The Dragonfly Telephoto Array

The Dragonfly Telephoto Array

Pieter van DokkumBob AbrahamAllison MerrittJielai Zhang

Imaging the night sky

•Over past ~40 years factor of ~100 improvement for point sources:

• Imaging mag 25 —> 30

• Spectroscopy mag 21 —> 26

•However, low surface brightness sky remains relatively unexplored: limits have not changed much in past 30+ years!

Tal et al 2009 - low surface brightness survey of nearby ellipticals

28-29 mag/arcsec2

David Malin - early 1980s (?)

Low surface brightness science

•Faint galaxies: dwarfs have very low surface brightness

Koposov et al 2015 - satellites of Magellanic Clouds


•Stellar halos and substructure / tidal debris around galaxies

Bullock & Johnston


•Lot of other things!

Comets

Light echos

Galactic cirrus

Intra-cluster light

Why so little progress?

•Building bigger telescopes / more sensitive instruments does not help

•Limitations have to do with design choices that make large, fast telescopes possible

VISTA telescope

What is needed?

•Signal from structures that are much bigger than the seeing scale: high sensitivity requires small focal ratio

•Night sky, and many objects in it, are >1000x brighter than the hoped-for regime: need high dynamic range and therefore well-behaved PSF

Small focal ratio

•Fast telescopes exist: modern reflectors have f=1-1.5, to provide a wide enough field for their large aperture

LSST optical design

Sloan telescope

Small focal ratio

•Moves energy to the wings of the PSF

COST: large secondary mirror —> large obstruction in light path

•Support structure causes diffraction

Mirrors are bad, too

•Mirrors themselves also cause scattering, because of micro-roughness and dust

2. Characterization of scattering in an optical resonator

-0.85o

1.00o

-0.64o

0o

0.71o

0o

Figure 2.2: Figure consisting out of 25 CCD images of the scatter from a single mirror.In the center an obscuration blocks the direct beam. The speckles are formed by scatterdue to surface roughness of the mirror.

The speckle pattern is not caused or influenced by edge-diffraction of the mirror as thediameter of the spot (Airy-disk) on the mirror (at L1 = 36 cm away from the pinhole) is small(1.22λL1/D ≈ 2 mm) as compared to the size of the mirror. Furthermore, the spot on themirror is also small as compared to the relevant dimensions of Fig. 2.2, so we can neglect thefinite size of the illuminated area and treat it approximately as a point scatterer in our analysisof the angle dependence of the scatter.

The standard way to quantify the distribution and the total amount of scatter of a mirror isexpressed by the so-called Bidirectional Reflectance Distribution Function (BRDF) and theTotal Integrated Scatter (TIS) [18, 29], respectively. The BRDF is defined as

BRDF =1P0

dPdΩcosθs

, (2.1)

where dP is the optical power scattered into a projected solid angle dΩcosθs, θs is the scat-tering angle, and P0 is the incident energy from the surface. The cosθs-term is a correctionto adjust the illuminated area on the mirror to its apparent size when viewed from the scat-ter direction. When the BRDF is integrated over the solid angle, where θs ranges from 0 toπ/2 and φ from 0 to 2π , the TIS is found. A correction for the cosθ -term is made in thisintegration. The connection between the TIS and the RMS surface roughness σ , is givenby [29]

TIS =!4πσ

λ

"2, (2.2)

assuming that the light is normally incident on the surface. As the scatter was observed to benicely rotational symmetric, we can use data from one radial direction only. To calculate theBRDF over a larger angular range than found in Fig. 2.2 some additional images were made.To limit the fluctuations in the offset (to ∼ 10 units on the 216 scale of the 16 bit camera)

8

Klassen 2006

PSF of conventional telescope

•Net effect: reflecting telescopes produce complex PSFs, which limit low surface brightness studies to ~29 mag/arcsec2

axis from bright stars. The ghost is the result of light from a starreflecting off the CCD, traveling back up the telescope, reflect-ing off the corrector, and ultimately returning to the CCD. Thisghost was masked out of all calibration and science images. Themost distinct reflections beyond 2′ are caused by the dewar win-dow and the filter, and appear as bumps in the PSF at roughly2.5′ and 10′ from the star. These are the reflections that will bemodeled and removed.

The sources of the reflections can be confirmed by using thesize of the reflections to determine the extra path length traveledby the reflected light. The size of the two largest reflections cor-respond to a reflecting surface 3.6 cm away from the CCD,which matches the position of the filter. The bottom surfaceof the filter (the side closest to the CCD) causes the brightestreflections, and can be easily seen in all the example images.The top surface of the filter produces a much fainter reflection,which is barely apparent (see Fig. 4) as a ring slightly beyondthe bright annulus of the bottom surface’s reflection. There isalso a reflection off the bottom surface dewar window, whichlies between the CCD and the filter, and a fainter reflectionoff the top of the dewar window. The size of the bottom reflec-

tion indicates that the window is 0.6 cm away from the CCD asexpected. The reflection off the top of the dewar window wasfaint enough to be far below the noise level on realistic fore-ground stars near the science fields, and so we do not attemptto model it. Similarly, multiple bounces between the reflectingsurfaces and the CCD will contribute a small amount of light atlarge radii, but using the measured reflectivities, we calculatethat all of the secondary reflections will have surface bright-nesses fainter than μV ¼ 30 mag arcsec"2 in our Arcturusimages, well below our detection limit. A summary of the re-flecting surfaces and the brightness of their reflections is pre-sented in Table 1.

These reflections have been modeled in Zemax, an opticalray-tracing program, to confirm their source and the causesof the offset between stars and their reflections. The model ofthe telescope included all optical surfaces in the light path, in-cluding the Schmidt corrector plate, primary and Newtonianmirrors, filter, dewar window, and CCD. In the Burrell Schmidtthere is also a contribution to the shift from the convex curvatureof the detector, which is on the order of a 100 μm heightdifference between the center and the edges of the CCD. The

FIG. 4.—Two 450 s exposures of Arcturus, with the star positioned in opposite corners of the field of view. The optical center is toward the top right of the image on theleft, and toward the bottom left on the right image. The images saturate to black at μV ¼ 21:2.

TABLE 1

SUMMARY OF THE REFLECTIONS THAT ARE MODELED AND REMOVED

Reflection Surface . . . . . . . . .Distance to Surface

(cm) ReflectivityReflection Radius

(arcmin)Surface Brightness(mag arcsec"2)

Dewar window bottom . . . . . 0.64 0.4% 2.7 20.7Filter bottom . . . . . . . . . . . . . . . 3.3 1.6% 17.0 23.0Filter top . . . . . . . . . . . . . . . . . . . 4.2 0.2% 19.5 25.2

NOTE.—These numbers use a CCD reflectivity averaged over the filter passband of 10% (Lesser, M., 2009,private communication). The listed surface brightness is for a 450 s exposure of Arcturus (MV ¼ "0:04).

1270 SLATER, HARDING, & MIHOS

2009 PASP, 121:1267–1278

This content downloaded from 128.100.89.97 on Tue, 18 Jun 2013 11:18:52 AMAll use subject to JSTOR Terms and Conditions

the top of the Figure 7. These are not necessarily the exact posi-tions of the removed features, since the reflections are offsetfrom the center of the star and hence are not radially symmetricabout the star. This also causes the radial profile to show asmooth transition instead of the sharp cutoff at the reflectionedges as seen in the images.

3. ATMOSPHERIC AND INTERNAL SCATTERING

After removing the modeled reflections from each pointing,our bright star images were co-added to obtain a mean point-spread function for all positions on the detector, which is shownby the middle line in Figure 7. We expect that point-spread func-tion after reflection removal to be independent of position in thefield of view. This is attested by the small scatter between theprofiles of the different pointings, as shown by the light grayshading.

The PSF can be divided at 5′ into two regions with signifi-cantly different behavior. Inside of 5′, the behavior of the profileis determined by the effects of diffraction, turbulent scattering inthe atmosphere, and complex internal reflections in the correc-tor. In this region the profile exhibits a steep dependence ondistance from the star that becomes shallower further out. Ap-proximating different parts of the profile inside of 5′ as powerlaws produces varying results, with power law slopes rangingfrom !3 to !2:4.

Inside of 10″, the profile should be dominated by the behav-ior of turbulent scattering in the atmosphere, along with diffrac-tion. Assuming the turbulence follows Kolmogorov, statisticsthis inner PSF should be fit by a Moffat function (Racine 1996).However, in our data the 1.45″ pixels do not adequately samplethe PSF in this inner region and we can only observe the outerwings of the turbulent scattering.

Between roughly 10–200″, a number of bright reflectionsappear in the profile. These reflections are caused by multiplebounces within the corrector lens, and exhibit complex andvarying shapes due to the figuring of the corrector. This part ofthe radial profile has been measured in many other telescopeswith results often indicating an r!2 behavior (King 1971; Surmaet al. 1990; Racine 1996) or slightly steeper (Gonzalez et al.2005; and see Bernstein 2007 for a more complete review).

FIG. 6.—Same images as in Fig. 4 after removal of the reflections and PSF. The images saturate at μV ¼ 22:1, roughly half the brightness of the images in Fig. 4.

FIG. 7.—Radial profile of Arcturus in the original images (top solid line), afterremoving reflections (middle dotted line), and after removing both reflectionsand scattered light (bottom solid line). In the top two profiles, the line indicatesthe mean of the profiles of individual images, while the bottom line is the profileof co-added image. The shaded regions indicate the maximum scatter betweenindividual observations. The horizontal lines indicate the sizes of the reflections,although the exact area affected will vary slightly since the reflections are notconcentric about the star. The read noise level for single exposures is equivalentto 27:6 mag arcsec!2.


2009 PASP, 121:1267–1278


is contaminated by high-latitude galactic dust, which reflectslight from the Galactic disk (e.g., Sandage 1976; Witt et al.2008). It is unlikely that any of the green-colored structurein this area is intracluster light. Faint ICL structures can indeedbe seen emanating from M87, as was previously observed inMihos et al. (2005).

To emphasize the impact of the scattered star light, we per-formed one reduction of the data without any reflection removalor star subtraction. This is shown on the left in Figure 8. Thedifference between the properly star-subtracted image andthe reduction without star subtraction is shown in Figure 9. Themost significant contribution to the excess light comes fromthree eighth-magnitude stars to the west of M87. In total, theflux from foreground stars in the field is equivalent to a singlestar of magnitude MV ¼ 3:7. As discussed in § 3, since at least1.2% of the star light is scattered to radii beyond 2.4′, there is

FIG. 8.—Virgo mosaic without any star subtraction is on the left, and on the right is the mosaic with full reflection modeling and star subtraction. North is up and east isto the left, and M87 is in the upper left. The mosaic is ∼2:5° on each side. In both images red is μV < 27:1 mag arcsec"2, green is between 27.8 and 28.0, and blue isfainter than 28:0 mag arcsec"2.

FIG. 9.—This image is the difference between the reduction with no star sub-traction and the image with the fully modeled reflection and PSF subtraction.Black indicates an excess of light in the image without star subtraction, andsaturates at 2.5 ADU (μV ¼ 27:8).

FIG. 10.—Cumulative histogram of the scattered light observed from stars inthe mosaic of Virgo. The solid line indicates the fraction of pixels that receivedscattered light at a given surface brightness. The dashed line shows the fractionof pixels that are masked due to contamination by bright stars.


2009 PASP, 121:1267–1278


Slater, Harding, & Mihos 2009

Ideal telescope

•Ideal telescope has no mirrors and an unobstructed light path refractor!

Refractors

•Except for solar telescopes, refractors have been dead for astronomy for a century

•But they are alive and well in the real world!

Refractors



Refractors



Superbly coated, perfectly baffled, optically-fast (f/2.8) refractor

Latest generation of telephoto lenses

•Many optical elements (bad) …

Latest generation of telephoto lenses

•Many optical elements (bad) …

•… BUT have special coatings (good)

the first element’s image side (shown by the broken line in Figure 7), so we decided to form a microcrystalline alumina film on this surface.

Unfortunately, we could not obtain adequate anti-reflection results when we formed a microcrystalline alumina film directly on the first element’s image-side surface. The reason was that the first lens element used glass with a high refractive index (nd) of 1.84. Since the microcrystalline alumina film’s refractive index profile transitioned continuously from 1.4 to 1.0, a very large refractive index gap remained at the lens-film boundary and produced large-amplitude reflected waves. The many small, phase-shifted reflected waves created within the microcrystalline alumina film had no chance of canceling out these larger reflected waves.

The ideal solution would be to extend the microcrystalline alumina film’s refractive index profile so that it started transitioning from 1.84, the same refractive index of the first element’s glass material. The hot-water deposition of alumina microcrystals, however, depends on chemical properties inherent to alumina; therefore, substituting another substance with a higher refractive index would not produce the same kind of microcrystallization. Moreover, even if we were able, hypothetically, to develop a microcrystalline film whose refractive index began transitioning from 1.84, when it came time to apply the film to a lens with a different refractive index, we would be forced to develop yet another custom microcrystalline film with a different refractive index profile. By no means is this a feasible solution when coping with many different lens refractive indexes.

Realizing this, we tried reducing the reflectance by inserting an intermediate layer between the first lens element and the microcrystalline alumina film. The idea is that the microcrystalline alumina film cancels reflections arising from the refractive index transition from 1.0 to 1.4, and the intermediate layer, functioning as a single-layer anti- reflective coating, reduces reflections arising from the refractive index transition from 1.4 to 1.84. The benefit of this approach is that the intermediate layer’s refractive index and thickness can be adjusted to match different lens refractive indexes.

Figure 8 shows the refractive index profile of the anti-reflective coating formed on the first lens element. In order to match the glass’s 1.84 refractive index, the microcrystalline alumina film was designed with a thickness

Fig.8 Refractive index profile of SWC

68nm 220nm

Lens1.4

Ref

ract

ive

Inde

x

Thickness (nm)

MicrocrystallineAlumina

1.0 Inte

rmed

iate

laye

r

1.56

1.84

68nm 220nm

Lens1.4

Ref

ract

ive

Inde

x

Thickness (nm)


1.0 Inte

rmed

iate

laye

r

1.56

1.84


68nm 220nm

Lens1.4

Ref

ract

ive

Inde

x

Thickness (nm)


1.0 Inte

rmed

iate

laye

r

1.56

1.84

68nm 220nm

Lens1.4

Ref

ract

ive

Inde

x

Thickness (nm)


1.0 Inte

rmed

iate

laye

r

1.56

1.84

Fig.9ޓMeasured reflectance of (a) SWC and (b) Multi–coating

(a) (b)

0.00.20.40.60.81.01.21.41.61.82.0

400 450 500 550 600 650 700Wavelength (nm)

Ref

lect

ance

(%)

0°15°30°45°

Angle ofincidence

0.00.20.40.60.81.01.21.41.61.82.0

400 450 500 550 600 650 700Wavelength (nm)

Ref

lect

ance

(%)

0°15°30°45°

Angle ofincidence


(a) (b)

0.00.20.40.60.81.01.21.41.61.82.0

400 450 500 550 600 650 700Wavelength (nm)

Ref

lect

ance

(%)

0°15°30°45°

Angle ofincidence

0.00.20.40.60.81.01.21.41.61.82.0

400 450 500 550 600 650 700Wavelength (nm)

Ref

lect

ance

(%)

0°15°30°45°

Angle ofincidence

SPIE-OSA/ Vol. 7652 765203-6

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/01/2013 Terms of Use: http://spiedl.org/terms

the first element’s image side (shown by the broken line in Figure 7), so we decided to form a microcrystalline alumina film on this surface.

Unfortunately, we could not obtain adequate anti-reflection results when we formed a microcrystalline alumina film directly on the first element’s image-side surface. The reason was that the first lens element used glass with a high refractive index (nd) of 1.84. Since the microcrystalline alumina film’s refractive index profile transitioned continuously from 1.4 to 1.0, a very large refractive index gap remained at the lens-film boundary and produced large-amplitude reflected waves. The many small, phase-shifted reflected waves created within the microcrystalline alumina film had no chance of canceling out these larger reflected waves.

The ideal solution would be to extend the microcrystalline alumina film’s refractive index profile so that it started transitioning from 1.84, the same refractive index of the first element’s glass material. The hot-water deposition of alumina microcrystals, however, depends on chemical properties inherent to alumina; therefore, substituting another substance with a higher refractive index would not produce the same kind of microcrystallization. Moreover, even if we were able, hypothetically, to develop a microcrystalline film whose refractive index began transitioning from 1.84, when it came time to apply the film to a lens with a different refractive index, we would be forced to develop yet another custom microcrystalline film with a different refractive index profile. By no means is this a feasible solution when coping with many different lens refractive indexes.

Realizing this, we tried reducing the reflectance by inserting an intermediate layer between the first lens element and the microcrystalline alumina film. The idea is that the microcrystalline alumina film cancels reflections arising from the refractive index transition from 1.0 to 1.4, and the intermediate layer, functioning as a single-layer anti- reflective coating, reduces reflections arising from the refractive index transition from 1.4 to 1.84. The benefit of this approach is that the intermediate layer’s refractive index and thickness can be adjusted to match different lens refractive indexes.

Figure 8 shows the refractive index profile of the anti-reflective coating formed on the first lens element. In order to match the glass’s 1.84 refractive index, the microcrystalline alumina film was designed with a thickness


68nm 220nm

Lens1.4

Ref

ract

ive

Inde

x

Thickness (nm)


1.0 Inte

rmed

iate

laye

r

1.56

1.84

68nm 220nm

Lens1.4

Ref

ract

ive

Inde

x

Thickness (nm)


1.0 Inte

rmed

iate

laye

r

1.56

1.84


68nm 220nm

Lens1.4

Ref

ract

ive

Inde

x

Thickness (nm)


1.0 Inte

rmed

iate

laye

r

1.56

1.84

68nm 220nm

Lens1.4

Ref

ract

ive

Inde

x

Thickness (nm)


1.0 Inte

rmed

iate

laye

r

1.56

1.84


(a) (b)

0.00.20.40.60.81.01.21.41.61.82.0

400 450 500 550 600 650 700Wavelength (nm)

Ref

lect

ance

(%)

0°15°30°45°

Angle ofincidence

0.00.20.40.60.81.01.21.41.61.82.0

400 450 500 550 600 650 700Wavelength (nm)

Ref

lect

ance

(%)

0°15°30°45°

Angle ofincidence


(a) (b)

0.00.20.40.60.81.01.21.41.61.82.0

400 450 500 550 600 650 700Wavelength (nm)

Ref

lect

ance

(%)

0°15°30°45°

Angle ofincidence

0.00.20.40.60.81.01.21.41.61.82.0

400 450 500 550 600 650 700Wavelength (nm)

Ref

lect

ance

(%)

0°15°30°45°

Angle ofincidence

SPIE-OSA/ Vol. 7652 765203-6

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/01/2013 Terms of Use: http://spiedl.org/terms

Standard multi-coating“Sub-wavelength” coating

The Dragonfly Telephoto Array

•First test: in basement

toy flashlight

paperclip

M51, with single telephoto lens (Megantic dark sky preserve)

NGC7626 group 400mm f/2.8 II (20 minutes)

standard lensNGC7626 group 400mm f/2.8 II (20 minutes)

0 10 20 30 40 5010-6

10-4

0.01

1

Radius (arcmin)

Log 1

0(Normalized

Flux)

Dragonfly Array

Burrell Schmidt

0.1 0.5 1.0 5.0 10.0 50.0

28

26

24

22

20

18

16

14

12

Log10 Radius (arcmin)

Surface

Brightness

(mag/arcsec2)

Dragonfly Array

Burrell Schmidt

Canon f400 f/2.8 II lenses have ~10x less scatter than the (superb) Burrell Schmidt

Abraham & van Dokkum 2014 [Dragonfly overview paper] Also (independent confirmation): Sandin et al 2014

•Dragonfly is currently a 0.46m, f/0.89 refractor

•2x3 degree field of view

•2.9’’ pixels

•Fully robotic - operates every clear night

•Dragonfly is currently a 0.46m, f/0.89 refractor

•2x3 degree field of view

•2.9’’ pixels

•Fully robotic - operates every clear night

Dragonfly can* reach ~32 mag/arcsec2No stellar halo around M101 13

a b

M101 M31

van Dokkum et al 2014

* After subtracting stars and significant binning. May not apply to all areas. Observe responsibly: results obtained in the past offer no guarantee for the future.

Early science results (not for this meeting..)

No stellar halo around M101

Seven dwarfs in the M101 field (Merritt et al 2014)

Ultra-diffuse galaxies in the Coma cluster

Plans

•Completed survey of ~10 nearby luminous spiral galaxies - results very soon (Merritt et al, Zhang et al)

•Upgrade! Will be going from 10 to 50 lenses

•Upgraded array will be equivalent to refractor with aperture of 1 m and focal ratio of 0.4

Summary

•Telephoto lenses offer way to (finally!) image the sky at surface brightness levels >29 mag/arcsec2

•Dragonfly Telephoto Array has been operating for ~1 year; already several surprising results. Upgrade under way!

Documents

The Dragonfly Telephoto Array