1: \documentclass[preprint2]{aastex}

2: 

3: \shorttitle{Atmospherically Induced Ellipticities}

4: \shortauthors{De Vries et al.}

5: \renewcommand{\textfraction}{0.00}

6: \renewcommand{\topfraction}{1.00}

7: 

8: \newcommand{\ro}{r$_{\rm o}$\,}

9: 

10: 

11: \begin{document}

12: 

13: \title{Image Ellipticity from Atmospheric Aberrations}

14: 

15: \author{W. H. de Vries\altaffilmark{1,2}, S. S. Olivier\altaffilmark{3}, S. J. Asztalos\altaffilmark{3}, L. J. Rosenberg\altaffilmark{3,4}, and K. L. Baker\altaffilmark{3}}

16: 

17: \altaffiltext{1}{University of California, Department of Physics, 1 Shields Ave, Davis, CA 95616}

18: \altaffiltext{2}{Institute for Geophysics and Planetary Physics, LLNL, L-413, 7000 East Avenue, Livermore, CA 94550}

19: \altaffiltext{3}{Lawrence Livermore National Laboratory, 7000 East Avenue, Livermore, CA 94550}

20: \altaffiltext{4}{University of Washington, Department of Physics, Box 35160, Seattle, WA 98195}

21: \email{devries1@llnl.gov}

22: 

23: 

24: \begin{abstract}

25: 

26: We investigate the ellipticity of the point-spread function (PSF)

27: produced by imaging an unresolved source with a telescope, subject to

28: the effects of atmospheric turbulence. It is important to quantify

29: these effects in order to understand the errors in shape measurements

30: of astronomical objects, such as those used to study weak

31: gravitational lensing of field galaxies.  The PSF modeling involves

32: either a Fourier transform of the phase information in the pupil plane

33: or a ray-tracing approach, which has the advantage of requiring fewer

34: computations than the Fourier transform. Using a standard method,

35: involving the Gaussian weighted second moments of intensity, we then

36: calculate the ellipticity of the PSF patterns. We find significant

37: ellipticity for the instantaneous patterns (up to more than

38: 10\%). Longer exposures, which we approximate by combining multiple

39: ($N$) images from uncorrelated atmospheric realizations, yield

40: progressively lower ellipticity (as $1/\sqrt{N}$).  We also verify

41: that the measured ellipticity does not depend on the sampling interval

42: in the pupil plane using the Fourier method.  However, we find that

43: the results using the ray-tracing technique do depend on the pupil

44: sampling interval, representing a gradual breakdown of the geometric

45: approximation at high spatial frequencies.  Therefore, ray tracing is

46: generally not an accurate method of modeling PSF ellipticity induced

47: by atmospheric turbulence unless some additional procedure is

48: implemented to correctly account for the effects of high spatial

49: frequency aberrations.  The Fourier method, however, can be used

50: directly to accurately model PSF ellipticity, which can give insights

51: into errors in the statistics of field galaxy shapes used in studies

52: of weak gravitational lensing.

53: 

54: \end{abstract}

55: 

56: \keywords{atmospheric effects -- gravitational lensing}

57: 

58: \section{Introduction}

59: 

60: Statistical analyses of weak gravitational lensing of field galaxies

61: \citep[e.g.,][]{wittman00,vanwaerbeke00,bacon00} are being used as

62: probes of cosmology and are expected to provide some of the strongest

63: cosmological tests in future, large astronomical survey projects, such

64: as the Large Synoptic Survey Telescope \citep[LSST;

65: e.g.,][]{tyson01,tyson02}. These surveys will allow for the precise

66: determination of various cosmological parameters, such as the matter

67: density distribution $\Omega_m$, the cosmological constant

68: $\Omega_\Lambda$, the equation of state $w$ of the dark energy, and

69: its time derivative. This is done by accurately analyzing large

70: numbers of background galaxies as their shapes are sheared by

71: intervening large-scale structure through weak gravitational lensing,

72: the results of which are then combined with, e.g., the very accurate

73: measurements of the cosmic microwave background radiation by the WMAP

74: satellite.

75: 

76: A critical part in these analyses is the accuracy to which one can

77: measure and correct the shape of the Point Spread Function (PSF) as it

78: varies across the detector \citep[see, e.g.,][]{hoekstra04}. This PSF

79: anisotropy is largely induced by the atmosphere

80: \citep[e.g.,][]{wittman05}, and cannot easily be modeled without

81: incorporating an explicit atmosphere. It is possible to mimic the

82: effects of the atmosphere by convolving either artificially generated,

83: or high resolution HST images with a suitable PSF (see, e.g., Heymans

84: et al. 2006a for the former, and Bacon et al.  2001 for the latter

85: approach), but that still does not include effects of PSF

86: anisotropy. We therefore set out to model the behavior of the PSF as

87: it gets folded through a realistic atmosphere and telescope

88: system. Since ray-tracing methods are commonly used to simulate the

89: shearing signal of weak lensing by large scale structure

90: \citep[e.g.,][]{jain00,vale03,heymans06b}, we include both ray-tracing

91: and Fourier transform methods to calculate what the PSF should be

92: based on the phase and intensity information in the pupil plane of the

93: telescope. The ray-tracing method has the advantage of computational

94: speed compared with the Fourier transform method. However, the Fourier

95: transform method correctly treats the effects of interference, which

96: are ignored by the geometric ray-tracing method. Because of its

97: relative computational efficiency, it is useful to understand whether

98: the ray tracing approach gives adequate PSF ellipticity information.

99: More importantly, it is crucial to understand the elliptical

100: properties of the PSF in order to enable quantitative analysis of the

101: errors in statistical studies of field galaxy shapes.

102: 

103: Our modeled observational setup is characterized as follows: a generic

104: 8 m-class telescope, a turbulent atmosphere with a Kolmogorov power

105: spectrum, and a single, on-axis point-source located at infinity. The

106: ellipticity and its direction are assumed to be representative for a

107: single region over which these quantities do not vary. The angular

108: extent of these regions can be, depending on observing conditions,

109: larger than an arcminute \citep[e.g.,][]{asztalos06}. Since the

110: typical LSST exposure will be 15 seconds, we also investigate the time

111: dependence of the atmospherically induced ellipticities, and how it

112: imposes limits on the ability to measure them.

113: 

114: \subsection{Layout of the paper}

115: 

116: The paper is organized as follows. In \S~\ref{simmethod} and

117: \S~\ref{calc}, we discuss the atmospheric simulations and the methods

118: applied to calculate the PSF ellipticity. Then in \S~\ref{results}, we

119: describe the results for both the ray-tracing method (RTM) and the

120: Fourier transform method (FTM). The main difference between these two

121: is the way one models image formation by the telescope. The RTM traces

122: the geometric path of rays from the pupil plane onto the focal plane,

123: whereas the FTM applies a Fourier transform to the complex

124: electromagnetic field in the pupil plane in order to calculate the

125: resulting image. The latter correctly incorporates effects of

126: interference, unlike the RTM. The geometric approximation of the RTM

127: has implications for shape measurements in the image plane, which we

128: quantify in this paper.

129: 

130: The results are subdivided into the effects of pupil plane sampling

131: (\S~\ref{sampling}), the variation of ellipticity as a function of

132: exposure time and the presence of wind (\S~\ref{wind}), and seeing

133: (\S~\ref{seeingEff}). The latter is approximated by using varying

134: ratios of D / \ro (where D is the aperture size of the telescope, and

135: \ro is the coherence length of the atmosphere). Longer exposure times

136: are simulated by increasing the number of completely independent phase

137: screens (which is a function of aperture diameter and wind-speed). In

138: order to check the accuracy of this approximation, we also evaluate a

139: model that includes intermediate phase screens, i.e., screens that

140: are not completely decorrelated from the previous one, but are

141: translated by a small fraction of the aperture size along the wind

142: direction.  In each of these sections we investigate the differences

143: between the RTM and FTM, which are summarized in \S~\ref{conclusion}.

144: 

145: \section{Simulations and Methods}\label{simmethod}

146: 

147: We generated random phases for the electromagnetic field in the pupil

148: of the telescope using Kolmogorov statistics to represent the effects

149: of atmospheric turbulence. The inner turbulence scale of the

150: simulations is set by the pixel size used in the simulations, to a

151: fraction of \ro. The outer scale of the turbulence has been fixed to a

152: value larger than the simulation box ($> 800$ m for the large screens

153: in \S~\ref{wind}, for instance), so that effectively the simulations

154: see a Kolmogorov turbulence spectrum with an infinite outer scale.

155: Assuming a constant magnitude in the circular pupil, we propagated the

156: field to the focal plane using a Fourier transform, and by squaring

157: the resulting magnitude, created a representation of the focal plane

158: image intensity. These images contain distinct speckle patterns for

159: different realizations of the atmosphere.

160: 

161: The ray tracing algorithm used an idealized model of a reflecting

162: telescope. A Kolmogorov phase screen was placed in the aperture of the

163: telescope and a uniform distribution of rays in the pupil was

164: assumed. The $x$ and $y$ derivatives of the phase screen were used to

165: determine the atmospherically induced deviations of the rays as they

166: propagated toward the focus of the telescope. At the focus, the

167: incoming rays are mapped onto a fiducial 2D detector grid which

168: determines the intensity distribution in the focal plane.

169: 

170: \section{Calculating Ellipticity}\label{calc}

171: 

172: \begin{figure}[t]

173: \plotone{f1.eps}

174: \caption{Representative point-spread functions, using the Fourier

175: method (left two images), and the ray-tracing method (right two

176: images). The top images are for an instantaneous realization of the

177: atmosphere, whereas the bottom images are the means for 256 such

178: instances. Notice that the top left Fourier image displays a prominent

179: speckle pattern due to interference. This pattern gets washed out over

180: time.}

181: \label{psfGal}

182: \end{figure}

183: 

184: 

185: We calculate the ellipticity of an object in the pupil plane as

186: follows. Assume we have an image for which the pixel coordinates are

187: given by $x = 1 .. N, y = 1 .. N$; the intensity in each pixel is

188: given by $I(x,y)$; the seeing is given by $\sigma$ (FWHM=2.355$\sigma$

189: for a Gaussian), and the central point source is located at $(x_c,

190: y_c)$. We then define Gaussian weights as follows:

191: 

192: \begin{equation}

193: w(x,y) = \frac{1}{\sqrt{(2\pi\sigma^2)}}

194: e^{-\frac{1}{2\sigma^2}(x-x_c)^2} . \frac{1}{\sqrt{(2\pi\sigma^2)}}

195: e^{-\frac{1}{2\sigma^2}(y-y_c)^2}

196: \end{equation}

197: 

198: 

199: \noindent and the Gaussian weighted moments: 

200: \small

201: \begin{eqnarray}

202: S_{w} = \sum_{x,y=1}^N (I(x,y) w(x,y)), \, S_{xx} = \sum_{x,y=1}^N (x^2 I(x,y) w(x,y))\nonumber\\ 

203: S_{x} = \sum_{x,y=1}^N (x I(x,y) w(x,y)), \,  S_{yy} = \sum_{x,y=1}^N (y^2 I(x,y) w(x,y))\nonumber\\

204: S_{y} = \sum_{x,y=1}^N (y I(x,y) w(x,y)), \, S_{xy} = \sum_{x,y=1}^N (x y I(x,y) w(x,y))\nonumber\\

205: \end{eqnarray}

206: \normalsize

207: 

208: \noindent This allows us to define the following quantities (all

209: weighted by $I(x,y)w(x,y)$):

210: 

211: \begin{eqnarray}

212: x_c = S_{x} / S_{w}, \quad y_c = S_{y} / S_{w}\nonumber\\

213: \left(\begin{array}{c}rxx \\ryy\\rxy \end{array}\right) = \left(\frac{1}{S_{w}S_{w}}\right) \left(\begin{array}{c}S_{xx}S_{w} - S_{x}S_{x} \\S_{yy}S_{w} - S_{y}S_{y}\\ S_{xy}S_{w} - S_{x}S_{y}\end{array} \right)

214: \end{eqnarray}

215: 

216: \noindent From these, one can calculate the image Gaussian scale

217: length $\sigma$ and the ellipticity $\epsilon$:

218: 

219: \begin{eqnarray}

220: \sigma = \sqrt{rxx+ryy}\nonumber\\

221: \epsilon = \left( \frac{((rxx-ryy)^2+(2rxy)^2}{(rxx+ryy)^2} \right)^{1/2}\nonumber\\

222: \epsilon_1 = \frac{rxx-ryy}{rxx+ryy}, \quad \epsilon_2 = \frac{2rxy}{rxx+ryy}

223: \end{eqnarray}

224: 

225: \noindent This is the method used by \citet[][KSB]{kaiser95} - see

226: also \citet{heymans06} for an overview of the different weak lensing

227: pipelines.  The weights in eqn.~1 are dependent on initial values of

228: $\sigma$, and the source centroid $(x_c, y_c)$. However, one can

229: iterate from initial guesses for the unknowns.  This algorithm

230: converges quickly (typically within a few steps), and yields values

231: for $\sigma$ and the source position that are readily verifiable. We

232: terminate the iterations when the changes in $\sigma$ are less than

233: 0.01 pixel.

234: 

235: \section{Results}\label{results}

236: 

237: \begin{figure*}[t]

238: \epsscale{1.8}

239: \plottwo{f2a.eps}{f2b.eps}

240: \caption{Ellipticity as function of the number of independent phase

241: screens and sampling rates for D / \ro = 40 (\ro = 21cm). The left

242: panel shows the results for ray-tracing, with the pupil-plane sampling

243: rates color-coded as: \ro / 2 green, \ro / 4 blue, \ro / 5 cyan, \ro /

244: 8 purple, and \ro / 16 yellow. The panel on the right shows the same

245: results (except \ro / 5) for the Fourier method. The latter method

246: clearly illustrates the expected behavior: ellipticity should be

247: independent of pupil plane sampling rates beyond Nyquist rates. The

248: progressive lowering of the curves for higher samping rates in the

249: left panel, therefore, is unphysical. The dashed line in the right

250: panel shows a $1/\sqrt{N}$ decline.}

251: \label{samp}

252: \end{figure*}

253: 

254: 

255: Most of the figures in this paper show the ellipticity behavior as a

256: function of the number $N$ of independent phase screens. A single

257: phase screen ($N=1$) therefore represents the {\it instantaneous}

258: ellipticity of a particular representation of the atmosphere (see

259: Fig.~\ref{psfGal}). In all of our subsequent discussions we assume

260: that the telescope is perfect, i.e., it does not induce image

261: aberrations.

262: 

263: Whether we apply the RTM or the FTM approach, we first create stacks

264: of 500 completely uncorrelated instances of the atmosphere (actually

265: phase-screens in the pupil plane). The resulting focal plane images

266: are then either ray-traced or calculated using an appropriate Fourier

267: transform. After this, we randomly select $N$ frames out of the 500,

268: which are then stacked, averaged, and have their ellipticity

269: calculated. We repeat this $1024/N$ times. Since there are not

270: $1024/N$ fully independent stacks present (for $N>2$), some smoothing

271: occurs, especially for the larger $N$ stacks. The figures show the

272: mean ellipticity for these $1024/N$ stacks, and the error-bars on the

273: means are approximated by the rms of the distribution divided by the

274: square root of the number of stacks ($1024/N$).

275: 

276: \subsection{Constant ratio D / \ro, varying sampling in pupil plane}\label{sampling}

277: 

278: The value of D / \ro has been fixed at 40 (\ro = 21 cm) for an assumed

279: 8.4 m pupil diameter with a central obscuration\footnote{These are the

280: current parameters for the LSST design} of 5.4 m, while the (phase)

281: sampling rate in the pupil plane increases from \ro / 2 to \ro /

282: 16. The ellipticity $\epsilon$ as function of the number of

283: independent phase screens $N$ is shown in Fig.~\ref{samp}. The left

284: panel illustrates the results for the RTM method. While each

285: individual sampling rate falls off as $1/\sqrt{N}$, they are offset in

286: ellipticity as the sampling rate increases. An increase from \ro / 2

287: to \ro / 16 more than halves the measured ellipticity for a given

288: number of independent phase screens. This is clearly not physical.

289: 

290: The following example may illuminate this behavior. Assume, for

291: instance, a simple one-dimensional cosine phase wave with frequency

292: $k$. The Fourier transform of this function produces two delta

293: functions located at $\pm k$ on the x-axis. However, since ray-tracing

294: uses the derivatives of the phase to calculate where the rays will go,

295: it produces a spread of points due to the range in derivatives. An

296: increase in the sampling rate will increase the likelihood high angle

297: rays will be modeled.  Furthermore, the derivative range becomes

298: larger for higher spatial frequencies (it goes from $-k$ to $+k$)

299: resulting in progressively more aberrant rays, regardless of the

300: sampling rate. While perhaps an extreme example, it does underline the

301: fact that 2D ray-tracing will produce a more spread-out image due to

302: rays being deflected into unphysical angles. This broadening of the

303: image then results in a lowered ellipticity as it decreases the local

304: asymmetry (remember that the ellipticity contribution of a point is

305: weighted by its distance, see Eqn. 1).

306: 

307: The FTM, on the other hand, does display the correct behavior (for the

308: exact same sets of phase screens): the ellipticity is independent of

309: the sampling rate (provided it is at least \ro / 2).  Except for the

310: sampling dependency, both methods exhibit the following

311: characteristics. First, ellipticities decrease linearly (in log-log)

312: as the number of frames increases. The slope is consistent with a $1 /

313: \sqrt{N}$ decline ($\alpha=-0.5$), as indicated by the dashed line in

314: the right panel. And second, ellipticities of individual speckle

315: images ($N=1$) are $\sim$9\% for D / \ro = 40 (\ro = 21 cm).

316: 

317: It should also be noted that there are $\sim20 - 40$ independent

318: instances of the atmosphere for an 8.4 meter telescope aperture with

319: wind-speeds of $\sim10-20$ m/s (typical, turbulence-weighted values

320: for many astronomical sites) and a 15 second exposure. Therefore,

321: these simulations predict that the raw ellipticity of a 15-second

322: exposure for a point source image through an 8.4 meter telescope with

323: D / \ro = 40 is $\sim2$\%. We explore the effects of wind in more

324: detail in \S~\ref{wind}.

325: 

326: \subsection{Binning in the image plane}\label{binning}

327: 

328: Obviously, no astronomical instrument designed for seeing-limited

329: observations will sample the PSF at the Nyquist interval for the

330: telescope diffraction pattern (we measure $\sigma\sim 60$ pixels), so

331: the next step is to see what happens to the ellipticities if one

332: progressively rebins the images of Fig.~\ref{psfGal}. The results are

333: listed in Table~1.  The ellipticities are for a stack of 32 random

334: frames with D / \ro = 40. This stack gets increasingly rebinned down

335: to scales where the PSF is barely resolved ($\sigma \sim 1$).

336: 

337: Given that the values of $\epsilon$ do not change significantly over a

338: large range of binning, it is clear that the ellipticity measurements

339: are robust and do not depend on the pixel scale. For comparison, the

340: rms spread in the value of $\epsilon$ for distinct random stacks of 32

341: images is about 60\% of the mean value of $\epsilon$, whereas the

342: listed relative range in column 3 is only $\sim 5\%$ under rebinning.

343: 

344: %table 1

345: \begin{deluxetable}{cccc}

346: \tablewidth{9cm}

347: \tablenum{1}

348: \label{bintab}

349: \tablecaption{Pixel binning effects on ellipticity}

350: \tablehead{

351:   \colhead{Binning $n$} &

352:   \colhead{$\sigma_n$ [pix]} &

353:   \colhead{$\epsilon_n$} &

354:   \colhead{Ratio $\epsilon_n / \epsilon_1$ [\%]}

355: }

356: \startdata

357: 1 & 61.82 & 0.02505 &  \\

358: 2 & 30.92 & 0.02507 & 100.1 \\

359: 4 & 15.47 & 0.02504 & 100.0 \\

360: 8 &  7.74 & 0.02513 & 100.3 \\

361: 16 & 3.89 & 0.02492 &  99.5 \\

362: 32 & 1.97 & 0.02393 &  95.5 \\

363: 64 & 1.04 & 0.02443 &  97.5 \\

364: \enddata

365: \tablecomments{The ellipticity is calculated on a random stack of 32

366: speckle patterns, with a sampling of \ro / 2, and a ratio of D / \ro =

367: 40 (\ro = 21 cm). There is no significant dependency on pixel

368: size. Note that $\sigma\sim 1$ to $2$ are typical astronomical seeing

369: disk sampling ratios.}

370: \end{deluxetable}

371: 

372: 

373: \subsection{Atmospheric model with wind}\label{wind}

374: 

375: \begin{figure}[t]

376: \epsscale{1.0}

377: \plotone{f3.eps}

378: \caption{Effect of wind on ellipticity behavior for D / \ro = 40 (\ro

379: = 21 cm). The blue dashed curve is for phase screens which are

380: translated across the aperture, and the red curve is for independent

381: instances of the atmosphere. Both these curves are calculated with the

382: Fourier method. The wind-speed $v$ is needed to convert the number of

383: frames $N$ into an elapsed time $t$ ($= ND / v$, with $D$ = 8.4 m, and

384: $v$ in units of m s$^{-1}$). Therefore, a typical 15 s LSST exposure,

385: with a wind-speed of 10 m s$^{-1}$, contains 18 frames. }

386: \label{windplot}

387: \end{figure}

388: 

389: 

390: So far we have only considered the behavior of ellipticity as function

391: of the number of uncorrelated instances of the atmosphere. As

392: described in \S~\ref{sampling}, the evolution of the PSF with

393: increasing exposure time can be estimated using the results from these

394: uncorrelated screens by associating each screen with a unit of time

395: equal to the aperture diameter divided by the wind speed, i.e., the

396: time it would take for the wind to translate a screen completely out

397: of the aperture.  However, a more realistic treatment of the PSF

398: evolution involves a more continuous translation of a Kolmogorov phase

399: screen across the telescope aperture.

400: 

401: For this purpose, we generated 3 large phase screens which contain 95

402: aperture-clearings each. The translation offset\footnote{This offset

403: should not be confused with the wind-speed $v$. All we want to make

404: sure is that we have enough numerical resolution (hence the 20 steps)

405: as the atmosphere translates across the aperture. How long it takes

406: for the atmosphere to clear an aperture does not matter for this

407: calculation. See also the caption to Fig.~\ref{windplot}.} is such

408: that an aperture-clearing takes 20 steps; from each phase screen we

409: therefore generate 1900 pupil images using the FTM. The ellipticity is

410: then calculated on combined stacks of 20 pupil images, yielding one

411: value per aperture-clearing. The results are plotted in

412: Fig.~\ref{windplot}. The blue dashed line shows the mean ellipticity

413: values for all $3\times95 / N$ independent aperture-clearings

414: (``frames''), with as error-bars the error in this mean. At $N=5$, for

415: instance, we calculated the mean of all uncorrelated instances of

416: $5\times20$ {\it consecutive} pupil images. On the other hand, the red

417: curve shows the ellipticity behavior of the sum of $N$ uncorrelated

418: instances of the atmosphere, using the same pupil plane sampling and

419: value of D / \ro = 40 (\ro = 21 cm).

420: 

421: A few things stand out. Below about 10 frames or so, the curves behave

422: differently. In case of the red line, the ellipticity jump from $N=1$

423: to $N=2$ is due to the fact that there is no ``image motion'' in the

424: $N=1$ case, whereas for $N=2$, two pupil images have been combined

425: with different PSF centroids. This raises the ellipticity beyond what

426: is there in a single PSF (one would need to shift-and-add to remove

427: this effect). It subsequently takes a few more co-added frames for

428: this centroid-offset effect to cancel out (on average the PSF has to

429: align with the optical axis since we put the source there).

430: 

431: The blue line does not suffer from this centroid-offset problem since

432: the phase-screens are continuous (only 1/20$^{\rm th}$ gets shifted

433: out between pupil images) and the PSF centroid cannot move around

434: discontinuous as a consequence. However, we do see another effect

435: present in the blue curve. Because the Kolmogorov phase screen will

436: generally have low-spatial-frequency correlations that are larger than

437: the telescope aperture, the ellipticity of the PSF is expected to

438: decrease more slowly with increasing $N$ than for the discontinuous

439: model using multiple independent phase screens. After some time (or

440: equivalently, for larger values of $N$), these low-spatial-frequency

441: correlations disappear and the slope of the blue curve steepens to

442: that of the red curve. This appears to be happening between $N=10$ and

443: $N=20$.

444: 

445: The main result of this exercise, however, is the confirmation that

446: our method of using uncorrelated instances timed at a rate equal to an

447: aperture clearing time is a valid approximation for the ellipticity

448: behavior in a statistical sense, as long as we are in the

449: long-exposure, $N > 20$ domain.

450: 

451: \subsection{Varying \ro}

452: \label{seeingEff}

453: 

454: \begin{figure}[t]

455: \epsscale{1.1}

456: \plotone{f4.eps}

457: \caption{Speckle patterns for individual phase screens at various

458: ratios of D / \ro (\ro = 21, 42, and 84 cm, respectively). Note that

459: with increasing values of \ro (left to right), the number of speckles

460: decreases, while their intensities go up. Also, the size of the

461: pattern decreases with increasing \ro.}

462: \label{speckle}

463: \end{figure}

464: 

465: We also investigated the effect of increasing the value of \ro.  The

466: expectation is that for larger values of \ro (i.e., better seeing

467: conditions) the number of speckles goes down (no atmosphere = no

468: speckles, diffraction pattern only) while their individual brightness

469: goes up (due to the conservation of flux). This is easily verified in

470: the individual speckle patterns (see Fig.~\ref{speckle}). It is not

471: clear, however, what the behavior of the RTM method with respect to

472: the FTM will be. In \S~\ref{sampling}, we noticed that the RTM method

473: significantly underestimates the actual ellipticity depending on the

474: sampling rate. If the relative offsets are constant then one might be

475: able to come up with a particular sampling rate for the ray-tracing

476: case that best matches the actual ellipticities (for the D / \ro = 40

477: case the best matching sampling looks to be about \ro / 3, see

478: Fig.~\ref{samp}).

479: 

480: The results are presented in Fig.~\ref{seeing}, with the solid lines

481: representing the FTM results, and the dashed lines are for the RTM

482: method. If we focus on the solid lines first, it is clear that the

483: ellipticities increase with increasing values of \ro at the same

484: number of stacked frames.  This can be qualitatively understood in

485: terms of the decreasing number of speckles distributing themselves in

486: a progressively less circular pattern due to the smaller number

487: statistics.

488: 

489: \begin{figure}[t]

490: \epsscale{1.0}

491: \plotone{f5.eps}

492: \caption{Measured ellipticities $\epsilon$ (see eqn. 4) for different

493: values of \ro, with values from top to bottom of 42 (red, D / \ro

494: =20), 21 (green, D / \ro = 40), and 10.5 cm (blue, D / \ro = 80),

495: respectively. The solid line curves have been calculated using the

496: Fourier method, and have a fixed pupil plane sampling rate of \ro /

497: 4. The dashed lines are calculated using ray-tracing, and are

498: color-coded and sampled similarly. The ellipticity for a given number

499: of independent phase screens depends on the size of \ro: large values

500: of \ro have larger ellipticities. Also note that low number statistics

501: are affecting the data-points toward large frame counts causing the

502: curves to cross eachother.}

503: \label{seeing}

504: \end{figure}

505: 

506: Based on this plot, it is also apparent that, even though the RTM

507: underestimates the ellipticity compared to the FTM, it does so more or

508: less independently of the value of \ro (which is a proxy for

509: seeing). This might open up the possibility that one either selects a

510: computationally efficient RTM sampling rate (say, \ro / 2) and apply

511: an appropriate (fixed) correction factor to the ellipticity results,

512: or adjust the sampling rate such that the RTM and FTM results are in

513: good agreement ($\sim$\ro / 3). Another approach that is under

514: investigation (G. Jernigan, private communication) is to roll off the

515: atmospheric power spectra at high spatial frequencies.  Further study

516: is needed to assess the accuracy of any of these approaches. For

517: instance, the numerical correction factors derived from the results

518: shown in Fig. 5, for the particular sampling ratio of \ro / 4, are

519: 1.32, 1.15, and 1.21, for \ro = 42 cm, \ro = 21 cm, and \ro = 10.5 cm,

520: respectively. Whether this variation is due solely to the statistical

521: errors in our modeling, or includes a systematic dependence on \ro is

522: not known. Clearly if one requires accurate modeled ellipticities,

523: then the computationally more expensive FTM is currently preferred.

524: 

525: \section{Conclusions}\label{conclusion}

526: 

527: Based on this analysis, we reach the following conclusions:

528: 

529: \begin{enumerate}

530: \item{Instantaneous speckle patterns have ellipticities of $\sim$10\%

531: for D / \ro = 40 (\ro = 21 cm).}

532: \item{Co-adding multiple patterns results in a linearly decreasing

533: ellipticity (on a log-log plot), consistent with a $\sqrt{N}$ slope of

534: $\alpha=-0.5$. }

535: \item{Modeling phase screen transport across the aperture (i.e., wind)

536: does not change the ellipticity results obtained from adding

537: uncorrelated phase screens in the limit of long exposures ($N>20$

538: aperture clearings).}

539: \item{Ellipticity values are robust over a large range of pixel

540: binning.}

541: \item{We expect no ellipticity dependency on sampling density in the

542: pupil plane (above sampling of \ro / 2). This is confirmed for the

543: Fourier method, but not for ray-tracing. The latter has a strong

544: dependency on sampling rate, in the sense that the higher the sampling

545: rate, the lower the resulting ellipticity. This can be understood as

546: the result of a breakdown in the geometric approximation for high

547: spatial frequency aberrations.}

548: \item{Ellipticities grow (for a given $N$) as the value of \ro

549: increases. However, since the average {\it size} of the PSF goes down

550: as \ro increases (see Fig.~\ref{speckle}), the ability to measure

551: precise ellipticities actually improves (for a given resolved object).}

552: \end{enumerate}

553: 

554: \noindent In summary, the effects of interference must be included in

555: order to comprehensively model point-source ellipticities induced by

556: the atmosphere.  Therefore, care has to be taken that the geometric

557: optics approximation to image formation by the telescope (i.e., ray

558: tracing) produces the same modeling results, as this is not true in

559: general.

560: 

561: \acknowledgments

562: 

563: The authors like to thank Don Phillion for generating the large phase

564: screens of \S~\ref{wind}, and the anonymous referee for helpful

565: comments.  This work was performed under the auspices of the

566: U.S. Department of Energy, National Nuclear Security Administration by

567: the University of California, Lawrence Livermore National Laboratory

568: under contract No.  W-7405-Eng-48.

569: 

570: \begin{thebibliography}{}

571: 

572: \bibitem[Asztalos et al.(2006)]{asztalos06} Asztalos, S., de Vries,

573: W., Rosenberg, L. J., Treadway, T., Burke, D., Claver, C., Saha, A.,

574: \& Puxley, P. 2007, \apj, 659, 1

575: \bibitem[Bacon et al.(2000)]{bacon00} Bacon, D. J., Refregier, A. R.,

576: \& Ellis, R. S. 2000, \mnras, 318, 625

577: \bibitem[Bacon et al.(2001)]{bacon01} Bacon, D. J., Refregier, A. R.,

578: Clowe, D., \& Ellis, R. S. 2001, \mnras, 325, 1065

579: \bibitem[Heymans et al.(2006a)]{heymans06} Heymans, C., Van Waerbeke,

580: L., Bacon, D., et al. 2006, \mnras, 368, 1323

581: \bibitem[Heymans et al.(2006b)]{heymans06b} Heymans, C., White, M.,

582: Heavens, A., Vale, C., \& Van Waerbeke, L. 2006, \mnras, in press

583: (astro-ph/0604001)

584: \bibitem[Hoekstra(2004)]{hoekstra04} Hoekstra, H. 2004, \mnras, 347,

585: 1337

586: \bibitem[Jain et al.(2000)]{jain00} Jain, B., Seljak, U., \& White,

587: S. 2000, \apj, 530, 547

588: \bibitem[Kaiser, Squires, \& Broadhurst(1995)]{kaiser95} Kaiser, N.,

589: Squires, G., \& Broadhurst, T. 1995, \apj, 449, 460

590: \bibitem[Tyson \& Angel(2001)]{tyson01} Tyson, A., \& Angel, R. 2001,

591: ``The New Era of Wide Field Astronomy", ASP conference series,

592: Vol. 232, R. G. Clowes, A. J. Adamson, \& G. E. Bromage, eds, p347

593: \bibitem[Tyson(2002)]{tyson02} Tyson, J. A., ``Survey and Other

594: Telescope Technologies and Discoveries'', Proceedings of the SPIE,

595: Tyson, J. Anthony; Wolff, Sidney., eds, Vol. 4836, p10

596: \bibitem[Vale \& White(2003)]{vale03} Vale, C., \& White, M. 2003,

597: \apj, 592, 699

598: \bibitem[Van Waerbeke et al.(2000)]{vanwaerbeke00} Van Waerbeke, L.,

599: Mellier, Y., Erben, T., et al. 2000, \aap, 358, 30

600: \bibitem[Wittman et al.(2000)]{wittman00} Wittman, D. M.,Tyson, J. A.,

601: Kirkman, D., Dell'Antonio, I., \& Bernstein, G. 2000, Nature, 405, 143

602: \bibitem[Wittman(2005)]{wittman05} Wittman, D. 2005, \apj, 632, L5

603: 

604: \end{thebibliography}

605: 

606: \end{document}

607: 				   

608: