Thersttworestri tions(2.9and2.10)retrievestarstypi allyasso iatedwith
thehalo,inparti ulardistantmainsequen eFstars(seeTable3fromCoveyetal.
(2007)). This sele tion however, an be signi antly ontaminated by quasars
and white dwarf-M dwarf pairs, whi h are abundant in (but not restri ted to)
the
−0.2 < g − r < 0.3
range (seeFigure 2.3). To redu e thepresen e of theseinterlopersandsele tthebulkoftheFstarspopulation,weapplyrestri tions2.11
(basedonTable4in Coveyetal.(2007))and2.12. Constraint2.13ensuresthat
the nal sour es are at most as metal ri h as the Sun (to a ount for possible
ontributionsfrommetal-ri hsatellites)andnotmoremetal-poorthan0.003times
theSun.
The de rease in interlopers attained by applying restri tions 2.11, 2.12, and
2.13 omparedtoonlyapplyingrestri tions2.9and2.10isillustratedinFigure2.3,
wherethered dots indi ate thenalsele tion ofhalo near-MSTOstarsand the
bla k dots represent the whole atalogue of star-like sour es. It is lear that
thenalsele tionof near-MSTOstars doesnotspanthewhole rangeofsour es
en ompassedbetween
g − r = 0.2
andg − r = 0.3
. Theee t ofthe[ F e/H]
andM r
sele tionisfurther illustratedin Figure2.4.Using the estimated absolute brightness, we al ulate the distan e modulus
and the helio entri distan e for all the near-MSTO stars. We dene distan e
modulusbinsofsize
∆ µ = 0.2
magand∆ µ = 0.4
mag,and ountthenumber ofnear-MSTOstarsper binforea hgroup of elds (A,B,C,...). The hoi e of
dis-tan ebinsismotivatedbya ompromisebetweenmaximisingtheradial distan e
resolutionandminimisingthePoissonnoiseinthestellarnumber ounts. Wetest
this ompromise byexploring two distan emodulusbinsizes, whi h orrespond
todistan ebinsizesoftheorderof
10 2
p and10
kp ,respe tively.Wethen al ulatethenumberdensityper binanditsun ertaintyasfollows:
ρ l,b,D = N l,b,∆µ
0 .2 · ln(10) · D hC 3 · ∆Ω · ∆µ ,
(2.14)E ρ =
r ( ρ
√ N
) 2 + ( ρ
√ n f ields
) 2 ,
(2.15)where
∆Ω
isthearea overed byea h group,D hC
is thehelio entri distan e,l
and
b
arethegala ti oordinatesandN l,b,∆µ
isthenumberofstarsper bininagivendire tionofthesky. Parti ularly,
∆Ω = 4 π
41253 Σ(deg 2 )
(2.16)and the area of ea h group (
Σ
) depends on the individual area of ea h eld ontributingtoit(Table2.1).The resultsfor these number density al ulations an beseen in Figure 2.5,
whereweplotthelogarithmi numberdensityagainstthegala to entri distan e 3
,
R GC
, forea h group (orline of sight). For thisand thesubsequentanalysis, weonly onsiderbinswith
R GC > 5kpc
,|z| > 10
kp (toavoidtheinner regionsoftheGalaxy)andadistan emodulusof
µ ≤ mag lim −4.5
(toguaranteea ompletesampleofthefaintestnear-MSTOstars 4
).
Figure2.5showsthat thedensityprolesde reasequitesmoothlyfor
40 − 60
kiloparse sandformostofthelinesofsight.
2.3.2 Fitting pro edure
Wet several models of theGala ti stellar number density distribution to the
data,rangingfromabasi axisymmetri powerlawtomore omplexmodelswith
triaxiality and a break in thepower law. Themodels take the following
math-emati alforms, with
x
,y
, andz
being the artesian gala to entri oordinates withtheSunat(8,0,0) kp (Malkin2012):- Axisymmetri model
ρ(x, y, z) = ρ 0 ·
x 2 + y 2 + z 2 q 2
n/2 ,
(2.17)where
q = c/a
isthepolaraxis ratioortheoblatenessofthehalo;- Triaxial model
ρ(x, y, z) = ρ 0 ·
x 2 + y 2
w 2 + z 2 q 2
n/2 ,
(2.18)where
w = b/a
istheratiobetweentheaxesintheGala ti plane;- Brokenpower law,withvaryingpower indexat
R break
ρ(x, y, z) =
ρ 0 · (R ellip ) n in , R ellip < R break
ρ 0 · (R ellip ) n out · R n break in −n out , R ellip ≥ R break
(2.19)R ellip =
x 2 + y 2 + z 2 q 2
1/2 ;
3
R GC = p R 2 + z 2
where
R
andz
aretheradialandverti al oordinatesonthe ylindri algala to entri referen e system.4
This onstraintguaranteesthattherearenodistan e ompletenessissuesduetoourspe i
typeofstellartra ersandduetothedierentdepthsofourelds. Theonlysubsetae tedby
in ompletenessisthatof
mag lim − 5.0 < µ < mag lim − 4.5
forthestarsinthe4.5 < M r < 5.0
range;anditsaveragelossisof
20%
overthetotalnumberofnear-MSTOstars(−2.0 < M r <
5.0
)inthesamedistan e range. Several testson dierentupperdistan ethresholds forthedensity proles show that thedistan e modulus onstraint of
µ ≤ mag lim − 4.5
isenoughtoguaranteethatallthelinesofsight ontributerobustdensitymeasurementsatthefurthest
distan esandthatthein ompletenessin
mag lim −5.0 < µ < mag lim −4.5
forthe4.5 < M r < 5.0
near-MSTOstarshasnostatisti allysigni antee tonthebesttparameters.
Figure2.5: Logarithmi stellardensityprolesversusdistan e forthenearMain
Sequen e turno point stars (near-MSTO) from the elds in groups A (green
ir les),B( yansquares),C(bluedownwardtriangles),D(yellowupward
trian-gles),E(redpentagons),F(pinkhexagons),G(purplediamonds)andH(orange
leftwardtriangles). Theirsymbolsmat h thoseinFigure2.1.
- Brokenpower law,withvaryingpower indexandoblatenessat
R break
ρ(x, y, z) =
ρ 0,in ·
x 2 + y 2 + z 2
q 2 in
n in /2 , R GC ≤ R break
ρ 0,out ·
x 2 + y 2 + z 2
q out 2
n out /2 , R GC > R break ,
(2.20)
wheretheinner power lawistto datathat meets
R GC ≤ R break
andtheouterpower lawisappliedtodatathat meets
R GC > R break
.Wetallthesemodelstothedatausingthe" urve-t"methodfromPython's
S ipy.optimize, whi h uses the Levenberg-Marquardt algorithm for non-linear
least squares tting. The obje tive fun tion takes the form of a
χ 2
, and wealso al ulatearedu ed
χ 2
foranalysispurposes,χ 2 =
N data
X
i=1
ρ data,i − ρ model,i E ρ,i
2
,
(2.21)χ 2 red = χ 2 N data − N params
,
(2.22)where
N data
andN params
arethenumber ofdata pointsandthenumber offreeparameters,respe tively.
The inuen e of the photometri un ertainties on the density proles and
the best t parameters is evaluated through a set of Monte Carlo simulations
that randomly modify the
g
,r
,i
,u
magnitudes of ea h star within the limits ofthephotometri un ertainties. Through this method wend that the variation
oftheMonte Carlo bestt parameters aligns with theun ertainties of ourbest
t parameters (derived from the se ond derivative of thets by the " urve-t"
method). The entre ofthesevariationsiswithin
1 σ
ofourdire t ndings.Wet all models tofour data sets: with and without[known℄ substru tures
andbinnedin
0 .2
and0 .4
magnitude ells.Inthiswaywe an he ktherobustnessofourresultstodierentbinningoptionsandweareableto omparewhatwould
betheee tofsubstru tureonourunderstandingofthesmoothhalo ifwewere
to ignore it or unable to re ognize it as su h. Spe i ally, we ut the distan e
binsat
R GC < 25
kp in groupEto avoid ontributionsbythestru turesinthe dire tionofthegala ti anti entre(theMono erosring,theAnti entreStru tureandtheEasternBandStru ture),thedistan ebinswithin
15 < D hC < 40
kp ingroupGto avoid ontributionsbytheSagittariusstream, and thedistan ebins
within
20 kpc < D hC < 60
kp in group H to avoid ontributions againby the Sagittariusstream.2.3.3 Results
Thebesttparametersforea hmodelresultingfromttingthesefourdatasets
aresummarizedin Tables 2.2to2.5. Table2.2 ontainstheresultsofttingthe
∆ µ = 0.2
magbinneddataex ludingregionswithsubstru ture,whereasTable2.3 ontainstheresultsofttingto allthe0 .2
mag bins. SimilarlyTable2.4 oversthetsto
∆ µ = 0.4
magdatawithoutsubstru turebins,andTable2.5,toall0 .4
magbins. Theredu ed
χ 2
andtheinitialparametershavealsobeenre ordedinthesetables.
We ompare the tting results for the four dierent data sets re orded in
Tables2.2to2.5andndthatthetsforwhi hthesubstru turehasbeenmasked
signi antlyoutperformthosethathavebeenallowedtotalltheavailabledata.
Thedieren eon
χ 2 red
forallthesemodelsandbinsizesisinevery aseatleastafa torof
2 .3
orlarger. Wendthatallowingthemodelstotdatathat ontainssubstru turedoesnotae tlargelymostofthestru turalparameters(polaraxis
ratiosare ompatible within the un ertainties and power law indi es have lose
values)ex eptthatitde reasesthediskaxisratio
w
byatleast10%
,suggestinga strongdeparturefromtheaxisymmetri modelthatisnotimpli itin theltereddata sets. Hen eforth we will restri t the remaining dis ussion to the results
derivedfromthe leanestdatasets.
Comparingtheparametersresultingfromthebesttstothemasked
0 .2
magand
0 .4
mag data, wend that thets to0 .2
mag binned data performbetterforallthemodels(
χ 2 red
ratiooftwo). Nonetheless,allthemeasurementsforthe dierentstru turalparametersinthetwodatasetsare ompatiblewithea hotherwithintheun ertainties. Thebesttsforthefourmodelsandtheirresidualsfor
oureightlinesofsightareshowninFigures2.6aand2.6bforthemasked
0 .2
magbinneddata. Itis learthatthedieren esbetweenthettedmodelsalongthese
sightlinesaresmall.
Ourdataarein on lusiveregardingtriaxiality,butare ompatiblewitheither
amildlytriaxialhaloorwithnotriaxiality. Forthe
0 .2
magdataset, thetriaxialmodel tsslightly better thanthe axisymmetri model and returns
w = 0.87 ± 0 .09
. For the0 .4
mag data set, however, the axisymmetri model ts slightly betterandthetriaxialmodelreturnsadiskaxisratio ompatiblewith1
. Inboth data sets the other best-tting parameters are pra ti ally identi al for the twomodels. Thisindi ates that the ostoftheextraparameter isnotsupportedby
the
0 .4
magdata. Thus,itishardtoderiveapre isevalueforthediskaxisratioandto on ludeifitistrulytriaxial,butaweightedaverageof
w
andthegeneralanalysisshow ondentlythat
w > 0.8
.Wein reasethe omplexityoftheaxisymmetri modelbyaddingtwodegrees
offreedomand onsideringa hangeinthepower lawindex
n
at aspe i breakdistan e
R break
(a brokenpower law). For this purpose, weusea grid ofvaluestoexplorealltheparametersex eptthedensitys alefa tor
ρ 0
,whi hweleftfreeto t (see below for the grid hara terization). This model de reases the
χ 2 red
inboththe
0 .2
andthe0 .4
magbinned ases,indi atingthat ourdataisbettert by a broken power law than by a simple axisymmetri model or a triaxial
model. Itturnsthesinglepowerlawindexfrom
n = −4.26 ± 0.06
intoalesssteepinner index
n in = −2.50 ± 0.04
and a steeper outer indexn out = −4.85 ± 0.04
(measurementshereareforweightedaveragesbetweenthe
0 .2
and0 .4
magdata).ewersurveyoftheGala ti halofromdeepCFHTandINTima
distan emodulus ells.
Model
χ 2 red ρ 0 (pc −3 ) · 10 −3 R break (kpc) n n in n out q q in q out w
axisymmetri
1.90 14 ± 6
−4.31 ± 0.09
0.79 ± 0.06
triaxial
1.86 14 ± 6
−4.28 ± 0.09
0.77 ± 0.06
0.87 ± 0.09
brokenp.l.
n 1.52 0.071 ± 0.003 19.0 ± 0.5
−2.40 ± 0.05 −4.80 ± 0.05 0.77 ± 0.03
brokenp.l.
n, q 1.99, 1.51 1 ± 3 19fixed
−3.3 ± 0.6 −4.9 ± 0.2
0.7 ± 0.2 0.88 ± 0.07
initialparameters
0.001 40.0 −3.00 −3.00 −3.50 0.70 0.70 0.8 1.00
Table2.3: SameasinTable2.2butthistimettingalltheavailabledata(in ludingthoseregions ontainingstellar ountsfromknown
substru turesanddete tedoverdensities).
Model
χ 2 red ρ 0 (pc −3 ) · 10 −3 R break (kpc) n n in n out q q in q out w
axisymmetri 4.71
8 ± 3
−4.15 ± 0.08
0.83 ± 0.06
triaxial 4.59
7 ± 2
−4.07 ± 0.08
0.82 ± 0.06
0.77 ± 0.07
brokenp.l.
n
4.240.17 ± 0.01 21.0 ± 0.5
−2.80 ± 0.05 −4.80 ± 0.05 0.84 ± 0.03
brokenp.l.
n, q
3.36,4.791 ± 2 21fixed
−3.3 ± 0.4 −5.0 ± 0.2
0.7 ± 0.2 0.89 ± 0.08
initialparameters
0.001
40.0 -3.00 -3.00 -3.50 0.70 0.70 0.8 1.00Table2.4: SameasinTable2.2butthistimettingthedatabinnedin
0.4
magdistan emodulus ells.Model
χ 2 red ρ 0 (pc −3 ) · 10 −3 R break (kpc) n n in n out q q in q out w
axisymmetri 3.89
12 ± 4
−4.26 ± 0.08
0.77 ± 0.05
triaxial 3.97
12 ± 5
−4.25 ± 0.08
0.77 ± 0.06
0.9 ± 0.1
brokenp.l.
n
2.610.11 ± 0.01 20.0 ± 0.5
−2.60 ± 0.05 −4.90 ± 0.05 0.81 ± 0.03
brokenp.l.
n, q
4.95,2.341 ± 1 20fixed
−3.2 ± 0.4 −5.0 ± 0.3
0.7 ± 0.2 0.82 ± 0.08
initialparameters
0.001
40.0 -3.00 -3.00 -3.50 0.70 0.70 0.8 1.00Table2.5: SameasinTable2.4butthistimettingalltheavailabledata(in ludingthoseregions ontainingstellar ountsfromknown
substru turesanddete tedoverdensities).
Model
χ 2 red ρ 0 (pc −3 ) · 10 −3 R break (kpc) n n in n out q q in q out w
axisymmetri 9.13
7 ± 2
−4.10 ± 0.07
0.81 ± 0.05
triaxial 9.19
7 ± 2
−4.07 ± 0.07
0.81 ± 0.06
0.86 ± 0.09
brokenp.l.
n
7.740.058 ± 0.005 20.0 ± 0.05
−2.40 ± 0.05 −4.8 ± 0.05 0.84 ± 0.03
brokenp.l.
n, q
6.05,9.20.6 ± 0.9 20fixed
−3.1 ± 0.4 −4.9 ± 0.2
0.7 ± 0.2 0.86 ± 0.07
(a)Fitteddensityprolesforthe
0.2
magbinneddata.Figure2.6: Densityprolesin de imallogarithmi s aleandthebest tmodels
fromTable2.2 (ttedto masked
0 .2
binned data). Thedierent linesrepresenttheaxisymmetri (bla k solid line), the triaxial (greendashed line), the broken
powerlawwithvaryingpower index (reddottedline)and thebrokenpower law
withvaryingpowerindexandoblateness(bluedashed-dotted-dottedline)models.
Thegreyareasdenotedatathathavebeenmaskedfromthettingtoa ountfor
thepresen eofsubstru ture.
(b)Data-to-modelresidualsforthe
0.2
magbinneddata.Figure2.6: Residualsbetweenthedata andthebestt modelsfrom panel2.6a.
Thedierentlinesandtheshadedareasfollowthesame olourandsymbol ode.
Italsoin reasesthe entralvalueofthepolaraxisratio
q
withintheun ertainties, fromaweightedq = 0.77 ± 0.04
toa weightedq = 0.79 ± 0.02
. Globally,thediskaxisratioseemsto bethemoststableparameterthroughout thedierentmodel
tstoourdata, returninga moderatelyoblatehalo.
Finally we x the break distan e at the best t value found by the broken
power law model (
R break = 19
kp and20
kp for the0 .2
and0 .4
mag binneddata,respe tively)andaddanotherparametertoit,allowingnotonly
n
,butalsoq
to hangeatthebreakdistan e. Wendthatthebesttstothismodelreturnsu h largeerrorbarsfortheinner halo that,inpra ti e, ityieldsun onstrained
measurements:
∆ ρ 0 ≤ ρ 0
,∆ n in
is12-18% ofn in
and∆ q in
is30%ofq in
.Weexplore ea h modelto investigatepossible parameter degenera ies,
toler-an e rangesand potentiallo al minima in our best ts. For this we x all the
parametersinthefourmodelsex eptthedensitys alefa tor
ρ 0
,andwerunthetsa rossa gridofparameter values. Inparti ular, thegridsarebuiltfollowing
q 2 , w 2 ∈ [0.1, 2.0; δ = 0.05]
,n ∈ [−5.0 − 1.0; δ = 0.1]
,n in ∈ [−4.0, −1.0; δ = 0.1]
,n out ∈ [−7.0, −3.0; δ = 0.2]
andR br ∈ [15, 50; δ = 1]
, whereδ
is thein remental stepforea hparameter. Wendthatthereisadegenera ybetweenR br
andn in
forthesimplebrokenpowerlawmodelforbothbinnings(seeFigure2.7).
Finallyourmeasurementsforthedensitys alefa tor
ρ 0
(ρ
atR GC = 1
kp ) aretheresultoflargeextrapolationsandmerelyserveas normalizationsforourts. Forthat reasonwedonotdis ussthese valuesin detail.