1:
2: \documentclass[useAMS,usenatbib]{mn2e}
3: \usepackage{epsfig}
4:
5: \title{Can large-scale structure probe CMB-constrained non-Gaussianity?}
6: \author[X. Kang, P. Norberg and J. Silk]
7: {X. Kang$^1$\thanks{Email: kangx@astro.ox.ac.uk},
8: P. Norberg$^2$ and Joseph Silk$^1$ \\
9: $^1$Department of Astrophysics, University of Oxford, Keble
10: Road, OX1 3RH, UK\\
11: $^2$SUPA\thanks{The Scottish Universities Physics Alliance},
12: Institute for Astronomy, University of Edinburgh, Royal
13: Observatory, Blackford Hill, Edinburgh, EH9 3HJ,UK}
14:
15: \date{Draft version \today}
16:
17: \def \lstar {L$^{*}_{\rm b_{\rm J}}$}
18: \def \hmpc {h$^{-1}$ Mpc}
19: \def \hmpccube {h$^{-3}$ Mpc$^{3}$}
20: \def \kms {s$^{-1}$ km}
21: \def \etal {et~al.}
22: \def \lcdm {$\Lambda$CDM}
23: \def\simlt{\lower.5ex\hbox{$\; \buildrel < \over \sim \;$}}
24: \def\simgt{\lower.5ex\hbox{$\; \buildrel > \over \sim \;$}}
25: \def\simpropto{\lower.2ex\hbox{$\; \buildrel \propto \over \sim \;$}}
26:
27:
28: \begin{document}
29:
30: \maketitle
31:
32: \label{firstpage}
33:
34: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
35:
36: \begin{abstract}
37: The first year Wilkinson Microwave Anisotropy Probe (WMAP) set
38: quantitative constraints on the amplitude of any primordial
39: non-Gaussianity. We run a series of dark matter-only N-body
40: simulations with the WMAP constraints to investigate the effect of the
41: presence of primordial non-Gaussianity on large scale structures. The
42: model parameters can be constrained using the observations of
43: protoclusters associated with Ly-$\alpha$ emitters at high redshift
44: ($2 \leq z \leq 4$), assuming the galaxy velocity bias can be modelled
45: properly. High redshift structure formation potentially provides a
46: more powerful test of possible primordial non-Gaussianity than does
47: the CMB, albeit on smaller scales. Another constraint is given by the
48: local galaxy density probability distribution function (PDF), as
49: mapped by the 2 degree Field Galaxy Redshift Survey (2dFGRS). The PDF
50: of 2dFGRS \lstar\ galaxies is substantially higher than the standard
51: model predictions and requires either a non-negligible bias between
52: galaxy and dark matter on $\sim 12$~\hmpc\ scales or a stronger
53: non-Gaussianity than allowed by the WMAP year one data. The latter
54: interpretation is preferred since second-order bias corrections are
55: negative. With a lower normalisation of the power spectrum
56: fluctuations, $\sigma_8=0.74$, as favoured by the WMAP 3 year data,
57: the discrepancy between the Gaussian model and the data is even
58: larger.
59:
60:
61: \end{abstract}
62:
63: \begin{keywords}
64: cosmology: theory -- large-scale structure of the Universe --
65: galaxies: clusters: general
66: \end{keywords}
67:
68:
69: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
70:
71:
72: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
73:
74: \section[]{Introduction}
75: \label{sec:Intro}
76:
77: The best limits on possible initial non-Gaussianities in the
78: primordial density fluctuation distribution have hitherto been imposed
79: by the CMB (Komatsu \& Spergel 2001). There are many models which can
80: generate such non-Gaussianities, including multiple field inflation or
81: more speculative models such as those assuming non-homogeneous
82: reheating processes or those seeded with topological defects.
83: Moreover such non-Gaussianities may be scale-dependent, and it is
84: therefore important to complement CMB searches with smaller scale
85: probes, on cluster and galaxy scales.
86:
87:
88: Our approach here is phenomenological, following that of Mathis, Diego
89: \& Silk (2004), and focuses on the large-scale galaxy distribution.
90: Tracers of primordial non-Gaussianities include possible large-scale
91: power excesses in the nearby galaxy distribution as mapped by 2dFGRS
92: and SDSS redshift surveys, the early formation of protoclusters up to
93: $z\sim4$ as traced by radio galaxies surrounded by Ly-$\alpha$
94: emitters and Lyman-break galaxies, and the possible excess of angular
95: power in CMB temperature fluctuations seen by CBI and ACBAR on cluster
96: scales (which is most likely due to the thermal SZ effect). With the
97: claimed reduction in normalisation of the fluctuation power spectrum
98: to $\sigma_8=0.74\pm 0.05$ (Spergel \etal\ 2006), such questions take
99: on renewed urgency.
100:
101: Our approach here is to study the implications via dark matter N-body
102: simulations of a simple non-Gaussian model, which modifies the
103: frequency of rare peaks in the primordial density field that defines
104: the locations where massive clusters (or superstructures) are
105: formed. If the probability distribution function of the density field
106: shows an excess (or deficit) of high peaks compared to the Gaussian
107: case, protoclusters of given mass form earlier (or later), and systems
108: above a given mass will be more (or less) abundant at fixed
109: redshift. Sadeh, Rephaeli \& Silk (2006) presented analytical models,
110: whereas here we simulate and evolve a suite of N-body simulations of
111: 300~\hmpc\ on the side.
112:
113:
114: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
115:
116:
117: \section[]{The case for non-Gaussianity}
118: \label{sec:Simus}
119:
120: We consider a simple non-Gaussian model, where the amplitude
121: of non-Gaussianity in the primordial fluctuations is parametrised by a
122: non-linear coupling parameter, $f_{NL}$ (Komatsu \etal\ 2003). We refer
123: to this non-Gaussian model as the $f_{NL}$ model. The Bardeen curvature
124: perturbations $\Phi$ at any given position $x$ can be written as
125: \begin{equation}
126: \Phi(x) = \Phi_{L}(x) + f_{NL}[\Phi_{L}^{2}(x)-<\Phi_{L}^{2}(x)>]
127: \label{eq:phi}
128: \end{equation}
129: where $\Phi_{L}(x)$ are Gaussian linear perturbations with zero
130: mean. This parameterisation is useful for setting quantitative
131: constraints on the amount of non-Gaussianity allowed by the CMB
132: data. Komatsu \etal\ (2003) used first year WMAP data (WMAP-1) to
133: establish limits on $f_{NL}$ and obtained $-58<f_{NL}<134$ at 95\%
134: confidence.
135:
136: We produce the initial conditions for the $f_{NL}$ simulations using
137: the conventional grid method, but when calculating the displacement
138: using the Zel'dovich approximation, we use the perturbed newtonian
139: potential, which is $-1$ times the curvature perturbation
140: given by Equation~\ref{eq:phi}. We then run a series of dark matter
141: N-body simulations using the publicly available code GADGET (Springel,
142: Yoshida \& White 2001). We consider three values for $f_{NL}$
143: (-58,0,134) consistent with WMAP1 constraints, where $f_{NL}=0$
144: corresponds to the standard Gaussian model. For these simulations, we
145: impose the initial power spectrum to be given by an adiabatic CDM
146: transfer function with spectral slope $n_{s}=1$. We use the `old`
147: concordance cosmological model, {\it i.e.} $\Omega_{m}=0.3,
148: \Omega_{\Lambda}=0.7$, $h=0.7$ and $\sigma_{8}=0.9$. Recent results
149: from WMAP-3 prefer a lower value for $\sigma_{8}$ (Spergel \etal\
150: 2006). We therefore run another series of simulations using the WMAP-3
151: cosmological parameters (ie. $\Omega_{m}=0.238, \Omega_{\Lambda}
152: =0.762$, $h=0.73$, $\sigma_{8}=0.74$), both for $f_{NL}=0$ and
153: $f_{NL}=134$. All N-body simulations are started at redshift
154: $z_{init}=70$ and we follow $128^3$ particles in a cube of 300~\hmpc\
155: aside, implying a particle mass of $\rm \sim 10^{12}h^{-1}M_{\odot}$,
156: which, for the present study, gives high enough resolution as shown in
157: Section~\ref{sec:Res}. For each of the above models, we present the
158: mean results from ten statistically equivalent
159: realizations. Hereinafter, we conservatively discuss the results for
160: the cosmological model with $\sigma_{8}=0.9$ unless otherwise stated.
161:
162:
163: \section{Results}
164: \label{sec:Res}
165: \subsection{Halo mass function in non-Gaussian models}
166:
167: As with any simulation run, it is essential to check if our analysis
168: is affected by the chosen mass resolution. In Fig.~\ref{fig:compare},
169: we show a comparison between our simulations and results from either
170: other high-resolution N-body simulations or analytical models. The
171: upper panel compares the halo mass function of our Gaussian
172: simulations with the fitting formula of Jenkins et al. (2001), which
173: is based on a series of high-resolution N-body simulations with
174: similar cosmology. Our estimate of the halo mass function is in good
175: enough agreement with the Jenkins et al. (2001) fit. In the lower
176: panel, we show the probability distribution function (PDF) of the
177: density contrast smoothed with a spherical top-hat filter of radius
178: 12~\hmpc. Kayo et al. (2001) have shown that the PDF can be well
179: described by a lognormal distribution. Here we show the mean PDF from
180: our 10 Gaussian simulations as the solid line, and the errors
181: correspond to the estimated variance. The dashed line shows the
182: analytic relation given by Kayo et al. (2001), which is shown to be
183: accurate at $\delta \geq -0.5$. It can be seen that our simulation
184: matches well the analytic prediction. The main point of the plot is
185: to show that our mass resolution is good enough to properly resolve
186: the over-dense tail of the probability distribution function, for a
187: typical smoothing scale used later on in our analysis.
188:
189:
190: \begin{figure}
191: \begin{minipage}{8cm}
192: \epsfig{file=fig1.eps,width=8cm}
193: \caption{Upper panel: mass function of dark matter halos at z=0 from
194: simulations with Gaussian initial conditions. The solid line shows the mean
195: mass function from 10 simulations with Gaussian initial conditions,
196: and the dashed line shows the Jenkins et al. (2001) halo mass function.
197: Lower panel: the probability distribution function of the density contrast.
198: Again, the solid line shows the mean from our 10 Gaussian simulations and the
199: dashed line is the corresponding analytic prediction of Kayo et al. (2001).}
200: \label{fig:compare}
201: \end{minipage}
202: \end{figure}
203:
204:
205: Our main motivation is to see whether the type of non-Gaussian models
206: considered here, ie. f$_{NL}$ models, can produce more massive
207: structures than does the standard Gaussian model. In Fig.~\ref{fig:MF}
208: we present the evolution of the dark matter halo mass function as a
209: function of redshift. The upper panel shows the results with WMAP-1
210: constrained f$_{NL}$ model parameters. A distinctive feature is that a
211: positive f$_{NL}$, which gives a more strongly disturbed gravitational
212: field, results in the formation of increased numbers of virialized
213: structures. This effect is also much stronger at higher redshifts, as
214: expected from the theoretical models of Matarrese et al. (2000). In
215: the lower panel of Fig.~\ref{fig:MF}, we show the ratio of halo mass
216: functions of non-Gaussian, $dn/dM(NG)$, and Gaussian models, $dn/dM(G)$,
217: by the thick lines with error bars from our ten statistically
218: equivalent realizations. Even though the simulation predicts the ratio
219: of halo mass function to increase with halo mass and with increasing
220: redshift, there is still a large discrepancy with the analytical model
221: expectations of Matarrese et al. (2001). Our simulations produce a
222: much larger number density of dark matter halos for non-Gaussian
223: models than expected by Press-Schechter-based model predictions. The
224: main difference between the Press-Schechter halo mass functions and
225: the N-body mass functions can be attributed to a different treatment
226: of the dynamics of halo formation (cf. Komatsu et al. 2003). It is
227: valid to ask whether primordial non-Gaussianity also changes the
228: dynamics of halo formation. This issue is unfortunately beyond our
229: current study, as following the dynamics of halo formation requires
230: much higher-resolution simulations than the ones we have run, together
231: with more regularly spaced simulation outputs. We therefore delay this
232: aspect of non-Gaussianity to future work.
233:
234:
235: \begin{figure}
236: \begin{minipage}{8cm}
237: \epsfig{file=fig2.eps,width=8cm}
238: \caption{Upper panel: dark matter halo mass functions with $f_{NL}$
239: parameters 134 (red dashed), 0 (black solid), -58 (blue
240: dotted). Lower panel: the ratio of halo mass functions of
241: non-Gaussian and Gaussian models. Thick lines with error bars are
242: for our suite of DM simulations and thin lines are the analytic
243: model predictions of Matarrese et al. (2000), for the same set of
244: $f_{NL}$ parameters (same line styles and colours as in the upper
245: panel).}
246: \label{fig:MF}
247: \end{minipage}
248: \end{figure}
249:
250:
251: \subsection{Protoclusters at z $\sim$ 2-4}
252:
253: As previously shown, there are significant differences between the
254: Gaussian model and the $f_{NL}$ models at high redshift, in particular
255: for large positive $f_{NL}$ values, which results in more structure
256: being formed. Miley \etal\ (2004) have reported a protocluster at
257: $z=4.1$ around a Ly-$\alpha$ emitter galaxy. They found 21
258: spectroscopically confirmed galaxies around the Ly-$\alpha$ emitter
259: galaxy, and the 1-dimensional velocity dispersion among these galaxies
260: is about 325~\kms. Kurk \etal\ (2004) also observed a protocluster at
261: $z=2.16$ with a velocity dispersion of $\sim 360$ s$^{-1}$ km over 3 to 5
262: Mpc, corresponding to co-moving sphere radii of 5 to
263: 9~\hmpc. Following Robinson \& Baker (2000) and Mathis, Diego \& Silk
264: (2004), we sample such structures in our simulations by throwing down
265: a large number of spheres with radii of 7~\hmpc\ and calculate their
266: 1-D dark matter velocity dispersion.
267:
268: \begin{figure}
269: \begin{minipage}{8cm}
270: \epsfig{file=fig3.eps,width=8cm}
271: \caption{Cumulative probability distribution function of the
272: one-dimensional velocity dispersion measured within spheres of radii
273: 7~\hmpc. Upper (lower) panel show the $z=4$ ($z=2$) results. Vertical
274: dashed lines indicate the observed velocity dispersion by Miley \etal\
275: (2004) and Kurk \etal\ (2004).}
276: \label{fig:PDV}
277: \end{minipage}
278: \end{figure}
279:
280: In Fig.~\ref{fig:PDV} we show the cumulative probability distribution
281: function of the measured 1-D dark matter velocity dispersion. In the
282: top panel, we compare our findings with the measurement of Miley
283: \etal\ (2004) at $z\sim4$ . In the bottom panel, we plot the
284: equivalent curves at $z=2$ and compare with the velocity dispersion
285: measurement of Kurk \etal\ (2004). Contrary to Mathis \etal\ (2004),
286: we find that our Gaussian simulation with $\sigma_{8}=0.9$ can explain
287: the existence of structures with velocity dispersions larger than
288: 320~\kms, such as the protocluster of Miley \etal\ (2004). Although
289: the simulation of Mathis \etal\ (2004) used the same cosmological
290: parameters as our simulation, they adopt too small a simulation volume
291: (box of 100~\hmpc\ aside), with the consequence that the cut-off in
292: the input power spectrum on large scales depresses the velocity
293: dispersion in their simulation.
294:
295: Fig.~\ref{fig:PDV} shows that for all simulations with $\sigma_8=0.9$,
296: it is possible to {\it observe} a $z\sim4$ protocluster with 1-D
297: velocity dispersion of about 320~\kms. On the other hand in a low
298: $\sigma_{8}$ universe (like one with $\sigma_{8}=0.74$, as favoured by
299: WMAP-3), the chance of {\it observing} such protoclusters is very
300: unlikely, unless there is a strongly disturbed primordial non-Gaussian
301: field (such as one with $f_{NL}=134$). We note that these are not in
302: any case very strong constraints, as we have ignored any velocity bias
303: between the galaxies and the dark matter, which to some extent could
304: solve the apparent discrepancy. Therefore we have not attempted to
305: quantify the predicted number density of such objects at $z\sim4$. Only
306: detection of a large number of protoclusters of similar size or bigger
307: at $z\sim4$ could actually put real constraints on this simple model,
308: together with some theoretical constraints on the strength of velocity
309: bias, as would be measurable in higher resolution N-body simulations.
310: The lower panel of Fig.~\ref{fig:PDV} shows that, at $z\sim2$, it is
311: rather easy to find, in all of the simulations considered, regions
312: with velocity dispersion larger than 400~\kms. This highlights the
313: fact that only very high redshift protoclusters are likely to set
314: constraints on the cosmological parameters using the 1-D velocity
315: dispersion.
316:
317: \begin{figure}
318: \begin{minipage}{8cm}
319: \epsfig{file=fig4.eps,width=8cm}
320: \caption{Top panel: Probability distribution function of the
321: mass (galaxy) density contrast within spheres of 12~\hmpc\ radii:
322: four sets of dark matter PDFs from Gaussian and non-Gaussian models
323: (see labels in panel) and, in magenta, the PDF of 2dFGRS \lstar\
324: galaxies. Lower panel: the ratio of the PDFs shown in the top panel
325: to the Gaussian PDF. This explicitly shows that the PDF of
326: 2dFGRS \lstar\ galaxies is distinctively different to any of the DM
327: models presented and highlights particularly well that a lower
328: $\sigma_8$ value increases the discrepancy between data and models
329: even further.}
330: \label{fig:PDF_z0}
331: \end{minipage}
332: \end{figure}
333:
334: \subsection{Large scale structures in the 2dFGRS}
335: \label{sec:lss}
336:
337: A clear limitation of the preceding results is the sparseness and
338: quality of the data that we are comparing to. Hence we consider in
339: this section constraints from local large-scale structure results.
340: We estimate the galaxy probability distribution
341: function (PDF) using a 2dFGRS \lstar\ volume-limited
342: sample (Norberg \etal\ 2002; Baugh \etal\ 2004), comparable in volume
343: to a simulation cube of $\sim$~200~\hmpc\ on the side. The PDF is
344: measured by Fast-Fourier-Transform (FFT) smoothing of the galaxy
345: distribution with a spherical top-hat filter of radius $R.$ For the
346: smoothing scales considered (4 to 16~\hmpc), we recover the results
347: of Croton \etal\ (2004), who used a count-in-cells (CiC) method to
348: measure the galaxy PDF. The FFT-based method, once we accurately
349: account for edge effects (particularly important when
350: smoothing over large scales), is several orders of magnitude
351: faster than standard CiC.
352:
353: In the top panel of Fig.~\ref{fig:PDF_z0}, we show the 2dFGRS \lstar\
354: galaxy PDF, smoothed over 12~\hmpc, together with the dark matter PDF
355: of the $f_{NL}$models specified in the caption. In the bottom panel,
356: we show the ratio between the PDFs presented in the top panel and the
357: PDF of the Gaussian simulation. Clearly the 2dFGRS galaxy PDF lies
358: well above all model predictions, and in particular it is very
359: discrepant with models using a low $\sigma_8$ value, as currently
360: favoured by WMAP-3 data.
361:
362: \begin{figure}
363: \begin{minipage}{8cm}
364: \epsfig{file=fig5.eps,width=8cm}
365: \caption{Top panel: Probability distribution function of the
366: mass (galaxy) density contrast within spheres of 12~\hmpc\ radii: in
367: black, the PDF of a $f_{NL}=500$ model; in blue, a biased \lcdm\
368: PDF; in magenta, the PDF of 2dFGRS \lstar\ galaxies. Both models
369: assume $\sigma_8=0.90$ cosmology. Lower panel: the ratio of the PDFs
370: shown in the top panel to the Gaussian \lcdm\ PDF. For a high
371: $\sigma_8$ value, both models seem able to reproduce the tail of
372: the data PDF rather well.}
373: \label{fig:PDF_galaxy}
374: \end{minipage}
375: \end{figure}
376:
377: Two possible approaches may help reconcile the models with the
378: observations. Firstly, in Fig.~\ref{fig:PDF_z0}, we compare the galaxy
379: PDF with the mass PDF, ignoring any bias between them. We consider a
380: simple linear bias model (b $\simeq 1.2$) between galaxy and dark
381: matter on the weakly non-linear scales of 12~\hmpc\ in the Gaussian
382: model. Secondly, we consider a more extreme positive non-Gaussian
383: model with $f_{NL}=500$, which predicts a significantly higher
384: probability in the tail of the PDF. The constraints from WMAP on
385: $f_{NL}$ are only for very large scales, and the non-Gaussianity may
386: be scale-dependent. In Fig.~\ref{fig:PDF_galaxy}, we compare these two
387: alternative models with the 2dFGRS \lstar\ PDF.
388:
389:
390:
391: In the following discussion, we assume that the data extracted from
392: the 2dFGRS is a fair representation of the Universe, {\it i.e.} we
393: require our model to easily reproduce the observations, something one
394: potentially expects from a volume close to $8\times
395: 10^{6}$~\hmpccube. Clearly, if $\sigma_8 = 0.9$, then both a standard
396: $\Lambda$CDM with relatively strong linear bias on non-linear scales
397: ($b \simeq 1.2$) or a strong non-Gaussian model ($f_{NL}=500$) without
398: bias reproduce the 2dFGRS PDF rather well. In the case of a low
399: $\sigma_8$ Universe, the bias needed for a pure Gaussian model would
400: be uncomfortably large ($b \simeq 1.46$), especially when accounting
401: for the most recent determination of galaxy bias on weakly non-linear
402: scales using the same 2dFGRS samples (Gazta$\tilde{\rm n}$aga \etal\
403: 2005). Alternatively, a strong positive non-Gaussianity ($f_{NL}\sim
404: 500$) coupled with a moderate galaxy bias could resolve the
405: discrepancy between data and models. In fact, a linear bias model on
406: the scales of 12~\hmpc\ may not be realistic. Gazta$\tilde{\rm n}$aga
407: et al. (2005) have recently measured the galaxy bias on weakly
408: non-linear scales (ie. $\sim$~10 to $\sim$~30~\hmpc) using the 2dFGRS
409: samples and found that the bias is non-linear in over-dense regions
410: ($\delta \gg 1$). However the second order corrections are found to
411: be negative, lowering the galaxy density, and hence favouring the
412: non-gaussian interpretation.
413:
414:
415: Two issues are worth mentioning in this analysis: we focus in
416: particular on the \lstar\ volume-limited sample, because of the
417: presence of two very large coherent superstructures in each 2dFGRS
418: regions
419: \footnote{The Northern one is part of the SDSS great wall}, which
420: strongly influences all higher-order statistics (Baugh \etal\ 2004;
421: Croton \etal\ 2004; Gazta$\tilde{\rm n}$aga \etal\ 2005; Nichol \etal\
422: 2006), and even to some extent the 2-point correlation function. The
423: other reason is that \lstar\ galaxies are expected to only be very
424: mildly biased tracers of the underlying dark matter (eg. Verde \etal\
425: 2002; Gazta$\tilde{\rm n}$aga \etal\ 2005). Results from
426: a larger volume-limited sample such as the $\sim2.5$~\lstar\
427: sample are more difficult to interpret, mostly because the galaxy
428: density is much lower, making the tail of the PDFs harder to
429: measure.
430:
431: \section[]{Conclusions}
432: \label{sec:CCL}
433:
434: In this paper, we have studied a simple non-Gaussian model in which
435: the initial potential has a small perturbation relative to the
436: Gaussian random field and can be described by a simple non-linear
437: coupling parameter (Komatsu \etal\ 2003). We produce a series of
438: N-body simulations to model structure formation with both Gaussian and
439: non-Gaussian initial conditions. We estimate the predicted halo mass
440: function as a function of redshift, and make comparison with
441: observations for the PDF of the 1-D DM velocity dispersion as well as
442: for the PDF of the DM overdensity.
443:
444: We find that, compared to the Gaussian model, a non-Gaussian model
445: with deeper potential wells (ie. positive $f_{NL}$) produces larger
446: structures, whereas models with negative $f_{NL}$ values do the
447: opposite. The differences are very pronounced at high redshift, but
448: are less evident at $z \sim 0$. The presence of large coherent
449: structures at high redshift is a better indicator of early structure
450: formation than similar objects at low redshift. Under the assumption
451: of no velocity bias between galaxy and mass, our results show that it
452: would be difficult to observe the $z\sim4.1$ protocluster of Miley
453: \etal\ (2004) in a $\sigma_{8}=0.74$ Universe with Gaussian initial
454: conditions, but becomes much easier with a strong non-Gaussian field
455: (eg. $f_{NL}=134$). Protoclusters at high redshift can put strong
456: constraints on non-Gaussian models, once the effects of bias and
457: sample selections are properly addressed. High redshift structure
458: formation potentially provides a more powerful test of possible
459: primordial non-Gaussianity than does the CMB, albeit on smaller
460: scales.
461:
462: This of course assumes that we live in a typical region of the
463: universe. However the large-scale coherent structures observed in the
464: 2DFGRS (Baugh et al. 2005) that dominate the higher order statistics,
465: and also seen in the SDSS, might indicate that a larger volume of the
466: universe must be sampled to robustly test non-Gaussianity.
467:
468:
469: A detailed comparison between the 2dFGRS \lstar\ galaxy PDF and the
470: PDFs from a series of dark matter models show how difficult it is to
471: reproduce the observed high tail in the galaxy PDF, when smoothed on
472: 12~\hmpc\ scales. In a $\sigma_8=0.90$ Universe, we argued that only a
473: Gaussian \lcdm\ model with strong linear bias between galaxies and
474: mass ($b \sim 1.2$) or a very strongly non-Gaussian model, with
475: $f_{NL} \sim 500$, can reconcile the results mapped by the 2dFGRS. The
476: bias constraints obtained by Gazta$\tilde{\rm n}$aga \etal\ (2005)
477: using the same 2dFGRS samples favour the latter option. However the
478: complication of galaxy bias on weakly non-linear scales in over-dense
479: regions ($\delta \gg 1$) makes it difficult to constrain the model
480: parameters based on a simple linear bias model. Nevertheless, in a low
481: $\sigma_8$ Universe as favoured by recent WMAP-3 results, the case for
482: primordial non-Gaussianity is strengthened. Further tests, with
483: realistic galaxy formation recipes, are needed to properly address
484: these issues. Our conclusions complement those in a recent
485: theoretical paper by Hikage, Komatsu \& Matsubara (2006), namely that
486: non-Gaussianity can be well constrained by high redshift galaxy
487: surveys and that non-linear galaxy bias makes it difficult to detect
488: primordial non-Gaussianity at low redshift.
489:
490: \section{Acknowledgements}
491: The authors would like to thank the referee for useful comments.
492: XK acknowledges a Royal Society Chinese Fellowship for financial
493: support. PN acknowledges receipt of a PPARC post-doctoral fellowship
494: held at the IfA. XK and PN are grateful for the hospitality of the
495: Physics Department of the University of Oxford during respective
496: stays and visits.
497:
498: \bsp \bibliographystyle{mnras}
499:
500:
501: \begin{thebibliography}{44}
502: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi
503: \bibitem[Baugh et al. 2004]{baugh04}
504: Baugh, C. M., et al. 2004, MNRAS, 351L, 44
505: \bibitem[Croton et al. 2004]{croton04}
506: Croton, D. J., et al. 2004, MNRAS, 352, 1232
507: \bibitem[Gaztanaga et al. 2005]{gaztanaga05}
508: Gazta$\tilde{\rm n}$ga, E., Norberg, P., Baugh, C. M., \& Croton, D. J., 2005, MNRAS, 364, 620
509: \bibitem[Hikage et al. 2006]{hikage06}
510: Hikage, C., Komatsu, E., \& Matsubara, T., 2006, ApJ submitted, astro-ph/0607284
511: \bibitem[Jenkins et al. 2001]{Jenkins01}
512: Jenkins, A., et at., 2001, MNRAS, 321, 372
513: \bibitem[Kayo et al. 2001]{Kayo01}
514: Kayo, I., Taruya, A., \& Suto, Y., 2001, ApJ, 561, 22
515: \bibitem[Komatsu and Spergel 2001]{kom01}
516: Komatsu, E., \& Spergel, D. N. 2001, Phys.Rev.D,63,063002
517: \bibitem[Komatsu et al. 2003]{kom03}
518: Komatsu, E., et al. 2003, ApJS, 148, 119
519: \bibitem[Kurk et al. 2004]{kurk04}
520: Kurk, J. D., Pentericci, L., Overzier, R. A., R$\ddot{\rm o}$ttgering, H. J. A., \& Miley, R. K., 2004, A\&A, 428, 817
521: \bibitem[Matarrese et al. 2000]{matar00}
522: Matarrese, S., Verde, L., Jimenez, R., 2000, ApJ 541, 10
523: \bibitem[Mathis, Diego \& Silk(2004)]{mat04}
524: Mathis, H., Diego, J. and Silk, J. 2004, MNRAS, 353, 681.
525: \bibitem[MIley et al. 2004]{miley04}
526: Miley, G. K., et al. 2004, Nature, 427, 47
527: \bibitem[Nichol et al. 2006]{nichol06}
528: Nichol, R. C., et al. 2006, MNRAS, 368, 1507
529: \bibitem[Norberg et al. 2002]{peder02}
530: Norberg, P., et al. 2002, MNRAS, 336, 907
531: \bibitem[Robinson et al. 2004]{robinson04}
532: Robinson, J., \& Baker, J. E., 2000, MNRAS, 311, 781
533: \bibitem[Sadeh, Rephaeli and Silk(2006)]{sad06}
534: Sadeh, S., Rephaeli, Y. \& Silk, J. 2006, MNRAS, 368, 1583
535: \bibitem[Spergel et al (2006)]{spe06}
536: Spergel, D. N., et al. 2006, astro-ph/0603449
537: \bibitem[Springel et al. 2001]{vilker01}
538: Springel, V., Yoshida, N., \& White, S. D. M., 2001, New Astronomy, 6, 79
539: \bibitem[Verde et al. 2002]{verde02}
540: Verde, L., et al. 2002, MNRAS, 335, 432
541: \end{thebibliography}
542:
543: \end{document}
544:
545:
546: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
547: