IPP Software Navigation Tools IPP Links Communication Pan-STARRS Links

Changeset 40102


Ignore:
Timestamp:
Jul 18, 2017, 5:01:37 PM (9 years ago)
Author:
eugene
Message:

much additional writing

File:
1 edited

Legend:

Unmodified
Added
Removed
  • trunk/doc/release.2015/systematics.20140411/systematics.tex

    r40099 r40102  
    1 \documentclass[iop,floatfix]{emulateapj}
     1% \documentclass[iop,floatfix]{emulateapj}
     2\documentclass[10pt,preprint]{aastex}
    23% \pdfoutput=1
    34
     
    2223\def\plotext{ps}
    2324
    24 \def\picdir{/home/eugene/chipresid.20140404}
    25 %\def\picdir{/data/kukui.2/eugene/chipresid.20140404}
     25%\def\picdir{/home/eugene/chipresid.20140404}
     26\def\picdir{/data/kukui.2/eugene/chipresid.20140404}
    2627
    2728% Pick a terse version of the title here;
    28 \shorttitle{Systematics in PS1}
     29\shorttitle{Charge Diffusion Variations in PS1}
    2930\shortauthors{E.A. Magnier et al}
    3031\begin{document}
    31 \title{Systematic Effects in Pan-STARRS1 Photometry and Astrometry}
     32\title{Charge Diffusion Variations in Pan-STARRS\,1 CCDs}
    3233
    3334% this is a crude trick to get the order of affiliations right.  These
     
    5051%PS Builder List
    5152% W.~S. Burgett,\altaffilmark{\IfA}
    52 % K.~C. Chambers,\altaffilmark{\IfA}
     53K.~C. Chambers,\altaffilmark{\IfA}
    5354% L. Denneau,\altaffilmark{\IfA}
    5455% P. Draper,\altaffilmark{\DUR}
     
    8586\begin{abstract}
    8687
    87 Lorem ipsum dolor sit amet, consectetur adipiscing elit. Vestibulum
    88 bibendum nisi id tristique posuere. Duis eu mollis nulla. Maecenas est
    89 turpis, mattis tempor urna vitae, placerat rhoncus sem. Lorem ipsum
    90 dolor sit amet, consectetur adipiscing elit. Sed quis velit
    91 nisl. Aliquam erat volutpat. Cras lacinia, nisl tristique auctor
    92 molestie, dolor nulla rhoncus purus, ac accumsan nunc nunc ac
    93 nibh. Maecenas vitae mollis mauris. Ut sollicitudin pulvinar purus,
    94 eget luctus lorem tincidunt vitae. Vestibulum eu mattis neque. Nulla
    95 in tortor id urna dapibus gravida a vel leo.
    96 
     88Thick back-illuminated deep-depletion CCDs have superior quantum
     89efficiency over previous generations of thinned and traditional thick
     90CCDs.  As a result, they are being used for major wide-field imaging
     91cameras in several projects.  We use observations from the Pan-STARRS
     92$3\pi$ survey to characterize the behavior of the deep-depletion
     93devices used in the Pan-STARRS\,1 Gigapixel Camera.  We have
     94identified systematic variations in the photometric behavior and
     95stellar profiles which are similar to the so-called tree rings
     96identified in devices used by other wide-field cameras (DECam and
     97Hypersuprime Camera).  The tree-ring features identified in these
     98other cameras result from lateral electric fields which displace the
     99electrons as they are transported in the silicon to the pixel
     100location.  In contrast, we show that the photometric and morphological
     101modifications observed in the GPC1 detectors are caused by variations
     102in the vertical charge transportation range and resulting charge
     103diffusion variations.
    97104\end{abstract}
    98105
     
    102109\section{INTRODUCTION}\label{sec:intro}
    103110
    104 \begin{verbatim}
    105 * early CCDs were thick, but low resistivity Si had low cross-section to red photons
    106 * thinning was used to improve the blue sensitivity, at the cost of further reducing the red sensitivity
    107 * by the early 2000s, high-resistivity Si was used to make thick "deep-depletion" devices with good red and blue response.
    108 * voltages?
    109 * sky-scraper pixels
    110 * Plazas et al and other effects
    111 *
    112 \end{verbatim}
     111CCD detectors have evolved greatly since they were first introduced
     112for astronomical imaging in the mid 1970s.  In addition to the
     113well-known increases in the size of CCDs over the past 4 decades, CCD
     114architecture has gone through three major evolutionary stages. 
     115
     116The first generation of CCDs used a silicon substrate a few hundred
     117microns thick on top of which gate structures were deposited to define
     118the pixels.  A positive voltage applied to the gate layers would
     119create a shallow region (\approx 10 microns thick) in which the holes
     120were depleted.  This ``depletion region'' acted as a potential well to
     121trap electrons, specifically those generated by absorbed photons.  The
     122thick silicon substrate required illumination from the ``front'' side
     123with the thin gate structures to allow the photons to reach the
     124depletion region and be detected.  These early CCDs had modest quantum
     125efficiency as photons were easily absorbed by the several micron thick
     126gate structures.  For an excellent review of the history of CCD
     127development, see \cite{1992ASPC...23....1J}.
     128
     129Thinned, backside-illuminated CCDs such as the TI 3PCCD
     130\citep{1981SPIE..290....6B} were developed to address the quantum
     131efficiency limitations of the first generation thick CCDs.  The
     132silicon substrate was removed using a chemical process, leaving a
     133delicate device only \approx 10 - 20\micron\ thick, exposing the
     134depletion region on the backside.  Photons entering the backside of
     135the device are not blocked by the gate structures and thus more easily
     136absorbed and detected.  Thinned backside-illuminated CCDs have high
     137quantum efficiency to blue photons.  However, as the wavelength
     138increases beyond \approx 800 nm, the silicon becomes more transparent
     139to the photons, with a corresponding drop in quantum efficiency for
     140red photons.  In addition, thin film interference between the entering
     141photons and those reflecting off the front side of the CCD result in
     142``fringe'' patterns for redder photons.
     143
     144Early generations of CCDs were made of low-resistivity (\approx 10 -
     14550 $\Omega$-cm) silicon.  Following experiments beginning in the early
     1461990s \citep{Holland.1996}, CCDs made from thick, high-resistivity ($
     147> 10 k\Omega$-cm) silicon were developed for astronomical instruments
     148in the early 2000s\citep{Holland.2003}.  The high-resistivity of the
     149silicon allows for depletion regions of hundreds of microns in depth,
     150compared to \approx 10\micron\ for the low-resistivity silicon.  This
     151modification allows for a back-illuminated CCD with a relatively thick
     152silicon subtrate of 75 - 300\micron.  Blue photons impinging on the
     153back of the device are absorbed near the back surface of the device
     154and are caried through the depletion region to the gates on the front
     155side.  The thick silicon allows red photons to have a greater chance
     156to be absorbed, increasing quantum efficiency in the red.  Because
     157these thick, deep-depletion devices have near-unity quantum efficiency
     158across the whole a very wide spectral range, they have become the
     159design of choice for many modern, large-scale CCD cameras (e.g.,
     160Pan-STARRS GPC1, \citealt{2009amos.confE..40T}; Subaru Hypersuprime
     161Camera, \citealt{2010SPIE.7735E..3FK}; Dark Energy Survey Camera,
     162\citealt{2015AJ....150..150F}).
     163
     164While these deep-depletion CCDs seem to be ideal, they do have
     165features which can cause challenges for precise measurements.  As a
     166result of the ``Brighter-Fatter Effect''
     167\citep{2014JInst...9C3048A,2015JInst..10C5032G}, the profile of bright
     168stars are measured to be wider than the profiles of faint stars.  The
     169accepted interpretation is that the electric fields produced by the
     170electrons accumulated from a star repel successive incoming electrons,
     171with the repulsion increasing the more electrons have accumulated.
     172
     173The effects of lateral electric fields are likewise identified as the
     174cause of the so-called ``Tree-Rings'' observed in the flat-field,
     175astrometry, and photometry response of thick deep depletion detectors
     176\citep{2014PASP..126..750P}.  These tree-ring patterns have been noted
     177in the flat-field response of deep depletion devices since their early
     178testing \citep[see, e.g., Figure 2 in][]{2010SPIE.7735E..1RE} and were
     179initially considered to be a sensitivity response which could be
     180removed with a flat-field.  As discussed in detail by
     181\cite{2014PASP..126..750P}, these Tree Rings are more correctly
     182interpretted as variations in the effective pixel area due to
     183migration of the electrons pushed by lateral electric fields induced
     184by small changes in the doping used to set the resistivity of the
     185silicon.  The changes in the effective area result in changes to the
     186apparent flat-field response as well as the astrometric response of
     187the detector.  More subtly, the flat-field response changes, since
     188they do not reflect actual variations in sensitivity, can lead to
     189systematic photometry errors for astronomical sources if the
     190flat-field images are used in the standard fashion.
     191
     192In this paper, we examine the behavior of an apparently-similar kind
     193of Tree Ring observed in the Pan-STARRS GPC1 CCDs.  Although we also
     194observe the pixel effective area changes caused by lateral electric
     195fields as described by \cite{2014PASP..126..750P}, we show below a
     196second effect which is more important in driving systematic photometry
     197errors.  We find that variations in charge diffusion, also resulting
     198from changes in the silicon doping structures, affect both the
     199observed stellar profiles as well as the photometry measured with
     200profile fitting techniques.  In Section~\ref{sec:PS1}, we discuss the
     201Pan-STARRS telescope, camera, and survey data used in this analysis.
     202In Section~\ref{sec:tree.rings}, we present the Tree-Ring-like
     203patterns as observed in several different types of measurements:
     204flat-field response, systematic photometry residuals, systematic
     205astrometric residuals, and stellar profile shape variations.  In
     206Section~\ref{sec:discussion}, we discuss the interpretation of
     207patterns we observe and present a simple model to explain the observed
     208behavior.  We conclude with a discussion of the implications of this
     209effect on astronomical measurements from deep depletion instruments
    113210
    114211\section{Pan-STARRS1}
    115 
    116 \note{tidy up this section}
     212\label{sec:PS1}
    117213
    118214The 1.8m Pan-STARRS\,1 telescope (PS1), located on the summit of
     
    121217March 2014, PS1 was run under the aegis of the Pan-STARRS Science
    122218Consortium to perform a set of wide-field science surveys; since March
    123 2014, the telescope is operated by the Pan-STARRS New Science
     2192014, the telescope has been operated by the Pan-STARRS New Science
    124220Consortium (PSNSC).  Under the PS1SC, the largest survey, both in
    125 terms of area of the sky covered and fraction of observing time
    126 (56\%), was the \TPS\ in which the entire sky north of Declination
    127 $-30$\degrees\ was imaged up \approx 80 times over the 4 years.  These
    128 observations were distributed over five filters, \grizy, and have been
    129 astrometrically and photometrically calibrated to good precision
    130 \citep{magnier2017.calibration}.
     221terms of area of the sky covered ($3\pi$ steradians) and fraction of
     222observing time (56\%), was the \TPS\ in which the entire sky north of
     223Declination $-30$\degrees\ was imaged up \approx 80 times over 4
     224years.  These observations were distributed over five filters, \grizy,
     225and have been astrometrically and photometrically calibrated to good
     226precision \citep{magnier2017.calibration}.
    131227
    132228% 2004SPIE.5489..667H == PS1.optics
     
    138234\citep[GPC1][]{2009amos.confE..40T}, with low distortion and generally
    139235good image quality.  The median seeing for the \TPS\ data vary
    140 somewhat by filter, with (\grizy) = (XXXX).  Routine observations are
    141 conducted remotely from the Advanced Technology Research Center in
    142 Kula, the main facility of the University of Hawaii's Institute for
    143 Astronomy operations on Maui.
    144 
    145 GPC1 \citep{2009amos.confE..40T}, currently the largest astronomical camera in
    146 terms of number of pixels, consists of a mosaic of 60 edge-abutted
    147 $4800\times4800$ pixel detectors, with 10~$\mu$m pixels subtending
    148 0.258~arcsec. These \note{OTA51} detectors, manufactured by Lincoln
    149 Laboratory, are \note{75$\mu$m}-thick back-illuminated CCDs with a
    150 readout time of 7 seconds for a full unbinned image. \note{details
    151   about the voltages?}  Initial performance assessments are presented
    152 in \cite{2008SPIE.7014E..0DO}. The active, usable pixels cover $\sim 80$\% of the
    153 FOV.
     236somewhat by filter: (\grizy) = (1.31, 1.19, 1.11, 1.07, 1.02)
     237arcseconds.  Routine observations are conducted remotely from the
     238Advanced Technology Research Center in Kula, the main facility of the
     239University of Hawaii's Institute for Astronomy operations on Maui.
     240
     241GPC1 \citep{2009amos.confE..40T}, currently the largest astronomical
     242camera in terms of number of pixels, consists of a mosaic of 60
     243edge-abutted $4800\times4800$ pixel detectors, with 10~$\mu$m pixels
     244subtending 0.258~arcsec. These CCID58 detectors, manufactured by
     245Lincoln Laboratory, are 75\micron-thick back-illuminated CCDs
     246\citep{Tonry.2006,Tonry.2008}.  Initial performance assessments are
     247presented in \cite{2008SPIE.7014E..0DO}. The active, usable pixels
     248cover \approx 80\% of the FOV.
    154249
    155250\subsection{Data Processing and Calibration}
     
    161256
    162257Images obtained by PS1 are processed by the Pan-STARRS Image
    163 Processing Pipeline (IPP; \citealp{PS1_IPP,magnier2017.datasystem}).  All observations are processed
    164 nightly, with results sent to groups within the science consortium
    165 (i.e., PS1SC during the \TPS) performing short-term science projects
    166 (e.g., searching for transient and moving objects).  In addition, the
    167 \TPS\ dataset has been re-processed several times with improved
    168 calibration and analysis techniques.  To date (2017 July), 3
    169 re-processings starting from raw pixel data have been performed.  The
    170 labels PV0, PV1, PV2, PV3 are used identify the nightly processing and
    171 successive re-processing versions.  PV3 has been used for the public
    172 release of the Pan-STARRS \TPS\ data via the {\it Barbara A. Mikulski
    173   Archive for Space Telescopes} (MAST) at the Space Telescope Science
    174 Institute.\footnote{http//panstarrs.stci.edu}
     258Processing Pipeline (IPP;
     259\citealp{2006amos.confE..50M,magnier2017.datasystem}).  All
     260observations are processed nightly, with results sent to groups within
     261the science consortium (i.e., PS1SC during the \TPS) performing
     262short-term science projects (e.g., searching for transient and moving
     263objects).  In addition, the \TPS\ dataset has been re-processed
     264several times with improved calibration and analysis techniques.  To
     265date (2017 July), 3 re-processings starting from raw pixel data have
     266been performed.  The labels PV0, PV1, PV2, PV3 are used identify the
     267nightly processing and successive re-processing versions.  PV3 has
     268been used for the public release of the Pan-STARRS \TPS\ data via the
     269{\it Barbara A. Mikulski Archive for Space Telescopes} (MAST) at the
     270Space Telescope Science Institute.\footnote{http//panstarrs.stci.edu}
    175271
    176272The data processing and calibration operations are discussed in detail
     
    207303factors which may make the flat-field image inconsistent with stellar
    208304photometry, e.g., SED, filter band-pass variations, etc
    209 \citep[see][]{waters2017,2004PASP..116..449M,magnier.belgium}.  This
    210 correction was made on a relatively coarse grid across the focal plane
    211 in order to accumulate sufficient statistics from the stars in the
    212 relatively small number of images available at the time.  We have
     305\citep[see][]{waters2017,2004PASP..116..449M,2007ASPC..364..153M}.
     306This correction was made on a relatively coarse grid across the focal
     307plane in order to accumulate sufficient statistics from the stars in
     308the relatively small number of images available at the time.  We have
    213309found that a single flat-field set can be used for all PS1
    214310observations to yield photometric consistency at the level of \approx
    215 2\% \note{use the ubercal flat stdev as a statistic}.  PS1 benefits in
    216 this regard from the stability of having a single instrument which is
    217 rarely removed. 
     3112\%.  PS1 benefits in this regard from the stability of having a
     312single instrument which is rarely removed.
    218313
    219314Photometry of the PS1 images is performed using a
    220315point-spread-function (PSF) model as well as multiple kinds of
    221 apertures \citep{magnier2017.analysis}.  In this analysis, we
    222 refer to aperture photometry performed using an aperture defined based
    223 on the image quality observed for a given chip.  The aperture diameter
    224 is set to be \note{XXX} times the FWHM for the image.
     316apertures \citep{magnier2017.analysis}.  In this analysis, we refer to
     317aperture photometry performed using an aperture defined based on the
     318image quality observed for a given chip.  The aperture diameter is set
     319to be \approx 3.75 times the FWHM for the image.
    225320
    226321To improve the photometric systematic errors beyond the level achieved
     
    228323photometry is re-calibrated within the databasing system based on the
    229324properties of the measured photometry.  The calibration process is
    230 discussed by \cite{2012ApJ...756..158S,2013ApJS..205...20M,magnier2017.calibration}.
     325discussed by
     326\cite{2012ApJ...756..158S,2013ApJS..205...20M,magnier2017.calibration}.
    231327As part of this process, several flat-field corrections have been
    232328determined.  For the PV2 analysis discussed here, a flat-field
    233329correction determined during the ubercal analysis
    234 \citep[see][]{2012ApJ...756..158S} consisted of an $8\times 8$ grid of corrections
    235 for each GPC1 chip and filter for each of 4 seasons.  The boundaries
    236 of those seasons are \note{tentatively} identified with modifications
    237 to the baffle structures or the system optics.  The critical point
    238 here is that the final effective flat-field image for the PV2 dataset
    239 is based on a dome-flat at the highest resolution, with very low
    240 resolution corrections based on photometry, resulting in photometric
    241 calibration with roughly 1 millimag consistency for each measurement
    242 \note{better number from ubercal?}.
     330\citep[see][]{2012ApJ...756..158S} consisted of an $8\times 8$ grid of
     331corrections for each GPC1 chip and filter for each of 4 seasons.  The
     332boundaries of those seasons are tentatively identified with
     333modifications to the baffle structures or the system optics.  The
     334critical point here is that the final effective flat-field image for
     335the PV2 dataset is based on a dome-flat at the highest resolution,
     336with very low resolution corrections based on photometry, resulting in
     337photometric systmatic uncertainties in the range 7 - 12
     338millimagnitudes, depending on the filter \citep{2013ApJS..205...20M}.
    243339
    244340For all objects, positions are measured from the PSF model for the
     
    252348
    253349\section{Tree-Ring-Like Patterns}
     350\label{sec:tree.rings}
    254351
    255352\begin{table}
     
    274371For many of the GPC1 OTA CCDs, we observe a pattern in the photometric
    275372residuals which is similar in appearence to the Tree Rings described
    276 in the Dark Energy Camera (DECam) by \cite{plazas.2014}.  This pattern
    277 consists of systematic deviations which are consistent in a set of
    278 circular arcs centered on the corner of the CCD, as shown in
     373in the Dark Energy Camera (DECam) by \cite{2014PASP..126..750P}.  This
     374pattern consists of systematic deviations which are consistent in a
     375set of circular arcs centered on the corner of the CCD, as shown in
    279376Figure~\ref{fig:psfmags.by.filter}.  The details of the analysis used
    280377to generate Figure~\ref{fig:psfmags.by.filter} are given below.  For
     
    282379circular silicon wafer into 4 inscribed squares.  Thus the corners of
    283380the CCDs lie in the center of the silicon boule, just as the center of
    284 the circular Tree Rings described by \cite{plazas.2014} match the
    285 center of the boule from which they came.  This gives the impression
    286 that a similar mechanism is responsible for the pattern observed in
    287 the PS1 photometry and the DECam photometry, namely the diffusive
    288 effects of lateral electric field variations in the detectors.  In the
    289 next section, we will make the case that the patterns observed in the
    290 PS1 residuals are {\em not} caused by this mechanism, but are instead
    291 caused by variations in the {\em vertical} electric field (the field
    292 direction perpendicular to the CCD surface). 
     381the circular Tree Rings described by \cite{2014PASP..126..750P} match
     382the center of the boule from which they came.  This gives the
     383impression that a similar mechanism is responsible for the pattern
     384observed in the PS1 photometry and the DECam photometry, namely the
     385diffusive effects of lateral electric field variations in the
     386detectors.  In the next section, we will make the case that the
     387patterns observed in the PS1 photometry residuals are {\em not} caused
     388by this mechanism, but are instead caused by variations in the {\em
     389  vertical} electric field (the field direction perpendicular to the
     390CCD surface).
    293391
    294392First, in this section, we will describe how we have measured the
     
    296394For all of these examples, we use a single GPC1 CCD (XY40) to
    297395illustrate the effects in detail, but a similar set of effects are
    298 seen in \note{many? most?} GPC1 detectors.  First, we show the
    299 residual PSF photometry.  Second, we show the residual Aperture
    300 photometry.  Third, we show the astrometric residual patterns.
    301 Fourth, we show the patterns observed in the flat-field images.
    302 Finally, we show measurements derived from the second-moments of the
    303 stars.
     396seen in many of the GPC1 detectors.  First, we show the residual PSF
     397photometry.  Second, we show the residual Aperture photometry.  Third,
     398we show the astrometric residual patterns.  Fourth, we show the
     399patterns observed in the flat-field images.  Finally, we show
     400measurements derived from the second-moments of the stars.
    304401
    305402For all effects discussed below, we are measuring the mean value of
     
    308405represents the same range of true GPC1 XY40 pixels regardless of the
    309406type of measurement.  To generate the photometry, astrometry, or
    310 second-moment measurements were extracted from the \note{PV0} DVO
     407second-moment plots, measurements were extracted from the PV0 DVO
    311408database for observations covering the region ($\alpha$,$\delta$) =
    312409(90\degree\ -- 150\degree, -25\degree\ -- 10\degree).  This region of
     
    358455
    359456Figure~\ref{fig:psfmags.by.filter} shows the 2D patterns of PSF
    360 photometric residuals.  In this case, we select PSF magnitude
     457photometry residuals.  In this case, we select PSF magnitude
    361458measurements for detections of stars which fall in the given
    362459superpixel.  We subtract each measurement from the average magnitude
     
    378475is comparable to the amplitude of the correlated structures, so we
    379476need to integrate along the radial structures to make stronger
    380 statements about these patterns. \note{hanging statement?}
     477statements about these patterns.
    381478
    382479Figure~\ref{fig:apmags.by.filter} shows the equivalent measurement for
     
    462559then observed by the PS1 telescope.  These flat-field images were
    463560obtained 2011 Feb 09 as part of a campaign to study the PS1 system
    464 response \citep{2012ApJ...750...99T}.  Flats were obtain in a set of 4nm steps,
    465 with \note{XXnm} band-pass.  To enhance the signal-to-noise, we have
    466 median-combined a set of 6 flats at the center of the corresponding filter.
     561response \citep{2012ApJ...750...99T}.  Flats were obtain in a set of
     5624nm steps.  To enhance the signal-to-noise, we have median-combined a
     563set of 6 flats at the center of the corresponding filter.
    467564
    468565In order to mask pixels which do not flatten well, we generate a
     
    535632multiple detections).  The second moments are measured with a Gaussian
    536633weighting function, with the $\sigma_{w}$ scaled by the PSF size so
    537 that the $\sigma$ measured for PSF stars is \approx 60\% of
     634that the $\sigma$ measured for PSF stars is \approx 65\% of
    538635$\sigma_{w}$.  (Note that, since the measured $\sigma$ of stellar
    539636objects is biased down by the weighting function, this is not quite
     
    541638discussion in \citealt{magnier2017.analysis}).  For each stellar
    542639detection, we extract the values $M_{xx,xy,yy} = \sum F_i w_i (x^2, x
    543 y, y^2) / \sum F_i w_i$.  For each exposure, we find the mean second
    544 moments ($\bar{M_{xx,xy,yy}}$) for PSF objects on this chip (XY40) and
    545 subtract that mean value from the instantaneous measurements of
    546 $M_{xx,xy,yy}$.  We then determine the median of the residual second
    547 moments for each superpixel, resulting in 3 images for each filter.
    548 
    549 \note{write out this math, check out psLibADD}
     640y, y^2) / \sum F_i w_i$.  For each exposure, we find the median second
     641moments for PSF objects on this chip (XY40) and subtract that median
     642value from the instantaneous measurements of $M_{xx,xy,yy}$.  We then
     643determine the median of the residual second moments for each
     644superpixel, resulting in 3 images ($\delta M_{xx,xy,yy}$) for each
     645filter.
    550646
    551647Using the second moment images, we can construct certain interesting
     
    559655related to the shape of the elliptical contour as follows:
    560656\begin{eqnarray}
    561 e_0 & = & \sigma^2_a  + \sigma^2_b \\
    562 e_1 & = & (\sigma^2_a  - \sigma^2_b) \cos (2 \theta) \\
    563 e_2 & = & \sigma^2_a  - \sigma^2_b
     657e_0 & = & \sigma^2_{\mbox{major}}  + \sigma^2_{\mbox{minor}} \\
     658e_1 & = & (\sigma^2_{\mbox{major}}  - \sigma^2_{\mbox{minor}}) \cos (2 \theta) \\
     659e_2 & = & \sigma^2_{\mbox{major}}  - \sigma^2_{\mbox{minor}}
    564660\end{eqnarray}
    565 Where $\sigma_a$ and $\sigma_b$ are the major and minor axis
     661Where $\sigma_{\mbox{major}}$ and $\sigma_{\mbox{minor}}$ are the major and minor axis
    566662dimensions of the ellipse and $\theta$ is the position angle. 
    567663Thus, $e_0$ is a measurement of the change in the size of the stellar
     
    570666can determine the angle of the PSF ellipticity from the $e_1$ term.
    571667
    572 Figure~\ref{fig:smear.by.filter} shows the spatial trend of the {\em
    573   smear}, $\sigma^2_{major} + \sigma^2_{minor} = \delta M_{xx} +
    574 \delta M_{yy}$.  This value corresponds to the increase or decrease in
     668Figure~\ref{fig:smear.by.filter} shows the spatial trend of $e_0$, the {\em
     669  smear}.  This value corresponds to the increase or decrease in
    575670the circularly-symmetric component of the image size.  The dynamic
    576671range of these images is -0.3 to +0.3 pixel$^2$. A tree-ring-like
     
    579674can also be seen.
    580675
    581 We can also construct a measurement of the change in ellipticity
    582 $\sigma^2_{major} - \sigma^2_{minor} = (M_{xx} - M_{yy})^2 + 4
    583 M_{xy}$.  This value is plotted in Figure~\ref{fig:shear.by.filter}.
    584 This value is positive definite and is plotted with a color scale
    585 ranging from -0.02 to 0.22 pixel$^2$.  We can also determine the
    586 orientation of the corresponding ellipse.  Overlayed on
     676Figure~\ref{fig:shear.by.filter} shows the spatial trend of $e_2$, the
     677{\em shear}.  This value is positive definite and is plotted with a
     678color scale ranging from -0.02 to 0.22 pixel$^2$.  We can also
     679determine the orientation of the corresponding ellipse.  Overlayed on
    587680Figure~\ref{fig:shear.by.filter} is a set of vectors representing the
    588681ellipse orientation as a function of postion.  The length of the
    589682vectors corresponds to the value of $\sigma^2_{major} -
    590 \sigma^2_{minor}$.  The tree-ring-like structure is {\em not} apparent in this
    591 figure for any filter.  The spatial variations are low-frequency and
    592 unrelated to the radial trend from the upper-left corner.
     683\sigma^2_{minor}$.  The tree-ring-like structure is {\em not} apparent
     684in this figure for any filter.  The spatial variations are
     685low-frequency and unrelated to the radial trend from the upper-left
     686corner.
    593687
    594688\subsection{Correlations Between Tree-Ring-Like Patterns}
     
    715809radial component of the astrometric residuals: $\frac{\partial
    716810  (\sigma^2_{major} + \sigma^2_{minor})}{\partial radius} \sim \delta
    717 R$ (see Figure~\ref{fig:dsmear.vs.astrom}
     811R$ (see Figure~\ref{fig:dsmear.vs.astrom}).
    718812
    719813Finally, the radial derivative of the radial component of the
     
    730824residual values without a derivative.  We are convinced that we have
    731825the sense of the derivative correct by examination of specific
    732 features in each imaage (e.g., \note{give example}).
     826features in each imaage.
    733827
    734828\begin{table}
     
    781875
    782876\section{Discussion}
     877\label{sec:discussion}
    783878
    784879These trends help to illuminate the underlying causes of these
    785880different effects. 
    786 
    787 \note{summarize what pure lateral electric fields would do}
    788881
    789882First, if we consider the smear pattern
     
    829922The slope of our relationship is \approx 0.5 in normalized units.
    830923Thus the observed trends appear to be too weak by a factor of \approx
    831 2.  \note{looks like a slope of 1.0 would not be excluded by these
    832   plots}
    833 
    834 \note{I need to use the relationship between the astrometry and the
    835   flat-field to calculate the amplitude of the lateral electric
    836   fields.}
     9242, but otherwise exhibits the expected behavior.
    837925
    838926The fact that the PSF ellipticity changes are {\em not} correlated
     
    846934magnitudes.
    847935
     936Finally, the correlation between the smear structures and the
     937astrometry residuals shows that these two effects are connected.  The
     938underlying connection is the pattern of the resistivity variations.
     939Regions with high (or low) resistivity show relatively high (or low)
     940amounts of smear; astrometric deviations follow the gradient between
     941these regions. 
     942
     943We interpret the changes in the {\em smear} term as changes in the
     944amount of charge diffusion.  The blue filters exhibit the strongest
     945changes in the amount of smear.  These are also the filters for which
     946the detected electrons have travelled the longest distance in the
     947silicon, and are thus most affected by diffusion effects. 
     948
     949\note{add more quantitative discussion of the variations in $E_y$ vs $E_x$?}
     950
    848951\section{Conclusion}
    849952
    850 The tree rings are showing (at least?) two effects, though they must
    851 be related.  First, the images are experiencing circularly-symmetric
    852 changes in the PSF size correlated with the tree-ring pattern.  These
    853 PSF size changes drive errors in the PSF photometry which the are also
    854 correlated with the tree ring pattern on the scale of a few
    855 millimagnitudes.  These PSF size changes are consistent with changes
    856 in the charge diffusion, which also introduces a circularly symmetric
    857 smearing.
    858 
    859 In addition, there are radial plate-scale changes
    860 correlated with the tree rings.  These plate-scale changes introduce a
    861 flat-field errors on the scale of \approx 1 millimagnitude and
    862 astrometric errors in the scale of 2-3 milliarcseconds.  The observed
    863 relationship between the flat-field deviations and the radial
    864 derivative of the astrometric deviations confirms that these two
    865 measurements are caused by the same effect. 
    866 
    867 There must be some common cause for both the smearing (charge
    868 diffusion) and the radial plate-scale changes since the astrometric
    869 deviations are correlated with the radial derivative of the smearing.
     953The tree rings observed in the Pan-STARRS GPC1 data show (at least)
     954two effects, though they are related.  First, the images are
     955experiencing circularly-symmetric changes in the PSF size correlated
     956with the tree-ring pattern.  These PSF size changes drive errors in
     957the PSF photometry which the are also correlated with the tree ring
     958pattern on the scale of a few millimagnitudes.  These PSF size changes
     959are consistent with changes in the charge diffusion, which also
     960introduces a circularly symmetric smearing.
     961
     962In addition, there are radial plate-scale changes correlated with the
     963tree rings.  These plate-scale changes introduce a flat-field errors
     964on the scale of \approx 1 millimagnitude and astrometric errors in the
     965scale of 2-3 milliarcseconds.  The observed relationship between the
     966flat-field deviations and the radial derivative of the astrometric
     967deviations confirms this interpretation \citep[see discussion
     968  in][]{2014PASP..126..750P}.
     969
     970The vertical diffusion variations and the lateral charge migration are
     971both driven by the same variations in the doping structures.  This
     972point is clear from the spatial correlation of the gradient in the
     973smear variations and the astrometric variations.
     974
     975% The small-scale variations in the charge diffusion observed in these
     976% devices has not been reported for DECam, Hypersuprime Cam, or
     977% prototype LSST sensors. 
     978
     979The small-scale variations in the charge diffusion observed in the
     980Pan-STARRS detectors represents a new type of systematic effect in
     981deep depletion devices.  This feature, if present in other detectors,
     982could manifest in systematic errors in several ways.  Like in the
     983Pan-STARRS analysis example, the charge diffusion variations result in
     984fine-structure in the observed stellar point-spread functions.  For
     985very precise photometry or morphological analysis, it will be
     986necessary for the PSF models to account for the extra charge
     987diffusion.  Unlike the non-uniform pixel-size effects, correction of
     988the PSF photometry cannot simply be performed as an average flat-field
     989correction on the measurements after they have been processed. 
     990The additional smearing acts as a convolution with a Gaussian kernel
     991of fixed size for a given filter.  The photometry bias is a function
     992of the fractional change of the PSF size.  Thus, the introduced error
     993depends on the average PSF for the image in question: an image with
     994good image quality will suffer larger PSF model errors than an image
     995with poor image quality.  To account for this effect in a rigorous
     996way, the analysis should use the measured diffusion variations to
     997modify the model PSFs as a function of position before they are used
     998for the image analysis.
     999
     1000The charge diffusion variations may also have an impact on
     1001spectroscopic measurements.  Modern, precise spectroscopic
     1002measurements rely on precise measurements of the stellar line
     1003profiles.  If such an analysis ignores variations in the charge
     1004diffusion, the measured line widths may be systematically biased.
     1005
     1006This analysis points to the importance of careful instrumental
     1007characterization, especially for those instruments which are used for
     1008large-scale surveys with largely automatic data analysis systems and
     1009stringent precision goals.
    8701010
    8711011\acknowledgments
     
    8931033\end{document}
    8941034
    895 Notes for paper re-work:
    896 
    897 * Paper focus is now only on the diffusion variations
    898   * strip out the discussion of other systematic effects
    899   * strip down the PS1 introduction discussion
    900 
    901 * tentative title:
    902   Evidence for Small-Scale Charge-Diffusion Variations in Pan-STARRS CCDs
    903 
    904 * outline
    905 
    906  1. introduction
    907     * thick CCDs
    908     * tree rings == transverse field effects (see Plazas et al)
    909     * we see something else
    910 
    911  4 model : diffusion variations due to E|| field variations
    912 
    913  5 discussion (how to treat in calibration / analysis)
    914 
    915  6 conclusions
    916 
    917 some possible refs to tree rings / charge diff:
    918 
    919 * http://adsabs.harvard.edu/abs/2016SPIE.9904E..2CW (Woods et al 2016; TESS)
    920 * https://arxiv.org/pdf/1605.01001.pdf : plazas et al
    921 * http://ieeexplore.ieee.org/document/1225293/?part=1 Altmannshofer et al 2003 (about thick Si)
    922 
    923 * plazas et al 2014 outline
    924 
    925   1. intro: thick CCDs, transverse electric fields
    926   2. DES / DECam
    927 
    928   2.1 flat-field tree rings (discussion of flat-field tree rings
    929   starting from the premise that they know the answer).
    930  
    931   3 impact on astrometry and photometry
    932 
    933   4 improving calibrations given tree rings
    934 
    935   5 summary and conclusions
    936 
     1035%% Some refs to be added as appropriate:
     1036% Bernstein DEC astrometry : arxiv 1703.01679
     1037% Baumer et al arxiv 1706.07400 (Flat-fielding)
Note: See TracChangeset for help on using the changeset viewer.