Shot noise in clustering power spectra
Nicolas Tessore, Alex Hall · 2025
Read Shot noise in clustering power spectra by Nicolas Tessore, Alex Hall (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-03749.txt.
Introduction
Nicolas Tessore and Alex Hall argue that shot noise in galaxy clustering maps is not noise at all. Their twenty twenty-five paper reworks a standard correction in cosmology. When points sample a field, pairs of a point with itself add a fixed term to the power spectrum.
That term, they show, cannot take the off-values often blamed on sampling quirks. Odd values instead signal real small-scale clustering or its lack. In simulations, how tracers are drawn from the field creates exactly those shifts. The cleanup matters: mislabeling signal as noise biases the cosmic measures that galaxy surveys exist to deliver, from dark energy to the growth of structure.
Catalog record
| CREATOR | Nicolas Tessore, Alex Hall |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-03749
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Shot noise in clustering power spectra Nicolas Tessore, Alex Hall Abstract We show that the ‘shot noise’ bias in angular clustering power spectra observed from discrete samples of points is not noise, but rather a known additive contribution that naturally arises due to degenerate pairs of points. In particular, we show that the true shot noise contribution cannot have a ‘non-Poissonian’ value, even though all point processes with non-trivial two-point statistics are non-Poissonian. Apparent deviations from the ‘Poissonian’ value can arise when significant correlations or anti-correlations are localised on small spatial scales. However, such deviations always correspond to a physical difference in two-point statistics, not a difference in noise. In the context of simulations, if clustering is treated as the tracer of a discretised underlying density field, any sub- or super-Poissonian sampling of the tracer induces such small-scale modifications and vice versa; we show this explicitly using recent innovations in angular power spectrum estimation from discrete catalogues. Finally, we show that the full covariance of clustering power spectra can also be computed explicitly: it depends only on the two-, three- and four-point statistics of the point process. The usual Gaussian covariance approximation appears as one term in the shot noise contribution, which is non-diagonal even for Gaussian random fields and Poisson sampling. \journalinfo The Open Journal of Astrophysics \submitted July 2025 1 Introduction The observed clustering of galaxies is an important probe of cosmological information, either as angular clustering on the sphere ( Dark Energy Survey Collaboration et al. 2022 ; Dvornik et al. 2023 ; More et al. 2023 , recently) or as 3D clustering in redshift space ( DESI Collaboration et al. 2025 , recently) . The two-point statistics of galaxy clustering are principally measured in two ways: as correlation functions in real (or configuration) space, or as power spectra in harmonic (or Fourier) space. The latter are attractive for several reasons. On a practical level, measuring power spectra is typically much faster than measuring correlation functions. In addition, power spectra are typically less correlated between scales compared with their configuration space counterparts, and allow a cleaner separation of non-linear scales. However, clustering power spectra measured from sets of observed points contain an obvious contamination, commonly called ‘shot noise’. Shot noise is an additive contribution to the power spectra with a constant value at all scales. This constant shot noise contribution must therefore in general be subtracted to obtain an estimate of the actual power spectra of interest. It would appear that shot noise arises in the following manner: there is an underlying density field, which contains the two-point statistics of interest, and of which the observed points are ‘tracers’. These tracers are drawn randomly from the underlying density field, following some sampling distribution, and this random sampling introduces additional variance into the measurement of the power spectra. The additional variance is a delta function at zero lag in the correlation function, which translates to a constant additive term — shot noise — in the power spectrum. In this picture, shot noise is closely related to the random sampling of the tracers. Therefore, one might worry that a model for the clustering of the points on very small scales, or indeed the distribution of number counts in very small cells, is required in order to accurately model and subtract the shot noise contribution from measured power spectra. In this work, we show that this is not the case: the shot noise contribution to the power spectrum depends only on the observed number of points, and not on their sampling distribution. The shot noise contribution thus always has the ‘Poissonian’ value, and any apparent deviation from said value corresponds to a change in the underlying two-point statistics. These two-point statistics describe real physical effects ( Baldauf et al. 2013 ; Ginzburg et al. 2017 ; Lacasa 2018 , e.g.) that have non-noise power spectra, even if this is not apparent at the scales at which the spectra are measured. In what follows, we establish a series of facts about point processes and shot noise. • The distribution of clustered points is always non-Poissonian (Section 2 ). • The ‘shot noise’ contribution to clustering power spectra is always ‘Poissonian’ (Section 3 ). • Any apparent deviation from the Poissonian shot noise contribution comes from a difference in two-point statistics, not shot noise (Section 4 ). • The shot noise contribution to the covariance of observed angular power spectra can be computed exactly (Section 5 ). A short summary of the conclusions from these facts is given in Section 6 . 2 Clustering is always non-Poissonian For a homogeneous and isotropic point process such as galaxy clustering, the joint probability distribution of a pair of points \hat{u} and \hat{u}^{\prime} can be written in terms of the two-point correlation function w as ( Peebles 1973 ) p(\hat{u},\hat{u}^{\prime})\propto 1+w(\theta)\;, (1) where the angle \theta=\arccos(\hat{u}\cdot\hat{u}^{\prime}) is the angular distance between \hat{u} and \hat{u}^{\prime} . The joint probability distribution ( 1 ) fully describes the two-point statistics of a homogeneous and isotropic point process, and no further assumptions about the nature of the point process, such as being a Poisson process, are necessary. In fact, a correlated point process with w\not\equiv 0 cannot be Poissonian. This appears to contradict the usual recipe for simulations: generate a density field \delta with correlation function w , then draw the number of points in a small region \mathrm{d}\hat{u} from a Poisson distribution with rate \bar{n}\,[1+\delta(\hat{u})]\,\mathrm{d}\hat{u} , where \bar{n} is the mean density of the field. By construction, the resulting number density is Poissonian — conditional on \delta . Over realisations of \delta , the covariance between the numbers of points N_{1} and N_{2} in a pair of non-overlapping regions \Omega_{1} and \Omega_{2} is \langle N_{1}N_{2}\rangle-\langle N_{1}\rangle\,\langle N_{2}\rangle=\bar{n}_{1}\bar{n}_{2}\int\limits_{\mathclap{\Omega_{1}\times\Omega_{2}}}w(\theta)\,\mathrm{d}\hat{u}_{1}\,\mathrm{d}\hat{u}_{2}\;. (2) Complete independence would require the integral to vanish for arbitrary \Omega_{1}\times\Omega_{2} . The point process can therefore only be Poissonian if w\equiv 0 ( Layzer 1956 ) . 3 Shot noise is always Poissonian If the joint probability distribution ( 1 ) fully describes the two-point statistics of the point process, the distribution of the points – Poissonian or not – should not enter any observable two-point quantity. In particular, the ‘shot noise’ contribution to the observed power spectrum cannot depend on the distribution of the points. This can be checked explicitly by computing the value of the shot noise ( Euclid Collaboration: Tessore et al. 2025 ) . First, write the observed number density n as a sum of Dirac delta functions in the observed positions \hat{u}_{1},\hat{u}_{2},\ldots , n(\hat{u})=\sum_{i}\delta(\hat{u}-\hat{u}_{i})\;. (3) Next, transform the observed number density n from real space to harmonic space, n_{\ell m}=\int\!n(\hat{u})\,Y_{\ell m}^{*}(\hat{u})\,\mathrm{d}\hat{u}\;, (4) where Y_{\ell m} are the spherical harmonics. Applying the transform ( 4 ) to the number density ( 3 ) and using the defining property of the delta function yields coefficients n_{\ell m}=\sum_{i}Y_{\ell m}^{*}(\hat{u}_{i})\;. (5) In other words, the transform of a discrete set of points are the sums of function values of Y_{\ell m} in the observed points. Finally, consider the observed two-point statistics in harmonic space. Given two not necessarily distinct samples A and B of points, the observed angular power spectrum C_{\ell}^{AB} is C_{\ell}^{AB}=\frac{1}{2\ell+1}\sum_{m}n_{\ell m}^{A*}n_{\ell m}^{B}\;. (6) Importantly, the