1: \section{Minkowski Functionals}
2: \label{sec:sec5}
3:
4: \subsection{Minkowski Functionals in Cosmology}
5: It is well known that, in contrast to a Gaussian random field
6: (hereafter, GRF), a fully evolved nonlinear density field cannot be
7: fully quantified in terms of its two-point correlation function; the
8: latter being simply the lowest and first of an infinite hierarchy of
9: correlation functions describing the galaxy distribution. Furthermore,
10: the bias in galaxies viz a viz dark matter can be nontrivial in
11: nature, and can give rise to different clustering properties of
12: galaxies and the underlying mass density field. Consequently the
13: correlation function $\xi_{gg}(r)$ determined either for galaxies or
14: for dark matter does not on its own validate a given cosmological
15: model unless support is provided by other, statistically indepdend
16: measures of clustering. It therefore becomes a challenge to compare
17: the theoretical predictions for LSS with real observational data.
18:
19: If we knew all $n-$point correlation functions, we would have a
20: complete description of the galaxy clustering process. However,
21: estimating $\xi_{gg}(r)$ for a sample of $N$ galaxies requires knowing
22: all pairs of galaxies in this sample, whereas calculating the 3-point
23: function implies taking all triplets. The amount of computation
24: escalates rapidly with $n$ -- the order in the $n-$point correlation
25: function -- especially for a large value of $N$. Besides, it is difficult
26: to extract intuitively useful information from these statistics. The
27: present article focuses on another class of statistics which
28: complement the correlation function approach and which have the
29: advantage of providing a physically appealing interpretation for the
30: evolving density field. These are the {\em Minkowski Functionals}.
31:
32: For an excursion set involving particles embedded in
33: $n-$dimensions, the Minkowski Functionals (hereafter, MFs) are defined
34: on an ($n-1$)-dimensional hypersurface. There are $n+1$ MFs in $n$ dimensions. In this
35: article, we shall be concerned with the 3-dimensional distribution of dark
36: matter and/or galaxies. Hence, MFs will be defined on a 2-dimensional
37: surface and they will reflect the physical properties of this surface
38: (in the given instance, an isodensity contour referring to a
39: supercluster of galaxies or a void). For a given surface the four MFs \cite{blaschke36,
40: min1903, meckwag94} are, respectively
41: \begin{enumerate}
42: \item Volume V,
43: \item Surface area S,
44: \item Integrated Mean Curvature C,
45: \begin{equation}
46: C = {1\over2}\oint\left({1\over R_1} + {1\over R_2}\right)dS.
47: \end{equation}
48: $R_1$ and $R_2$ are the two principal radii of curvature of the
49: surface in a given local neighbourhood and the integral is taken
50: over the entire closed surface.
51: \item Integrated Gaussian Curvature (or Euler characteristic) $\chi$,
52: \begin{equation}
53: \chi = {1\over2\pi}\oint\left({1\over R_1R_2}\right)dS.
54: \end{equation}
55: A related quantity which is more popular in Cosmology is the genus $G$. The
56: genus is related to $\chi$ by
57: \begin{equation}
58: G = 1 - {\chi\over2}.
59: \end{equation}
60: The 3-dimensional genus of an object is a topological invariant. It
61: can be interpreted in terms of the connectivity of the surface. In
62: simple terms it can be viewed as the number of independent cuts
63: which one can make to the surface {\em without} breaking it into two
64: separate pieces. A torus can be cut {\em once} and yet remain in one
65: piece. Hence, its genus is 1. However, a sphere would separate into
66: two pieces if a cut were made. Hence, its genus is 0. Two objects with
67: the same value of genus are topologically similar: one can be
68: obtained by continuously deforming the other. Thus, a sphere and
69: a cube are topologically equivalent.
70: More concretely, genus of an object is the number of handles that
71: the object has, in excess of the number of holes which it encloses
72: (e.g., see \cite{matsub03}). Thus,
73: $$
74: {\rm G = [\# ~of ~handles ~to ~the ~surface] - [\# ~of ~holes ~enclosed ~by ~the ~surface].}
75: $$
76: According to this definition, a sphere has no handle, a torus has
77: a single handle (equivalent to a sphere with one handle), and a
78: pretzel has two handles. Introducing a hole or a bubble {\em inside}
79: the surface {\em reduces} its genus by 1, whereas adding a handle to
80: the surface {\em increases} its genus by 1.
81: \end{enumerate}
82:
83: An attractive feature of MFs is that they depend upon
84: the entire hierarchy of correlation functions \cite{schmalthesis}.
85: Thus, MFs can indirectly help us capture the effect of $n-$point
86: functions. Higher order $n-$point functions gradually become
87: important as a primordial, featureless GRF evolves to develop
88: nonlinear structures. This reflects in the behaviour of
89: MFs for the corresponding density field. MFs are additive in nature,
90: i.e., they can be studied for individual objects (say, clusters or
91: superclusters defined using some prescription) as well as for the
92: entire ensemble of such objects.
93:
94: Of the 4 MFs listed above, the genus G was already known to the
95: cosmology community \cite{dorosh70, gott86, melotopo90}. Tools like
96: percolation analysis which can also be related to MFs, had earlier
97: been introduced in cosmology by \cite{zes82}. There were strong
98: reasons why this was the case.
99:
100: It is well known that a system evolving under gravitational
101: instability becomes progressively more non-Gaussian. (The hypothesis
102: of a primordial spectrum of density perturbations which is distributed
103: in the manner of a Gaussian random field is supported by observations
104: of the cosmic microwave background carried out by WMAP and other
105: experiments.) For CDM-like cosmological models this results in more
106: matter being concentrated in filamentary and pancake-like
107: distributions whose coherence scale evolves with time. As a result
108: the filling fraction at percolation progressively decreases, from
109: % Jatush please insert the reference below and check whether reference
110: % to figure is correct.
111: $16\%$ for the initial Gaussian random field to $\sim 5\%$ for LCDM
112: \cite{ks93,sss97,shyes98,sh-sh-s04}. This implies that more matter is
113: being transferred to regions which occupy a progressively smaller
114: amount of space and are also spatially anisotropic and therefore
115: percolate easily. Figuratively this corresponds to the prominance
116: acquired by superclusters of galaxies over time -- see Figure
117: \ref{fig:web}. This theoretical picture which described the emergence
118: of structure in an initially featureless medium agreed well with what
119: observations were telling us about our universe. The emerging picture
120: of LSS taken from redshift surveys shows us that galaxies are
121: distributed preferentially along filaments and sheets which seem to
122: encircle vast voids. The distribution of galaxies does not occupy much
123: volume, and because of the presence of superclusters, percolates
124: easily. Percolation theory can therefore be used to describe an
125: important quality of the observed galaxy distribution -- its
126: connectivity. At the percolation threshold one can visit the far
127: corners of the cosmic web by following filamentary/sheet-like
128: overdensities. The cosmic web is therefore a connected structure and
129: this is clearly revealed in surveys such as Figure \ref{fig:web}.
130: Connectivity is of course a mathematical notion, and it is possible to
131: precisely quantify connectivity of an object or a system by evaluating
132: its percolation properties and its genus. Doroshkevich(1970) gave
133: analytic formula for the genus of a GRF which showed that at the
134: median density threshold, GRF would exhibit a sponge-like topology
135: \cite{dorosh70}. Simulations of LSS within CDM-cosmogonies provided
136: similar visual impression when smoothed on sufficiently large scales.
137: Like percolation, the genus too has come to be considered as a useful
138: probe of the initial Gaussianity (as well as the final
139: non-Gaussianity) of the cosmic density field; see for instance
140: \cite{melotopo90, colley97,sss97}.
141:
142: %Jatush please place new percolation/genus figure here.
143: %%%
144: \begin{figure}
145: \begin{center}
146: \centering
147: \includegraphics[width=4.2in]{perc_genus.ps}
148: \caption{Percolation (left panels) and genus (right panels) curves
149: are plotted as functions of the density contrast $\delta$ for
150: scale free models of gravitational clustering with the
151: perturbation spectral index $n=-2$. Solid and dashed curves in
152: the left panels correspond to the percolation curve for the
153: largest cluster and void respectively. Vertical solid/dashed
154: lines mark the threshold describing percolation between opposite
155: faces of the cube for clusters/voids respectively. Figure courtesy of
156: \cite{sss97}.}
157: \label{fig:perc_gen}
158: \end{center}
159: \end{figure}
160: %%%
161: An important indicator of non-Gaussianity in a distribution is the
162: {\em percolation curve}. The percolation curve describes the {\it
163: volume fraction} (or filling factor -- FF) in the largest structure
164: (cluster/void) as a function of the density contrast threshold
165: $\delta$ \cite{fn7}. Salient features of the percolation curve are
166: illustrated in Fig. \ref {fig:perc_gen} in which density fields (with
167: an initial power spectrum $P(k)\sim k^{-2}$) evolve from an epoch when
168: the scale of nonlinearity is $k_{\rm NL}= 64k_f, 16k_f, 4k_f$, ($k_f$
169: is the fundamental mode corresponding to the box-size of our N-body
170: simulation). In Fig. \ref {fig:perc_gen} percolation curves for
171: clusters (thick solid lines) and voids (thick dashed lines) are shown
172: as functions of the density contrast $\delta.$ Starting from a high
173: density threshold (small $FF$) we find several isolated clusters
174: (corresponding to peaks of the density field). Lowering the density
175: threshold further results in the {\it merger} of clusters leading to a
176: rapid growth in the percolation curve and to the onset of percolation.
177: A further lowering of the threshold to very small values results in
178: the merger of almost all clusters so that $FF \rightarrow 1$. An
179: identical procedure followed for underdense regions by gradually
180: increasing the density contrast threshold (increasing $FF$) results in
181: a similar functional form for the volume fraction in the largest void.
182: In our samples the largest cluster percolates between opposite faces
183: of the cube when its filling factor is about half the total $FF$. In
184: most cases percolation also coincides with the highest jump in the
185: volume of the largest cluster which was used by \cite{delgh91} as a
186: working definition of the percolation threshold.
187:
188: The solid/dashed vertical line in Fig. \ref {fig:perc_gen}
189: represents the density contrast threshold $\delta_C$ below/above which
190: clusters/voids percolate. We note that {\it both} clusters and voids
191: percolate over a range of overlapping density contrasts --- a feature
192: that is only possible in three or more dimensions --- and corresponds
193: to what is commonly called a {\it sponge} topology for the density
194: distribution.
195:
196: As the simulation evolves
197: $\delta_C$ increases monotonically
198: for $n=-2$, as structures form and align on increasingly larger scales.
199: From Fig. \ref {fig:perc_gen} we see that voids
200: find it easier to percolate as the simulation evolves, as a result
201: the range in densities when both phases percolate initially increases,
202: enhancing the extent of {\it sponge-like topology} in the distribution.
203:
204: The right panel in Fig. \ref {fig:perc_gen} shows the evolution of the
205: genus curve for the same simulation. The top panel corresponding to an
206: early epoch shows the genus curve largely retaining its bell-shaped
207: form which it has for the primordial Gaussian random field $G(\nu) = A
208: (1 - \nu^2) \exp(-\nu^2/2)$ (\cite{hgw86,gwm87}). During later epochs
209: the density distribution grows progressively more non-Gaussian and
210: this is reflected in the change in both shape as well as amplitude of
211: the genus curve in Fig. \ref {fig:perc_gen}.
212:
213: Although a study of the genus gives us a useful handle on the
214: connectivity of a density field, we still lack information about the
215: morphology of LSS. For example, if the domain of information is
216: solely restricted to knowing genus, a filament with one handle would
217: be considered to be identical to a sphere with one handle or a pancake
218: with one handle. Therefore, to gain objective insights into the
219: nature of the supercluster-void network, we must complement genus of
220: an object with quantities which in some way, characterise geometry of
221: that object. This is precisely the role played by the first three
222: MFs, Volume V, Surface Area S and Integrated Mean Curvature C. These
223: MFs change when local deformations are applied to the surface, and
224: hence, these may be useful to glean information about the typical size
225: of an object. The final goal here would be to use this geometric
226: information and give an objective meaning to generic sizes and shapes
227: of superclusters and voids belonging to LSS. In this article we will
228: show that MFs can be utilised to characterise the geometry and
229: topology of the cosmic density field, and eventually can be employed
230: to get information about the morphology of the supercluster-void
231: network in LSS. Because MFs depend on the full hierarchy of
232: correlation functions, we can use them to compare and distinguish two
233: rival cosmological models. Recently, Matsubara (2003) derived
234: semianalytical expressions of MFs for weakly nonlinear cosmic density
235: field by using a second-order perturbative formalism \cite{matsub03}.
236: The results obtained by Matsubara can be used in conjunction with the
237: methods reviewed here to obtain insight into the dynamics of the
238: clustering process of a density field smoothed on sufficiently large
239: scales. The role of bias on large scales can also be directly probed
240: using such methods.
241:
242: \subsection{Morphology of LSS with Shapefinders}
243:
244: Sahni, Sathyaprakash \& Shandarin (1998) showed that the
245: MFs could be used to evaluate both the size as well as the shape of a
246: three-dimensional object such as a supercluster or a void \cite{sss98}.
247: The size is given in terms of
248: three Shapefinders which have dimensions of length. These are defined
249: as ratios of MFs, and are conviniently termed as Length (${\cal L}$),
250: Breadth (${\cal B}$) and Thickness (${\cal T}$).
251: %%%
252: \begin{eqnarray}
253: {\cal T} & = & {3\times V\over S}\\
254: {\cal B} & = & {S\over C} \\
255: {\cal L} & = & {C\over 4\pi G}.
256: \end{eqnarray}
257:
258: The three Shapefinders are defined using spherical normalization, so
259: that for a sphere of radius $R$, ~${\cal T} = {\cal B} = {\cal L} = R$.
260: %Later in Chapter 3, we shall see that for structures defined as
261: %connected overdense regions at or above the percolation threshold, the
262: %three Shapefinders convey the morphological information about
263: %superclusters in a generic cosmic density field remarkably well.
264:
265: The above measures quantify the {\em size} of the object in question. To
266: further quantify the {\em shape} of these objects, two dimensionless
267: Shapefinders have been defined as follows. These are Planarity (${\cal
268: P}$) and Filamentarity (${\cal F}$).
269: %%%
270: \begin{eqnarray}
271: {\cal P} &=& {\cal {B - T\over B + T}}\\
272: {\cal F} &=& {\cal {L - B\over L + B}}.
273: \end{eqnarray}
274: %%%
275:
276: Sahni et al.(1998) showed that (${\cal P,F}$) quantify the shape of
277: both simple as well as complicated objects. Thus, for an oblate ellipsoid its
278: planarity is relatively large and its filamentarity is small. For a
279: prolate spheroid, the reverse is true. The statistics respond
280: monotonically to deformations of these surfaces.
281:
282: If we include the genus then, the triplet of numbers (${\cal P,F,G}$),
283: can define a three-dimensional
284: {\em Shape-Space} which may be used to represent the distribution of
285: shapes of superclusters and their topologies in a given cosmological
286: density field.
287: %Jatush, kindly add correct figure reference below.
288: Note that the inclusion of genus is significant, since most
289: superclusters at moderate density thresholds can have a spongy texture
290: with several loops and/or branches emanating from a central body (see
291: Figure \ref{fig:perclcdm} for an illustration).
292: \subsection{Methods of Estimation}
293: In cosmology the distribution of galaxies in space can frequently be regarded as
294: being an example of a point process.
295: There have been several attempts of evaluating MFs for a given point
296: process. For example, the Boolean grain model \cite{meckwag94,
297: schmalthesis} would decorate the input set of points with spheres of
298: varying radii, and would study the morphology of the structures which
299: contain overlapping spheres. According to another approach
300: \cite{Krofton97} one smooths the distribution of points
301: using a suitable window function, and defines the density field on a
302: grid. MFs can be evaluated for connected structures defined on a grid
303: by using Crofton's formulae or Koenderink Invariants \cite{Krofton97}.
304:
305: A {\em new} approach for determining MFs has been reported in
306: \cite{sh-s-s-sh03}. Rather than working with a grid-approximation for
307: connected structures, these workers go a step forward, and model
308: surfaces for these objects. Minkowski Functionals are evaluated for
309: the isodensity surfaces {\em online}, i.e., while these are being
310: modelled using an elaborate surface triangulation technique. Sheth et
311: al.(2003) describe the algorithms for surface construction and for the
312: evaluation of MFs on these surfaces. Sheth (2004) discusses the
313: method of implementation of these algorithms which has culminated into
314: a robust and accurate software SURFGEN (short for ``SURFace
315: GENerator'')\cite{sheth04}. SURFGEN has been tested for accuracy
316: against Gaussian random fields and against a variety of simply and
317: multiply connected eikonal surfaces. SURFGEN evaluates the geometric
318: MFs to better than 1\% accuracy in most cases, while the genus is
319: evaluated exactly \cite{sh-s-s-sh03}. SURFGEN has since been employed
320: to study geometry, topology and morphology of LSS {\em both} in dark
321: matter and galaxies within a variety of cosmogonies.
322: (SURFGEN could also be relevant in areas of science where the properties of
323: surfaces becomes important such as in medical imaging (tomography) and in
324: condensed matter physics.)
325:
326: In the next section we shall discuss a set of new results obtained by
327: applying MF-based techniques to LSS-data.
328:
329: