-
PDF
- Split View
-
Views
-
Cite
Cite
Matteo Biagetti, Kwan Chuen Chan, Vincent Desjacques, Aseem Paranjape, Measuring non-local Lagrangian peak bias, Monthly Notices of the Royal Astronomical Society, Volume 441, Issue 2, 21 June 2014, Pages 1457–1467, https://doi.org/10.1093/mnras/stu680
- Share Icon Share
Abstract
We investigate non-local Lagrangian bias contributions involving gradients of the linear density field, for which we have predictions from the excursion set peak formalism. We begin by writing down a bias expansion which includes all the bias terms, including the non-local ones. Having checked that the model furnishes a reasonable fit to the halo mass function, we develop a one-point cross-correlation technique to measure bias factors associated with χ2-distributed quantities. We validate the method with numerical realizations of peaks of Gaussian random fields before we apply it to N-body simulations. We focus on the lowest (quadratic) order non-local contributions |$-2\chi _{10}(\mathrm{\boldsymbol k}_1\cdot \mathrm{\boldsymbol k}_2)$| and |$\chi _{01}[3(\mathrm{\boldsymbol k}_1\cdot \mathrm{\boldsymbol k}_2)^2-k_1^2 k_2^2]$|, where |$\mathrm{\boldsymbol k}_1$|, |$\mathrm{\boldsymbol k}_2$| are wave modes. We can reproduce our measurement of χ10 if we allow for an offset between the Lagrangian halo centre-of-mass and the peak position. The sign and magnitude of χ10 is consistent with Lagrangian haloes sitting near linear density maxima. The resulting contribution to the halo bias can safely be ignored for M = 1013 M⊙ h−1, but could become relevant at larger halo masses. For the second non-local bias χ01 however, we measure a much larger magnitude than predicted by our model. We speculate that some of this discrepancy might originate from non-local Lagrangian contributions induced by non-spherical collapse.
1 INTRODUCTION
Understanding the clustering of dark matter haloes has been a topic of active research for many years. A number of analytic approaches have been developed to tackle this issue such as the peak model (Bardeen et al. 1986, hereafter BBKS), the excursion set framework (Bond et al. 1991) or perturbation theory (see e.g. Bernardeau et al. 2002, for a review). Heuristic arguments like the peak-background split (Kaiser 1984) and approximations like local bias (Fry & Gaztanaga 1993) have been very helpful for modelling the clustering of dark matter haloes. Nevertheless, improvements in computational power and numerical algorithms as well as the advent of large-scale galaxy surveys have considerably increased the need for an accurate description of halo clustering. Until recently, however, it was unclear how the peak approach, which is thus far the only framework in which biased tracers form a discrete point set, relates to the more widespread excursion set theory, local bias approximation or peak-background split argument.
Working out this connection has been the subject of several recent papers. Desjacques (2013), building on earlier work by Desjacques et al. (2010), showed that correlation functions of discrete density peaks can be computed using an effective (i.e. which does not involve measurable counts-in-cells quantities) generalized bias expansion in which all the bias parameters, including those of the non-local terms,1 can be computed from a peak-background split. In parallel, Paranjape & Sheth (2012) demonstrated how the peak formalism, which deals with statistics of density maxima at a fixed smoothing scale, can be combined with excursion set theory, whose basic building block is the density contrast at various filtering scales. Similar ideas can already be found in the early work of Bond (1989). Paranjape, Sheth & Desjacques (2013a, hereafter PSD) subsequently computed the mass function and linear bias of haloes within this excursion set peak (ESP) approach and showed that it agrees very well with simulation data.
The focus of this work is on the second-order non-local bias terms predicted by the ESP approach. These generate corrections to the Fourier peak bias of the form |$-2\chi _{10}(\mathrm{\boldsymbol k}_1\cdot \mathrm{\boldsymbol k}_2)$| and |$\chi _{01}[3(\mathrm{\boldsymbol k}_1\cdot \mathrm{\boldsymbol k}_2)^2-k_1^2 k_2^2]$| (Desjacques 2013). What makes them quite interesting is the fact that there are related to χ2 rather than normally distributed variables. Here, we will show how one can measure their amplitude in the bias of dark matter haloes without computing any correlation function. Of course, this technique can also be applied to measure non-local Lagrangian bias contributions induced by e.g. the tidal shear, but this will be the subject of future work.
This paper is organized as follows. In the first part, we will advocate a slight modification of the original ESP formulation of PSD in order to easily write down the corresponding effective bias expansion (Section 2). Next, we will explain how the cross-correlation technique proposed by Musso, Paranjape & Sheth (2012), which has already been successfully applied to the bias factors associated with the density field (PSD; Paranjape et al. 2013b), can be extended to measure the second-order non-local bias factors χ10 and χ01 that weight the two quadratic, non-local bias contributions (Section 3). Finally, we will validate our method with peaks of Gaussian random fields before measuring χ10 and χ01 for dark matter haloes (Section 4). We conclude in Section 5.
2 EXCURSION SET PEAKS
In this section, we apply the excursion set approach to the peak model in the case of a moving barrier to get a prediction of the halo mass function which we compare to simulations. We then get expressions for bias parameters, generalizing results in Desjacques (2013) and Desjacques, Gong & Riotto (2013). We also point out a few changes to PSD. We show that, as far as the mass function is concerned, these modifications do not make much difference (only few per cent, in agreement with what PSD found), but they affect first- and second-order bias parameters, as new terms arise.
2.1 Notation
2.2 First-crossing and moving barrier
2.2.1 Summary of previous results
Let us first summarize the basic ideas behind the ESP approach introduced by Paranjape & Sheth (2012) and further developed in PSD and Desjacques et al. (2013).
The excursion set approach states that a region of mass M has virialized when the overdensity δ(R), where R ∼ M1/3 is the filtering scale associated with the perturbation, reaches the spherical collapse threshold δc provided that, for any R′ > R, the inequality δ(R) < δc holds. This last condition formally implies an infinite set of constraints (one at each smoothing scale). However, as was shown in Musso & Sheth (2012), the requirement δ(R + ΔR) < δc with ΔR ≪ 1 furnishes a very good approximation. This follows from the fact that the trajectory described by δ(R) as a function of R is highly correlated for large radii. As a result, if δ crosses δc at R, then it is almost certainly below the threshold at any larger radius.
This first-crossing condition can be combined with the peak constraint, so that peaks on a given smoothing scale are counted only if the inequality above is satisfied. In this case, the effective peak bias expansion introduced in Desjacques (2013) is modified through the presence of a new variable μ (equation 2) which, as was shown in Desjacques et al. (2013), reflects the dependence of bias to the first-crossing condition.
2.2.2 Modifications to PSD
We made a couple of modifications to the approach of PSD, which we will now describe in more detail.
First, PSD used the fact that μ ≡ u when Gaussian filtering is also applied to the density field, so that the first-crossing condition can be accounted for with the variable u only. When δ is smoothed with a tophat filter however, one should in principle deal explicitly with μ and, therefore, consider the trivariate normal distribution |${\cal N}(\nu ,u,\mu )$|. We will proceed this way.
2.3 Comparison with numerical simulations
To test the validity of our approach, we compare the ESP mass function with that of haloes extracted from N-body simulations. For this purpose, we ran a series of N-body simulations evolving 10243 particles in periodic cubic boxes of size 1500 and 250 h−1 Mpc. The particle mass thus is 2.37 × 1011 and 1.10 × 109 M⊙ h−1, respectively. The transfer function was computed with CAMB (Lewis, Challinor & Lasenby 2000) assuming parameter values consistent with those inferred by WMAP7 (Komatsu et al. 2011): a flat ΛCDM cosmology with h = 0.704, Ωm = 0.272, Ωb = 0.0455, ns = 0.967 and a normalization amplitude σ8 = 0.81. Initial conditions were laid down at redshift z = 99 with an initial particle displacement computed at second order in Lagrangian perturbation theory with 2LPTic (Crocce, Pueblas & Scoccimarro 2006). The simulations were run using the N-body code GADGET-2 (Springel 2005) while the haloes were identified with the spherical overdensity halo finder AHF (Knollmann & Knebe 2009) assuming an overdensity threshold Δc = 200 constant throughout redshift.
![Halo mass function measured from N-body simulation at redshift z = 0 (left-hand panel) and z = 1 (right-hand panel) with different box sizes as indicated in the figures. The error bars are Poisson. The data are compared to the theoretical prediction (equation 15) based on the ESP formalism and the fitting formula of Tinker et al. (2008). We also show the fractional deviation of the Tinker et al. (2008) and the measured halo mass function relative to our theoretical prediction.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/441/2/10.1093/mnras/stu680/2/m_stu680fig1.jpeg?Expires=1722373133&Signature=rUsR1TvaM1DbUw6S73u1FXH2F5IeuU6Fn0ojAloxalkOqMJn2lYGyx59fTth28JlQW0teYaf5x1ChVa61w2hbi~m4kvd0FQsfbs6MOOzUOYZejBkDMNTyum29G7igcNFG~2hzvXpLRrWpM-Vfu-S7wIjSO~Wpdnfq4twgRxaSiwE0Bx2aEoPhBuBcWvjrD-9L0PzLSB8TP9I3udkF7yGGxYsKtJ1QPn7d6C9L1GM90uW5mjI3IZ2vOV9jgb-DCt1kObOGDHA2ZSPfcZVzjp67FFpkLMuiyJ9r5aM5nGWuc4-xG9NyQ6qfGQfKkbNv~Hh3UkbNpWPgiiu0fTyLY5IWA__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Halo mass function measured from N-body simulation at redshift z = 0 (left-hand panel) and z = 1 (right-hand panel) with different box sizes as indicated in the figures. The error bars are Poisson. The data are compared to the theoretical prediction (equation 15) based on the ESP formalism and the fitting formula of Tinker et al. (2008). We also show the fractional deviation of the Tinker et al. (2008) and the measured halo mass function relative to our theoretical prediction.
2.4 Bias parameters
The behaviour of the bias factors bij0 and χkl as a function of halo mass is similar to that seen in fig. 1 of Desjacques (2013). The bias factors bijk with k ≥ 1 weight the contributions of μk terms to the clustering of ESP that are proportional to derivatives of the tophat filter w.r.t. the filtering scale RT. Similar contributions appear in the clustering of thresholded regions (Matsubara 2012; Ferraro et al. 2013) since their definition also involve a first-crossing condition.
3 BIASES FROM CROSS-CORRELATION: EXTENSION TO χ2 VARIABLES
In this section, we will demonstrate that the bias factors χij can be measured with a one-point statistics. We will test our method on density peaks of a Gaussian random field before applying it to dark matter haloes.
3.1 Bias factors bijk: Hermite polynomials
Before we generalize this approach to the chi-squared bias factors χij, we emphasize that, in this cross-correlation approach, the smoothing scale Rl can take any value as long as it is distinct from the halo smoothing scale. Paranjape et al. (2013b) chose Rl ≫ Rs in the spirit of the peak-background split but this requirement is, in fact, not necessary as long as the correlation between the two scales is taken into account. In any case, we will stick with the notation Rl for convenience.
3.2 Bias factors χij: Laguerre polynomials
4 TEST WITH NUMERICAL SIMULATIONS
In this section, we first validate our predictions based on peaks of Gaussian random fields with measurements extracted from random realizations of the Gaussian linear density field, and then move on to calculate χ10 and χ01 for M ≳ M⋆ haloes, where M⋆ is the characteristic mass of the haloes.
4.1 Peaks of Gaussian random fields
We generate random realizations of the Gaussian, linear density field with a power spectrum equal to that used to seed the N-body simulations described above. To take advantage of Fast Fourier Transforms (FFTs), we simulate the linear density field in periodic, cubic boxes of side 1000 h−1 Mpc. The size of the mesh along each dimension is 1536. We smooth the density field on scale Rs = 5 h− 1 Mpc with a tophat filter and find the local maxima by comparing the density at each grid point with its 26 neighbouring values.
![Sections for νl, $3 \eta _{\rm l}^2$ and $5 \zeta _{\rm l}^2$ (from left to right). A filtering scale of Rl = 5 and 10 h−1 Mpc is used for the first and second row, respectively. Note that a tophat kernel is applied for νl, while a Gaussian window is used for $\eta _{\rm l}^2$ and $\zeta _{\rm l}^2$. In each panel, the dimension of the section is 200 × 200 h−2 Mpc2.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/441/2/10.1093/mnras/stu680/2/m_stu680fig2.jpeg?Expires=1722373133&Signature=NHoO7qBKMo-QkkSXTiw1EIEdHLUik9I2HhmVl0NDGUMcIOxZD4t8IogwkZtjIlwlbifh3Z230PCIkcRMjviqKRFPFccFgUAFjZhX5q-yuG8RDNb2Apoegh7FP19h~84fEsEG97zxiO7zcTiOL~H2cNqhdW~SVEtArBip7CIXxrGkY9fG9mbqwCOE1hTpZCFhnZGL2aD4k0Hmgdx0mFKvVShvw2UL-ryFarJIR5jehwmIhl1TPgiQYq7orBwAWgXJnqFEAfxrP6ySinZ00SibfFW2s4KvsQf9V31A2hxZt~MvQe6M~W4nms-fmp9m148oh1L3qxJdNgJGztRebmqYWA__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Sections for νl, |$3 \eta _{\rm l}^2$| and |$5 \zeta _{\rm l}^2$| (from left to right). A filtering scale of Rl = 5 and 10 h−1 Mpc is used for the first and second row, respectively. Note that a tophat kernel is applied for νl, while a Gaussian window is used for |$\eta _{\rm l}^2$| and |$\zeta _{\rm l}^2$|. In each panel, the dimension of the section is 200 × 200 h−2 Mpc2.
![Conditional probability distribution for the variables $3\eta _{\rm l}^2$ (left-hand panel) and $5\zeta _{\rm l}^2$ (right-hand panel) measured at the position of maxima of the linear density field smoothed with a Gaussian filter on scale R = 5 h−1 Mpc. Left-hand panel: histograms indicate the results for Rl = 10, 15 and 20 h−1 Mpc, which leads to ϵ1 = 0.71, 0.44 and 0.29 as quoted in the figure. Right-hand panel: histograms show the results for a fixed Rl = 10 h− 1 Mpc (which implies ϵ2 = 0.57) but several peak height intervals. In all cases, the solid curves are the theoretical prediction (see text) whereas the dashed (green) curves represents the unconditional distribution $\chi _k^2(y)$.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/441/2/10.1093/mnras/stu680/2/m_stu680fig3.jpeg?Expires=1722373133&Signature=XnN-v2mqEKeRTEnti5~wH9IEQK-2RNhU4vccURFqu-jrlx9QowfxJrczDa-sAGXB5~yuaaUqvbWH53yd4cIQkwnlhkIKVHAt9f71qozaj7Iz4DMI579oVOg9Fa9kp-SdcTx~roVLE8Q-VdwSCEKAW7S0Z-DeBP6R3E4~jrqO5gveIAj0hbAWkmUF25iMf5Vg~dBGOpx9AWfvAWqhRfdMHP4v485WQw4pbJeK-Et8A-FzjX5XESNk6-Y2D-eu7d4nQb-7Yzl8pt7aE3u3k8fJjEDHHkpZG-VwBJtRWBO6dFOoStY5jHp~GEnTDs9M6u1yJj94yQTfx9Cxa9aR3fFrtg__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Conditional probability distribution for the variables |$3\eta _{\rm l}^2$| (left-hand panel) and |$5\zeta _{\rm l}^2$| (right-hand panel) measured at the position of maxima of the linear density field smoothed with a Gaussian filter on scale R = 5 h−1 Mpc. Left-hand panel: histograms indicate the results for Rl = 10, 15 and 20 h−1 Mpc, which leads to ϵ1 = 0.71, 0.44 and 0.29 as quoted in the figure. Right-hand panel: histograms show the results for a fixed Rl = 10 h− 1 Mpc (which implies ϵ2 = 0.57) but several peak height intervals. In all cases, the solid curves are the theoretical prediction (see text) whereas the dashed (green) curves represents the unconditional distribution |$\chi _k^2(y)$|.
4.2 Dark matter haloes
Having successfully tested the theory against numerical simulations of Gaussian peaks, we will now attempt to estimate the bias factors χ10 and χ01 associated with dark matter haloes. For this purpose, we first trace back all dark matter particles belonging to virialized haloes at redshift z = 0 to their initial position at z = 99. We then compute the centre-of-mass positions of these Lagrangian regions and assume that they define the locations of protohaloes. We can now proceed as for the Gaussian peaks and compute ν, |$\eta _{\rm l}^2$| and |$\zeta _{\rm l}^2$| at the position of protohaloes.
To predict the value of RG given RT, we followed PSD and assumed that RG(RT) can be computed through the requirement that 〈δG|δT〉 = δT. This yields a prediction for the value of the cross-correlation coefficients ϵ1 and ϵ2 as a function of halo mass, which we can use as an input to |$\chi _k^2(y|x)$| and only fit for x. However, we found that using the predicted ϵ1 leads to unphysical (negative) values for x when one attempts to fit |$P(3\eta _{\rm l}^2|{\rm halo})$|. Therefore, we decided to proceed as follows.
Estimate both ϵ1 and x = 〈3η2|halo〉 by fitting the model |$\chi _3^2(y|x;\epsilon _1)$| to the measured |$P(3\eta _{\rm l}^2|{\rm halo})$|.
Compute ϵ2 assuming that the same RG enters the spectral moments.
Estimate |$x=\langle 5\zeta _{\rm l}^2|{\rm halo}\rangle$| by fitting the theoretical model |$\chi _5^2(y|x;\epsilon _2)$| to the simulated |$P(5\zeta _{\rm l}^2|{\rm halo})$|.
We considered data in the range |$0<3\eta _{\rm l}^2<8$| and |$0<5\zeta _{\rm l}^2<12$| and gave equal weight to all the measurements (assuming Poisson errors does not affect our results significantly). Table 1 summarizes the best-fitting values obtained for four different halo bins spanning the mass range 1013-1015 M⊙ h− 1, whereas the measured probability distributions together with the best-fitting models are shown in Fig. 4. The data are reasonably well described by a conditional χ2-distribution, but the fit is somewhat poorer when the cross-correlation coefficient is close to unity.
![Conditional probability distribution for $3\eta _{\rm l}^2$ (top panel) and $5\zeta _{\rm l}^2$ (bottom panel) measured at the centre-of-mass position of protohaloes. The filter is Gaussian with Rl = 10 h− 1 Mpc. The various curves show the best-fitting theoretical predictions for the halo mass bins considered here. Halo mass range is in unit of 1013 M⊙ h−1. Poisson errors are much smaller than the size of the data points and, therefore, do not show up in the figure.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/441/2/10.1093/mnras/stu680/2/m_stu680fig4.jpeg?Expires=1722373133&Signature=cJuKdPWmUQntYw8CYvc0oGzQaaJBkwtyDf~U6BRW0Ip0knOvORyN~UGpMg~UCjFaLUqfDbhL70pf5MVqylRoI8AHdwJz6HQT1bbfZ3uRWAPYB0fBu2OOsNUMK4g3o9yMT8zU2ADuPJWeyydNP5MAMLnT4LxSLqoWQ6U-vccSN66dtrvjzGy~FTg5G2cRzBHHvdbfKbMj1lr-lqHBo7NaN1~0JL0A3aNMUt1s1L8PE77YRXyacJ9jPDRMEbpEWYeOlbbSnUi~khHJIrA8GAtZYGHq4mmWTIAD4Z6C03exyLUHYYeYQR4YKvcX-hBOsXYhP6Xc2DT3opyOpT-mTB4C8w__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Conditional probability distribution for |$3\eta _{\rm l}^2$| (top panel) and |$5\zeta _{\rm l}^2$| (bottom panel) measured at the centre-of-mass position of protohaloes. The filter is Gaussian with Rl = 10 h− 1 Mpc. The various curves show the best-fitting theoretical predictions for the halo mass bins considered here. Halo mass range is in unit of 1013 M⊙ h−1. Poisson errors are much smaller than the size of the data points and, therefore, do not show up in the figure.
Best-fitting parameter values as a function of halo mass. The latter is in unit of 1013 M⊙ h−1. Note that we also list the values of ϵ2 even though it is not directly fitted to the data (see text for details).
Halo mass . | 〈3η2|halo〉 . | ϵ1 . | 〈5ζ2|halo〉 . | (ϵ2) . |
---|---|---|---|---|
M > 30 | 0.71 | 0.80 | 2.98 | (0.70) |
10 < M < 30 | 1.24 | 0.66 | 4.49 | (0.52) |
3 < M < 10 | 1.62 | 0.54 | 5.82 | (0.37) |
1 < M < 3 | 1.94 | 0.49 | 6.12 | (0.31) |
Halo mass . | 〈3η2|halo〉 . | ϵ1 . | 〈5ζ2|halo〉 . | (ϵ2) . |
---|---|---|---|---|
M > 30 | 0.71 | 0.80 | 2.98 | (0.70) |
10 < M < 30 | 1.24 | 0.66 | 4.49 | (0.52) |
3 < M < 10 | 1.62 | 0.54 | 5.82 | (0.37) |
1 < M < 3 | 1.94 | 0.49 | 6.12 | (0.31) |
Best-fitting parameter values as a function of halo mass. The latter is in unit of 1013 M⊙ h−1. Note that we also list the values of ϵ2 even though it is not directly fitted to the data (see text for details).
Halo mass . | 〈3η2|halo〉 . | ϵ1 . | 〈5ζ2|halo〉 . | (ϵ2) . |
---|---|---|---|---|
M > 30 | 0.71 | 0.80 | 2.98 | (0.70) |
10 < M < 30 | 1.24 | 0.66 | 4.49 | (0.52) |
3 < M < 10 | 1.62 | 0.54 | 5.82 | (0.37) |
1 < M < 3 | 1.94 | 0.49 | 6.12 | (0.31) |
Halo mass . | 〈3η2|halo〉 . | ϵ1 . | 〈5ζ2|halo〉 . | (ϵ2) . |
---|---|---|---|---|
M > 30 | 0.71 | 0.80 | 2.98 | (0.70) |
10 < M < 30 | 1.24 | 0.66 | 4.49 | (0.52) |
3 < M < 10 | 1.62 | 0.54 | 5.82 | (0.37) |
1 < M < 3 | 1.94 | 0.49 | 6.12 | (0.31) |
The second-order bias factors χ10 and χ01 of the dark matter haloes at z = 0 can be readily computed from equation (39) using the best-fitting values of 〈3η2|halo〉 and 〈5ζ2|halo〉. The results are shown in Fig. 5 as the data points. Error bars indicate the scatter among the various realizations and, therefore, likely strongly underestimate the true uncertainty. The dashed curves indicate the predictions of the ESP formalism. The measurements, albeit of the same magnitude as the theoretical predictions, quite disagree with expectations based on our ESP approach, especially χ01 which reverses sign as the halo mass drops below 1014 M⊙ h−1.
![The bias factors $\sigma _1^2\chi _{10}$ and $\sigma _2^2\chi _{01}$ of dark matter haloes identified in the N-body simulations at z = 0 are shown as filled (green) circle and (blue) triangle, respectively. Error bars indicate the scatter among six realizations. The horizontal dashed (green) line at −3/2 and the dashed (blue) curve are the corresponding ESP predictions. The dotted (blue) curve is $\sigma _2^2\chi _{01}$ in a model where haloes are allowed to collapse in filamentary-like structures. The solid curves are our final predictions, which take into account the offset between peak position and protohalo centre-of-mass (see text for details).](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/441/2/10.1093/mnras/stu680/2/m_stu680fig5.jpeg?Expires=1722373133&Signature=rWM0a5c9nbS44WVKe~407ZH5qorGGCGT~kql3vWxsgPrHmCXbWI4S~l8opD4wk7~C~-kMVg0bFu2s7xVNAV7w1p29eZOObKbjED7LvgF7A8JHDl1sdr0-vTIO32LpQFgYJ5-4nDmQBmto8CpQI1NokrXbv2vvAHIB7trESPSHqvKmGnDVhTvshNyCRstxK2wmYZhtWP3-fdiU3aJnDKTzGNlOEt8ca59YDRpo7wgNU71pA-EpvPwXeNAzBbeVtoqxSWjqdA9kJGxj07uFZgEx~PZUisDCHdBpzvM1wN5jzlM6fWAkL9vZ9tevBE77oGfQvpvoMjKPZTtEIi7SdNQmA__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
The bias factors |$\sigma _1^2\chi _{10}$| and |$\sigma _2^2\chi _{01}$| of dark matter haloes identified in the N-body simulations at z = 0 are shown as filled (green) circle and (blue) triangle, respectively. Error bars indicate the scatter among six realizations. The horizontal dashed (green) line at −3/2 and the dashed (blue) curve are the corresponding ESP predictions. The dotted (blue) curve is |$\sigma _2^2\chi _{01}$| in a model where haloes are allowed to collapse in filamentary-like structures. The solid curves are our final predictions, which take into account the offset between peak position and protohalo centre-of-mass (see text for details).
4.3 Interpretation of the measurements
To begin with, we note that, if haloes were forming out of randomly distributed patches in the initial conditions, then both χ10 and χ01 would be zero since 〈3η2〉 = 3 and 〈5ζ2〉 = 5 for random field points.
5 CONCLUSION
Dark matter haloes and galaxies are inherently biased relative to the mass density field, and this bias can manifest itself not only in n-point statistics such as the power spectrum or bispectrum, but also in simpler one-point statistics. In this work, we took advantage of this to ascertain the importance of certain non-local Lagrangian bias factors independently of a two-point measurement. We extended the cross-correlation technique of Musso et al. (2012) to χ2-distributed variables, focusing on the quadratic terms |$\eta ^2(\mathrm{\boldsymbol x})$| and |$\zeta ^2(\mathrm{\boldsymbol x})$| (see equation 20) which arise from the peak constraint and for which we have theoretical predictions. In principle, however, our approach could be applied to measure the Lagrangian bias factor associated with any χ2-distributed variable such as the tidal shear for instance. We validated our method with peaks of Gaussian random field before applying it to a catalogue of dark matter haloes with mass M > 1013 M⊙ h−1. Including an offset between the protohalo centre-of-mass and the peak position in the modelling (motivated by the analysis of Ludlow & Porciani 2011), we were able to reproduce our measurements of the non-local bias |$\sigma _1^2\chi _{10}$|. Our result χ10 < 0 is consistent with the findings of Ludlow & Porciani (2011), who demonstrated that protohaloes with M > 1013 M⊙ h−1 preferentially form near initial density peaks (χ10 ≡ 0 for a random distribution). However, we were unable to explain the measurements of |$\sigma _2^2\chi _{01}$|, even with the additional assumption that a fraction of the haloes collapse from filamentary-like structures rather than density peaks. We speculate that a dependence of the halo Lagrangian bias on |$s_2({\boldsymbol x})$| might be needed to explain this discrepancy.
The dependence on |$\eta ^2({\boldsymbol x})$| induces a correction |$-2\chi _{10}(\mathrm{\boldsymbol k}_1\cdot \mathrm{\boldsymbol k}_2)$| to the halo bias which, for collinear wavevectors |$\mathrm{\boldsymbol k}_1$| and |$\mathrm{\boldsymbol k}_2$| of wavenumber 0.1 h−1 Mpc, is Δb ≈ 0.02 (0.05) and ≈0.30 (0.88) for haloes of mass M = 1013 and 1014 M⊙ h−1 at redshift z = 0 (1), respectively. Relative to the evolved, linear halo bias |$b_1^{\rm E}\equiv 1 + b_{100}$|, the fractional correction is |$\Delta b/b_1^{\rm E}\sim$|2 per cent and ∼15 per cent for the same low and high halo mass in the redshift range 0 < z < 1. Hence, this correction can safely be ignored for M = 1013 h−1 Mpc, but it could become relevant at larger halo masses.
We also refined the ESP approach of PSD so that clustering statistics can be straightforwardly computed from the (effective) bias expansion equation (19) (following the prescription detailed in Desjacques 2013). We checked that the predicted halo mass function, from which all the bias factors can be derived, agrees well with the numerical data. However, some of the model ingredients, especially the filtering of the density field, will have to be better understood if one wants to make predictions that are also accurate at small scales.
VD would like to thank the Perimeter Institute for Theoretical Physics and CCPP at New York University for their hospitality while some of this work was being completed there. MB, KCC and VD acknowledge support by the Swiss National Science Foundation.
To facilitate the comparison with other studies, we will call non-local terms all contributions to Lagrangian clustering that are not of the form |$\delta ^n(\mathrm{\boldsymbol x})$|, where |$\delta (\mathrm{\boldsymbol x})$| is the linear mass density field.
We thank Marcello Musso for pointing this out to us.
REFERENCES
APPENDIX A: THE CURVATURE FUNCTION OF DENSITY PEAKS
APPENDIX B: BIVARIATE χ2 DISTRIBUTIONS
![Conditional chi-squared distribution $\chi _k^2(y|x;\epsilon )$ for 3 and 5 d.o.f. Results are shown for several values of x and a fixed cross-correlation coefficient ϵ = 0.7. The dashed (green) curve represents the unconditional distribution $\chi _k^2(y)$.](https://cdn.statically.io/img/oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/mnras/441/2/10.1093/mnras/stu680/2/m_stu680figa1.jpeg?Expires=1722373133&Signature=lWpMVzFKdo24tUzuJKjAvBrIdPF2WDVBPzttuQsX74kNRB-OkYmnbjtCTBKSQj7agEzprLIUd-aTrRoQtJgPIX0EjCGPEZFWcrIlozfo-~mamWaNzUfMFb8YHbpaJHs-ZdKZxJytEEgErwcOlxCU6GxKSXL1Etr4NdQMO11mJ9r0XtMj9UaDorENecQ6fBMkrlaVwWs2VNQFLpqYDnaqU1~WCRhBPK1EdORww0MpiYYRmcyKtZUgr8tksmVrxPOMUm435dEIgVNdOlvpIVBmHcsDnSYs8wlprHhEbpIsEBwE2J5xgELJ1Z2BkzwSFuG5pgdXWYZNIDHLkYzP9j~AnQ__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Conditional chi-squared distribution |$\chi _k^2(y|x;\epsilon )$| for 3 and 5 d.o.f. Results are shown for several values of x and a fixed cross-correlation coefficient ϵ = 0.7. The dashed (green) curve represents the unconditional distribution |$\chi _k^2(y)$|.