) + c$.
637: However, both the $t$-test for the regression coefficients
638: and the $F$-test for the overall fit indicate no significance
639: at 95\% confidence level. This result supports our previous
640: finding in simulated fractal distributions: the number of
641: HII regions in a galaxy affects the quality of the estimated
642: fractal dimension by introducing a systematic bias and a noise.
643: The presence of the variable $N(\mathrm{HII})$ in the model is
644: simply the result of the existence of this bias.
645: If we exclude those objects with $N(\mathrm{HII}) < 200$ and
646: repeat the analysis, we obtain that any variable is able to
647: significantly explain the observed variance in the fractal
648: dimension.
649:
650: \subsection{Extrinsic variables}
651:
652: Now we will consider the variables $T$, $A$, $M_B$, $R_{25}$,
653: $V_{rot}$, and $N(\mathrm{HII})/R_{25}^2$, and we will restrict
654: the analysis to galaxies having $N(\mathrm{HII}) > 200$.
655: Figure~\ref{fig:extrinsecas}
656: shows $D_c$ as a function of each one of these variables. First
657: we construct an initial linear model incorporating all of these
658: variables. The application of the task {\sl step} (within the
659: package {\sl stats} of {\bf R}) yields that the variables $M_{B}$
660: and $R_{25}$ represent the best liner model fitting the data,
661: this is
662: \begin{equation}
663: \label{eq:fit}
664: D_c= -(0.069 \pm 0.025) M_{B} -(0.006 \pm 0.003) R_{25} +0.456 \ \ .
665: \end{equation}
666: The overall fit gives $F=4.0$ that for degrees of freedom 2 and 43
667: is significant at the confidence level of 95\%. The regression
668: coefficients of the variables $M_{B}$ and $R_{25}$ are significant
669: at the confidence levels of 99\% and 95\% (marginally significant),
670: respectively.
671:
672: According to our results $D_c$ correlates with $M_{B}$ and, to a
673: lesser extent, with $R_{25}$. Brightest galaxies have fractal
674: dimensions higher than faintest ones (Figure~\ref{fig:extrinsecas}c).
675: A slight trend is apparent in Figure~\ref{fig:extrinsecas}c in
676: which the correlation with $M_{B}$ appears to saturate for galaxies
677: brighter than $M_{B} \simeq -21$, but the large scatter in this
678: $M_{B}$ interval prevents us from drawing a general conclusion.
679: \citet{Fei87} find a pure scatter diagram when comparing $D_c$ and
680: $M_B$ which, together with the lack of correlation with other
681: properties is used as argument favoring a universal $D_c$ value
682: in spiral galaxies. The dependence on $M_B$ that we observe is
683: probably hidden by their imprecise determination of $D_c$
684: (uncertainties of the order of $\sim 0.3$). In contrast,
685: \citet{Par03} report the same trend in their sample of
686: dwarf irregulars, i.e. the highest fractal dimensions for
687: the brightest galaxies. They applied a correction factor
688: to the calculated $D_c$ to take account of differences in
689: the number of HII regions. However, their results have to be
690: taken with caution because, as mentioned before, it is a
691: little disturbing that they have only five galaxies with
692: $N(\mathrm{HII}) \gtrsim 200$ for which the correlation
693: does not exist at all. \citet{Ode06} analyzed in detail the
694: distribution of resolved young stars in star-forming dwarf
695: galaxies. Unfortunately she used only four galaxies, but
696: she obtained that the brightest galaxy had the highest
697: fractal dimension and the faintest galaxy the lowest.
698: The interpretation given by \citet{Par03} is that
699: a small $D_c$ value implies a higher intragalactic
700: gas porosity which reduces the (massive) star formation
701: rate \citep{Sil97}. The same argument remains valid for
702: spiral galaxies if the presence of density waves does
703: not greatly affect the star formation process \citep{Elm86}.
704: For the irregular galaxies in our sample, we find that the
705: average fractal dimension ($\simeq 1.79$) is similar within
706: the uncertainties to that found for spirals with
707: $N(\mathrm{HII}) \gtrsim 200$ ($\simeq 1.81$).
708: Moreover, it is higher than that for spirals with
709: $N(\mathrm{HII}) < 200$ ($\simeq 1.65$) although,
710: statistically, there is no significative difference.
711: Curiosity prompted us to check what happens when irregular
712: galaxies are included in the previous statistical analyses,
713: and what we find is that the correlation disappears.
714: Dwarf irregular galaxies are characterized by fractal
715: dimensions similar to those of the brightest spiral
716: galaxies, so that any possible correlation between
717: $D_c$ and $M_B$ is removed when both types of
718: galaxies are considered. Even if we try to correct
719: the small sample size bias in the irregulars, the
720: resulting $D_c$ values would be higher and this
721: effect would be enhanced. Therefore, in spite of
722: the small number of irregular galaxies studied,
723: we can conclude that if $D_c$ is a function of
724: $M_B$ in irregulars \citep[as suggested by][]{Par03}
725: this correlation does not follow the same law as for
726: spirals; and the fractal dimension for irregulars
727: is, on average, higher than for spirals with the same
728: absolute magnitude.
729:
730: \subsection{Possible evolutionary effects}
731:
732: If we assume that star formation takes place mainly along
733: the spiral arms then the inter-arm regions should typically
734: exhibit older HII regions. This kind of evolutionary effect
735: have been proposed to explain variations in the luminosity
736: functions of HII regions in the spiral arms versus the
737: inter-arm regions \citep[e.g.][]{Oey98}.
738: We wonder whether this kind of effect could introduce some
739: bias because in principle one would expect that only
740: the brightest regions are visible for the farthest
741: (and faintest) galaxies. However, we didn't detect any
742: correlation between the calculated $D_c$ value and the
743: galaxy's distance so that our results would seem to be
744: unaffected by this possible effect.
745: In any case, we have analyzed the dependence of $D_c$ on
746: the brightness of the HII regions for those galaxies
747: with enough available data. We first selected galaxies
748: from the sample for which we have data on HII region
749: brightness and a sample size large enough
750: ($N(\mathrm{HII}) > 600$) to calculate $D_c$ in a
751: reliable way for the brightest 1/3 of the HII regions.
752: There are a total of 9 galaxies fulfilling these
753: requiriments:
754: NGC~628, NGC~3344, NGC~3486, NGC~3631, NGC~3726,
755: NGC~4254, NGC~4303, NGC~4321, and NGC~6946
756: \citep[the reference for this data is][]{Bra06}.
757: We divided the sample in three equal subsamples
758: ordered in descending brightness: the brightest
759: HII regions (high-brightness), the medium bright
760: ones (medium-brightness), and the faintest ones
761: (low-brightness).
762: Figure~\ref{fig:NGC6946} shows, as an illustrative
763: example, the resulting distributions in NGC~6946
764: for which we have little more than $\sim 500$ HII
765: regions in each case.
766: We recalculated $D_c$ for each case and the results
767: are shown in Table~\ref{table5}.
768: The general trend is that $D_c$ is smaller for
769: the brightest HII regions than for the rest of
770: the data. This can be seen by eye from inspection
771: of Figure~\ref{fig:NGC6946} where the left-side
772: panel exhibits, clearly, a more clumpy morphology.
773: The only exceptions to this general trend are
774: NGC~4254 that exhibits the opposite behavior and
775: NGC~4303 that does not show any significant
776: variation. The rest of the galaxies have a
777: smaller $D_c$ value for the brightest HII
778: regions. These regions should reflect, in a
779: first approximation, the initial distribution
780: of star-forming regions in each galaxy. It
781: seems likely from our results that some kind
782: of evolutionary effect tends to randomize
783: (homogenize) in some degree the initial
784: distributions of HII regions.
785:
786: Does this result affect the correlations we found
787: in the previous sections? We have repeated the
788: analysis replacing the full sample $D_c$ values
789: by the high-brightness values for the 9 galaxies
790: in Table~\ref{table5}. First we use all the galaxies
791: having more than $200$ HII regions (46 galaxies).
792: In this case, the analysis yields exactly the same
793: variables for the best fit (eq.~\ref{eq:fit}) with
794: nearly the same $F$ value ($4.1$) and a confidence level
795: of 99\% for the overall fit. The coefficients for $M_{B}$
796: and $R_{25}$ are $-0.069$ and $-0.007$ with confidence
797: levels of 99\% and 95\%, respectively. We have also
798: tested what happens if we take only the galaxies
799: having more than $200$ and less than $600$ HII
800: regions and the high-brightness values for the
801: 9 galaxies in Table~\ref{table5}. The idea behind
802: this selection criterion is to avoid galaxies having
803: a ``mixed" (bright and faint) population of HII regions.
804: Again, the analysis yields $M_{B}$ and $R_{25}$ as the
805: variables providing the best fit to the data. The fit
806: gives $F=4.6$ that for degrees of freedom 2 and 37 is
807: significant at the 95\%. The coefficients obtained in
808: this case for $M_{B}$ and $R_{25}$ are $-0.079$ and
809: $-0.006$ with confidence levels of 99\% and 90\%,
810: respectively.
811: Summarizing, when only the bright HII regions are
812: considered the trend shown in eq.~(\ref{eq:fit})
813: is preserved, i.e. the correlation strength with
814: $M_{B}$ remains the same (or even increases) whereas
815: the correlation with $R_{25}$ is, in the best scenario,
816: marginally significant.
817:
818: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
819: % AQUI EMPIEZAN LAS CONCLUSIONES
820: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
821:
822: \section{Conclusions}
823: \label{sec:conclusiones}
824:
825: The study of the distribution of HII regions in galaxias may
826: provide important clues to understanding the processes involved
827: in the star formation at galactic scales. Our approach in this
828: work has been apply a precise and accurate technique to calculate
829: the degree of self-similar clumpiness (the correlation dimension
830: $D_c$) in a large enough sample of disk galaxies. All the obtained
831: dimensions are shown in Table~\ref{table3}. Our most reliable results
832: are distributed in the range $1.5 \lesssim D_c \lesssim 2.0$ with an
833: average value $\langle D_c \rangle = 1.81$. This value corresponds
834: to a three-dimensional dimension of $\langle D_f \rangle = 2.73$.
835: The similarity between this value and the value $D_f \simeq 2.7 \pm
836: 0.1$ measured in interstellar molecular clouds \citep{San07Df} may
837: seem, in principle, unexpected given the very different spatial
838: scales involved. There are many energy sources that can drive
839: the turbulence and structure the interstellar medium \citep{Elm04},
840: but this structure can be modified at a galactic level due to the
841: action of other physical processes acting at this larger spatial
842: scales. The fractal behavior we have observed span a range of
843: scales from $\sim 10$ pc to $\sim 25$ kpc, and we have not found
844: any dependence on the spatial scales involved.
845:
846: The uncertainties of the most reliable $D_c$ values are small
847: enough ($\sim 0.03$ on average) to allow us to conclude that
848: there are significant variations among galaxies. This conclusion
849: is incompatible with a universal picture in which the fractal
850: dimension is approximately a constant for the spiral galaxies
851: \citep{Fei87}. We have found that $D_c$ exhibits a weak but
852: significant correlation with $M_{B}$ and, to a lesser extent,
853: with $R_{25}$. The faintest galaxies tend to distribute the
854: HII regions in a more clumpy or clustered way. It is the
855: first time that this behavior is reported in spiral galaxies,
856: but it is in agreement with similar results suggested for
857: irregular galaxies \citep{Par03,Ode06}.
858: Moreover, the fractal dimension for the brightest HII regions
859: tends to be smaller than for the faintest ones. This behavior
860: could be the result of some kind of evolutionary effect that
861: tends to randomize in some degree the initial distribution
862: of HII regions \citep{Oey98}. However, if only the brightest
863: HII regions are taken into account the observed correlation
864: with $M_{B}$ and $R_{25}$ remains unchanged.
865: We have not found correlations with other galactic properties
866: (morphological type, spiral arm class, rotation velocity, and
867: surface density of HII regions).
868: There are many galactic properties from which $D_c$ could depend,
869: such as star formation activity, mass, age, metallicity, or a
870: combination of them. It has been suggested that the fractal
871: dimension of the distribution of star-forming sites could be
872: increased during the star formation process
873: \citep[for example,][and this work]{Fue06}.
874: A more complete analysis including more galactic variables
875: and a wider and more diverse sample of galaxies would be
876: necessary in order to obtain a clearer picture.
877:
878: \acknowledgments
879: We acknowledge D.~Hunter for providing the data on irregular
880: galaxies. We also thank an anonymous referee for his/her
881: comments, which improved this paper and led us to add
882: Section~5.4 to an earlier version of this paper.
883: This research has made use of NASA's Astrophysics
884: Data System and of HyperLeda and VizieR databases.
885: We acknowledge financial support from MEC of Spain through grant
886: AYA2007-64052 and from Consejer\'{\i}a de Educaci\'on y Ciencia
887: (Junta de Andaluc\'{\i}a) through TIC-101.
888:
889: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
890: % BIBLIOGRAFIA
891: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
892:
893: \begin{thebibliography}{}
894: \bibitem[Akaike(1974)]{Aka74} Akaike, H. 1974,
895: IEEE Transactions on Automatic Control, 19, 716
896: \bibitem[Bastian et al.(2007)]{Bas07} Bastian, N., Ercolano, B., Gieles,
897: M., Rosolowsky, E., Scheepmaker, R.~A., Gutermuth, R., \&
898: Efremov, Y. 2007, \mnras, 379, 1302
899: \bibitem[Begun et al.(2006)]{Beg06}
900: Begun, A., Chengalur, J. N., \& Bhardwaj, S. 2006,
901: \mnras, 372, L33
902: \bibitem[Bergin \& Tafalla(2007)]{Ber07}
903: Bergin, E. A. \& Tafalla, M. 2007, \araa 45, 339
904: \bibitem[Bradley et al.(2006)]{Bra06}
905: Bradley, T. R., Knapen, J. H., Beckman, J. E., \& Folkes S. L.
906: 2006, \aap, 459, L13
907: \bibitem[Chappell \& Scalo(2001)]{Cha01}
908: Chappell, D. \& Scalo, J. 2001, \apj, 551, 712
909: \bibitem[Courtes et al.(1993)]{Cou93}
910: Courtes, G., Petit, H., Hua, C. T., Martin, P., Blecha, A.,
911: Huguenin, D., \& Golay, M. 1993, \aap, 268, 419
912: \bibitem[de la Fuente Marcos \& de la Fuente Marcos(2006)]{Fue06}
913: de la Fuente Marcos, R., \& de la Fuente Marcos, C. 2006,
914: \mnras, 372, 279
915: \bibitem[Dutta et al.(2008)]{Dut08} Dutta, P., Begum, A., Bharadwaj, S.,
916: \& Chengalur, J. N. 2008, \mnras, 384, L34
917: \bibitem[Eddy(1977)]{Edd77}
918: Eddy, W. F. 1977, {\it Trans. Math. Soft.}, 3, 398
919: \bibitem[Elmegreen \& Elmegreen(1987)]{Elm87}
920: Elmegreen, D. M. \& Elmegreen, B. G. 1987, \apj, 314, 3
921: \bibitem[Elmegreen \& Elmegreen(1986)]{Elm86} Elmegreen, B. G., \&
922: Elmegreen, D. M. 1986, \apj, 311, 554
923: \bibitem[Elmegreen \& Elmegreen(2001)]{Elm01}
924: Elmegreen, B. G. \& Elmegreen, D. M. 2001, \aj, 121, 1507
925: \bibitem[Elmegreen et al.(2006)]{Elm06} Elmegreen, B. G., Elmegreen, D.
926: M., Chandar, R., Whitmore, B., \& Regan, M. 2006, \apj, 644, 879
927: \bibitem[Elmegreen et al.(2003)]{Elm03} Elmegreen, B. G., Elmegreen, D.
928: M., \& Leitner, S. N. 2003, \apj, 590, 271
929: \bibitem[Elmegreen \& Scalo(2004)]{Elm04}
930: Elmegreen, B. G. \& Scalo, J. 2004, \araa 42, 211
931: \bibitem[Evans et al.(1996)]{Eva96}
932: Evans, I. N., Koratkar, A. P., Storchi-Bergmann, T., Kirkpatrick,
933: H., Heckman, T. M., \& Wilson A. S. 1996, \apjs, 105, 93
934: \bibitem[Falconer(1990)]{Fal90}
935: Falconer, K. J. 1990, Fractal Geometry: Mathematical
936: Foundations and Applications (London: Wiley)
937: \bibitem[Federrath et al.(2007)]{Fed07}
938: Federrath, C., Schmidt, W., \& Klessen, R. S. 2007,
939: preprint (arXiv:0710.1359)
940: \bibitem[Feinstein(1997)]{Fei97}
941: Feinstein, C. 1997, \apjs, 112, 29
942: \bibitem[Feitzinger \& Galinski(1987)]{Fei87}
943: Feitzinger, J. V., \& Galinski, T. 1987, \aap, 179, 249
944: \bibitem[Garcia-Gomez \& Athanassoula(1991)]{Gar91}
945: Garcia-Gomez, C., \& Athanassoula, E. 1991,
946: \aaps, 89, 159
947: \bibitem[Garcia-Gomez et al.(2002)]{Gar02}
948: Garcia-Gomez, C., Athanassoula, E., \& Barbera, C. 2002,
949: \aap, 389, 68
950: \bibitem[Gonzalez-Delgado et al.(1997)]{Gon97}
951: Gonzalez-Delgado, R. M., Perez, E., Tadhunter, C., Vilchez,
952: J. M., \& Rodriguez-Espinosa, J. M. 1997, \apjs, 108, 155
953: \bibitem[Grassberger \& Procaccia(1983)]{Gra83}
954: Grassberger, P., \& Procaccia, I. 1983, \prl, 50, 346
955: \bibitem[Hodge(1985)]{Hod85} Hodge, P. 1985, \pasp, 97, 688
956: \bibitem[Hodge et al.(1990)]{Hod90}
957: Hodge, P., Gurwell, M., Goldader J. D., \& Kennicutt, R. C.
958: Jr. 1990, \apjs, 73, 661
959: \bibitem[Hodge et al.(1999)]{Hod99}
960: Hodge, P. W., Balsley, J., Wyder, T. K., \& Skelton, B. P.
961: 1999, \pasp, 111, 685
962: \bibitem[Kim et al.(2003)]{Kim03}
963: Kim, S., Staveley-Smith, L., Dopita, M. A., Sault, R. J.,
964: Freeman, K. C., Lee, Y., \& Chu, Y.-H. 2003, \apjs, 148, 473
965: \bibitem[Knapen et al.(1993)]{Kna93}
966: Knapen, J. H., Arnth-Jensen, N., Cepa, J., \& Beckman, J. E.
967: 1993, \aj, 106, 56
968: \bibitem[Lin et al.(2003)]{Lin03}
969: Lin, W., Zhou, X., Burstein, D., Windhorst, R. A., Chen, J.,
970: Chen, W.-P., Jiang, Z., Kong, X., Ma, J., Sun, W.-H., Wu H.,
971: Xue, S., \& Zhu J. 2003, \aj, 126, 1286
972: \bibitem[Ochsenbein et al.(2000)]{Och00}
973: Ochsenbein, F., Bauer, P., Marcout, J. 2000, \aaps, 143, 221
974: \bibitem[Odekon(2006)]{Ode06}
975: Odekon, M. C. 2006, \aj, 132, 1834
976: \bibitem[Oey \& Clarke(1998)]{Oey98}
977: Oey, M. S. \& Clarke, C. J. 1998, \aj, 115, 1543
978: \bibitem[Paladini et al.(2004)]{Pal04}
979: Paladini, R., Davies, R. D., \& DeZotti, G. 2004,
980: \mnras, 347, 237
981: \bibitem[Parodi \& Binggeli(2003)]{Par03}
982: Parodi, B. R., \& Binggeli, B. 2003, \aap, 398, 501
983: \bibitem[Paturel et al.(2003)]{Pat03}
984: Paturel, G., Petit, C., Prugniel, Ph., Theureau, G., Rousseau,
985: J., Brouty, M., Dubois, P., \& Cambresy, L. 2003, \aap, 412, 45
986: \bibitem[Pellet et al.(1978)]{Pel78}
987: Pellet, A., Astier, N., Viale, A., Courtes, G., Maucherat, A.,
988: Monnet, G., \& Simien, F. 1978, \aaps, 31, 439
989: \bibitem[Petit et al.(1996)]{Pet96}
990: Petit, H., Hua, C. T., Bersier, D., \& Courtes G. 1996,
991: \aap, 309, 446
992: \bibitem[Petit(1998)]{Pet98}
993: Petit, H. 1998, \aaps, 131, 317
994: \bibitem[Pfenniger \& Combes(1994)]{Pfe94} Pfenniger, D., \& Combes,
995: F. 1994, \aap, 285, 94
996: \bibitem[R Development Core Team(2008)]{R08}
997: R Development Core Team 2008,
998: {\it R: A Language and Environment for Statistical Computing},
999: R Foundation for Statistical Computing, Vienna, Austria,
1000: URL http://www.R-project.org, ISBN 3-900051-07-0
1001: \bibitem[Roye \& Hunter(2000)]{Roy00}
1002: Roye, E. W., \& Hunter, D. A. 2000, \aj, 119, 1145
1003: \bibitem[Rozas et al.(1999)]{Roz99}
1004: Rozas, M., Zurita, A., Heller, C. H., \& Beckman, J. E. 1999,
1005: \aaps, 135, 145
1006: \bibitem[Rozas et al.(2000)]{Roz00}
1007: Rozas, M., Zurita, A., \& Beckman, J. E. 2000, \aap, 354, 823
1008: \bibitem[Silk(1997)]{Sil97} Silk, J. 1997, \apj, 481, 703
1009: \bibitem[S\'anchez et al.(2005)]{San05}
1010: S\'anchez, N., Alfaro, E. J., \& P\'erez, E. 2005,
1011: \apj, 625, 849
1012: \bibitem[S\'anchez et al.(2007a)]{San07GB}
1013: S\'anchez, N., Alfaro, E. J., Elias, F., Delgado, A. J., \&
1014: Cabrera-Ca\~no, J. 2007a, \apj, 667, 213
1015: \bibitem[S\'anchez et al.(2007b)]{San07Df}
1016: S\'anchez, N., Alfaro, E. J., \& P\'erez, E. 2007b,
1017: \apj, 656, 222
1018: \bibitem[Stanimirovic et al.(1999)]{Sta99}
1019: Stanimirovic, S., Staveley-Smith, L., Dickey, J. M., Sault,
1020: R. J., \& Snowden, S. L. 1999, \mnras, 302, 417
1021: \bibitem[Tassis(2007)]{Tas07} Tassis, K. 2007, \mnras, 382, 1317
1022: \bibitem[Tsvetanov \& Petrosian(1995)]{Tsv95}
1023: Tsvetanov, Z. I., \& Petrosian A. R. 1995, \apjs, 101, 287
1024: \bibitem[Westpfahl et al.(1999)]{Wes99}
1025: Westpfahl, D. J., Coleman, P. H., Alexander, J., \& Tongue,
1026: T. 1999, \aj, 117, 868
1027: \bibitem[Willett et al.(2005)]{Wil05} Willett, K. W., Elmegreen, B. G.,
1028: \& Hunter, D. A.\ 2005, \aj, 129, 2186
1029: \end{thebibliography}
1030:
1031: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1032: % AQUI VAN LAS TABLAS
1033: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1034:
1035: \clearpage
1036:
1037: \input{tab1.tex}
1038:
1039: \clearpage
1040:
1041: \input{tab2.tex}
1042:
1043: \clearpage
1044:
1045: \input{tab3.tex}
1046:
1047: \clearpage
1048:
1049: \input{tab4.tex}
1050:
1051: \clearpage
1052:
1053: \input{tab5.tex}
1054:
1055: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1056: % AQUI VAN LAS FIGURAS
1057: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1058:
1059: \clearpage
1060:
1061: \begin{figure}[th]
1062: \epsscale{0.9}
1063: \plotone{f1.eps}
1064: \caption{Calculated dimension $D_c$ as a function of the
1065: fractal dimension $D_f$ for three different values of the
1066: slice thickness: $f=1$ (open circles), $f=10^{-2}$ (open
1067: triangles), and $f=10^{-4}$ (open squares). The number of
1068: points for the fractals is $N=1000$. Each point is the
1069: result of calculating the average of $50$ random fractals,
1070: and the bars indicate the standard deviations. The solid
1071: and dashed lines show the theoretical results for a
1072: projection and a thin slice, respectively.}
1073: \label{fig:dcn1000}
1074: \end{figure}
1075:
1076: \clearpage
1077:
1078: \begin{figure}[th]
1079: \epsscale{0.9}
1080: \plotone{f2.eps}
1081: \caption{Calculated dimension $D_c$ as a function of the
1082: sample size $N$ for the case $f=10^{-4}$. Open circles
1083: correspond to fractals with $D_f=3$ and open squares
1084: to fractals with $D_f=2$. Each point is the average
1085: of $50$ random realizations and the bars show the
1086: standard deviations. Dashed horizontal lines show
1087: the expected results for thin slices.}
1088: \label{fig:dcfslice}
1089: \end{figure}
1090:
1091: \clearpage
1092:
1093: \begin{figure}[th]
1094: \epsscale{.85}
1095: \plotone{f3a.eps}
1096: \caption{Positions of the HII regions in the galaxy sample
1097: according to references given in Table~\ref{table2}.
1098: The numbers in parenthesis are the corresponding fractal
1099: dimensions (Table~\ref{table3}).
1100: The axis units are kpc. Dashed line circles indicate the
1101: radius $R_{25}$. Points surrounded by circles are not
1102: taken into account when calculating $D_c$ (see text).}
1103: \label{fig:galaxias}
1104: \end{figure}
1105: \clearpage
1106: {\plotone{f3b.eps}}\\
1107: \centerline{Fig. 3. --- Continued.}
1108: \clearpage
1109: {\plotone{f3c.eps}}\\
1110: \centerline{Fig. 3. --- Continued.}
1111: \clearpage
1112: {\plotone{f3d.eps}}\\
1113: \centerline{Fig. 3. --- Continued.}
1114: \clearpage
1115: {\plotone{f3e.eps}}\\
1116: \centerline{Fig. 3. --- Continued.}
1117: \clearpage
1118: {\plotone{f3f.eps}}\\
1119: \centerline{Fig. 3. --- Continued.}
1120: \clearpage
1121: {\plotone{f3g.eps}}\\
1122: \centerline{Fig. 3. --- Continued.}
1123: \clearpage
1124: {\plotone{f3h.eps}}\\
1125: \centerline{Fig. 3. --- Continued.}
1126: \clearpage
1127: {\plotone{f3i.eps}}\\
1128: \centerline{Fig. 3. --- Continued.}
1129:
1130:
1131: \begin{figure}[th]
1132: \epsscale{.9}
1133: \plotone{f4.eps}
1134: \caption{Calculated dimension $D_c$ as a function of the
1135: number of available data $N(\mathrm{HII})$ for spiral
1136: galaxies (blue solid circles) and irregular galaxies
1137: (red open circles) in the sample. The error bars indicate
1138: the uncertainties obtained from bootstrapping. Superposed
1139: (black squares) are the mean $D_c$ values with their
1140: standard deviations as a function of the mean number of HII
1141: regions for four bins having the same number of galaxies
1142: (Table~\ref{table4}).}
1143: \label{fig:dcnhii}
1144: \end{figure}
1145:
1146: \clearpage
1147:
1148: \begin{figure}[th]
1149: \plotone{f5.eps}
1150: \caption{Calculated dimension $D_c$ as a function of (a)
1151: morphological type $T$, (b) arm class $A$, (c) absolute magnitude
1152: $M_B$, (d) radius $R_{25}$, (e) rotation velocity $V_{rot}$,
1153: and (f) average surface density of star forming regions
1154: $N(\mathrm{HII})/R_{25}^2$. Only galaxies having
1155: $N(\mathrm{HII}) > 200$ have been plotted.}
1156: \label{fig:extrinsecas}
1157: \end{figure}
1158:
1159: \clearpage
1160:
1161: \begin{figure}[th]
1162: \plotone{f6.eps}
1163: \caption{Positions of the HII regions in the galaxy
1164: NGC~6946 for high, medium, and low brightness regions
1165: (see text). The resulting fractal dimensions were
1166: $D_c = 1.64$ (high-brightness), $D_c = 1.82$
1167: (medium-brightness), and $D_c = 1.79$ (low-brightness).}
1168: \label{fig:NGC6946}
1169: \end{figure}
1170:
1171: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1172: % FIN DEL DOCUMENTO
1173: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1174:
1175: \end{document}
1176:
1177:
1178:
1179:
1180: