Search

We calculate the abundance of UV-bright galaxies in the presence of ultralight axion (ULA) dark matter (DM), finding that axions suppress their formation with a non-trivial dependence on redshift and luminosity. We set limits on axion DM using both Planck cosmic microwave background (CMB) and UV luminosity function (UVLF) data. We exclude a single axion as all the DM for $m_{ax} < 10^{-21.6}$ eV and limit axions with $-26 < \log( m_{ax}/{eV}) < -23$ to be less than $22\%$ of the DM (both limits at $95\%$ credibility). These limits use UVLF measurements from 24,000 sources from the Hubble Space Telescope (HST) that probe small-scale structure at redshifts $4 < z < 10$. We marginalize over a parametric model that connects halo mass and UV luminosity that has been shown to match hydrodynamical simulations. Our results bridge a window in axion mass and DM fraction previously unconstrained by cosmological data, between large-scale CMB and galaxy clustering and the small-scale Lyman-$\alpha$ forest. These high-$z$ measurements provide a powerful consistency check of low-$z$ tests of axion DM, which include the recent hint for a sub-dominant ULA DM fraction in Lyman-$\alpha$ forest data. We also consider a sample of 25 spectroscopically-confirmed high-$z$ galaxies from the James Webb Space Telescope (JWST). We find that these data are consistent with the HST UVLF assuming $\Lambda$CDM and our flexible parametric model of UV luminosity. Combining HST and JWST UVLF data does not improve our constraints beyond HST alone, but future JWST measurements have the potential to improve these results significantly. We also find an excess of low-mass halos ($< 10^9 M_\odot$) at $z < 3$, which could be probed by sub-galactic structure probes (e.g., stellar streams, satellite galaxies and strong lensing)., Comment: 23 pages, 20 figures. Minor changes to match version submitted to journal, references added

We find that combined Planck cosmic microwave background, baryon acoustic oscillations and supernovae data analyzed under $\Lambda$CDM are in 4.9$\sigma$ tension with eBOSS Ly$\alpha$ forest in inference of the linear matter power spectrum at wavenumber $\sim 1 h\,\mathrm{Mpc}^{-1}$ and redshift = 3. Model extensions can alleviate this tension: running in the tilt of the primordial power spectrum ($\alpha_\mathrm{s} \sim -0.01$); a fraction $\sim (1 - 5)\%$ of ultra-light axion dark matter (DM) with particle mass $\sim 10^{-25}$ eV or warm DM with mass $\sim 90$ eV. The new DESI survey, coupled with high-accuracy modeling, will help distinguish the source of this discrepancy., Comment: 15 pages, 9 figures. Minor changes to match version submitted to journal, references added

The era of precision cosmology allows us to test the composition of the dark matter. Mixed ultralight or fuzzy dark matter (FDM) is a cosmological model with dark matter composed of a combination of particles of mass $m\leq 10^{-20}\;\mathrm{eV}$, with an astrophysical de Broglie wavelength, and particles with a negligible wavelength sharing the properties of cold dark matter (CDM). In this work, we simulate cosmological volumes with a dark matter wave function for the ultralight component coupled gravitationally to CDM particles. We investigate the impact of a mixture of CDM and FDM in various proportions $(0\%,\;1\%,\;10\%,\;50\%,\;100\%)$ and for ultralight particle masses ranging over five orders of magnitude $(2.5\times 10^{-25}\;\mathrm{eV}-2.5\times 10^{-21}\;\mathrm{eV})$. To track the evolution of density perturbations in the non-linear regime, we adapt the simulation code AxioNyx to solve the CDM dynamics coupled to a FDM wave function obeying the Schr\"odinger-Poisson equations. We obtain the non-linear power spectrum and study the impact of the wave effects on the growth of structure on different scales. We confirm that the steady-state solution of the Schr\"odinger-Poisson system holds at the center of halos in the presence of a CDM component when it composes $50\%$ or less of the dark matter but find no stable density core when the FDM accounts for $10\%$ or less of the dark matter. We implement a modified friends-of-friends halo finder and find good agreement between the observed halo abundance and the predictions from the adapted halo model axionHMCode., Comment: Added references

We search for ultra-light axions as dark matter (DM) and dark energy particle candidates, for axion masses $10^{-32}\,\mathrm{eV} \leq m_\mathrm{a} \leq 10^{-24}\,\mathrm{eV}$, by a joint analysis of cosmic microwave background (CMB) and galaxy clustering data -- and consider if axions can resolve the tension in inferred values of the matter clustering parameter $S_8$. We give legacy constraints from Planck 2018 CMB data, improving 2015 limits on the axion density $\Omega_\mathrm{a} h^2$ by up to a factor of three; CMB data from the Atacama Cosmology Telescope and the South Pole Telescope marginally weaken Planck bounds at $m_\mathrm{a} = 10^{-25}\,\mathrm{eV}$, owing to lower (and theoretically-consistent) gravitational lensing signals. We jointly infer, from Planck CMB and full-shape galaxy power spectrum and bispectrum data from the Baryon Oscillation Spectroscopic Survey (BOSS), that axions are, today, $< 10\%$ of the DM for $m_\mathrm{a} \leq 10^{-26}\,\mathrm{eV}$ and $< 1\%$ for $10^{-30}\,\mathrm{eV} \leq m_\mathrm{a} \leq 10^{-28}\,\mathrm{eV}$. BOSS data strengthen limits, in particular at higher $m_\mathrm{a}$ by probing high-wavenumber modes ($k < 0.4 h\,\mathrm{Mpc}^{-1}$). BOSS alone finds a preference for axions at $2.7 \sigma$, for $m_\mathrm{a} = 10^{-26}\,\mathrm{eV}$, but Planck disfavours this result. Nonetheless, axions in a window $10^{-28}\,\mathrm{eV} \leq m_\mathrm{a} \leq 10^{-25}\,\mathrm{eV}$ can improve consistency between CMB and galaxy clustering data, e.g., reducing the $S_8$ discrepancy from $2.7 \sigma$ to $1.6 \sigma$, since these axions suppress structure growth at the $8 h^{-1}\,\mathrm{Mpc}$ scales to which $S_8$ is sensitive. We expect improved constraints with upcoming high-resolution CMB and galaxy lensing and future galaxy clustering data, where we will further assess if axions can restore cosmic concordance., Comment: 53 pages, 22 figures. Minor changes to match version accepted for publication in JCAP

Three-dimensional correlations of the Lyman-$\alpha$ (Ly$\alpha$) forest and cross correlations between the Ly$\alpha$ forest and quasars have been measured on large scales, allowing a precise measurement of the baryon acoustic oscillation (BAO) feature at redshifts $z>2$. These 3D correlations are often modelled using linear perturbation theory, but full-shape analyses to extract cosmological information beyond BAO will require more realistic models capable of describing non-linearities present at smaller scales. We present a measurement of the Ly$\alpha$ forest flux power spectrum from large hydrodynamic simulations -- the Sherwood simulations -- and compare it to different models describing the small-scale deviations from linear theory. We confirm that the model presented in Arinyo-i-Prats et al. (2015) fits the measured 3D power up to $k=10\, h\rm{Mpc^{-1}}$ with an accuracy better than 5%, and show that the same model can also describe the 1D correlations with similar precision. We also present, for the first time, an equivalent study for the cross-power spectrum of halos with the Ly$\alpha$ forest, and we discuss different challenges we face when modelling the cross-power spectrum beyond linear scales. We make all our measured power spectra public in \url{https://github.com/andreufont/sherwood_p3d}. This study is a step towards joint analyses of 1D and 3D flux correlations, and towards using the quasar-Ly$\alpha$ cross-correlation beyond BAO analyses., Comment: 31 pages, 19 figures

The nature of dark matter is one of the major puzzles of fundamental physics, integral to the understanding of our universe across almost every epoch. The search for dark matter takes place at different energy scales, and use data ranging from particle colliders to astrophysical surveys. We focus here on CMB-S4, a future ground-based Cosmic Microwave Background (CMB) experiment, which is expected to provide exquisite measurements of the CMB temperature and polarization anisotropies. These measurements (on their own and in combination with other surveys) will allow for new means to shed light on the nature of dark matter., Comment: Contribution to Snowmass 2021. 20 pages, 4 figures

We set the strongest limits to-date on the velocity-independent dark matter (DM) - proton cross section $\sigma$ for DM masses $m = 10\,\mathrm{keV}$ to $100\,\mathrm{GeV}$, using large-scale structure traced by the Lyman-alpha forest: e.g., a 95% lower limit $\sigma < 6 \times 10^{-30}\,\mathrm{cm}^2$, for $m = 100\,\mathrm{keV}$. Our results complement direct detection, which has limited sensitivity to sub-GeV DM. We use an emulator of cosmological simulations, combined with data from the smallest cosmological scales used to-date, to model and search for the imprint of primordial DM-proton collisions. Cosmological bounds are improved by up to a factor of 25., Comment: 12 pages, 5 figures. Minor changes to match version accepted for publication in PRL

Gravitational weak lensing by dark matter halos leads to a measurable imprint in the shear correlation function of galaxies. Fuzzy dark matter (FDM), composed of ultralight axion-like particles of mass $m\sim 10^{-22}\text{ eV}$, suppresses the matter power spectrum and shear correlation with respect to standard cold dark matter. We model the effect of FDM on cosmic shear using the optimised halo model \textsc{HMCode}, accounting for additional suppression of the mass function and halo concentration in FDM as observed in $N$-body simulations. We combine Dark Energy Survey year 1 (DES-Y1) data with the \emph{Planck} cosmic microwave background anisotropies to search for shear correlation suppression caused by FDM. We find no evidence of suppression compared to the preferred cold DM model, and thus set a new lower limit to the FDM particle mass. Using a log-flat prior and marginalising over uncertainties related to the non-linear model of FDM, we find a new, independent 95\% C.L. lower limit $\log_{10}m>-23$ combining \emph{Planck} and DES-Y1 shear, an improvement of almost two orders of magnitude on the mass bound relative to CMB-only constraints. Our analysis is largely independent of baryonic modelling, and of previous limits to FDM covering this mass range. Our analysis highlights the most important aspects of the FDM non-linear model for future investigation. The limit to FDM from weak lensing could be improved by up to three orders of magnitude with $\mathcal{O}(0.1)$ arcmin cosmic shear angular resolution, if FDM and baryonic feedback can be simultaneously modelled to high precision in the halo model., Comment: 25 pages,16 figures, updated discussion of redshift evolution, massive neutrinos, version accepted for publication in MNRAS, v3: minor corrections

Ultralight axions and other bosons are dark matter candidates present in many high energy physics theories beyond the Standard Model. In particular, the string axiverse postulates the existence of up to $\mathcal{O}(100)$ light scalar bosons constituting the dark sector. Considering a mixture of axions and cold dark matter, we obtain upper bounds for the axion relic density $\Omega_a h^2 < 0.004$ for axions of mass $10^{-31}\;\mathrm{eV}\leq m_a \leq 10^{-26}\;\mathrm{eV}$ at 95% confidence. We also improve existing constraints by a factor of over 4.5 and 2.1 for axion masses of $10^{-25}$ eV and $10^{-32}$ eV, respectively. We use the Fourier-space galaxy clustering statistics from the Baryon Oscillation Spectroscopic Survey (BOSS) and demonstrate how galaxy surveys break important degeneracies in the axion parameter space compared to the cosmic microwave background (CMB). We test the validity of the effective field theory of large-scale structure approach to mixed ultralight axion dark matter by making our own mock galaxy catalogs and find an anisotropic ultralight axion signature in the galaxy quadrupole. We also observe an enhancement of the linear galaxy bias from 1.8 to 2.4 when allowing for 5% of the dark matter to be composed of a $10^{-28}$ eV axion in our simulations. Finally, we develop an augmented interpolation scheme allowing a fast computation of the axion contribution to the linear matter power spectrum leading to a 70% reduction of the computational cost for the full Monte Carlo Markov chains analysis., Comment: Updated references

Interpreting observations of the Lyman-$\alpha$ forest flux power spectrum requires interpolation between a small number of expensive simulations. We present a Gaussian process emulator modelling the 1D flux power spectrum as a function of the amplitude and slope of the small-scale linear matter power spectrum, and the state of the intergalactic medium at the epoch of interest ($2 < z < 4$). This parameterisation enables the prediction of the flux power spectrum in extended cosmological models that are not explicitly included in the training set, eliminating the need to construct bespoke emulators for a number of extensions to $\Lambda$CDM. Our emulator is appropriate for cosmologies in which the linear matter power spectrum is described to percent level accuracy by just an amplitude and slope across the epoch of interest, and in the regime probed by eBOSS/DESI data. We demonstrate this for massive neutrino cosmologies, where the emulator is able to predict the flux power spectrum in a $\Sigma m_\nu=0.3$ eV neutrino cosmology to sub-percent accuracy, without including massive neutrinos in the training simulations. Further parameters would be required to describe models with sharp features in the linear power, such as warm or light axion dark matter. This work will facilitate the combination of upcoming DESI data with observations of the cosmic microwave background, to obtain constraints on neutrino mass and other extensions to $\Lambda$CDM cosmology., Comment: 26 pages, 11 figures. Minor changes to match version published in JCAP

We present a general framework for obtaining robust bounds on the nature of dark matter using cosmological $N$-body simulations and Lyman-alpha forest data. We construct an emulator of hydrodynamical simulations, which is a flexible, accurate and computationally-efficient model for predicting the response of the Lyman-alpha forest flux power spectrum to different dark matter models, the state of the intergalactic medium (IGM) and the primordial power spectrum. The emulator combines a flexible parameterization for the small-scale suppression in the matter power spectrum arising in "non-cold" dark matter models, with an improved IGM model. We then demonstrate how to optimize the emulator for the case of ultra-light axion dark matter, presenting tests of convergence. We also carry out cross-validation tests of the accuracy of flux power spectrum prediction. This framework can be optimized for the analysis of many other dark matter candidates, e.g., warm or interacting dark matter. Our work demonstrates that a combination of an optimized emulator and cosmological "effective theories," where many models are described by a single set of equations, is a powerful approach for robust and computationally-efficient inference from the cosmic large-scale structure., Comment: 23 pages, 11 figures. Minor changes to match version accepted for publication in PRD. Companion letter: arXiv:2007.12705 "Strong bound on canonical ultra-light axion dark matter from the Lyman-alpha forest" (K. Rogers & H. Peiris)

We present a new bound on the ultra-light axion (ULA) dark matter mass $m_\text{a}$, using the Lyman-alpha forest to look for suppressed cosmic structure growth: a 95% lower limit $m_\text{a} > 2 \times 10^{-20}\,\text{eV}$. This strongly disfavors ($> 99.7\%$ credibility) the canonical ULA with $10^{-22}\,\text{eV} < m_\text{a} < 10^{-21}\,\text{eV}$, motivated by the string axiverse and solutions to possible tensions in the cold dark matter model. We strengthen previous equivalent bounds by about an order of magnitude. We demonstrate the robustness of our results using an optimized emulator of improved hydrodynamical simulations., Comment: 8 pages, 4 figures. Minor changes to match version accepted for publication in PRL

Using a suite of hydrodynamical simulations with cold dark matter, baryons, and neutrinos, we present a detailed study of the effect of massive neutrinos on the 1-D and 3-D flux power spectra of the Lyman-$\alpha$ (Ly$\alpha$) forest. The presence of massive neutrinos in cosmology induces a scale- and time-dependent suppression of structure formation that is strongest on small scales. Measuring this suppression is a key method for inferring neutrino masses from cosmological data, and is one of the main goals of ongoing and future surveys like eBOSS, DES, LSST, Euclid or DESI. The clustering in the Ly$\alpha$ forest traces the quasi-linear power at late times and on small scales. In combination with observations of the cosmic microwave background, the forest therefore provides some of the tightest constraints on the sum of the neutrino masses. However there is a well-known degeneracy between $\Sigma m_{\nu}$ and the amplitude of perturbations in the linear matter power spectrum. We study the corresponding degeneracy in the 1-D flux power spectrum of the Ly$\alpha$ forest, and for the first time also study this degeneracy in the 3-D flux power spectrum. We show that the non-linear effects of massive neutrinos on the Ly$\alpha$ forest, beyond the effect of linear power amplitude suppression, are negligible, and this degeneracy persists in the Ly$\alpha$ forest observables to a high precision. We discuss the implications of this degeneracy for choosing parametrisations of the Ly$\alpha$ forest for cosmological analysis., Comment: 18 pages, 7 figures

We present methods for interpolating between the 1-D flux power spectrum of the Lyman-$\alpha$ forest, as output by cosmological hydrodynamic simulations. Interpolation is necessary for cosmological parameter estimation due to the limited number of simulations possible. We construct an emulator for the Lyman-$\alpha$ forest flux power spectrum from $21$ small simulations using Latin hypercube sampling and Gaussian process interpolation. We show that this emulator has a typical accuracy of 1.5% and a worst-case accuracy of 4%, which compares well to the current statistical error of 3 - 5% at $z < 3$ from BOSS DR9. We compare to the previous state of the art, quadratic polynomial interpolation. The Latin hypercube samples the entire volume of parameter space, while quadratic polynomial emulation samples only lower-dimensional subspaces. The Gaussian process provides an estimate of the emulation error and we show using test simulations that this estimate is reasonable. We construct a likelihood function and use it to show that the posterior constraints generated using the emulator are unbiased. We show that our Gaussian process emulator has lower emulation error than quadratic polynomial interpolation and thus produces tighter posterior confidence intervals, which will be essential for future Lyman-$\alpha$ surveys such as DESI., Comment: 28 pages, 10 figures, accepted to JCAP with minor changes. Fixes a typographical error in eq. 2.3

The Lyman-alpha forest provides strong constraints on both cosmological parameters and intergalactic medium astrophysics, which are forecast to improve further with the next generation of surveys including eBOSS and DESI. As is generic in cosmological inference, extracting this information requires a likelihood to be computed throughout a high-dimensional parameter space. Evaluating the likelihood requires a robust and accurate mapping between the parameters and observables, in this case the 1D flux power spectrum. Cosmological simulations enable such a mapping, but due to computational time constraints can only be evaluated at a handful of sample points; "emulators" are designed to interpolate between these. The problem then reduces to placing the sample points such that an accurate mapping is obtained while minimising the number of expensive simulations required. To address this, we introduce an emulation procedure that employs Bayesian optimisation of the training set for a Gaussian process interpolation scheme. Starting with a Latin hypercube sampling (other schemes with good space-filling properties can be used), we iteratively augment the training set with extra simulations at new parameter positions which balance the need to reduce interpolation error while focussing on regions of high likelihood. We show that smaller emulator error from the Bayesian optimisation propagates to smaller widths on the posterior distribution. Even with fewer simulations than a Latin hypercube, Bayesian optimisation shrinks the 95% credible volume by 90% and, e.g., the 1 sigma error on the amplitude of small-scale primordial fluctuations by 38%. This is the first demonstration of Bayesian optimisation applied to large-scale structure emulation, and we anticipate the technique will generalise to many other probes such as galaxy clustering, weak lensing and 21cm., Comment: 23 pages, 4 figures. Minor changes to match version published in JCAP

We test a method to reduce unwanted sample variance when predicting Lyman-$\alpha$ (ly$\alpha$) forest power spectra from cosmological hydrodynamical simulations. Sample variance arises due to sparse sampling of modes on large scales and propagates to small scales through non-linear gravitational evolution. To tackle this, we generate initial conditions in which the density perturbation amplitudes are {\it fixed} to the ensemble average power spectrum -- and are generated in {\it pairs} with exactly opposite phases. We run $50$ such simulations ($25$ pairs) and compare their performance against $50$ standard simulations by measuring the ly$\alpha$ 1D and 3D power spectra at redshifts $z=2$, 3, and 4. Both ensembles use periodic boxes of $40$ Mpc/h containing $512^3$ particles each of dark matter and gas. As a typical example of improvement, for wavenumbers $k=0.25$ h/Mpc at $z=3$, we find estimates of the 1D and 3D power spectra converge $34$ and $12$ times faster in a paired-fixed ensemble compared with a standard ensemble. We conclude that, by reducing the computational time required to achieve fixed accuracy on predicted power spectra, the method frees up resources for exploration of varying thermal and cosmological parameters -- ultimately allowing the improved precision and accuracy of statistical inference., Comment: 15 pages, 3 figures

Correlations measured in three dimensions in the Lyman-alpha forest are contaminated by the presence of the damping wings of high column density (HCD) absorbing systems of neutral hydrogen (HI; having column densities $N(\mathrm{HI}) > 1.6\times10^{17}\,\mathrm{atoms}\,\mathrm{cm}^{-2}$), which extend significantly beyond the redshift-space location of the absorber. We measure this effect as a function of the column density of the HCD absorbers and redshift by measuring 3D flux power spectra in cosmological hydrodynamical simulations from the Illustris project. Survey pipelines exclude regions containing the largest damping wings. We find that, even after this procedure, there is a scale-dependent correction to the 3D Lyman-alpha forest flux power spectrum from residual contamination. We model this residual using a simple physical model of the HCD absorbers as linearly biased tracers of the matter density distribution, convolved with their Voigt profiles and integrated over the column density distribution function. We recommend the use of this model over existing models used in data analysis, which approximate the damping wings as top-hats and so miss shape information in the extended wings. The simple 'linear Voigt model' is statistically consistent with our simulation results for a mock residual contamination up to small scales ($|k| < 1\,h\,\mathrm{Mpc}^{-1}$). It does not account for the effect of the highest column density absorbers on the smallest scales (e.g., $|k| > 0.4\,h\,\mathrm{Mpc}^{-1}$ for small damped Lyman-alpha absorbers; HCD absorbers with $N(\mathrm{HI}) \sim 10^{21}\,\mathrm{atoms}\,\mathrm{cm}^{-2}$). However, these systems are in any case preferentially removed from survey data. Our model is appropriate for an accurate analysis of the baryon acoustic oscillations feature. It is additionally essential for reconstructing the full shape of the 3D flux power spectrum., Comment: 13 pages, 11 figures. Minor changes to match version published in MNRAS

We measure the effect of high column density absorbing systems of neutral hydrogen (HI) on the one-dimensional (1D) Lyman-alpha forest flux power spectrum using cosmological hydrodynamical simulations from the Illustris project. High column density absorbers (which we define to be those with HI column densities $N(\mathrm{HI}) > 1.6 \times 10^{17}\,\mathrm{atoms}\,\mathrm{cm}^{-2}$) cause broadened absorption lines with characteristic damping wings. These damping wings bias the 1D Lyman-alpha forest flux power spectrum by causing absorption in quasar spectra away from the location of the absorber itself. We investigate the effect of high column density absorbers on the Lyman-alpha forest using hydrodynamical simulations for the first time. We provide templates as a function of column density and redshift, allowing the flexibility to accurately model residual contamination, i.e., if an analysis selectively clips out the largest damping wings. This flexibility will improve cosmological parameter estimation, e.g., allowing more accurate measurement of the shape of the power spectrum, with implications for cosmological models containing massive neutrinos or a running of the spectral index. We provide fitting functions to reproduce these results so that they can be incorporated straightforwardly into a data analysis pipeline., Comment: 11 pages, 6 figures. Minor changes to match version published in MNRAS

We present Spin-SILC, a new foreground component separation method that accurately extracts the cosmic microwave background (CMB) polarisation $E$ and $B$ modes from raw multifrequency Stokes $Q$ and $U$ measurements of the microwave sky. Spin-SILC is an internal linear combination method that uses spin wavelets to analyse the spin-2 polarisation signal $P = Q + iU$. The wavelets are additionally directional (non-axisymmetric). This allows different morphologies of signals to be separated and therefore the cleaning algorithm is localised using an additional domain of information. The advantage of spin wavelets over standard scalar wavelets is to simultaneously and self-consistently probe scales and directions in the polarisation signal $P = Q + iU$ and in the underlying $E$ and $B$ modes, therefore providing the ability to perform component separation and $E$-$B$ decomposition concurrently for the first time. We test Spin-SILC on full-mission Planck simulations and data and show the capacity to correctly recover the underlying cosmological $E$ and $B$ modes. We also demonstrate a strong consistency of our CMB maps with those derived from existing component separation methods. Spin-SILC can be combined with the pseudo- and pure $E$-$B$ spin wavelet estimators presented in a companion paper to reliably extract the cosmological signal in the presence of complicated sky cuts and noise. Therefore, it will provide a computationally-efficient method to accurately extract the CMB $E$ and $B$ modes for future polarisation experiments., Comment: 13 pages, 9 figures. Minor changes to match version published in MNRAS. Map products available at http://www.silc-cmb.org. Companion paper: arXiv:1605.01414 "Wavelet reconstruction of pure E and B modes for CMB polarisation and cosmic shear analyses" (B. Leistedt et al.)

We present new clean maps of the CMB temperature anisotropies (as measured by Planck) constructed with a novel internal linear combination (ILC) algorithm using directional, scale-discretised wavelets --- Scale-discretised, directional wavelet ILC or SILC. Directional wavelets, when convolved with signals on the sphere, can separate the anisotropic filamentary structures which are characteristic of both the CMB and foregrounds. Extending previous component separation methods, which use the frequency, spatial and harmonic signatures of foregrounds to separate them from the cosmological background signal, SILC can additionally use morphological information in the foregrounds and CMB to better localise the cleaning algorithm. We test the method on Planck data and simulations, demonstrating consistency with existing component separation algorithms, and discuss how to optimise the use of morphological information by varying the number of directional wavelets as a function of spatial scale. We find that combining the use of directional and axisymmetric wavelets depending on scale could yield higher quality CMB temperature maps. Our results set the stage for the application of SILC to polarisation anisotropies through an extension to spin wavelets., Comment: 15 pages, 13 figures. Minor changes to match version published in MNRAS. Map products available at http://www.silc-cmb.org