-
PDF
- Split View
-
Views
-
Cite
Cite
Aseem Paranjape, Ravi K. Sheth, Vincent Desjacques, Excursion set peaks: a self-consistent model of dark halo abundances and clustering, Monthly Notices of the Royal Astronomical Society, Volume 431, Issue 2, 11 May 2013, Pages 1503–1512, https://doi.org/10.1093/mnras/stt267
- Share Icon Share
Abstract
We describe how to extend the excursion set peaks framework so that its predictions of dark halo abundances and clustering can be compared directly with simulations. These extensions include: a halo mass definition which uses the TopHat filter in real space; the mean dependence of the critical density for collapse δc on halo mass m; and the scatter around this mean value. All three of these are motivated by the physics of triaxial rather than spherical collapse. A comparison of the resulting mass function with N-body results shows that, if one uses δc(m) and its scatter as determined from simulations, then all three are necessary ingredients for obtaining ∼10 per cent accuracy. For example, assuming a constant value of δc with no scatter, as motivated by the physics of spherical collapse, leads to many more massive haloes than seen in simulations. The same model is also in excellent agreement with N-body results for the linear halo bias, especially at the high mass end where the traditional peak-background split argument applied to the mass function fit is known to underpredict the measured bias by ∼10 per cent. In the excursion set language, our model is about walks centred on special positions (peaks) in the initial conditions – we discuss what it implies for the usual calculation in which all walks contribute to the statistics.
1 INTRODUCTION
The evolution of the abundance of virialized clusters is a powerful probe of the combined effects of the nature of the initial conditions (Gaussian or not?), the nature of gravity (general relativity or not?) and the expansion history of the Universe (cosmological constant or more complex?). This has motivated studies to understand and parametrize this dependence, so that data sets can more precisely constrain the physical models.
Working with a special subset of positions is motivated by measurements extracted from simulations showing that the excursion set prediction for the mass of the object in which a randomly chosen particle ends up is almost always incorrect (White 1996), whereas the prediction for the subset of particles which are at the centre of mass of the protohalo is almost always quite accurate (Sheth, Mo & Tormen 2001). Moreover, for this special subset of (centre-of-mass) particles, the predicted mass depends on whether one assumes a spherical or a triaxial collapse – and the prediction is more accurate if one uses the more physically appropriate (though by no means perfect!) triaxial collapse model. In other words, the physics and the statistics are simpler, and tell a consistent story only if one works directly with the special subset of positions around which collapse occurs. More recent work has shown that protohalo positions do indeed coincide with peaks in the initial field (Ludlow & Porciani 2011), and their initial velocities are biased with respect to that of the dark matter (Elia, Ludlow & Porciani 2012), as is expected for peaks (Desjacques & Sheth 2010). This provides additional motivation for the ESP approach.
Therefore, the main goals of this work are to include TopHat smoothing in the ESP formalism and incorporate the effects of more complicated collapse models which are seen in simulations. All this is the subject of Section 2. Some testable predictions which elucidate the relation between the ESP approach and the usual excursion set analysis of all (rather than special) positions in space are discussed in Section 3. Section 4 presents implications for the spatial distribution of these objects – the ESP predictions for Lagrangian halo bias – which can be used to further test the model. A final section summarizes our results. We will show numerical results at redshift z = 0 for two Λ cold dark matter (ΛCDM) cosmologies: one with |$(\Omega _{\rm m},\Omega _\Lambda ,\sigma _8,n_s,h) = (0.3,0.7,0.9,1.0,0.7)$| which we denote Wilkinson Microwave Anisotropy Probe 1 (WMAP1) and the other with (0.25, 0.75, 0.8, 1.0, 0.7) which we denote WMAP3.
2 THE UNCONDITIONAL MASS FUNCTION
2.1 Notation
We will designate Gaussian filtered quantities with the subscript G. By contrast, TopHat filtered objects will have no subscript. The Fourier transforms of the TopHat and Gaussian filters are W(kR) = (3/kR)j1(kR) and |$W_{\rm G}(kR_{\rm G}) = {\rm e}^{-k^2R_{\rm G}^2/2}$|, respectively, with j1(y) being a spherical Bessel function.
We will frequently need the following integrals over the linearly extrapolated dimensionless matter power spectrum Δ2(k) ≡ k3P(k)/2π2.
2.2 Matching filter scales
The technical problem which we address in this subsection is how to ensure that the peaks in the Gaussian filtered density field δG will have an overdensity δTH = δc when smoothed with a TopHat, since essentially all measurements in simulations use TopHats only. In practice, we need a mapping between the scales RTH and RG of the two filters. All previous work accomplishes this either by matching the volumes: (4π/3) R3TH = (2π)3/2 RG3, or by matching the variances: 〈δ2TH〉 = 〈δ2G〉. In this second case, the relation between RTH and RG depends on the shape of the power spectrum. But neither of these conditions guarantee that a peak identified on scale RG satisfies δTH = δc.
For this reason, we construct the mapping between RTH and RG by finding that RG (at a given RTH) for which 〈 δG|δTH 〉 = δTH. This is equivalent to 〈δGδTH〉 = 〈δ2TH〉. This definition circumvents the technical complication arising from the variance of the Laplacian of δTH (which involves a divergent integral). For the ΛCDM P(k) we consider in this paper, RG ≈ 0.46RTH with a mild mass dependence which we account for, in the range we explore, 3 × 1010 < m/(h− 1 M⊙) < 3 × 1015.
Our choice of matching between RTH and RG, namely 〈 δG|δ 〉 = δ, means that p(δG|δ) is a Gaussian with mean δ and variance 〈δ2G〉 − s. However, 〈δ2G〉 ≈ s to better than 5 per cent and, thus, p(δG|δ) ≈ δD(δG − δ) (since the cross-correlation between δG and δ is very close to unity). This significantly simplifies our ESP analysis for the following reason. In principle, the ESP framework also requires us to keep track of the variable |$\tilde{x}\equiv \delta ^{\prime }/\sqrt{\langle\delta ^{\prime 2}\rangle }$|, where δ′ = dδ/ds. This scalar (isotropic) variable correlates with μ and x, but not with the other (anisotropic) variables used by Bardeen et al. (1986). Therefore, one should in principle work with the three-dimensional Gaussian vector |$(\mu ,x,\tilde{x})$|. However, because p(δG|δ) ≈ δD(δG − δ), it turns out that setting |$\tilde{x}=x$| yields an excellent approximation (note that this relation is exact for the Gaussian filter). We have indeed checked that our final answers for dn/dln m only change by a few per cent if we switch between the approximate and exact treatments. In the following, we will therefore always set |$\tilde{x}=x$| and only work with the two-dimensional Gaussian vector (μ, x).
Even though we smooth the density field with a TopHat filter, there is no compelling reason to use such a filter except to make a connection with the spherical collapse approximation. Numerical simulations indeed suggest that, while the Lagrangian volume occupied by the protohaloes is rather compact, it is more diffuse than a TopHat window (Porciani, Dekel & Hoffman 2002; Dalal et al. 2008; Despali, Tormen & Sheth 2013). However, since a deep understanding of halo collapse is still lacking, we will stick to the TopHat (and Gaussian) filters for simplicity.
2.3 Excursion set peaks with a constant barrier
Fig. 1 compares the ESP result for a constant barrier B(s) = δc = 1.686 (solid red) with the fit to N-body simulations from Tinker et al. (2008) (dashed green). For the latter we used parameters from table 2 of Tinker et al. (2008) appropriate for haloes identified using the spherical overdensity (SO) definition at 200 times the mean density of the universe. We used cosmological parameters for the WMAP1 cosmology mentioned in Section 1, for which the Tinker et al. (2008) function gives a good fit (cf. their table 2). Later, we will also show results for the WMAP3 cosmology which was also explored by Tinker et al. (2008) – changing cosmology does not affect our results significantly. For comparison, we also show the functional form of Sheth & Tormen (1999) (short dashed blue) with their (q, p) = (0.7, 0.26) which gives a good fit to the MICE simulations (Crocce et al. 2010).
![ESP mass function for the mixed filtering approach discussed in the text, using a constant barrier (solid red). This is compared with the fit to N-body simulations given by Tinker et al. (2008), appropriate for SO haloes that enclose 200 times the mean density of the universe (dashed green). For comparison, we also show the Sheth & Tormen (1999) mass function (short dashed blue) with their (q, p) = (0.7, 0.26), which gives a good fit to the MICE simulations (Crocce et al. 2010). The ESP+constant barrier mass function overpredicts the abundance at all scales of interest. All curves used the WMAP1 cosmology – we have checked that changing the cosmology does not significantly alter the comparison.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/431/2/10.1093/mnras/stt267/2/m_stt267fig1.jpeg?Expires=1723729622&Signature=jkhn3nYTtWN1l08WJk17pywxyVuv0EgkpYUBS4qimYZyMSEbD6E~UpdMLGeiGoZrBrTP04-GcW9cIiuWBKjcr49KPgYBVXCIZt7qbWVlpeDIJc3WxsZEK58vhSkdSij4tw9upRWWgtigU7xz3qhLAy~vf07q09IxByq49GF8OF96wDGIlHg~8JwES8IBbsApDQORfoMj8ls-pmXoX5FLwNgaxMwWiaBnWdvuNBIi5vA7hvrMSui9VPELH7AQSu~iZQ2LMs8bNfr7biK5B4gVAwaz6AjZgajGWQOEvfKBaHvXLGc5rWP2gYSxUiiU2u1CckK54YjVrW6mhQH1qP7c-Q__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
ESP mass function for the mixed filtering approach discussed in the text, using a constant barrier (solid red). This is compared with the fit to N-body simulations given by Tinker et al. (2008), appropriate for SO haloes that enclose 200 times the mean density of the universe (dashed green). For comparison, we also show the Sheth & Tormen (1999) mass function (short dashed blue) with their (q, p) = (0.7, 0.26), which gives a good fit to the MICE simulations (Crocce et al. 2010). The ESP+constant barrier mass function overpredicts the abundance at all scales of interest. All curves used the WMAP1 cosmology – we have checked that changing the cosmology does not significantly alter the comparison.
Clearly, equation (9) predicts far too many haloes at all interesting scales, including the high mass end (ν2 = 10 corresponds to m = 1.3 × 1015h− 1 M⊙ for this cosmology).
2.4 Beyond the spherical-collapse constant barrier
Before explaining how this can be done, we note that, although equation (10) is a reasonable approximation to the mass dependence of the critical density required for collapse (Sheth et al. 2001), there is substantial scatter around this value (Dalal et al. 2008; Sheth 2009). This scatter has been quantified by Robertson et al. (2009)1 who found that, over the range log10(ν2) ∈ (−0.3, 0.9), the value of B was Lognormally distributed around a mean value of ≃ 1.51 + 0.49σ0 with variance (0.09σ20) (cf. their table 1 and fig. 3). The physical origin of this scatter is unclear. Triaxial collapse predicts some scatter (Sheth et al. 2001; Despali et al. 2013), which is related to the statistics of the tidal field, but the measured scatter is considerably larger. In any case, the physical origin of this scatter is unimportant for what follows – though we believe that this is an issue which deserves further investigation.
2.5 Excursion set peaks with a square-root barrier and scatter
For ESP, the square-root barrier (11) with β > 0 dramatically lowers the halo counts at large masses (fig. 3 in Paranjape & Sheth 2012). While this goes in the right direction, introducing scatter in the value of β will counteract this decrease. This is because a scatter in β induces a scatter in the predicted mass of the halo. The steepness of the mass function implies that the scatter preferentially moves small mass objects to larger masses (the same effect that leads to Eddington bias), which in turn increases the abundance at large masses.
In practice, we choose p(β) to be Lognormal with mean and variance chosen such that the mass function fESP(ν) as well as the conditional distribution p(β|ν) are in reasonable agreement with the results of simulations. Having done this, we will see later that the clustering of haloes is also correctly predicted by our model, which is a non-trivial test of the formalism. We set 〈 β 〉 = 0.5 and Var(β) = 0.25 for the Lognormal p(β); we have checked that varying these numbers within a ∼10 per cent range does not affect our results significantly. We avoided fitting these numbers more accurately because, as mentioned earlier, their physical origin is at present unclear and a more accurate fit would not lead to any physical insights. Moreover, recent work, in which haloes were identified using an ellipsoidal rather than spherical halo finder, suggests that the mean value is closer to 1.68 (1 + 0.2σ0) with variance around this mean of 0.04σ20 (Despali et al. 2013).
Fig. 2 shows the results for the WMAP1 (solid red) and WMAP3 (dotted red) cosmologies. Note that the value of σ8 is ∼10 per cent smaller in the WMAP3 cosmology. We plot the ratio of the ESP prediction in equation (12) and the Tinker et al. (2008) fit. For comparison we also indicate the ratio of the MICE fit with the Tinker et al. fit (for the WMAP1 cosmology) as the short dashed blue curve. The horizontal dotted lines mark 5 per cent deviations from the Tinker et al. fit; the typical scatter in their measurements in this mass range is 5–10 per cent (cf. fig. 6 a of Tinker et al. 2008).
![Ratio of the ESP mass function for a square-root barrier with scatter in the barrier slope (equation 12) to the Tinker et al. (2008) fit, for WMAP1 (solid red) and WMAP3 (dotted red) cosmologies. In each case, we set δc = 1.686 and p(β) to be Lognormal with 〈 β 〉 = 0.5 and Var(β) = 0.25, for the reasons discussed in the text. For comparison, we also show the ratio of the MICE fit with the Tinker et al. fit for the WMAP1 cosmology (short dashed blue). Horizontal dotted lines mark 5 per cent deviations from the Tinker et al. fit; the typical scatter in their measurements in this mass range is 5–10 per cent.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/431/2/10.1093/mnras/stt267/2/m_stt267fig2.jpeg?Expires=1723729622&Signature=DVohvJozL~dZ5G~MhuqQCEG-boAyWCS-Qo~0OtUpQZinkZeIW7aynOVrmmCyEFwFxarXcr5-jp~9cE82C~TvI8vGD08JV7iQ7wh1ARoG45AUo~XAaIaGRh09wx9SIsfdsZ3ctxBC-udsnu9BFovdw4qZrJK2fDP3nBGQYXSD~xgJppjsZX~x8VNEk0tKC68MhI16Gk46fqZfYhCaqtZ2kCCv7i28FlJXuMUOt52A3Em8Av5a2O7ve5EXqeyCN~YVzZ-Kj3VLlRAE83yJbxrqhnCehnOy~gnTPIu9HUBR2z0Tkr-JLirz-xTrRWqnw-NPuppJKaD6GXqc~XgdnkubSQ__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Ratio of the ESP mass function for a square-root barrier with scatter in the barrier slope (equation 12) to the Tinker et al. (2008) fit, for WMAP1 (solid red) and WMAP3 (dotted red) cosmologies. In each case, we set δc = 1.686 and p(β) to be Lognormal with 〈 β 〉 = 0.5 and Var(β) = 0.25, for the reasons discussed in the text. For comparison, we also show the ratio of the MICE fit with the Tinker et al. fit for the WMAP1 cosmology (short dashed blue). Horizontal dotted lines mark 5 per cent deviations from the Tinker et al. fit; the typical scatter in their measurements in this mass range is 5–10 per cent.
Fig. 3 shows the predicted mean 〈 β|ν 〉 and scatter Σβ(ν) of the conditional distribution p(β|ν) (equation 13) for the WMAP1 (red) and WMAP3 (blue) cosmologies. The black curves show the corresponding quantities measured by Robertson et al. (2009), assuming the fit Bfit = 1.506 + 0.487σ from their table 1 and using 〈 β|ν 〉Robertson + = (Bfit − 1.686)/σ in order to compare with our square-root barrier model in equation (11). The arrows indicate the approximate range over which Robertson et al. performed these fits for 〈 β|ν 〉 (dashed) and Σβ(ν) (solid). The ESP predictions are within 30 per cent of the measured quantities. Together with Fig. 2, this shows that our model works rather well.
![The conditional mean 〈 β|ν 〉 and scatter $\Sigma _\beta (\nu )\equiv \sqrt{{\rm Var}(\beta |\nu )}$ of the distribution p(β|ν) (equation 13) for the WMAP1 (red) and WMAP3 (blue) cosmologies. The black curves show the corresponding quantities measured by Robertson et al. (2009). The ESP curves used δc = 1.686, fESP(ν|β) from equation (14) and assumed p(β) to be Lognormal with 〈 β 〉 = 0.5 and Var(β) = 0.25. The curve for 〈 β|ν 〉 from Robertson et al. (2009) assumed the ‘linear’ fit from their table 1, Bfit = 1.506 + 0.487σ0, and shows 〈β|ν〉 = (Bfit − 1.686)/σ0. The arrows indicate the approximate range over which Robertson et al. performed these fits for 〈 β|ν 〉 (dashed) and Σβ(ν) (solid). The ESP predictions are within 30 per cent of the measured quantities.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/431/2/10.1093/mnras/stt267/2/m_stt267fig3.jpeg?Expires=1723729622&Signature=jYsexh6nuVMTkunHqm4UPXBoiaBmzlL05dOgK1t2-vLbXFee1G0tBkAjxUQjRuAJdrlNCV7iasxrQkiaBHAEKaPLi1~vUTSoPMXUQF4oPXzP2X6rNaMu1pNsb69HB371DBE-tPd~ZX6u2HCqWuxBzzgStC6~lPqjReNrJZAqox03~X4vfw94~MSOxFSrd7UBHVKPSVs0te7nT1c7CcKtU53j~Zfi20Sb9hf1N1QeLdIWHkvms8Y7s3dZcoOrHalLCWhYU798x1GBllYaDpretNi8YpT0cttp3c~NcrIYliZ1RbDphaDlcZvNApzYXMtaI855Ok5daoGCDBijSTu~yA__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
The conditional mean 〈 β|ν 〉 and scatter |$\Sigma _\beta (\nu )\equiv \sqrt{{\rm Var}(\beta |\nu )}$| of the distribution p(β|ν) (equation 13) for the WMAP1 (red) and WMAP3 (blue) cosmologies. The black curves show the corresponding quantities measured by Robertson et al. (2009). The ESP curves used δc = 1.686, fESP(ν|β) from equation (14) and assumed p(β) to be Lognormal with 〈 β 〉 = 0.5 and Var(β) = 0.25. The curve for 〈 β|ν 〉 from Robertson et al. (2009) assumed the ‘linear’ fit from their table 1, Bfit = 1.506 + 0.487σ0, and shows 〈β|ν〉 = (Bfit − 1.686)/σ0. The arrows indicate the approximate range over which Robertson et al. performed these fits for 〈 β|ν 〉 (dashed) and Σβ(ν) (solid). The ESP predictions are within 30 per cent of the measured quantities.
3 IMPLICATIONS FOR AVERAGES OVER ALL, RATHER THAN A SPECIAL SUBSET OF POSITIONS
The ESP model of the previous section is the first to self-consistently integrate the complex physics of non-spherical collapse with the fact that the physics is almost certainly simplest around the centre of mass of the protohalo. Although the agreement with the halo counts represents a significant and non-trivial success, and the next section shows that the predicted spatial distribution is also accurate, it makes a number of other testable predictions which are related to the logic of the excursion set approach.
For example, one can construct the centre-of-mass random walk associated with each halo and explicitly check, object by object, whether or not the corresponding barrier was crossed at a larger scale. This is, in fact, precisely the kind of test which was performed by Sheth et al. (2001). Revisiting this test is left to future work.
3.1 Distribution of δ for all particles in protohaloes of a given mass
We remarked in Section 1 that our ESP model explicitly averages only over special positions in the universe (see the text following equation 9 for exactly where this happens). If we treat these positions as the fundamental quantities – for which the physics of collapse is most easily applied – then by accounting for their initial density profiles, it is straightforward to estimate what would have happened had we averaged over all positions. Because our model has been calibrated for centre-of-mass positions solely, predicting the right distribution of δ values within each protohalo provides another test of our approach. We illustrate the idea below.
![Predicted distribution of δ for all particles in protohaloes of a given mass. Histogram shows our model, equation (16), which uses as input the smooth Lognormal distribution shown for p(δpk|Spk). Our model is well approximated by a Gaussian (solid curve); it has a slightly different mean and variance than do the measurements in simulations (the dashed Gaussian).](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/431/2/10.1093/mnras/stt267/2/m_stt267fig4.jpeg?Expires=1723729622&Signature=PFoCfu1zsK8YyS~6lyT6r-j9~P3OsNLcnpP~9OtLSVnsVBqxGxggTw~Ix6LDAO4B3BRAbxk2B~eVbKmArNclEpQytxASpnBHFbMuhLqiIjAX8~QmkH2-wKU~ofQMK-Xl-y8oztoBlgWOZ8bGsGd5YT0tUTS6MDKcPWZMjIWBxz5rbbhFR0nZo0Oxn2ie8LDkgDURbB~iw6n0ayKSawmNhCejKlsZ-uH0ceTLCMU1FZQDAu01t6vjgb7XqEGcTx6~azHSlJDKZlOvfmAPT13XhBqtRNyQQOrhe0bIJCoz8-A8H1dSGu46zpAZkRxx9ReixJqTUShkAbTj5-gtGRaJJA__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Predicted distribution of δ for all particles in protohaloes of a given mass. Histogram shows our model, equation (16), which uses as input the smooth Lognormal distribution shown for p(δpk|Spk). Our model is well approximated by a Gaussian (solid curve); it has a slightly different mean and variance than do the measurements in simulations (the dashed Gaussian).
The dashed curve shows a Gaussian distribution with mean (1.67 + 0.075 Spk)/1.35 and variance (0.35/1.35) Spk, which describes the measurements in simulations of this quantity rather well (Achitouv et al. 2013). In effect, our model is trying to predict this dashed curve. Because we use as input the Lognormal distribution of peak heights p(δpk|ν) that is seen in simulations (shown as the other solid curve which peaks at larger δ: it has mean |$\langle\delta _{\rm pk}|S_{\rm pk}\rangle = 1.506 + 0.487\sqrt{S_{\rm pk}}$|, and variance 0.09 Spk), our model has no free parameters, so it is perhaps remarkable that the model does as well as it does. Although our prediction is narrower than the measured distribution (predicted variance is 0.18 Spk whereas the measured value is 0.26 Spk) its mean, 1.35, is within 5 per cent of the measured value of 1.29.
Fig. 5 shows a similar analysis for Spk = 2, which corresponds to lower mass haloes (7.8 × 1012h− 1M⊙) for which the assumption that haloes form from peaks is less secure; clearly, the discrepancies between the model and the measurements are qualitatively similar even at these small masses. The predicted mean and variance are 1.43 and 0.136 Spk, whereas the measured values are 1.36 and 0.26 Spk.
![Same as previous figure, but now for Spk = 2.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/431/2/10.1093/mnras/stt267/2/m_stt267fig5.jpeg?Expires=1723729622&Signature=VxYLK~5~V5S1mOKSvhvDpfI2tHSRdwLuU4wh7DRJ9JlKItVwlKKFImea2xU-DYOyt5Ef9ejAyEXeQ7N26XryJnkL9jrOaacQ50OZVdhbHkN07MVVME6ZEWaxWjuwjIMCNDn2rdGQVdgoNQs6WxMUB5k6P3btQwOoNV4-hG-Aw16bHba2tWId2P6DHCk0F4FiPftZm27myqszeeOokkQhqtUmDr-CpdP0v2SrmCtWXV7~k2q4yrnk7pXQXzq26l3vzLIhhPKyWGZ35kiiclm51H6uYdZ38ttEWb6siCpefZV9sZq26yyeVK2AVjpH4OiZs6xQtKL86cn5ScWdmuewPA__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
3.2 Distribution of predicted mass for all particles in protohaloes of a given mass
The lower mean value implies that the mass which the excursion set approach predicts for the particles which are not at the centre of mass should be systematically smaller than the mass of the halo in which they end up – in qualitative agreement with one of the findings of Sheth et al. (2001, see their figs 2–4).
To see that this is indeed the case, we can use our ESP approach (walks centred on peaks in the initial field) to model what the usual excursion set estimate (walks centred on all positions in the initial field) really represents. Recall that, for the statistics of all positions, we know that if δc is independent of m then the excursion set prediction for the mass fraction in objects with mass m(s) is |$sf(s) = \nu \exp (-\nu ^2/2)/\sqrt{2\pi }$| with ν2 = δ2c/s. [Strictly speaking, to compare most directly with Sheth et al. (2001) we should account for the distribution in δc values at each s; but the constant deterministic δc discussion below captures the gist of the argument. Also, our expressions below are appropriate for a sharp k filter, but the logic, and so the qualitative trends, do not depend on this choice.]
For constant δc, figs 1 and 2 in Paranjape & Sheth (2012) show that SfESP(S) is rather well approximated simply by |$\nu f(\nu ) = \nu \exp (-\nu ^2/2)/\sqrt{2\pi }$| with ν2 = (Ξδc)2/S for some Ξ < 1. [Strictly speaking, they used the Sheth–Tormen functional form for νf(ν); however, at high masses, this form reduces to the simpler one given here with |$\Xi =\sqrt{0.7}$|.] Since this is the same functional form for S that appears on the left-hand side of equation (17) for s, it is easy to check that sf(s|S) should also be the same function of ν, but with ν2 = (δc − Ξδc)2/(s − S).
Thus, this calculation shows explicitly that the average over all walks will yield f which has a similar functional form to the one for the specially chosen subset, but the two will have different values of δc. Moreover, the distribution of predicted masses will be very different from a delta function centred on M. Rather, it will look just like the conditional excursion set prediction for the fraction of all walks which first cross δc on scale s given that they are known to pass through Ξδc < δc on scale S before first crossing δc. This latter point quantifies the explanation given by Sheth et al. (2001) when addressing the concerns raised by White (1996). It also expands on a point first made by Paranjape & Sheth (2012): the ESP framework naturally accounts for the fact that one must rescale |$\nu \rightarrow \sqrt{q}\nu$| with q ∼ 0.7 if one wants to use the usual excursion set expressions (based on the statistics of all positions) to model halo counts – but this rescaling is not necessary if one uses the statistics of the special positions for which the physics is simplest. Of course, because the physics of collapse refers to collapse around the halo centres of masses, and not around random positions in space, this discussion highlights the dangers of interpreting the barrier height in the random walk calculation with that for the physics of collapse around special positions in space.
4 HALO BIAS
In this section, we derive expressions for the predicted clustering of haloes in our model.
The presence of ϵ× in these expressions for real-space bias is an indication of k-dependent Fourier-space bias, as discussed at length by Musso et al. (2012).2 At large scales R0 → ∞, ϵ× → 0. Consequently, the coefficients bn0 in equation (20) correspond to the large-scale (k → 0) Fourier-space bias that is routinely measured in simulations. Further, Musso et al. (2012) pointed out that bn0 correspond to the usual peak-background split expressions for bias, bn0 = f(ν)− 1( − ∂/∂δc)nf(ν). In other words, the peak-background split bias parameters correspond to taking the large-scale limit R0 → ∞, or equivalently, setting ϵ× → 0 and S×/S0 → 1 in the coefficients bn.
Fig. 6 shows our ESP prediction for the linear peak-background split bias δcb10 (i.e. setting ϵ× = 0, S×/S0 = 1 in equation 29) as the solid red curve. The dotted black curve is the spherical collapse prediction ν2 − 1 (Mo & White 1996). The dashed green curve shows the fit presented by Tinker et al. (2010) for the same simulations analysed by Tinker et al. (2008). We have checked that in this case the results for the WMAP1 and WMAP3 cosmologies are nearly identical, so we only show the results for WMAP3. The ESP curve matches the N-body fit very well at large masses, which is where the ESP framework is expected to work well. Note that the curve for Tinker et al. (2010) is not simply the logarithmic derivative of the mass function calibrated by Tinker et al. (2008); that derivative would underestimate the large-scale bias by up to 10 per cent (Manera, Scoccimarro & Sheth 2010; Tinker et al. 2010). (For comparison, the short dashed blue curve shows the peak-background split prediction for the MICE mass function.)
![Scale-independent (peak-background split) linear Lagrangian halo bias δcb10 in the ESP framework for a square-root barrier with Lognormal scatter (solid red). This curve is in excellent agreement with the fit to N-body simulations from Tinker et al. (2010), the simulations being the same as those for which Fig. 2 showed a mass function comparison (see the text for a discussion). The dotted black curve is the spherical collapse prediction ν2 − 1 (Mo & White 1996), and the short dashed blue curve is the peak-background split prediction associated with the mass function fit to the MICE simulations.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/431/2/10.1093/mnras/stt267/2/m_stt267fig6.jpeg?Expires=1723729622&Signature=SHrj~~NYWOpWDiMCCvQv4tdr3xNPoLBpNN09NkDnTo2oKVFPzVqn-a0-Ta~xIPeJDWQZ9ah2aIpe5m-6JYluU-2MiHqdSSgynEtLdeOoIHyiSfOkYaPbfoGynqpbmepINSurm5SBbGTv86ZPkso-v1Z5HPDhLQZsOGP7eKIT6RjQ6Qx5I4mQ6f3kIf~WBB~WMzgctZcFLf8MeHInhliRgQKa5v~vGjHBdsg2G-qN6RYQnlh3xc444iVj0ZXUZFEUfXtntVvVikpY41-JebEmR6HK3vKxtYKpOHXcqWBmNY50173mktGzb~n01ka9k9OgjxuHIbuaoYFpo3D~eH9XxQ__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Scale-independent (peak-background split) linear Lagrangian halo bias δcb10 in the ESP framework for a square-root barrier with Lognormal scatter (solid red). This curve is in excellent agreement with the fit to N-body simulations from Tinker et al. (2010), the simulations being the same as those for which Fig. 2 showed a mass function comparison (see the text for a discussion). The dotted black curve is the spherical collapse prediction ν2 − 1 (Mo & White 1996), and the short dashed blue curve is the peak-background split prediction associated with the mass function fit to the MICE simulations.
Fig. 7 shows the corresponding result for the quadratic peak-background split coefficient δ2cb20 for ESP (solid red) and the spherical collapse result ν2(ν2 − 3) (Mo, Jing & White 1997, dotted black). The dashed blue curve shows the Mo et al. (1997) expression after rescaling |$\nu \rightarrow \sqrt{0.7}\nu$|. In this case, there are no measurements in simulations with which to compare – this is the subject of ongoing work – but we have included the figure to show that b20 behaves qualitatively like b10, transitioning from the usual (sharp k excursion set) expression rescaled by |$\nu \rightarrow \sqrt{0.7}\nu$| at ν ∼ 1 to lying above this rescaled value at larger ν.
![Same as previous figure, but now for the second-order bias coefficient. In this case, there are no measurements from simulations with which to compare.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/431/2/10.1093/mnras/stt267/2/m_stt267fig7.jpeg?Expires=1723729622&Signature=TY1g7JFucA6iulnDuJnznki2ViaZM0j8UL-HYO4k38E8bzz7hts0p1Gc2OEwYT752yulfrlm5njx9ULCotaip8xCYpJf2Ha2ZVFVIEgW2mm7TwaBImx7kmb-Au3keM5KFVKUpteRGhrLJ-~L08wAY5Bna2fxPi8WyTxDm3It6KRTa6CL~rdUwztP57YtTeioLvdoBdKHoJV3andx8d6o8jThezETLI1Rcj6WrfKs9OPqB5U6JL3IieYVHfN0kGdlk3PAyCtO0lAOtWB0RmBnRqGNBISYStpxn03c9TOYuUUqVVuGExPn2wfQtcSzhZAJz~hqxwmxWYRBW3CHk0sEQQ__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Same as previous figure, but now for the second-order bias coefficient. In this case, there are no measurements from simulations with which to compare.
These plots illustrate another point made by Paranjape & Sheth (2012): the ESP framework naturally accounts for the fact that the measured large-scale bias deviates from that predicted by a naive application of the peak-background split to a mass function that was fit using the rescaled variable |$\nu \rightarrow \sqrt{q}\nu$| with q ∼ 0.7.
4.1 Reconstructing the peak-background split parameters
In this subsection, we revisit the idea of reconstructing the peak-background split parameters bn0 from suitable real-space measurements of the parameters bn (Musso et al. 2012). In particular, we wish to address some subtleties that arise due to the stochasticity of the barrier height.
5 DISCUSSION
We have presented a derivation of equation (12), which represents the first analytic model of halo abundances which accounts self-consistently for the fact that the physics of collapse is best described around the halo centre of mass, is not spherical, and can vary substantially from one place in the universe to another. The comparison between equation (12) and halo counts in simulations (Fig. 2) is very encouraging. (Fig. 1 shows that accounting for the physics of non-spherical collapse is necessary.)
We also derived expressions for the associated large-scale bias factors b10 and b20 (these follow from setting ϵ× → 0 and S×/S0 → 1 in equations 29 and 30); the former is in very good agreement with measurements from simulations. We are in the process of determining if b20 is similarly accurate. Additionally, we showed how real-space measurements of b1 and b2 can be converted to practical estimates of b10 and b20 (equations 35 and 36), which extends the results of Musso et al. (2012) and Paranjape & Sheth (2012) to the case of stochastic barrier heights. Testing the accuracy of this reconstruction in N-body simulations is also work in progress.
Our ESP analysis highlighted the benefits of developing a model for the abundance of positions in space around which the physics of collapse is simplest. In Section 3, we showed that it can be used to understand what would have happened if one had instead studied all initial positions (Figs 4 and 5). This provides the basis for a number of other tests of the more formal aspects of our approach – as well as elucidating the relation between our ESP calculation and the usual excursion set approach.
Our model must take as input one physical quantity from simulations. This quantity describes how the typical density required for collapse depends on protohalo mass, and how much this density varies from one protohalo to the next. While the triaxial collapse model correctly predicts that this mean value and the scatter around it should increase as halo mass decreases, the predicted scatter is smaller than observed. We therefore hope that our work will motivate studies of the physical origin of p(β), as well as measurements in simulations of how this distribution evolves. If this distribution does not evolve, then our ESP model provides a particularly simple model for how the distribution of halo abundances, when expressed as a function of ν, should be nearly but not exactly universal (see discussion in Musso & Sheth 2012). Fig. 8 shows that the predicted departures from universality in this simple case are in reasonable agreement with measurements.
![Predicted departures from self-similarity in the ESP model, for the WMAP3 cosmology. Top panel shows the predicted halo abundances at z = 0 and 1, expressed as a function of y = [δcD(0)/D(z)σ0(m)]2. We have assumed that the distribution p(β) defined in the text does not evolve with redshift. Bottom panel shows the ratio of our predictions to the fitting formulae for these redshifts which Tinker et al. (2008) calibrated from halo abundances measured in their simulations.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/431/2/10.1093/mnras/stt267/2/m_stt267fig8.jpeg?Expires=1723729622&Signature=objuFv9zfFXyRetN2O0wqYb95ZiDemldlzIWTrd1pLwMdZLPZkhHSf8XVXKjnSUTi4gqAWCCjkWFoBKHGSBIfTCr~idfIZmm6fOpi8ec0Xi3QPXlcJFvU4oKJbjyybq-5kncccAJnAW39zelLqHedKu8wLsEG-6CmVq7m7keUdz4vXp9QPY5W9YtGNWssC2PzKIC0c3yzCHytNvBQ8ise-8MVL0zLXkKNWbSQR~WCbt1juT40kB47m4WqTVRkdhNGaWSsscVbZPjFgL7YYa-t1M5fGJyAWgmEUc50KdfCgdBCtxg4BfsC0IebjkkZacQxEbHaazykOIBQRe9k8Ictg__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Predicted departures from self-similarity in the ESP model, for the WMAP3 cosmology. Top panel shows the predicted halo abundances at z = 0 and 1, expressed as a function of y = [δcD(0)/D(z)σ0(m)]2. We have assumed that the distribution p(β) defined in the text does not evolve with redshift. Bottom panel shows the ratio of our predictions to the fitting formulae for these redshifts which Tinker et al. (2008) calibrated from halo abundances measured in their simulations.
Returning to the issue of the distribution of β, an obvious direction of future work is that peaks have non-trivial deformation and inertia tensors. Non-spherical initial shapes mean the critical density for collapse will now depend not just on the tidal field but on the shape and the misalignment angle between the principal axes of the two tensors. In principle, our ESP model contains a prescription for the distribution of shapes (and misalignments) for which one should run the ellipsoidal collapse model.
This does raise one caveat: in peaks theory (on which our ESP model is based), the axes of the two tensors are expected to be well aligned, with the direction of maximum compression (due to the tidal field) aligned with the shortest axis (van de Weygaert & Bertschinger 1996). Measurements in simulations do show strong alignment, but the direction of maximum compression lies instead along the longest axis (Porciani et al. 2002; Despali et al. 2013). On the other hand, these measurements are dominated by masses for which δc(m)/σ0(m) < 1, where our ESP model may not apply anyway: there is no physically compelling reason why a peak of (normalized) height 1 should dominate its surroundings. While this mass regime may be less interesting for studies which use the abundance of rich clusters to constrain cosmological models, understanding it is useful for understanding the formation and evolution of galaxies. So we hope our approach of identifying just what it is that is special about the positions around which gravitational collapse occurs, and then using only these positions to make inferences about halo assembly histories, will eventually provide insight into the low mass regime as well.
We thank Christophe Pichon for discussions, and for pointing out to us the work of Hanami (2001), and Ixandra Achitouv for discussions regarding barrier stochasticity. This work is supported in part by NSF-0908241 and NASA NNX11A125G (RKS and AP), and by the Swiss National Science Foundation (VD). AP thanks A. Réfrégier and A. Amara at ETH, Zürich for hospitality during a visit when this work was completed. VD also wishes to thank ICTP for hospitality when parts of this work were completed.
We use the results of Robertson et al. (2009) rather than those of Dalal et al. (2008) because the latter only show a plot whereas Robertson et al. provide a description of the full probability distribution. Both groups find a similar mean and rms for the barrier heights, however, which increase approximately linearly with σ0.
The factor (S×/S0)n is a consequence of the fact that the excursion set calculation lives in real space rather than Fourier space, and would be present even if the Fourier-space bias were constant.