Background Starlight Polarimetry in the Age of SOFIA and Gaia Dan - - PowerPoint PPT Presentation

Background Starlight Polarimetry in the Age of SOFIA and Gaia

Dan Clemens (Boston University)

NSF/AST 14‐12269 & 18‐14531 , USRA SOF_4‐0026, & NASA NNX15AE51G gratefully acknowledged

1

Findings First, Talk Second

1. The combination of Gaia DR2 with near‐infrared (NIR) Background Starlight Polarimetry (BSP) is transformative

– Distances to dark clouds to <3% uncertainty – Can classify stars as foreground, embedded, background, and deep background – Better star cluster stellar memberships

• Proper motions are better than distances

2. The HAWC+ far‐infrared (FIR) imaging polarimeter on SOFIA is fabulous

– Sensitive tool for performing FIR thermal emission polarimetry (TEP) – Combined with NIR BSP ‐> B‐field characterized over wide range of scale sizes

3. B‐field strengths from Davis‐Chandrasekhar‐Fermi (DCF) method ‐> too low

– Forward modeling of cluster star Position Angle dispersions (PA) reveals BSP noise bias – Bayesian posteriors for (B) also favor lower PA, higher B strengths – Foregrounds and deep backgrounds both contribute PA noise, masking target cloud value

2

BSP Review

3

Background Stars (Unpolarized Light*) Molecular (or atomic) cloud or zone with B‐field aligned dust grains

BPlane of Sky

Starlight with impressed Linear polarization signal (P || BPOS) Observer Sees: Andersson+’13 Estimate B‐field Strength as 4

∆ ∆

(Davis, Chandrasekhar & Fermi)

We use Mimir – NIR Imaging Polarimeter for BSP

1. Galactic Plane survey (GPIPS) of 3,237 10x10’ FOVs like this one 2. Individual Dark Clouds located off the Galactic plane, like L1544

4

• FOV = 10x10 arcmin at

1024x1024 pixels at 0.6”/pixel on the 1.8m Perkins telescope in Flagstaff, AZ

• H (1.6m) and K (2.2m) stellar

polarimetry.

• 3. Outer Galaxy fields towards

Perseus Spiral Arm with

• pen stellar clusters

Cluster Stars = Ideal, Same‐Distance, Multiple BSP Probes

• Must isolate cluster stars from field (fore + background) in the FOV
• Past techniques: spatial over‐density +/‐ Color‐Magnitude Diagram

placements

– Select “members” based on probability of being in the over‐density and falling into the Giant or Main Sequence branches (Mercer+’05) – Fit colors for age, metallicity, distance, reddening (Hoq & Clemens ’15) – Not very robust, nor accurate if Giant branch is poorly populated, esp when using NIR colors

5

Observer Foreground Stars

Target Cloud

Star Cluster Background Stars

(J‐K) mH

CMD of NGC663 in JHK with best‐fit Isochrone (H&C ’15)

Then came Gaia DR2 in April 2018…

• Parallax, proper motions, optical g‐band magnitudes

– to > 10 kpc, mg < 20 mag

• We fetched Gaia data for 30+ open cluster fields in Perseus

– Test B‐field changes from Perseus Spiral arm to interarm regions – Clusters selected to be in front of, within, and behind Perseus Arm

• Matched Gaia stars to Mimir‐observed NIR polarimetry stars (simple 1.5” cone)

– Incredible match rate

• On average, ~925 Mimir polarization‐measured stars per FOV  900 (97%) matched to Gaia stars
• More than half of the cluster FOVs had < 10 stars that did not match.
• Reiterate: Nearly every star with measured polarization now has a distance measure, also
• Created Bayesian Markov Chain Monte Carlo (MCMC) code to separate cluster

members from field stars

– Each star has likelihood of being drawn from the cluster or the field; greatest one kept – Cluster likelihood based on gaussians in 3‐D of parallax, proper motion in RA, pm in Dec

6

Cluster Parameters Were Recovered Stunningly Well

7

S/N ~ 80‐380

For FOV containing cluster NGC 1245: (left) Stars across FOV (right) parallax vs proper motion Field Stars

100 pc 1 kpc 10 kpc 100 kpc

Cluster Parameters Were Recovered Stunningly Well

8

S/N ~ 80‐380

For FOV containing cluster NGC 1245: (left) Stars across FOV (right) parallax vs proper motion Field Stars

100 pc 1 kpc 10 kpc 100 kpc

Cluster Member Stars

MCMC Cluster Separation Recovers the Color‐Mag. Diagram

• Color (J‐K) – magnitude (H)

diagram for one cluster

– Cluster member stars populate Giant and Main Sequence Branches well – Field stars look like, well, field stars – Polarization fraction uncertainty P below 5˚ to H=15.5

9

Gaia Parallaxes + NIR BSP – Cloud, Cluster Distances

• Classically, cloud distance methods plot

E(B‐V) or AV vs stellar distance

– Looking for a jump associated with the cloud – Difficult, if AV varies across the face of the cloud (which it does) – Polarization percentage, P, often has the same problem

• Try using Pol. PA as the changing property

– Versus Gaia DR2 distances (w/ uncertainties)

• Better, use Stokes U and Q simultaneously

– Look for jumps at the same distance(s) in both

10

GF9‐2 distance of 27010 pc from PA jump with parallax. Fit using MCMC (Clemens+’18)

Gaia Parallaxes + NIR BSP – Cloud, Cluster Distances

• Classically, cloud distance methods plot

E(B‐V) or AV vs stellar distance

– Looking for a jump associated with the cloud – Difficult, if AV varies across the face of the cloud (which it does) – Polarization percentage, P, often has the same problem

• Try using Pol. PA as the changing property

– Versus Gaia DR2 distances (w/ uncertainties)

• Better, use Stokes U and Q simultaneously

– Look for jumps at the same distance(s) in both

11

GF9‐2 distance of 27010 pc from PA jump with parallax. Fit using MCMC (Clemens+’18)

Gaia, NIR Polarimetry, and Stokes U, Q steps

• Used MCMC to find

number(s), location(s) of jumps in Stokes U, Q in

• pen cluster FOVs
• Field stars in blue
• Cluster members in red
• NGC 2262 located at

about 4.740.1 kpc

– Behind two distinct foreground layers – One with no P, one with significant P

12

Gaia, NIR Polarimetry, and Stokes U, Q steps

• Used MCMC to find

number(s), location(s) of jumps in Stokes U, Q in

• pen cluster FOVs
• Field stars in blue
• Cluster members in red
• NGC 2262 located at

about 4.740.1 kpc

– Behind two distinct foreground layers – One with no P, one with significant P

13

B‐field strengths – Davis‐Chandrasekhar‐Fermi method

• 4

∆ ∆ ; where V is gas velocity dispersion,  is gas mass

density, and PA is BSP PA dispersion.

• PA can be found many ways

– Unweighted frequentist dispersion – Weighted frequentist dispersion – Boostrap with resampling – Gaussian fitting of PA distributions

• PA can be corrected for observational uncertainty, if uniform in the sample

– Works for TEP; ∆ ∆

and is the same for each pixel

• Not so for BSP because of the wide range of stellar brightnesses ‐> range of

– Forward‐model from observations to uncover ∆ for BSP

14

Forward‐Modeling : Cluster Fields ‐> PA vs mH; Cluster N(mH)

• Thousand(s) of stars per Mimir FOV

– 16 HWP angles ‐> 4 U, Q sets and robust PA – Fit run of PA vs mH for each FOV

• Predictive noise function for model stars
• Each cluster ‐> apparent magnitude

function

– Binned by magnitude – Can fit, but that misses Giant vs Main Sequence numbers too much – Keep averages in bins for modeling

• Poisson draws, using these means

15

Forward Modeling Flow of Computation

16

• 1. Establish Cluster

Membership with Gaia + MCMC

• 2. Create apparent

magnitude bins

• 3. For the FOV, characterize polarization

uncertainty as function of apparent brightness

• 4. Draw Poisson‐based

numbers of stars for each mag bin to create mock cluster with same magnitude distribution

• 5. Form model Stokes

U, Q for each mock

• star. Use <U>, <Q>

from real cluster plus PA(mH) from fit.

• Compute PA for model cluster stars

– “model output PA” (I use bootstrap) – Find it is < PA(observed) for the real cluster

• Whoops! Forgot to include PA(true)

– The “astrophysical” dispersion ‐> DCF Method – Include a synthetic version of this dispersion

• Implemented as draws from a gaussian distribution of

PAs with G = PA (true)

• Future ‐> based on 2nd Order Structure Function?
• Sweep of PA(true) steps from 0 to 40˚ by 0.5˚
• (Re)Compute model output PA
• Do for 50,000 mock clusters for each PA(true)

– Generates distribution functions of model

• utput PA
• Each cluster shows different relation between

PA (true) and PA (model output)

– Reflects S/N of observations; stellar brightnesses

17

PA(true)

(model output)

Compare Cluster PA(obs.) to Infer Mock Cluster PA(true)

• PA(obs.) values select a range
• f likely PA (true) values

– True values < observed values

• Convolve PA (true)

distributions with gaussian modeled PA (obs.)

• Result is a posterior

distribution (PA(true))

– Very gaussian looking

18

NGC 869 7˚ 4˚

Orange = Convolved Values Blue = Gaussian Fit

(true)

Modeling findings

1. BSP PA values are noise‐biased

– Like Ricean, but more complex

2. Forward modeling recovers PA (true) distributions

– PA(true) < PA (obs) ‐> B(true) > B(obs+DCF)

3. ∝ ∆

+ ∆ ~ gaussian →

– Favors lower PA, higher B again

4. Realization – PA noise bias, both observational and astrophysical, will be added by foreground and deep background ISM

– Increasing PA (obs) and decreasing the DCF B‐strength we would have estimated – Either eliminate them, or must forward model their effects, too

19

target

1/PA

Foreground adds noise Deep background adds noise

It’s not hopeless – GF9‐2 Mimir + SOFIA/HAWC+

20

GF9‐2 YSO

• Quick progress can be made by choosing a

clean astrophysical laboratory

– No foreground ISM – No deep background ISM

• GF9 Filamentary Dark Cloud

– With GF9‐2 YSO – Distance = 27010 pc (our MCMC)

• Near‐Quiescent low‐mass early SF

– ~Lowest luminosity Class 0 YSO – 0.3 L – Minimal outflow found

• Is its B‐field still intact, or was it modified?
• B‐field Probes:

– Near‐IR BSP w/ Mimir – SOFIA HAWC+ E‐Band (216 m) Polarimetry – Optical I‐band BSP (Poidevin & Bastien ’06) – Submm 850m Planck Polarization (XIX ’15)

Follow the FIR Flux…

21

• Select synthetic beam placements that:
• 1. Are centered on the brightest Stokes I

emission (thin contours in the figure)

• 2. Are offset from each other so as to be

independent samples (the circles don’t cross)

• Select synthetic beam sizes so that:
• 1. Maximize the number of synthetic

beams with good PSNR

• 2. Don’t give away too much angular

resolution

• Tried a range of gaussian sizes, found best

at 4 HAWC+ pixels FWHM

• 6 beams with PSNR > 1.6 (PA < 18˚);

11 not detected

Blue = Mimir H‐band; Green = K‐band; Orange = HAWC+ Mapped Region; Black contours = 216m Intensity; Circles = synthetic beams – Black non‐detections and Magenta detections; Red = HAWC+ BPA (from Clemens+’18)

Combining all B‐field probes – Plane of Sky Magnetic Field Orientation vs Offset from YSO

• B‐field orientation is unchanged from ~3pc to ~6000 AU; No B‐field disruption near the YSO
• No strong evidence of B‐field orientation change with 

22

Comparisons with Models

23

• Hull+’17
• Simulations with

synthetic observations

• Multi‐scale
• GF9‐2 observations are

most consistent with strong B‐field case

• BPA uniform from 3pc

to 6000AU

• Last column, top two

rows

• Seems GF9‐2 YSO formed in

a fairly strong B‐field region

Findings (the final time…)

1. The combination of Gaia DR2 with near‐infrared (NIR) Background Starlight Polarimetry (BSP) is transformative

– Distances to dark clouds to <3% uncertainty – MCMC is a powerful tool here – Can classify stars as foreground, embedded, background, and deep background – Better star cluster stellar memberships – again, MCMC is powerful

• Proper motions are better than distances

2. The HAWC+ far‐infrared (FIR) imaging polarimeter on SOFIA is fabulous

– Sensitive FIR thermal emission polarimetry (TEP) – more so with smart smoothing – Combined with NIR BSP ‐> B‐field characterized over wide range of scale sizes

3. B‐field strengths from Davis‐Chandrasekhar‐Fermi (DCF) method ‐> too low

– Forward modeling of cluster star Position Angle dispersions (PA) reveals BSP noise bias – Bayesian posteriors for (B) also favor lower PA, higher B strengths – Foregrounds and deep backgrounds both contribute PA noise, masking target cloud value

24

Boston University Team

• From left:

– Catherine Cerny

• Joined in 2018 Summer – Gaia expert

– Genevieve Schroeder (U. Rochester)

• Summer 2017 REU intern
• Flew on SOFIA tomorrow in September for

HAWC+ observations of L1448

– Jordan Montgomery, now Dr. J. Montgomery

• At MIT Lincoln Labs, along with Sadia Hoq,

John Vaillancourt, Mike Pavel…)

• Flew on SOFIA 2016 December

– Sofia Kressy

• BU Senior Undergrad – lead on Mimir Orion

project (HRO, Gaia, …)

– Adham El‐Batal

• 3rd year grad student
• Just back from GAIA workshop in Heidelberg
• Flew on SOFIA 2016 December

–

• Dr. Thushara Pillai
• BU Senior Research Scientist

25