The Properties of Warm Dark Matter Haloes

Mon. Not. R. Astron. Soc. 000, ??–21 (2012) Printed 20 May 2014 (MN LATEX style file v2.
2)
The properties of warm dark matter haloes
Mark R. Lovell1⋆ , Carlos S. Frenk1 , Vincent R. Eke1, Adrian Jenkins1 , Liang Gao2,1,
and
1
Tom Theuns1,3
Institute for Computational Cosmology, Durham University, South Road, Durham, UK, DH1 3LE
arXiv:1308.1399v2 [astro-ph.CO] 19 May 2014
2 National Astronomical Observatories, Chinese Academy of Science, Beijing, 100012, China

3 Department of Physics, University of Antwerp, Groenenborgerlaan 171, B-2020 Antwerpen, Belgium
Accepted *** Received ***; in original form ***
ABSTRACT
Well-motivated elementary particle candidates for the dark matter, such as the sterile
neutrino, behave as warm dark matter (WDM). For particle masses of order a keV,
free streaming produces a cutoff in the linear fluctuation power spectrum at a scale
corresponding to dwarf galaxies. We investigate the abundance and structure of WDM
haloes and subhaloes on these scales using high resolution cosmological N-body simu-
lations of galactic haloes of mass similar to the Milky Way’s. On scales larger than the
free-streaming cutoff, the initial conditions have the same power spectrum and phases
as one of the cold dark matter (CDM) haloes previously simulated by Springel et al.
as part of the Virgo consortium Aquarius project. We have simulated four haloes with
WDM particle masses in the range 1.5 − 2.3 keV and, for one case, we have carried
out further simulations at varying resolution. N-body simulations in which the power
spectrum cutoff is resolved are known to undergo artificial fragmentation in filaments
producing spurious clumps which, for small masses (< 107 M⊙ in our case) outnum-
ber genuine haloes. We have developed a robust algorithm to identify these spurious
objects and remove them from our halo catalogues. We find that the WDM subhalo
mass function is suppressed by well over an order magnitude relative to the CDM
case for masses < 109 M⊙ . Requiring that there should be at least as many subhaloes
as there are observed satellites in the Milky Way leads to a conservative lower limit
to the (thermal equivalent) WDM particle mass of ∼ 1.5keV. WDM haloes and sub-
haloes have cuspy density distributions that are well described by NFW or Einasto
profiles. Their central densities are lower for lower WDM particle masses and none
of the models we have considered suffer from the “too big to fail” problem recently
highlighted by Boylan-Kolchin et al.
Key words: cosmology: dark matter – galaxies: dwarf
1 INTRODUCTION ligible thermal velocities during the era of structure forma-

tion.
The identity of the dark matter remains one of the central
unsolved problems in cosmology. Various lines of evidence, More recently, particle candidates that have appreciable
for example, data on the cosmic microwave background ra- thermal velocities at early times, and thus behave as warm,
diation, indicate that the dark matter is made up of non- rather than cold, dark matter have received renewed atten-
baryonic elementary particles (e.g. Larson et al. 2011), but tion. The best-known example is a sterile neutrino which, if
exactly which kind (or kinds) of particle are involved is not it occurs as a triplet, could explain observed neutrino oscil-
yet known. For the past thirty years or so attention has fo- lation rates and baryogenesis (e.g. Asaka & Shaposhnikov
cused on cold dark matter (CDM) (see Frenk & White 2012, 2005). This model is known as the neutrino minimal stan-
for a review), for which there are well-motivated candidates dard model (νMSM; Boyarsky et al. 2009c,b); a list of alter-
from particle physics, for example, the lightest supersym- native models may be found in Kusenko (2009). Warm parti-
metric particle or neutralino (Ellis et al. 1984), or the axion cles are relativistic when they decouple from the primordial
(Preskill et al. 1983). Cold dark matter particles have neg- plasma and become non-relativistic during the radiation-
dominated era. This causes the particles to free stream
out of small perturbations, giving rise to a cutoff in the
⋆ E-mail: [email protected] linear matter power spectrum and an associated suppre-
c 2012 RAS

2 M. R. Lovell et al.
sion of structure formation on small scales. When the par- which show that CDM haloes and their subhaloes have cen-
ticles collect at the centres of dark matter haloes, their tral cusps (Navarro et al. 1996b, 1997; Springel et al. 2005).
non-negligible thermal velocities reduce their phase-space Strigari et al. (2010) explicitly showed that it is always
density compared to the CDM case and this can result possible to find CDM subhaloes formed in the Aquarius
in the formation of a ‘core’ in the density profile whose high resolution simulations of galactic haloes (Springel et al.
size varies inversely with the velocity dispersion of the 2008a) that are consistent with these data, however the sub-
halo (Hogan & Dalcanton 2000). However, recent analyt- haloes that best fit the kinematical data for the bright satel-
ical and numerical work (Macciò et al. 2012; Shao et al. lites turn out not to be the most massive ones, as would
2013; Macciò et al. 2013) has shown that the resulting cores naturally be expected for these bright satellites. This sur-
are astrophysically uninteresting being, in particular, sig- prising result was explored in detail in the Aquarius simu-
nificantly smaller than the cores claimed to be present in lations by Boylan-Kolchin et al. (2011, 2012), who dubbed
dwarf satellites of the Milky Way (e.g. Gilmore et al. 2007; it the ‘too big to fail’ problem; it was also found in gasdy-
de Vega & Sanchez 2010). namic simulations of Aquarius haloes by Parry et al. (2012).
On comoving scales much larger than the free-streaming The discrepancy has attracted a great deal of attention be-
cutoff, the formation of structure proceeds in very similar cause it could potentially rule out the existence of CDM.
ways whether the dark matter is cold or warm and so cur- Possibly related problems include the paucity of galaxies
rent astronomical observations on those scales (larger than in voids (Tikhonov et al. 2009), and the local HI velocity
∼ 1Mpc) cannot distinguish between these two very differ- width function (Zavala et al. 2008; Papastergis et al. 2011)
ent types of dark matter particles. Successes of the CDM (but see Sawala et al. 2013).
paradigm, such as the remarkable agreement of its predic- A number of solutions to the ‘too big to fail’ problem
tions (in a universe dominated by a constant vacuum energy, have now been proposed. Within the CDM context, perhaps
Λ) with observations of temperature fluctuations in the cos- the simplest is that the virial mass of the Milky Way halo is
mic microwave background radiation (e.g. Komatsu et al. smaller than the average mass, M200 ∼ 1.4 × 1012 M⊙ , of the
2011) and the clustering of galaxies (e.g. Cole et al. 2005), Aquarius haloes (Vera-Ciro et al. 2013; Wang et al. 2012). A
carry over, for the most part, to a warm dark matter (WDM) somewhat more uncertain possibility is that the central den-
model. To distinguish between these two types of dark mat- sity of CDM subhaloes may have been reduced by the kind
ter using astrophysical considerations it is necessary to re- of explosive baryonic processes proposed by Navarro et al.
sort to observations on the scale of the Local Group. (1996a) which appear to occur in some recent hydrodynamic
Over the past decade, surveys such as SDSS (York et al. simulations (Pontzen & Governato 2012; Brooks & Zolotov
2000), PAndAS (Ibata et al. 2007) and Pan-STARRS 2014; Parry et al. 2012; Zolotov et al. 2012) but not in oth-
(Kaiser et al. 2010) have begun to probe the Local Uni- ers (di Cintio et al. 2011) which assume different prescrip-
verse in detail. A number of new dwarf spheroidal (dSph) tions for physics that are not resolved in the simulations.
satellite galaxies have been discovered around the Milky More radical solutions to the ‘too big to fail’ prob-
Way and M31 (e.g. Willman et al. 2005; Walsh et al. 2007; lem require abandoning CDM altogether. Vogelsberger et al.
Martin et al. 2009; Bell et al. 2011; Martin et al. 2013). (2012) show that simulations with a new class of ‘self-
Follow-up studies of stellar kinematics have been used to interacting’ dark matter could solve the problem. However,
investigate their dynamics and mass content (Walker et al. a solution is also possible with more conventional assump-
2009, 2010; Wolf et al. 2010; Tollerud et al. 2012). These tions. In particular, Lovell et al. (2012) show that simula-
data indicate that some dSphs have mass-to-light ratios of tions with WDM produce very good agreement with the
around 100, and are thus systems in which the properties dSph kinematical data. The absence of small-scale power
of dark matter may be most directly accesible. Analyses of in the initial fluctuation field causes structure to form later
the number and structure of dSphs should therefore provide than in the CDM case. Haloes of a given mass thus collapse
strong constraints on the nature of the dark matter. when the mean density of the universe is smaller and, as a re-
The luminosity function of satellites in the Local sult, end up with lower central densities (Avila-Reese et al.
Group has now been determined to quite faint magnitudes 2001). However, the WDM model they assumed was ‘too
(Koposov et al. 2008; Tollerud et al. 2008), confirming that warm’, in the sense that it assumed too low a particle mass
there are far fewer satellites around galaxies like the Milky (and thus too large a cut-off scale in the initial power spec-
Way than there are subhaloes in cosmological N-body sim- trum) and produced only 18 dark matter subhaloes within
ulations from CDM initial conditions (Diemand et al. 2005; 300 kpc of the main halo centre whereas observations sug-
Springel et al. 2005). This discrepancy is not new and can gest the actual number of satellites may be over an order of
be readily explained by the physics of galaxy formation magnitude greater (Tollerud et al. 2008).
because feedback processes are very efficient at suppress- This constraint from subhalo central densities is one
ing the formation of galaxies in small haloes (Bullock et al. of several that can be used to place bounds on the WDM
2000; Benson et al. 2002; Somerville 2002). Recent hydrody- particle mass. The measured clustering of the Lyman α for-
namic simulations have confirmed this conclusion originally est lines at high redshift sets a lower limit to the parti-
deduced from semi-analytical models of galaxy formation cle mass (Viel et al. 2005; Boyarsky et al. 2009b; Viel et al.
(Okamoto et al. 2010; Wadepuhl & Springel 2011). 2013) while the absence of X-rays from particle decay sets
Kinematical studies of the bright Milky Way satellites a (model dependent) upper limit to the mass of the sterile
can constrain the internal structure of their dark matter neutrino (whose decay rate into pairs of neutrinos and X-
subhaloes. Gilmore et al. (2007) argued that the data sup- ray photons scales with the mass of the sterile neutrino; see
port the view that dSphs have central cores, in appar- Kusenko 2009; Boyarsky et al. 2012, and references therein.)
ent contradiction with the results of N-body simulations The results of Lovell et al. (2012) and related results by
c 2012 RAS, MNRAS 000, ??–21

Properties of WDM haloes 3
Macciò et al. (2012); Shao et al. (2013); Macciò et al. (2013) (one CDM and four WDM), we ran simulations at differ-
raise the question of whether it is possible to find a range ent resolution. Our ‘high resolution’ suite corresponds to
of WDM particle masses that lead to ‘warm enough’ models level-2 in the original Aquarius notation; it has particle
that match satellite central densities but which are also ‘cold mass of 1.55 × 104 M⊙ , and gravitational softening length
enough’ to generate the observed number of satellite galax- of ǫ = 68.1pc. All haloes were also run at “low resolution”
ies (Polisensky & Ricotti 2011; Kamada et al. 2013). In this (level-4), with particle mass of 4.43 × 105 M⊙ and gravita-
work we examine both the number and structure of satellite tional softening of ǫ = 355.1pc. Finally, we ran an interme-
galaxies in simulations as a function of the WDM particle diate resolution version (level 3) of the warm dark matter
mass. models with the lightest and heaviest dark matter particles,
The first requirement is to be able to count accurately with particle mass 5.54 × 104 M⊙ and ǫ = 125.0pc, in order
the number of dark matter haloes formed in WDM cos- to facilitate convergence studies. All haloes were simulated
mologies. The first simulations of WDM models (Bode et al. from z = 127 to z = 0 using the gadget3 N-body code
2001) showed the halo mass function to be suppressed as (Springel et al. 2008a).
expected, but also found that at least 90 percent of haloes, To set up the initial conditions for the WDM runs we
depending on the choice of power spectrum cutoff, formed employed the transfer function, T (k), defined as
from the fragmentation of filaments and had masses below PWDM (k) = T 2 (k)PCDM (k). (1)
the smoothing scale. Wang & White (2007) examined this
effect in hot dark matter (HDM) simulations (which assume where P (k) denotes the power spectrum as a function of
a much larger power spectrum cutoff scale than in WDM) comoving wavenumber k. We adopted the fitting formula
and showed that the fragmentation of filaments depends on for T (k) given by Bode et al. (2001):
the resolution of the simulation, thus concluding that most
of the haloes in the Bode et al. (2001) simulations were due T (k) = (1 + (αk)2ν )−5/ν , (2)
to a numerical artifact.
In this paper we introduce a series of methods for iden- where ν and α are constants. Bode et al. (2001) and
tifying spurious haloes in simulations, and then use our Viel et al. (2005) find that ν can take values between 1 and
cleaned halo sample to examine the distribution and struc- 1.2 depending on the fitting procedure; we adopted ν = 1
ture of WDM haloes as a function of the power spectrum for simplicity. The position of the cutoff in the power spec-
cutoff. The paper is organised as follows. In Section 2 we trum is determined by the parameter α, such that higher
present our simulation set and in Section 3 we describe our values of α correspond to cutoffs at larger length scales. In
algorithm for removing spurious subhaloes. We then present principle, the initial conditions for WDM simulations should
our results in Section 4, and draw conclusions in Section 5. include thermal velocities for the particles (Colı́n et al. 2008;
Macciò et al. 2012; Shao et al. 2013). However, at the res-
olution of our simulations, the appropriate velocities would
have a negligible effect (Lovell et al. 2012) and are therefore
2 THE SIMULATIONS not included. All of our CDM and WDM initial conditions
employed a glass-like initial particle load (White 1994).
We begin by describing the details of our simulations, the For our four WDM models we adopted values of
procedure for generating initial conditions and a general α of 0.0199h−1 Mpc, 0.0236h−1 Mpc, 0.0297h−1 Mpc, and
overview. 0.0340h−1 Mpc respectively. The last of these corresponds
to the original WDM simulation presented in Lovell et al.
(2012) which, however, assumed the WMAP year-1 cos-
2.1 Simulation parameters mological parameters. That model was originally chosen
Our N-body simulation suite is based upon that of the as a thermal relic approximation to the M2L25 model of
Aquarius Project (Springel et al. 2008a), a set of six (Aq- Boyarsky et al. (2009b), the νMSM parameter combination
A through to Aq-F) galactic dark matter haloes simulated that has the largest effective free-streaming length that is
at varying resolution (levels 1-5, where level 1 corresponds to still consistent with bounds from the Lyman-α forest (but
the highest resolution). The Aquarius simulations assumed see also Viel et al. 2013).
cosmological parameter values derived from the W ilkinson Bode et al. (2001) related α to a generic thermal relic
Microwave Anisotropy Probe (WMAP) year 1 data. These warm dark matter particle mass, mWDM , using the formula:
have now been superseded and in this paper we use the cos-
0.15
mological parameter values derived from the W MAP year 7

0.05 mWDM −1.15 ΩWDM
data (Komatsu et al. 2011): matter density, Ωm = 0.272; α=
hMpc−1 1keV 0.4
dark energy density, ΩΛ = 0.728; Hubble parameter, h = 1.3
0.704; spectral index, ns = 0.967; and power spectrum nor- h gWDM −0.29
× , (3)
malization σ8 = 0.81. 0.65 1.5
Our main set of simulations follows the formation of where ΩWDM is the WDM contribution to the density pa-
four WDM galactic haloes with different effective WDM par- rameter; we have set the number of degrees of freedom,
ticle masses. The initial phases in the fluctuation spectrum gWDM = 1.5. We list the thermal relic masses for each of
are identical to those of the original CDM Aq-A halo but our models in Table 1, and use these masses as labels for the
the transfer function is that appropriate to WDM as de- models, namely m2.3 , m2.0 , m1.6 , and m1.5 ; we denote the
scribed below. In addition, we resimulated the level-2 Aq-A CDM simulation with W MAP year-7 parameters as CDM-
halo using the WMAP year-7 cosmology. For all five haloes W7. We also give the cutoff mass scale for each simulation,
c 2012 RAS, MNRAS 000, ??–21

Simulation mWDM [keV] α[h−1 Mpc] Mth [M⊙ ] mν=1.12

WDM [keV]
CDM-W7 – 0.0 – –
m2.3 2.322 0.01987 1.4 × 109 1.770
m2.0 2.001 0.02357 1.8 × 109 1.555
m1.6 1.637 0.02969 3.5 × 109 1.265
m1.5 1.456 0.03399 5.3 × 109 1.106
Table 1. Parameters of the simulations. The parameter α determines the power spectrum cutoff (Eqn. 2); mWDM is the thermal relic
mass corresponding to each value of α; and Mth is the cutoff mass scale defined using a top hat filter as described in the text. The final
column gives the particle masses that, when combined with the ν = 1.12 transfer function and mWDM − α relation of Viel et al. (2005),
give the best approximation to our ν = 1 transfer functions.
which we define as the mass within a top hat filter which,

when convolved with the CDM power spectrum, results in 2
a function that peaks at the same value of k as the WDM
power spectrum. 0
log10 [ ∆2=(k3P(k))/2π2 ]
In order to compare our study to that of Viel et al.
(2005) and Viel et al. (2013) we need to take into account
−2
that the transfer function that we use assumes ν = 1 in
Eqn. 2 while theirs assumes ν = 1.12. For values of k near
the power spectrum cutoff, the transfer function for a given −4
mWDM has a higher amplitude if ν = 1.12 than if ν = 1. To
CDM
match the power on this scale then requires a higher value of m2.3
mWDM if ν = 1 than if ν = 1.12. We can therefore derive an −6 m2.0
m1.6
‘equivalent ν = 1.12’ mass for each of our models which gives m1.5
the best approximation to the transfer function in our ν = 1 −8
simulations. These masses are listed in the final column in 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Table 1. (We carry out the comparison for T 2 (k) > 0.5 and log10(k/h Mpc−1)
use the equation relating mWDM and α given in Eqn. 7 of
Viel et al. 2005). Figure 1. The linear theory power spectrum used in the simula-
tions. The black line corresponds to the CDM model, CDM-W7,
The linear theory power spectra used to set up the ini- while the blue, green, orange and red lines correspond to the
tial conditions are plotted in Fig. 1. By construction, the m2.3 , m2.0 , m1.6 , and m1.5 WDM models respectively. The ar-
peak of the power spectrum moves to higher k as α decreases rows mark, in order of smallest to largest, the Nyquist frequency
(and the particle mass increases). For all WDM models the of our low, medium, and high resolution simulations.
initial power spectrum peaks at a value of k smaller than
the Nyquist frequency of the particle load in the simula-
tion. This will lead to the formation of spurious haloes as
mentioned in Section 1.
Simulation M200 [M⊙ ] r200 [kpc] M200b [M⊙ ] r200b [kpc]
Self-bound haloes were identified using the subfind al-
gorithm (Springel et al. 2001); they are required to contain CDM-W7 1.94×1012 256.1 2.53×1012 432.1
at least 20 particles. The largest subfind group is the galac- m2.3 1.87×1012 253.4 2.52×1012 431.4
m2.0 1.84×1012 251.7 2.51×1012 430.8
tic halo itself, to which we will refer as the ‘main halo’.
m1.6 1.80×1012 250.1 2.49×1012 429.9
Smaller haloes that reside within the main halo are known m1.5 1.80×1012 249.8 2.48×1012 429.0
as ‘subhaloes’, whereas those that are outside the main halo Aq-A2 1.84×1012 245.9 2.52×1012 433.5
are ‘independent haloes’. Most of the subhaloes will have
experienced gravitational stripping whilst most of the inde- Table 2. Properties of the main friends-of-friends halo in each
pendent haloes will have not. high resolution simulation. The radii r200 and r200b enclose re-
A first view of the simulations is presented in Fig. 2. gions within which the mean density is 200 times the critical and
The smooth component of the main haloes is very similar background density respectively. The masses M200 and M200b are
in all five models: in all cases, the haloes are similarly cen- those contained within these radii. We also reproduce data from
trally concentrated and elongated. The main difference is the original Aquarius Aq-A2 halo.
in the abundance of subhaloes. The myriad small subhaloes
evident in CDM-W7 are mostly absent in the WDM models.
For these, the number of subhaloes decreases as α increases
(and the WDM particle mass decreases). the background density (M200b ). There is a slight trend of
The apparent similarity of the main haloes displayed decreasing mass with increasing α, but the maximum change
in Fig. 2 is quantified in Table 2 which lists the masses and is only 7 percent for M200 and 2 percent for M200b . The
radii of the largest friends-of-friends halo in each simulation. change in cosmological parameters also makes only a small
The table gives their masses enclosed within radii of mean difference: M200 is 5 percent higher for CDM-W7 than for
density 200 times the critical density (M200 ) and 200 times the original Aquarius halo with WMAP year 1 parameters.
c 2012 RAS, MNRAS 000, ??–21

Figure 2. Images of our haloes at redshift z = 0. The panels show CDM-W7 (top), m2.3 , m2.0 , m1.6 , and m1.5 (left to right, then top
to bottom). The image intensity and hue indicate the projected squared dark matter density and the density-weighted mean velocity
dispersion respectively (Springel et al. 2008a). Each panel is 1.5Mpc on a side.
c 2012 RAS, MNRAS 000, ??–21

−1.0
108
CDM
6 m2.3 −1.5
10 m2.0
m1.6
dlogρ(r)/dlogr
m1.5 −2.0
ρ(r)/<ρ>
104
−2.5
102
CDM
m2.3
−3.0 m2.0
100 m1.6
m1.5
ρ(r)/ρC7(r)
1.4
−3.5
1.0 1 10 100
0.7 r [ kpc ]
1 10 100
r [ kpc ] Figure 4. Radial variation of the logarithmic slope of the density
Figure 3. Density profiles of the main haloes (including sub- profiles of the main haloes in the simulations. Line colours and
haloes) in the simulations normalised by the background matter plotting range are as in Fig. 3.
density. The line colours are as in Fig. 1. The profiles are plot-
ted only beyond the ‘Power radius’ (Power et al. 2003) at which
numerical convergence is expected. The bottom panel shows the with the corresponding region of a higher resolution simula-
profiles for the WDM simulations normalized to the profile for tion with the same initial conditions by plotting those parti-
the CDM-W7 model. cles that have collapsed into dark matter haloes. In both sim-
ulations there are two large haloes and several smaller ones.
2.2 The structure of the main haloes The large haloes have very similar sizes and positions in the
two simulations, and can be regarded as genuine objects. By
The density profiles of the main haloes (including substruc- contrast, the small haloes have different sizes and positions
tures) in our high resolution simulations are plotted in Fig. 3. in the two simulations; there are also more of them in the
There is good agreement amongst all the haloes at radii higher resolution case. As shown by Wang & White (2007),
(10-100) kpc, with the five profiles agreeing to better than increasing the resolution even by rather large factors is not
10 percent. At larger radii, systematic differences between sufficient to prevent the formation of these artificial haloes.
CDM-W7 and the WDM models begin to appear and these Using glass initial conditions, as we do for our simulations,
become increasingly pronounced for the warmer models. does not reduce this problem. Future N-body codes that
These differences are due to slight variations in the posi- use phase space smoothing techniques may be able to alle-
tion of large substructures in the outer parts. There are also viate this problem (Hahn et al. 2013; Shandarin et al. 2012;
small differences at much smaller radii (< 10kpc) which are Angulo et al. 2013). At present, however, the only practical
are not systematic and are thus likely due to stochastic vari- measure is to develop a reliable algorithm for identifying and
ations in the inner regions. removing these ‘spurious’ haloes from the halo catalogues.
The radial variation of the logarithmic slope of the den- We now introduce an algorithm for distinguishing be-
sity profile of each halo is plotted in Fig. 4. In all cases tween genuine and spurious subhaloes. It exploits three
the slope at the innermost point plotted approaches the properties of the artefacts – mass, resolution dependence
Navarro-Frenk-White (NFW) asymptotic value of −1 but and the shape of the initial particle distribution – to de-
there is no evidence that the slope is converging. There fine a series of cuts that isolate the artefacts. We present an
is a slight tendency in the inner parts, r < 4kpc, for the outline of the method in Section 3.1 and provide details in
slope in the WDM models to be shallower than in the CDM Section 3.2. Note that while the results presented here have
model, but there is no obvious trend with α, possibly be- been derived for subhaloes that have been accreted into an-
cause of stochastic effects in the inner regions. Thus, apart other halo, the algorithm is equally valid for independent
from minor differences, the structure of these ∼ 1012 M⊙ haloes.
haloes varies little with power spectrum cut off, as expected
for systems of mass ≫ Mth .
3.1 Outline of the methods
Previous simulations have shown that spurious haloes have
3 REMOVAL OF SPURIOUS HALOES
small masses at formation and outnumber genuine haloes on
One of the main aims of this study is to determine the mass those mass scales where they are present (Wang & White
function of subhaloes in WDM simulations. However, as we 2007). Thus, in principle, many spurious haloes can be sin-
discussed in Section 1, simulations in which the initial power gled out by applying a mass cut. This mass threshold, how-
spectrum has a resolved frequency cutoff can undergo spu- ever, is not well defined because the mass function of genuine
rious fragmentation of filaments. An example is shown in haloes overlaps that of the spurious haloes, so it is useful to
Fig. 5, where we compare a region in one of our simulations introduce additional criteria to ensure that, as far as pos-
c 2012 RAS, MNRAS 000, ??–21

800
500
600
400
kpc
300 400
kpc
200
200
100
0
0
0 100 200 300 400
0 100 200 300 400 800
kpc kpc
Figure 5. A region of a WDM simulation performed at two dif-

ferent resolutions. The particle mass for the high resolution sim- 600
ulation (right) is 29 times smaller than that of the low resolution
kpc
case (left). Only particles in bound structures at this snapshot are
shown. Particles are coloured according to the halo to which they 400
belong. The number of particles plotted in each panel is equal to
the number of bound-structure particles in the low resolution sim-
ulation; we have applied random sampling in the high resolution 200
case.
0
sible, all artificial haloes are identified and no genuine ones 0 200 400 600 800 1000
are removed. kpc
The resolution dependence of the spurious fragmenta-
tion can be used to refine the distinction between genuine Figure 6. The particles of Fig. 5 traced back to their positions
in the initial conditions. The low resolution simulation is shown
and artificial haloes. While genuine haloes in a simulation
in the top panel and the high resolution simulation in the bottom
at a given resolution are expected to be present in the same
panel. Note the highly flattened configurations of spurious haloes.
simulation at higher resolution, this need not be the case
for spurious haloes, as illustrated in Fig. 5. Springel et al.
(2008a) showed that it is possible to match haloes and sub- what follows, we restrict attention to subhaloes lying within
haloes between different resolution simulations by tracing r200b of the main halo centre at z = 0 except where we state
their particles back to the initial conditions and identifying otherwise.
overlapping Lagrangian patches in the two simulations. We
refer to the initial Lagrangian region of each halo, or more
precisely the unperturbed simulation particle load, as its 3.2 Application
‘protohalo’. The initial positions of the particles displayed
3.2.1 Protohalo shapes
in Fig. 5 are shown in Fig. 6. The two large objects originate
from protohaloes of similar size and location, but there are To determine the flattening of protohaloes we consider all
clear discrepancies in the number, location and mass of the the particles that make up a subhalo at some epoch (deter-
small objects. Thus, attempts to match small haloes in the mined below), find their positions in the unperturbed sim-
two simulations will often fail because spurious haloes in the ulation particle load and calculate the inertia tensor of the
low resolution calculation do not have a counterpart in the particle set:
high resolution simulation.
A third criterion exploits the most striking feature vis- X
ible in Fig. 6: the shapes of the protohaloes. Genuine pro- Iij = m(δij |x|2 − xi xj ), (4)
all particles
tohaloes are spheroidal, whereas spurious protohaloes have
much thinner, disc-like geometries. They can therefore be where δij is the Kronecker delta function, m is the particle
easily flagged as the progenitors of spurious haloes in the mass and x is the particle position relative to the protohalo
initial conditions. centre of mass. We take a > b > c to be the axis lengths
In this study we are interested in objects that become of the uniform, triaxial ellipsoid that has the same moment
subhaloes at the present day. We will apply these three cri- of inertia tensor as the protohalo. We can then calculate
teria to them in the following order. First, we identify a cut s = c/a, known as the sphericity. A disc-like (or, more rarely,
based on protohalo shape, rejecting from the catalogue all needle-like) spurious subhalo will have a major axis (disc
subhaloes flatter than a given threshold. Secondly, we apply diameter, a) much longer than its minor axis (disc thickness,
a mass cut; finally, we refine the mass cut using a match- c), and thus a small value of s. Genuine subhaloes, on the
ing procedure between simulations at different resolution. In other hand, are spheroidal and thus have higher values of s.
c 2012 RAS, MNRAS 000, ??–21

3.2.2 A first guess of the mass cut
For a first guess of the mass cut below which a major-
0.8
ity of subhaloes are spurious, we resort to the results of
Wang & White (2007). They showed that the characteristic
mass below which spurious subhaloes begin to dominate the
0.6
subhalo mass function is related to the matter power spec-
trum cutoff and the simulation resolution. The larger the
smean
0.4
value of the cutoff frequency and the higher the resolution
of the simulation, the smaller is the mass of the largest spu-
rious subhaloes. Wang & White (2007) derived an empirical
0.2
formula for the mass at which spurious subhaloes begin to
dominate:
0.0
106 107 108 109 1010 Mlim = 10.1ρ̄dkpeak
−2
, (5)
MMax [ MO• ]
where ρ̄ is the mean density of the Universe, d is the mean
Figure 7. Mean subhalo sphericities as a function of MMax for interparticle separation (a measure of resolution), and kpeak
the high resolution CDM-W7 (black) and the m1.5 (red) runs. is the wavenumber at which the dimensionless power spec-
The region between the upper and lower 99 percentiles of the trum, ∆2 (k), has its greatest amplitude. We can apply this
CDM distribution is shown in grey; the same region for the m1.5
formula to MMax to estimate a cut below which the majority
simulation is delineated by the red dotted lines.
of the subhaloes will be spurious. Some genuine haloes will
have MMax below this threshold but the mass limit can be
refined using the matching criterion.
We now need to choose an appropriate epoch at which
to identify the particles that make up the protohalo. This
should be well before the subhalo has fallen into a larger 3.2.3 Matching subhaloes between simulations
halo, after which its outer particles will be stripped. We
select the earliest simulation snapshot below which the A subhalo that is present in both a low resolution simulation
halo mass is more than half the maximum mass, the ‘half- (LRS) and in its high resolution counterpart (HRS) is likely
maximum mass snapshot’. The initial positions of the par- to be genuine. We can use this property to refine the mass
ticles in the object at this time are used to evaluate the cut. We set the cutoff mass to be Mmin = κMlim , where κ is
protohalo sphericity. a constant such that the number of LRS subhaloes of mass
The distributions of s for the subhaloes that survive to greater than Mmin is equal to the number of subhaloes with
z = 0 in the CDM-W7 and m1.5 simulations are illustrated matches in the HRS. We will assume that the value of κ
in Fig. 7, as a function of MMax . The mean sphericity is determined for the LRS subhaloes is also applicable to the
shown as a solid line and the 98 percent range is indicated by HRS catalogues.
the dotted lines in each case. The figure reveals two regimes. We now introduce an algorithm for finding high resolu-
For values of MMax > 109 M⊙ , the sphericity distributions in tion counterparts of the low resolution subhaloes. Genuine
the two simulations are consistent with each other. For lower haloes should originate from the same Lagrangian region
masses the protohaloes in the m1.5 simulation are much flat- regardless of resolution. Therefore, to match subhaloes we
ter than in CDM-W7. This clear dichotomy suggests that require a quantitative measure to compare these Lagrangian
most of the m1.5 subhaloes with MMax > 109 M⊙ are gen- regions in simulations of different resolution and check that
uine and most of those with MMax < 108 M⊙ are spurious. they overlap and have the same shape. These shapes are de-
We can use the CDM subhaloes to define a cut in protohalo fined by point-like particles. In order to develop a quantita-
sphericity above which WDM subhaloes are likely to be real. tive measure of the overlap we need to smooth these points.
We find that 99 percent of CDM subhaloes containing more We measure the degree to which a pair of objects in differ-
than 100 particles at the half-maximum mass snapshot have ent resolution simulations are the ‘same’ by comparing the
protohaloes with sphericity greater than ∼ 0.16 (depending entirety of the regions from which they form. We introduce
slightly on simulation resolution), which we denote scut . We a statistic:
exclude from our cleaned subhalo catalogue any WDM sub-
halo whose protohalo has sphericity less than scut , regard- 2
UAB
less of mass. This cut rejects between 86 percent (m2.3 ) and R= , (6)
UAA UBB
93 percent (m1.5 ) of the WDM subhaloes as spurious. We R
have checked, as we show later, that the subhaloes rejected where UXY = φX ρY dV , V is volume, and ρA/B and φA/B
by this criterion do not have clear counterparts in pairs of are the density of and gravitational potential due to the mat-
simulations of different resolution, where in this case the dif- ter distributions A/B respectively. It can be shown using
ference in resolution is a factor of 8. We find that varying Green’s Theorem that if the matter distribution of subhalo
scut by 20 percent changes the number of subhaloes identi- A is proportional everywhere to that of subhalo B, R = 1;
fied as genuine by less than 20 percent, which is within the for any other configuration R < 1. We apply this formula
2σ Poisson uncertainty in the number identified using our to our candidate LRS-HRS protohalo particle distributions,
chosen value of scut . representing each particle as a spherical shell of radius equal
c 2012 RAS, MNRAS 000, ??–21

by using our sphericity cut to determine the distribution of
1.0 R for spurious subhaloes. For each WDM model, we take
10000 subsamples of 100 subhaloes that fail the sphericity
cut (with replacement) and take the second highest R of
0.8 each subsample to be the threshold, Rmin , below which sub-
haloes are spurious. This result is not sensitive to the size of
0.6 our subsamples. The mean value of Rmin across the 10000
subsamples is found to be in the range 0.94-0.96 for each
R
of the four WDM models. For those subhaloes that instead

0.4 pass the sphericity cut, the mean value of Rmin is greater
CDM than 0.995 for all four models, showing that sphericity is a
m2.3
0.2 m2.0 robust and accurate diagnostic of whether or not an object
m1.6 is spurious.
m1.5
We now couple the matching and sphericity criteria to
0.0
determine the optimal cut in MMax . In Fig. 9, we plot s
107 108 109 1010
MMax [ MO• ] as a function of MMax for the LRS subhaloes in each of
our four WDM models, indicating their matching quality
Figure 8. R as a function of MMax for CDM and WDM LRS sub- by colour. We adopt Rmin = 0.94. We restrict attention to
haloes matched to HRS counterparts (those that fail the spheric- subhaloes that pass the sphericity cut and take a mass limit
ity cut are still included). The black dots denote CDM subhaloes, Mmin = κMlim such that the number of subhaloes with mass
blue m2.3 , green m2.0 , orange m1.6 , and red m1.5 (the same as
greater than Mmin is equal to the number of subhaloes with
Fig. 1.
R > Rmin . In Fig. 9 this is equivalent to the number of red
dots to the right of the mass cut being equal to the number
to the LRS mean interparticle separation and with infinitesi- of blue dots to the left. We find that this condition requires
mal thickness. The best match for the LRS subhalo will then values of κ between 0.4 and 0.6, given the uncertainty in
be the HRS halo with which it attained the highest value of Rmin . For simplicity, we will adopt κ = 0.5; we find that
R. We retain this value of R for each LRS subhalo as our this value provides a good compromise between rejecting
measure of its matching quality. A genuine LRS subhalo will low mass genuine objects and including high mass spurious
have a good match at high resolution and therefore have a subhaloes in all four models. Varying Rmin and κ in the
value of R close to 1, whereas a spurious subhalo will have range stated here makes a difference of ∼ 10 percent to
a poor match and a lower value of R. the number of subhaloes returned in the m1.5 model and
To find candidate matches, we first divide the simula- ∼ 5 percent in the other cases. The values of Mmin are then
tion volume into a grid of cells of comoving length > 1.5×108 M⊙ , 2.2×108 M⊙ , 3.2×108 M⊙ , and 4.2×108 M⊙ for
∼60kpc,
and, for a given low resolution protohalo, choose as candi- the m2.3 , m2.0 , m1.6 , and m1.5 models respectively in the low
date matches the high resolution protohaloes that occupy resolution simulations. For the high resolution simulations,
the same and neighbouring grid cells. It is computationally they decrease to 5.1 × 107 M⊙ , 7.0 × 107 M⊙ , 1.1 × 108 M⊙ ,
expensive to calculate R for the largest subhaloes, but we and 1.4 × 108 M⊙ .
found that random sampling of each halo with 10000 par- To summarize, we have used the mass, resolution de-
ticles returned values of R that did not vary systematically pendent, and Lagrangian region shape properties to iden-
with MMax for subhaloes of MMax > 109 M⊙ . We therefore tify spurious subhaloes in our subhalo catalogues. Having
adopt a threshold of 10000 particles. When attempting to derived values for scut and Mmin – the latter as a func-
match subhaloes between simulations, minor differences in tion of power spectrum cutoff and resolution – we can apply
which particles are assigned to each subhalo can have an these cuts to the high resolution simulations. We plot the re-
impact on R. We mitigate this problem by performing the sults in Fig. 10. Changing the value of κ in the range 0.4-0.6
calculation for both the maximum-mass and half-maximum produces a variation of < 5 percent in all four HRS models,
mass snapshots, selecting the higher value of the two for each and this does not affect our conclusions. In what follows we
subhalo. The resulting values of R are plotted as a function consider only those subhaloes that pass the cuts in each of
of MMax in Fig. 8. these panels.
At high masses, the CDM and WDM protohaloes have
R close to 1. As the protohalo mass decreases, R becomes
systematically lower and the decline is much steeper for the
WDM models, as expected in the presence of poorly match- 4 RESULTS
ing spurious subhaloes. Unfortunately, a small proportion of
4.1 The subhalo mass and Vmax functions
CDM subhaloes also attain low values of R and the demar-
cation between the distributions of R for WDM and CDM In Fig. 11 we present the cumulative distributions of subhalo
is much less clear cut than we found for the sphericity mea- mass, Msub , and Vmax at z = 0, where Vmax is defined as
surement, s. Were we to take the same approach for R as we p peak amplitude of the circular velocity profile Vcirc =
the
did for s, we would infer a cut in R of about 0.68. More than GM (< r)/r, with G the gravitational constant and M (<
half of the WDM subhaloes have a value of R closer to 1 than r) the mass enclosed within radius r. This is a useful proxy
this, and since the sphericity-based algorithm rejects ∼ 90 for mass that is insensitive to the definition of the edge of
percent of subhaloes, adopting this cut in R would return a the subhalo. The figure includes both genuine (solid lines)
heavily contaminated sample. We circumvent this problem and spurious (dashed lines) subhaloes. Overall, the spurious
c 2012 RAS, MNRAS 000, ??–21

0.8 m2.3 m2.0 nWDM /nCDM = (1 + Mhm M −1 )β , (7)

0.6 where Mhm is the mass associated with the scale at which
the WDM matter power spectrum is suppressed by 50 per-
s
0.4
cent relative to the CDM power spectrum, M is subhalo
mass and β is a free parameter. The best fit value is β
0.2
of 1.3, slightly higher than the value of 1.16 found by
0.0 Schneider et al. (2012) for friends-of-friends haloes (rather
0.8 than subfind subhaloes as in our case). A slightly better fit
m1.6 m1.5 is obtained by introducing an additional parameter, γ, such
0.6 that:
s
0.4
nWDM /nCDM = (1 + γMhm M −1 )β , (8)
0.2 with γ = 2.7 and β = 0.99. However, better statistics are
0.0 required to probe the subhalo mass function more precisely.
106 107 108 109 1010 106 107 108 109 1010 In principle, comparison of the abundance of subhaloes
MMax [ MO• ] MMax [ MO• ] shown in Fig. 11 with the population of satellite galaxies
observed in the Milky Way can set a strong constraint on
Figure 9. Dot plots of s and MMax for subhaloes in the four the mass of viable WDM particle candidates. Assuming that
different WDM models at low resolution. Blue points correspond every satellite possesses its own dark matter halo and that
to R > 0.94 and red points to R < 0.94. The horizontal, dashed the parent halo in our simulations has a mass comparable
line is scut and the vertical line is Mmin . All subhaloes are within to that of the Milky Way halo, a minimum requirement is
r200b of the main subhalo centre at redshift zero. that the number of subhaloes in the simulations above some
value of Msub or Vmax should exceed the number of Milky
Way satellite above these values. In practice, the compari-
son is not straightforward because: (i) the values of Msub
or Vmax for the observed population are not well known
0.8 m2.3 m2.0 and (ii) the total number of Milky Way satellites is un-
0.6 certain. Nevertheless, we can obtain a conservative limit on
the mass of the particle as follows. There are 22 satellites
s
0.4 in the Milky Way for which good quality kinematical data
exist (Walker et al. 2009; Wolf et al. 2010). Eleven of these
0.2 are ‘classical satellites’ and the remainder are SDSS satel-
lites. Of the classical satellites, eight are dwarf spheroidals
0.0 and the others are the large and small Magellanic clouds
0.8 m1.6 m1.5 (LMC and SMC) and Sagittarius. Wolf et al. (2010) have
estimated values of the mass (and line-of-sight velocity dis-
0.6 2
persion, σlos ) within the (deprojected 3D) half-light radius
for the eight classical and 11 SDSS dwarf spheroidals. These
s
0.4 are essentially insensitive to the velocity anisotropy of the

stellar populations. The circular velocity within this radius
0.2
is then given by:
0.0
106 107 108 109 1010 106 107 108 109 1010 q
2
MMax [ MO• ] MMax [ MO• ] Vcirc (r1/2 ) = 3σlos . (9)
Figure 10. Dot plots of s and MMax for subhaloes in the four
The values of Vcirc are lower limits to Vmax for each satel-
different WDM models at high resolution. The horizontal, dashed lite. Leo IV has the smallest circular velocity, Vcirc = 5.7 ±
line is scut and the vertical line is Mmin . All subhaloes are within 2.9 km s−1 , of the 22 studied by Wolf et al. (2010). We
r200b of the main subhalo centre at redshift zero. show in Appendix A that our simulations have converged
to better than 8 percent at this value of Vmax , showing
that our conclusions are not affected by resolution issues
(c.f. Polisensky & Ricotti 2011). As shown by Springel et al.
subhaloes outnumber the genuine ones by a factor of 10. (2008b), values of Vmax for subhaloes in Aquarius level 2 sim-
However, the mass function is dominated by genuine haloes ulations are converged to within ∼ 10 percent for Vmax >
beyond Msub ∼ (1 − 3) × 107 M⊙ , corresponding to Vmax ∼ 1.5 km s−1 . We have examined the convergence in our m2.3
(4−6) km s−1 , for the different models. The differential mass model and find that our L3 and L2 resolution Vmax func-
function (relative to the CDM mass function) for genuine tions are converged to within 2σ (Poisson) of each other for
haloes in the m2.3 case can be fit with the functional form Vmax > 4 km s−1 . This is more modest than for the CDM
given by Schneider et al. (2012): Aquarius simulations, but sufficient to resolve the Leo IV
c 2012 RAS, MNRAS 000, ??–21

type satellites. This result also gives us confidence that our
ability to count satellites is not impaired by the numerical 105
r < r200b
issues (c.f. Polisensky & Ricotti 2011).
The known number of satellites in the Milky Way halo, 104
22, is a lower limit to the total number within 280 kpc
of the galaxy’s centre, the distance to which the tip of
103
the red giant branch can be detected in the SDSS. This
N(>Msub)
is because although all the classical satellites (i.e. satel-
lites brighter than MV = −11) have probably been discov- 102
ered, SDSS surveyed only 20 percent of the sky [data re-
lease 5(DR5)]. Thus, a conservative lower limit to the WDM
particle mass is obtained by requiring that the simulation 101
should produce at least 22 satellites within this radius with
Vmax > 5.7 km s−1 . Our m1.5 simulation produced only 25 100
subhaloes with Vmax greater than this value within the larger 105 106 107 108 109 1010 1011
Msub [MO•]
radius, r200b = 429 kpc. Furthermore, the mass of the m1.5
halo, M200 = 1.80 × 1012 M⊙ , is towards the higher end of
acceptable values for the mass of the Milky halo; simula- 105
tions of haloes with lower mass would produce even fewer r < r200b
subhaloes. Finally, any residual contamination by spurious
subhaloes would artificially inflate the numbers in our sub- 104 CDM
m2.3
halo sample. Thus, we can safely set a conservative lower m2.0
limit to the mass of the WDM particle of mWDM = 1.5 keV. m1.6
103 m1.5
We can set a less conservative but still robust lower
N(>Vmax)
limit to mWDM by correcting the observed number of SDSS

satellites to take into account the area surveyed. A simple 102
extrapolation multiplying the observed number by a factor
of 5 has to be taken with caution because we know that the
classical satellites are not distributed isotropically but are 101
concentrated towards a plane, called the ‘Great pancake’
by Libeskind et al. (2005). However, from analysis of the 100
Aquarius simulations, Wang et al. (2012) have argued that 1 10 100
such flat configurations occur only for the most massive ∼ Vmax [ km/s ]
10 subhaloes and the anisotropy of the distribution falls off
Figure 11. Cumulative subhalo mass, Msub , (top panel ) and
rapidly with increasing sample size so that samples of ∼ 50
Vmax (bottom panel ) functions of subhaloes within r < r200b of
subhaloes follow quite close the overall shape of the halo. the main halo centre in the high resolution simulations at z = 0.
Based on this, we do not make any corrections for anisotropy Solid lines correspond to genuine subhaloes and dashed lines to
and conclude that the Milky Way contains at least 11 + 5 × spurious subhaloes. The black line shows results for CDM-W7
11 = 66 satellites with Vmax > 5.7km s−1 within 280 kpc. and the colours lines for the WDM models, as in Fig. 1. The
Using the same argument as before, counting out to a radius black cross in the lower panel indicates the expected number of
of 419 kpc in the simulations to be conservative, we find that satellites of Vmax > 5.7km s−1 as derived in the text.
only the m2.3 and CDM models produces enough satellites
to satisfy the limit.
To make an estimate of the halo-to-halo scatter, we
the number of subhaloes within r200b from each of our
make use of the result of Boylan-Kolchin et al. (2010) that
models as the distribution mean and compute the prob-
the intrinsic scatter in the abundance of CDM subhaloes,
2 ability that a given halo will have at least 66 subhaloes.
σscatter , can be fit by the sum of the Poisson, σP , and intrin-
2 This probability equals 22 percent for m2.0 and 0.30 per-
sic, σI , variances:
cent for m1.6 . Therefore, we conclude on this evidence that
mWDM > 1.6 keV1 . This is a more conservative limit than
2 2
σscatter = σP + σI2 , (10) found by Polisensky & Ricotti (2011), although our choice
2
of central halo is slightly more massive than theirs. A larger
where σP = hN i and σI2 2
= sI hN i . Here, sI is a constant, suite of WDM simulations is required to determine more
which Boylan-Kolchin et al. (2010) calibrate against their precisely the variation in WDM subhalo abundance at a
simulation results and thus obtain sI = 0.18. They also given host halo mass as well as the systematic variation of
found that the probability distribution for the number of abundance with host halo mass.
subhaloes N , given the mean hN i and intrinsic coefficient
sI , is well described by the negative binomial distribution:
1
Γ(N + r) To check whether this limit is sensitive to our choice of scut ,
P (N |r, p) = pr (1 − p)N , (11) we repeated the analysis lowering scut by 20 percent. In this case
Γ(r)Γ(N + 1)
the probability for the m1.6 model increases to 2.7 percent; thus
where p = [1 + s2I hN i]−1 and r = s−2
I . We then adopt this mass is still excluded at 95 percent confidence.
c 2012 RAS, MNRAS 000, ??–21

1000.00 100
CDM
−1
m2.3
100.00 10 m2.0
m1.6
m1.5
10.00 10−2
n(r) / <n>
sub (<r)
fcumul
1.00 10−3
0.10 10−4
0.01 10−5
10 100 1000
r [ kpc ] 10 100
r [ kpc ]
Figure 13. Cumulative mass fraction in substructures as a func-

1.4 tion of radius. The black line corresponds to the CDM model,
CDM CDM-W7, while the blue, green, orange and red lines correspond
m2.3
1.2 m2.0 to the m2.3 , m2.0 , m1.6 , and m1.5 WDM models respectively.
m1.6
1.0
m1.5
dfn(<r) / dlogr
mass. It may be that better statistics might reveal differ-

0.8 ences in the radial distribution of WDM subhaloes.
The subhalo number density profiles are shallower than
0.6 that of the halo dark matter. Springel et al. (2008a) found
that the subhalo profiles are well described by an Einasto
0.4
form (see Eqn. 13 below), with r−2 = 199kpc = 0.81r200 and
0.2
αein = 0.678. The lower panel of Fig. 12 shows that, as was
the case for CDM, subhaloes lie preferentially in the outer
0.0 parts of the halo, between 100 kpc and the virial radius, even
10 100 1000 though the number density is highest in the central regions.
r [ kpc ]
The cumulative mass fraction in subhaloes as a func-
Figure 12. The radial distribution of subhaloes. Top: the spher- tion of radius is depicted in Fig. 13. As expected from the
ically averaged number density of Msub > 108 M⊙ subhaloes nor- mass functions of Fig. 11, the subhalo mass fractions in the
malised to the mean overdensity at r200b for our four WDM and WDM models are lower than for CDM. At r200b , the mass
one CDM models. The dotted line indicates the CDM main halo fractions in WDM subhaloes are approximately 5 percent,
density profile from Fig. 3, renormalised to pass through the locus less than half the value in the CDM case. There is a small,
of radial distribution points at 250kpc. Bottom: the number frac- but systematic decrease in the mass fraction with decreasing
tion of subhaloes per logarithmic interval in radius, on a linear-log
WDM particle mass.
plot. The area under the curves is proportional to subhalo num-
ber, so this plot shows that subhaloes are preferentially found in
the outer parts of the halo. The black line corresponds to the
CDM model, CDM-W7, while the blue, green, orange and red
4.3 The internal structure of WDM subhaloes
lines correspond to the m2.3 , m2.0 , m1.6 , and m1.5 WDM mod- We now consider the internal structure of WDM haloes, par-
els respectively. ticularly their radial density profiles. We begin by perform-
ing a convergence test of the profiles.
4.2 The radial distribution of subhaloes

4.3.1 Convergence of the density profiles
The number density of subhaloes of mass Msub > 108 M⊙
as a function of radius, normalized to the mean number Springel et al. (2008a) carried out a careful study of the con-
density within r200b , is shown in the top panel of Fig. 12. vergence properties of the CDM Aquarius haloes upon which
The bottom panel shows the cumulative number fraction our set of WDM halo simulations is patterned. Here we carry
of subhaloes per logarithmic radial interval. The number out an analogous study of the WDM subhaloes. We focus
density profiles of subhaloes in the different WDM models on the most extreme case, m1.5 , since this differs most from
are very similar to one another and to the CDM case. This CDM. Fig. 14 shows the density profiles of the nine most
uniformity is suprising since, as we shall see below, the cen- massive subhaloes lying within 500 kpc in the m1.5 simula-
tral densities of WDM subhaloes decrease with decreasing tion at three different resolutions (levels 2, 3 and 4). For the
WDM particle mass, making them increasingly vulnerable to subhaloes of mass > 1 × 109 M⊙ , we find that the three real-
tidal disruption. This result is reminiscent of that found by izations agree extremely well at all radii satisfying the con-
Springel et al. (2008a) that the number density profiles of vergence criterion of Power et al. (2003). For those of lower
Aquarius subhaloes are essentially independent of subhalo mass, the low resolution (level 4) examples have fewer that
c 2012 RAS, MNRAS 000, ??–21

10000 particles and although this limits the range where the
convergence test is applicable, the convergence is still very 0.10
good. NFW
Einasto (αein = 0.18)
To emphasize the differences between subhaloes simu- 0.08
lated at different resolution, we plot, in Fig. 15, the ratios
of the intermediate and low resolution density profiles to
that of their high resolution counterparts. At the smallest 0.06
Qmedian
radius that satisfies the Power et al. (2003) criterion, the
level 3 simulations are converged to better than 10 percent;
0.04
in most cases the same is true of the level 4 simulations.
There are large excursions, however, in the outer parts, be-
yond ∼ 10 kpc. These are particularly noticeable for those 0.02
subhaloes that are closer than 100 kpc from the main halo
centre, and reflect the slightly different positions within the 0.00
main halo of each of the matched subhaloes. 1.5 2.0 2.5 CDM
We can determine the mass range where the density pro- mp [keV]
files are converged by considering the ratio of circular veloci-
ties at the convergence radius of Power et al. (2003) between Figure 16. Median value of the goodness of fit statistic, Q, for
matched subhaloes at different resolution. Demanding that Einasto (blue dots) and NFW (red dots) fits to all subhaloes
deviations from the level-2 simulation should not exceed 10 of Msub > 109 M⊙ , as a function of the WDM particle mass,
mWDM . In the Einasto fits, we have fixed αein = 0.18. The error
percent, we find that the structure of level-3 subhaloes is
bars indicate the upper and lower quartiles of the distribution.
well converged for subhalo masses > 108 M⊙ whereas for
The Einasto data points are slightly offset in mWDM for clarity.
level-4 subhaloes convergence is only achieved for masses
> 109 M⊙ .
Einasto profiles, in the latter case with fixed shape parame-
ter (αein = 0.18, following Springel et al. 2008a), by plotting
4.3.2 The density profiles of subhaloes the median value of Q for each of the different models as a
function of the thermal equivalent WDM particle mass. As
We now consider the spherically averaged radial density pro-
for CDM, we find that the Einasto profile is a marginally
files of subhaloes in all four different WDM models. For the
better fit to WDM subhaloes than the NFW profile. There
CDM case Springel et al. (2008a) found that the profiles of
is little variation in the quality of the Einasto fits for the
subhaloes are well fitted by either an NFW (Navarro et al.
different values of the particle mass, but the NFW fits seem
1996b, 1997) or an Einasto (Einasto 1965; Navarro et al.
to become slightly worse with increasing mass.
2004) functional form. The NFW profile is given by:
The density profiles of subhaloes vary systematically
with the WDM particle mass. Before performing a statistical
δc ρcrit comparison, we illustrate this variation with a few examples
ρ(r) = , (12)
(r/rs )(r/rs + 1)2 of subhaloes that we have been able to match across simula-
tions with different WDM particle masses. Such matches are
where δc is a characteristic overdensity (usually expressed
not trivial because the subhaloes have masses close to the
in units of the critical density) and rs is a spatial scale that
cutoff in the initial power spectrum and thus their formation
marks the transition between the asymptotic slopes of −1
histories can vary substantially from one case to another. In
and −3. The Einasto profile is given by:
Fig. 17 we show nine examples of subhaloes where, based
αein on their positions and masses, we have been able to identify
2 r likely matches. In Fig. 18 we show the ratio of the profiles
ρ(r) = ρ−2 exp − −1 , (13)
αein r−2 to that of their CDM counterpart.
The differences amongst the profiles tend, in most cases,
where r−2 is the scale (analogous to rs ) where the profile to be larger at smaller radii. As the WDM particle mass
attains a slope of −2, ρ−2 is the density at r−2 and αein decreases, the subhalo profiles tend to become shallower. At
is a shape parameter. Springel et al. find that Einasto fits the innermost converged point, the density of the subhalo
(which have an additional free parameter) are marginally with the smallest value of mWDM is generally a factor of
better than NFW fits for CDM subhaloes even when αein is several smaller than its CDM counterpart. For example, the
fixed to a constant. m1.5 keV subhalo in the central panel of the Figs. 17 and 18
Following Springel et al. (2008a) we define a goodness is a factor of ∼ 3 less dense at the innermost converged
of fit statistic for the functional fits to the subhalo profiles point than its CDM counterpart and a factor of ∼ 2 less
as: dense than the subhalo with m2.3 keV.
The trends seen in Figs 17 and 18 reflect the fact that,
1 X for fixed cosmological parameters, haloes of a given mass
Q2 = [ln ρi − ln ρmodel (ri )]2 , (14)
Nbins form later in WDM models than in CDM (Avila-Reese et al.
i
2001; Lovell et al. 2012). We can quantify the difference by
where ρi is the density measured at radius ri , and ρmodel is comparing, for example, the central masses of haloes in our
the model density evaluated at that same radius. In Fig. 16 various models. The masses enclosed within 300 pc and 2 kpc
we show how well our subhaloes can be fit by NFW and of the centre in field haloes and subhaloes in our simula-
c 2012 RAS, MNRAS 000, ??–21

106
105
104
103 dL = 433.9 kpc

ML = 1.44×1010MO•
dL = 285.3 kpc
ML = 1.20×1010MO•
dL = 317.4 kpc
ML = 0.87×1010MO•
MH/ML =1.13 MH/ML =0.99 MH/ML =0.91
102
106
105
ρ(r)/<ρ>
104
103 dL = 255.4 kpc

ML = 0.75×1010MO•
dL = 312.6 kpc
ML = 0.71×1010MO•
dL = 81.6 kpc
ML = 0.47×1010MO•
MH/ML =0.94 MH/ML =0.97 MH/ML =1.08
102
106
105
104
103 dL = 208.7 kpc

ML = 0.13×1010MO•
dL = 89.8 kpc
ML = 0.03×1010MO•
dL = 188.1 kpc
ML = 0.02×1010MO•
MH/ML =1.23 MH/ML =1.32 MH/ML =1.58
102
0.1 1.0 10.0 0.1 1.0 10.0 0.1 1.0 10.0
r [ kpc ]
Figure 14. Spherically averaged radial density profiles for subhaloes matched between the high (level 2), intermediate (level 3), and
low (level 4) resolution versions of the m1.5 simulation. Blue corresponds to high, red to intermediate, and green to low resolution.
The density profiles are shown by thick lines down to the smallest radius at which they satisfy the convergence criterion of Power et al.
(2003), and are continued by thin lines down to a radius equal to twice the softening length. In the legend, dL is the distance of the low
resolution subhalo from the main halo centre, ML is the subhalo mass, and ML /MH is the ratio between the masses of the low and high
resolution counterparts.
tions are plotted in Fig. 19 as a function of halo mass. For tidal stripping removes material from the outer regions, leav-
field haloes (left panel) there is a clear separation at both ing the central density largely unaffected. As a result, after
radii amongst the different models: at fixed mass, the WDM falling into their host halo, objects move primarily to the
haloes have lower central masses than their CDM-W7 coun- left in Fig. 19 but the change is comparatively greater for
terparts and the enclosed mass decreases with the WDM the less concentrated WDM subhaloes than for the CDM
particle mass. For (field) haloes of mass less than 5×109 M⊙ , subhaloes. Nevertheless, an offset amongst the WDM sub-
the masses enclosed within 300pc are lower relative to the haloes and amongst these and the CDM subhaloes remains,
CDM case by factors of ∼ 4 and ∼ 3 in the m1.6 model particularly at large masses.
and m2.3 models respectively. At higher masses the differ-
ences are smaller (by factors of 2 and 3 for the m2.3 and m1.6 Another measure of central mass is provided by the
cases respectively), thus the main halo density profiles varies value of Vmax which we plot as a function of mass for field
very little for this range of mWDM . The situation is some- haloes in Fig. 20. There is a marked difference between the
what different for subhaloes (right panel), largely because CDM-W7 and the WDM haloes which, at a given mass, have
a lower Vmax . As expected, these differences decrease with
c 2012 RAS, MNRAS 000, ??–21

1.3
1.2
1.1
1.0
0.9
0.8
0.7
1.3
1.2
1.1
ρ(r)/ρH
1.0
0.9
0.8
0.7
1.3
1.2
1.1
1.0
0.9
0.8
0.7
0.1 1.0 10.0 0.1 1.0 10.0 0.1 1.0 10.0
r [ kpc ]
Figure 15. Ratio of the intermediate (level 3; red) and low (level 4; green) resolution density profiles of the m1.5 subhaloes shown in
Fig. 14 to the density profile of their high resolution (level 2) counterparts. The blue dashed line indicates the convergence radius for the
high resolution subhaloes.
increasing halo mass. At 109 M⊙ the mean value of Vmax for trated than their CDM counterparts to begin with, they are
the m2.3 case is a factor of 1.33 smaller than for CDM-W7. more susceptible to stripping once they become subhaloes
The differences in the internal structure of haloes in the (see also Knebe et al. 2002). Thus, at fixed Vmax , the values
WDM and CDM cases can be further quantified by com- of rmax in the m2.3 case are now typically only 40 percent of
paring the relation between Vmax and rmax , the radius at the values for field haloes. Even so, since the typical values
which Vmax is attained. We plot these relations separately of rmax for subhaloes with Vmax > 10 km/s are greater than
for independent haloes and subhaloes in Fig. 21. Tidal strip- 1kpc (even in the models with the smallest WDM particle
ping of CDM subhaloes causes their value of Vmax to drop
less rapidly than their value of rmax , leading to an increase
in the concentration of the subhalo (Peñarrubia et al. 2008;
Springel et al. 2008a). As may be seen by comparing the top eters. For the Aquarius simulations (which assumed WMAP1
and bottom panels of Fig. 21, the values of rmax for CDM cosmological parameters), this number decreases to 62 percent
subhaloes at fixed Vmax are typically 70 percent of the val- (Springel et al. 2008a), as can be seen by comparing the dotted
ues for field haloes 2 . Since WDM subhaloes are less concen- lines in the two panels of Fig. 20. This difference is driven pri-
marily by the higher value of σ8 in the WMAP1 cosmology which
causes haloes of a given mass to collapse earlier and thus be more
2 This number depends on the choice of cosmological param- concentrated than their WMAP7 counterparts.
c 2012 RAS, MNRAS 000, ??–21

106
105
104
d1.5 = 382.0 kpc d1.5 = 466.3 kpc d1.5 = 599.5 kpc

103 M1.5 = 4.47×1010MO• M1.5 = 1.62×1010MO• M1.5 = 1.27×1010MO•
M1.5/MCDM =1.03 M1.5/MCDM =0.90 M1.5/MCDM =0.87
102
106
105
ρ(r)/<ρ>
104
d1.5 = 293.7 kpc d1.5 = 661.6 kpc d1.5 = 303.5 kpc

103 M1.5 = 1.19×1010MO• M1.5 = 0.81×1010MO• M1.5 = 0.79×1010MO•
102
106
105
104
d1.5 = 239.0 kpc d1.5 = 316.0 kpc d1.5 = 124.0 kpc

103 M1.5 = 0.70×1010MO• M1.5 = 0.69×1010MO• M1.5 = 0.50×1010MO•
102
0.1 1.0 10.0 0.1 1.0 10.0 0.1 1.0 10.0
r [ kpc ]
Figure 17. Spherically averaged radial density profiles of subhaloes in simulations of different WDM particle mass. The subhaloes have
been matched across simulations on the basis of their position and mass. However, it should be noted that in some cases the matches
are uncertain. The different colours correspond to different WDM particle masses: red, orange, green and blue to 1.5, 1.6, 2 and 2.3 keV
respectively, while black corresponds to the CDM case. In the legend, d1.5 is the distance of the subhalo from the main halo centre in
the mWDM = 1.5keV, M1.5 is the mass of the subhalo also in this case, and M1.5 /MCDM is the ratio of this mass to that of the CDM
counterpart. As in Fig. 14 the density profiles are shown by thick lines down to the smallest radius at which they satisfy the convergence
criterion of Power et al. (2003), and are continued by thin lines down to a radius equal to twice the softening length
mass), the majority of any dSphs residing in subhaloes like strongly on the mass of the Galactic halo and all but disap-
these would not show clear signs of tidal disruption. pears if the Milky Way’s halo has a mass of 1 × 1012 M⊙ ,.
Alternatively, Lovell et al. (2012) showed the the problem
is naturally solved in a WDM model even if the mass of
4.4 The abundance of the most massive subhaloes the Galactic halo is 2 × 1012 M⊙ . Their WDM model, cho-
Boylan-Kolchin et al. (2011, 2012) showed that the most sen to have a particle mass only just compatible with the
massive subhaloes in the Aquarius halo simulations are much Lyman-α constraints of Boyarsky et al. (2009a,b) (but not
too massive and concentrated to host the brightest dSph with the more recent constraint quoted by Viel et al. 2013)
satellites of the Milky Way. Parry et al. (2012) reached the is the m1.5 model of the current study.
same conclusion using gasdynamic simulations of the Aquar- The Milky Way contains three satellites, the LMC, SMC
ius haloes. This discrepancy was called the ‘too big to fail and Sagittarius, that are brighter than the brightest dSph,
problem’ by Boylan-Kolchin et al. Subsequently Wang et al. Fornax. The ‘too big to fail problem’ consists of having sub-
(2012) showed that the extent of the discrepancy depends stantially more than three massive subhaloes within 300 kpc
c 2012 RAS, MNRAS 000, ??–21

1.2
1.0
0.8
0.6
0.4
0.2
1.2
1.0
ρ(r)WDM/ρ(r)CDM
0.8
0.6
0.4
0.2
1.2
1.0
0.8
0.6
0.4
0.2
0.1 1.0 10.0 0.1 1.0 10.0 0.1 1.0 10.0

r [ kpc ]
Figure 18. Ratio of the density profiles of matched subhaloes in simulations of different WDM particle mass relative to the mass of the
CDM counterpart. The colours are as in Fig. 17 as is the use of thick and thin lines.
in the simulations whose properties are incompatible with 5 DISCUSSION AND CONCLUSIONS
the measured kinematics of the nine brightest dSphs, specif-
ically with the measured masses within their half-light radii Although the existence of dark matter was inferred in the
(where masses can be robustly measured from the data; 1930s, its identity remains one of the most fundamental un-
Walker et al. 2009; Wolf et al. 2010). In our WDM simula- solved questions in physics. The evidence points towards
tions we thus count the number of subhaloes within 300 kpc dark matter being made of as yet undiscovered elemen-
of the main halo centre that have circular velocity profiles tary particles. Over the past thirty years attention has fo-
of amplitude greater than the measured half-light circu- cused on cold dark matter (Peebles 1982; Davis et al. 1985;
lar velocities of the 9 brightest dSphs plus their 3σ errors Bardeen et al. 1986) but this is not the only possibility. For
(Walker et al. 2009; Wolf et al. 2010; Lovell et al. 2012). We example, the lightest sterile neutrino in the νMSM model
find 1, 1, 3 and 4 subhaloes in the m1.5 , m1.6 , m2.0 and m2.3 (Asaka & Shaposhnikov 2005) would behave as warm dark
WDM models respectively and 6 in CDM-W7. Thus, all our matter, generating very similar structures to CDM on scales
WDM simulations are free of the ‘too big to fail problem’ larger than bright galaxies but very different structures
even in a 2×1012 M⊙ Galactic halo. Note that if we knew the on smaller scales (Lovell et al. 2012; Macciò et al. 2012;
mass of the Milky Way halo precisely, this argument could, Schneider et al. 2012).
in principle, be used to set an upper limit on the (thermal) In this study we have carried out a series of high res-
WDM particle mass. olution N-body simulations of galactic haloes in universes
dominated by WDM, taking as the starting point one of the
c 2012 RAS, MNRAS 000, ??–21

109
108
M(<R) [ MO• ]
107
106
R=0.3kpc R=0.3kpc
R=2.0kpc R=2.0kpc
105
108 109 1010 108 109 1010 1011
Mhalo [ MO• ] Mhalo [ MO• ]
Figure 19. Central masses of field haloes (left) and subhaloes within r200b (right), evaluated within radii of 2 kpc (crosses) and 300pc
(circles) as a function of total mass. Different colours correspond to different simulations: black for CDM-W7, blue, green, orange and
red for models m2.3 , m2.0 , m1.6 , and m1.5 respectively.
the tension between the structure of CDM subhaloes orbit-

ing in haloes of mass ∼ 2 × 1012 M⊙ and the kinematical
haloes data for Milky Way satellites (Boylan-Kolchin et al. 2012;
Wang et al. 2012). We then performed a series of simula-
tions of WDM haloes, using as initial conditions the same
fluctuation phases and linear power spectrum of Aq-A, suit-
Vmax [ km/s ]
ably truncated to represent WDM with (thermal equivalent)

particle masses in the range 1.5 keV to 2.3 keV. Our main
simulations correspond to level-2 resolution in the notation
10 of Springel et al. (2008a), but we also ran simulations at
CDM
m2.3 lower resolution to establish convergence.
m2.0
m1.6 N-body simulations with a resolved cutoff in the initial
m1.5 power spectrum undergo artificial fragmentation in filaments
(Bode et al. 2001; Wang & White 2007). The resulting spu-
108 109 1010 rious structures need to be identified before the simulations
Mhalo [ MO• ] can be analyzed. This is best done in the initial conditions:
we found that the spurious fragments evolve from disc-like
Figure 20. Mhalo vs. Vmax for field haloes. The black dots show
structures that are much flatter than the progenitors of
the data for the CDM-W7 simulation and the black line repre-
sents the mean relation. The colour dots show data for the WDM genuine haloes. The sphericity of structures in the initial
simulations: blue, green, orange and red for models m2.3 , m2.0 , conditions therefore provides a robust flag for spurious ob-
m1.6 , and m1.5 respectively. The mean relation is shown only for jects which we supplement with a mass cut, Mmin , derived
the m2.3 WDM model in which the number of subhaloes is largest from the limiting mass for genuine haloes, Mlim , inferred by
and thus the least noisy. Wang & White (2007) from simulations of hot dark matter
models. We find that a cut of Mmin = κMlim , with κ = 0.5,
captures the results from a comparison of matched haloes in
haloes from the Aquarius project of simulations of CDM simulations of different resolution. The combined sphericity
galactic haloes carried out by the Virgo Consortium (“Aq- and mass cut criteria result in clean catalogues of genuine
A” in Springel et al. 2008a). As a prelude we resimulated haloes and subhaloes.
this CDM halo replacing the cosmological parameters from The spherically averaged density profile of the main halo
the WMAP year-1 values assumed by Springel et al. to is virtually indistinguishable in the CDM and all our WDM
the WMAP year-7 values (Komatsu et al. 2011). For CDM simulations but there are large differences in the abundance
this change has the effect of lowering the central densi- and structure of their subhaloes. For WDM, the subhalo
ties of galactic subhaloes, alleviating (but not eliminating) mass functions begin to diverge from the CDM case at
c 2012 RAS, MNRAS 000, ??–21

reflected in the main halo mass, Mhost − Msub , Msub −Vmax ,
and Vmax −rmax relations, such that, for a given mass, sub-
haloes haloes in warmer dark matter models have progressively
lower central densities, lower values of Vmax and higher val-
10.0 ues of rmax relative to CDM subhaloes. These differences af-
fect the evolution of subhaloes once they fall into the main
rmax [ kpc ]
halo since less concentrated haloes are more easily stripped.

Both the abundance and the structure of WDM sub-
haloes can be compared to observational data. The require-
1.0 CDM ment that the models should produce at least as many sub-
m2.3
m2.0 haloes as there are observed satellites in the Milky Way sets
m1.6 a lower limit to the WDM particle mass. This is a very
m1.5 conservative limit since feedback processes, arising from the
0.1 reionization of gas in the early universe and supernova en-
10 100 ergy, would prevent the formation of galaxies in small mass
Vmax [ km/s ]
haloes just as they do in CDM models (e.g. Benson et al.
2002). However, the number of subhaloes above a given mass
or Vmax depends, of course, on the host halo mass (Gao et al.
2004; Wang et al. 2012). For the case we have considered, in
subhaloes
which Mhost ∼ 1012 M⊙ , we find that the WDM particle
mass must be greater than 1.5 keV or 1.6 keV depending on
10.0 whether we simply consider the observed number of satellites
or apply a correction for the limited area surveyed by the
rmax [ kpc ]
SDSS. This limit is less stringent than that limit of 3.3 keV
(2σ) inferred by Viel et al. (2013) from the clumpiness of the
1.0 Lyman-α forest of a sample of quasars at redshift z > 4, al-
CDM
m2.3 though the two results are not directly comparable because
m2.0 Viel et al. (2013) use a slightly different transfer function. In
m1.6 principle it might also be possible to set an upper limit on
m1.5
the WDM particle mass by comparing the subhalo central
0.1
10 100 densities with those inferred for the brightest satellites of
Vmax [ km/s ] galaxies like the Milky Way. Current kinematical data are
insufficient for this test but they are compatible with the
properties of the most massive subhaloes in the four WDM
Figure 21. Vmax vs. rmax for independent haloes (top) and sub-
models we have considered none of which suffers from the
haloes (bottom). The black dots show the data for the CDM-W7
simulation and the black line represents the mean relation in the
‘too big to fail’ problem highlighted by Boylan-Kolchin et al.
case. The dotted line corresponds to a ΛCDM simulation using (2012).
the WMAP1 cosmological parameters. The colour dots show data WDM remains a viable alternative to CDM, along
for the WDM simulations: blue, green, orange and red for mod- with other possibilities such as self-interacting dark mat-
els m2.3 , m2.0 , m1.6 , and m1.5 respectively. The mean relation ter (Vogelsberger et al. 2012) and cold-plus-warm mix-
is shown only for the m2.3 WDM model in which the number of tures (Anderhalden et al. 2013). Further theoretical work,
subhaloes is largest and thus the least noisy. The solid lines of the including simulations and semi-analytical calculations
top panel are reproduced in the bottom panel as dashed lines. (Benson et al. 2013; Kennedy et al. 2013) combined with
better data for dwarf galaxies offer the prospect of ruling
out or validating these models.
masses between ∼ 2 × 109 M⊙ for the m2.3 (least extreme)
and ∼ 7 × 109 M⊙ for the m1.5 (most extreme) models.
The cumulative mass functions are well fit by fitting func-
ACKNOWLEDGEMENTS
tions given in §4.1: they become essentially flat for subhaloes
masses below ∼ 7 × 109 M⊙ . The mass fraction in substruc- ML acknowledges an STFC studentship (STFC grant ref-
tures within r200b is lower than in the CDM case by factors erence ST/F007299/1). CSF acknowledges ERC Advanced
between 2.4 (for m1.5 ) and 2 (m2.3 ). The radial distributions Investigator grant COSMIWAY. We would like to thank the
of subhaloes are very similar to the CDM case. anonymous referee for a careful reading of the text. The sim-
WDM haloes and subhaloes are cuspy (except in the ulations used in this paper were carried out on the Cosmol-
very inner regions - see Macciò et al. (2012) and Shao et al. ogy Machine supercomputer at the Institute for Computa-
(2013)) and are well fit by NFW profiles, and even better tional Cosmology, Durham. LG acknowledges support from
by Einasto profiles. However, the central density of WDM the One hundred talents Program of the Chinese Academy
haloes depends on the WDM particle mass: in those cases of Science (CAS), the National Basic Research Program of
where it is possible to identify the same subhalo in CDM China (program 973, under grant No. 2009CB24901), NSFC
and different WDM simulations, the density profiles have grant No. 10973018, and an STFC Advanced Fellowship.
systematically shallower slopes in the latter which become This work used the DiRAC Data Centric system at Durham
flatter for smaller particle masses. This change of slope is University, operated by the Institute for Computational
c 2012 RAS, MNRAS 000, ??–21

Cosmology on behalf of the STFC DiRAC HPC Facility Gao L., White S. D. M., Jenkins A., Stoehr F., Springel
(www.dirac.ac.uk). This equipment was funded by BIS Na- V., 2004, MNRAS, 355, 819
tional E-infrastructure capital grant ST/K00042X/1, STFC Gilmore G., Wilkinson M. I., Wyse R. F. G., Kleyna J. T.,
capital grant ST/H008519/1, and STFC DiRAC Operations Koch A., Evans N. W., Grebel E. K., 2007, ApJ, 663, 948
grant ST/K003267/1 and Durham University. DiRAC is Hahn O., Abel T., Kaehler R., 2013, MNRAS, 434, 1171
part of the National E-Infrastructure. This work was sup- Hogan C. J., Dalcanton J. J., 2000, Phys. Rev. D, 62,
ported in part by an STFC rolling grant to the ICC. The 063511
data featured in the figures in this paper are available on Ibata R., Martin N. F., Irwin M., Chapman S., Ferguson
request from ML. A. M. N., Lewis G. F., McConnachie A. W., 2007, ApJ,
671, 1591
This paper has been typeset from a TEX/ LATEX file prepared Kaiser N. et al., 2010, in Society of Photo-Optical In-
by the author. strumentation Engineers (SPIE) Conference Series, Vol.
7733, Society of Photo-Optical Instrumentation Engineers
(SPIE) Conference Series
Kamada A., Yoshida N., Kohri K., Takahashi T., 2013,
REFERENCES
JCAP, 3, 8
Anderhalden D., Schneider A., Macciò A. V., Diemand J., Kennedy R., Frenk C., Cole S., Benson A., 2013, ArXiv
Bertone G., 2013, JCAP, 3, 14 e-prints
Angulo R. E., Hahn O., Abel T., 2013, MNRAS, 434, 3337 Knebe A., Devriendt J. E. G., Mahmood A., Silk J., 2002,
Asaka T., Shaposhnikov M., 2005, Physics Letters B, 620, MNRAS, 329, 813
17 Komatsu E. et al., 2011, ApJS, 192, 18
Avila-Reese V., Colı́n P., Valenzuela O., D’Onghia E., Fir- Koposov S. et al., 2008, ApJ, 686, 279
mani C., 2001, ApJ, 559, 516 Kusenko A., 2009, Phys. Rep., 481, 1
Bardeen J. M., Bond J. R., Kaiser N., Szalay A. S., 1986, Larson D. et al., 2011, ApJS, 192, 16
ApJ, 304, 15 Libeskind N. I., Frenk C. S., Cole S., Helly J. C., Jenkins
Bell E. F., Slater C. T., Martin N. F., 2011, ApJ, 742, L15 A., Navarro J. F., Power C., 2005, MNRAS, 363, 146
Benson A. J. et al., 2013, MNRAS, 428, 1774 Lovell M. R. et al., 2012, MNRAS, 420, 2318
Benson A. J., Frenk C. S., Lacey C. G., Baugh C. M., Cole Macciò A. V., Paduroiu S., Anderhalden D., Schneider A.,
S., 2002, MNRAS, 333, 177 Moore B., 2012, MNRAS, 3204
Bode P., Ostriker J. P., Turok N., 2001, ApJ, 556, 93 Macciò A. V., Paduroiu S., Anderhalden D., Schneider A.,
Boyarsky A., Iakubovskyi D., Ruchayskiy O., 2012, Physics Moore B., 2013, MNRAS, 428, 3715
of the Dark Universe, 1, 136 Martin N. F. et al., 2009, ApJ, 705, 758
Boyarsky A., Lesgourgues J., Ruchayskiy O., Viel M., Martin N. F. et al., 2013, ApJ, 772, 15
2009a, JCAP, 5, 12 Navarro J. F., Eke V. R., Frenk C. S., 1996a, MNRAS, 283,
Boyarsky A., Lesgourgues J., Ruchayskiy O., Viel M., L72
2009b, Physical Review Letters, 102, 201304 Navarro J. F., Frenk C. S., White S. D. M., 1996b, ApJ,
Boyarsky A., Ruchayskiy O., Shaposhnikov M., 2009c, An- 462, 563
nual Review of Nuclear and Particle Science, 59, 191 Navarro J. F., Frenk C. S., White S. D. M., 1997, ApJ, 490,
Boylan-Kolchin M., Bullock J. S., Kaplinghat M., 2011, 493
MNRAS, 415, L40 Navarro J. F. et al., 2004, MNRAS, 349, 1039
Boylan-Kolchin M., Bullock J. S., Kaplinghat M., 2012, Okamoto T., Frenk C. S., Jenkins A., Theuns T., 2010,
MNRAS, 422, 1203 MNRAS, 406, 208
Boylan-Kolchin M., Springel V., White S. D. M., Jenkins Papastergis E., Martin A. M., Giovanelli R., Haynes M. P.,
A., 2010, MNRAS, 406, 896 2011, ApJ, 739, 38
Brooks A. M., Zolotov A., 2014, ApJ, 786, 87 Parry O. H., Eke V. R., Frenk C. S., Okamoto T., 2012,
Bullock J. S., Kravtsov A. V., Weinberg D. H., 2000, ApJ, MNRAS, 419, 3304
539, 517 Peñarrubia J., McConnachie A. W., Navarro J. F., 2008,
Cole S. et al., 2005, MNRAS, 362, 505 ApJ, 672, 904
Colı́n P., Valenzuela O., Avila-Reese V., 2008, ApJ, 673, Peebles P. J. E., 1982, ApJ, 263, L1
203 Polisensky E., Ricotti M., 2011, Phys. Rev. D, 83, 043506
Davis M., Efstathiou G., Frenk C. S., White S. D. M., 1985, Pontzen A., Governato F., 2012, MNRAS, 421, 3464
ApJ, 292, 371 Power C., Navarro J. F., Jenkins A., Frenk C. S., White
de Vega H. J., Sanchez N. G., 2010, MNRAS, 404, 885 S. D. M., Springel V., Stadel J., Quinn T., 2003, MNRAS,
di Cintio A., Knebe A., Libeskind N. I., Yepes G., 338, 14
Gottlöber S., Hoffman Y., 2011, MNRAS, 417, L74 Preskill J., Wise M. B., Wilczek F., 1983, Physics Letters
Diemand J., Moore B., Stadel J., 2005, Nature, 433, 389 B, 120, 127
Einasto J., 1965, Trudy Astrofizicheskogo Instituta Alma- Sawala T., Frenk C. S., Crain R. A., Jenkins A., Schaye J.,
Ata, 5, 87 Theuns T., Zavala J., 2013, MNRAS, 431, 1366
Ellis J., Hagelin J. S., Nanopoulos D. V., Olive K., Sred- Schneider A., Smith R. E., Macciò A. V., Moore B., 2012,
nicki M., 1984, Nuclear Physics B, 238, 453 MNRAS, 424, 684
Frenk C. S., White S. D. M., 2012, Annalen der Physik, Shandarin S., Habib S., Heitmann K., 2012, Phys. Rev. D,
524, 507 85, 083005
c 2012 RAS, MNRAS 000, ??–21

Shao S., Gao L., Theuns T., Frenk C. S., 2013, MNRAS, (m4 /m3 )n =Vmax (4)/Vmax (3) (where the numbers denote
430, 2346 the resolution level) we find n = 0.29. The high resolution,
Somerville R. S., 2002, ApJ, 572, L23 level 2, simulation has a particle mass 3.6 times smaller
Springel V. et al., 2008a, MNRAS, 391, 1685 than that of level 3. Therefore we expect this simulation to
Springel V. et al., 2008b, Nature, 456, 73 be complete to 10 percent at Vmax = 4.2km s−1 . A similar
Springel V. et al., 2005, Nature, 435, 629 analysis for the m1.5 simulation shows that this is already
Springel V., White S. D. M., Tormen G., Kauffmann G., complete at level 3 for Vmax =5.7km s−1 .
2001, MNRAS, 328, 726 We have checked the validity of this approach by
Strigari L. E., Frenk C. S., White S. D. M., 2010, MNRAS, analysing the original Aquarius Aq-A2 and Aq-A1 sim-
408, 2364 ulations. The Aq-A1 simulation has a particle mass of
Tikhonov A. V., Gottlöber S., Yepes G., Hoffman Y., 2009, 1.7 × 103 M⊙ , a factor of ∼ 8 smaller than the level 2 simula-
MNRAS, 399, 1611 tions. We find that at Vmax =5.7km s−1 the Aq-A2 subhalo
Tollerud E. J. et al., 2012, ApJ, 752, 45 Vmax function deviates by 8 percent from the Aq-A1 result.
Tollerud E. J., Bullock J. S., Strigari L. E., Willman B., The suppression of small subhaloes in WDM models should
2008, ApJ, 688, 277 result in better subhalo completeness in this case compared
Vera-Ciro C. A., Helmi A., Starkenburg E., Breddels M. A., to CDM in this mass range (c.f. convergence between levels 3
2013, MNRAS, 428, 1696 and 2 for m1.5 ). We therefore conclude that we have lost no
Viel M., Becker G. D., Bolton J. S., Haehnelt M. G., 2013, more than 8 percent of the ‘true’ number of subhaloes in the
Phys. Rev. D, 88, 043502 m2.3 simulation and even fewer in the warmer models.
Viel M., Lesgourgues J., Haehnelt M. G., Matarrese S.,
Riotto A., 2005, Phys. Rev. D, 71, 063534
Vogelsberger M., Zavala J., Loeb A., 2012, MNRAS, 3127
Wadepuhl M., Springel V., 2011, MNRAS, 410, 1975
Walker M. G., Mateo M., Olszewski E. W., Peñarrubia J.,
Wyn Evans N., Gilmore G., 2009, ApJ, 704, 1274
Walker M. G., Mateo M., Olszewski E. W., Peñarrubia J.,
Wyn Evans N., Gilmore G., 2010, ApJ, 710, 886
Walsh S. M., Jerjen H., Willman B., 2007, ApJ, 662, L83
Wang J., Frenk C. S., Navarro J. F., Gao L., Sawala T.,
2012, MNRAS, 424, 2715
Wang J., White S. D. M., 2007, MNRAS, 380, 93
White S. D. M., 1994, ArXiv Astrophysics e-prints
Willman B. et al., 2005, ApJ, 626, L85
Wolf J., Martinez G. D., Bullock J. S., Kaplinghat M.,
Geha M., Muñoz R. R., Simon J. D., Avedo F. F., 2010,
MNRAS, 406, 1220
York D. G. et al., 2000, AJ, 120, 1579
Zavala J., Okamoto T., Frenk C. S., 2008, MNRAS, 387,
364
Zolotov A. et al., 2012, ApJ, 761, 71
APPENDIX A: CONVERGENCE STUDY

For several dwarf spheroidal satellites of the Milky Way it
is possible to measure the circular velocity at the radius en-
compassing half the light in a relatively model-independent
way Walker et al. (2009); Wolf et al. (2010). The smallest
measured value is 5.7 km s−1 for Leo IV. The circular ve-
locity at the half-light radius is a lower bound on Vmax .
Therefore, to compare with Milky Way data, we need the
number of subhaloes in the simulations with Vmax greater
than 5.7km s−1 . It is important to check that the simula-
tions resolve all these subhaloes.
We have performed a convergence study using the level
4, level 3, and level 2 simulations for two of the WDM mod-
els. For the m2.3 model, the subhalo Vmax function at level
4 deviates by 10 percent from that in the corresponding
level 2 simulation at a value of Vmax = 11km s−1 ; the level
3 subhalo Vmax function deviates by the same amount at a
value of Vmax = 6km s−1 . The particle masses in the level 4
and level 3 simulations differ by a factor of 8. If we write
c 2012 RAS, MNRAS 000, ??–21

The Properties of Warm Dark Matter Haloes

Uploaded by

Copyright:

Available Formats

The Properties of Warm Dark Matter Haloes

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Properties of Warm Dark Matter Haloes

Uploaded by

Copyright:

Available Formats

Mon. Not. R. Astron. Soc. 000, ??–21 (2012) Printed 20 May 2014 (MN LATEX style file v2.

The properties of warm dark matter haloes

2 National Astronomical Observatories, Chinese Academy of Science, Beijing, 100012, China

Accepted *** Received ***; in original form ***

1 INTRODUCTION ligible thermal velocities during the era of structure forma-

c 2012 RAS, MNRAS 000, ??–21

c 2012 RAS, MNRAS 000, ??–21

Simulation mWDM [keV] α[h−1 Mpc] Mth [M⊙ ] mν=1.12

which we define as the mass within a top hat filter which,

c 2012 RAS, MNRAS 000, ??–21

c 2012 RAS, MNRAS 000, ??–21

c 2012 RAS, MNRAS 000, ??–21

Figure 5. A region of a WDM simulation performed at two dif-

c 2012 RAS, MNRAS 000, ??–21

c 2012 RAS, MNRAS 000, ??–21

of the four WDM models. For those subhaloes that instead

c 2012 RAS, MNRAS 000, ??–21

0.8 m2.3 m2.0 nWDM /nCDM = (1 + Mhm M −1 )β , (7)

0.4 are essentially insensitive to the velocity anisotropy of the

c 2012 RAS, MNRAS 000, ??–21

limit to mWDM by correcting the observed number of SDSS

c 2012 RAS, MNRAS 000, ??–21

Figure 13. Cumulative mass fraction in substructures as a func-

mass. It may be that better statistics might reveal differ-

4.2 The radial distribution of subhaloes

c 2012 RAS, MNRAS 000, ??–21

c 2012 RAS, MNRAS 000, ??–21

103 dL = 433.9 kpc

103 dL = 255.4 kpc

103 dL = 208.7 kpc

c 2012 RAS, MNRAS 000, ??–21

c 2012 RAS, MNRAS 000, ??–21

d1.5 = 382.0 kpc d1.5 = 466.3 kpc d1.5 = 599.5 kpc

d1.5 = 293.7 kpc d1.5 = 661.6 kpc d1.5 = 303.5 kpc

d1.5 = 239.0 kpc d1.5 = 316.0 kpc d1.5 = 124.0 kpc

c 2012 RAS, MNRAS 000, ??–21

0.1 1.0 10.0 0.1 1.0 10.0 0.1 1.0 10.0

c 2012 RAS, MNRAS 000, ??–21

the tension between the structure of CDM subhaloes orbit-

ably truncated to represent WDM with (thermal equivalent)

c 2012 RAS, MNRAS 000, ??–21

halo since less concentrated haloes are more easily stripped.

c 2012 RAS, MNRAS 000, ??–21

c 2012 RAS, MNRAS 000, ??–21

APPENDIX A: CONVERGENCE STUDY

c 2012 RAS, MNRAS 000, ??–21

You might also like

Accepted * Received *; in original form ***