1: 

2: %% Manuscript for the Journal P.A.S.P.  (Publications of the Astronomical

3: %% Society of the Pacific), using AASTeX v5.2 macros.

4: 

5: % Use emulateapj class to write draft in double-column ApJ format, then

6: % simply replace with \documentclass{aastex} before submission.

7: 

8: \documentclass{emulateapj}

9: %\documentclass[12pt,preprint]{aastex}

10: %% Note that manuscript produces a one-column, double-spaced document:

11: 

12: %% Sometimes a paper's abstract is too long to fit on the

13: %% title page in preprint2 mode. When that is the case,

14: %% use the longabstract style option.

15: 

16: %% \documentclass[preprint2,longabstract]{aastex}

17: 

18: %% If you want to create your own macros, you can do so

19: %% using \newcommand. Your macros should appear before

20: %% the \begin{document} command.

21: %%

22: %% If you are submitting to a journal that translates manuscripts

23: %% into SGML, you need to follow certain guidelines when preparing

24: %% your macros. See the AASTeX v5.x Author Guide

25: %% for information.

26: 

27: \newcommand{\xs}{XSPEC{}}

28: \newcommand{\xstw}{XSPEC {\tt 12.x}{}}

29: \newcommand{\isis}{ISIS{}}

30: \newcommand{\slirp}{SLIRP{}}

31: \newcommand{\slang}{S-Lang{}}

32: \newcommand{\tcl}{TCL{}}

33: \newcommand{\fort}{{Fortran}{}}

34: \newcommand{\initp}{{\tt initpackage}{}}

35: \newcommand{\bbp}{{Black Box Problem}{}}

36: \newcommand{\pgplot}{{\tt PGPLOT}{}}

37: \newcommand{\swig}{{\tt SWIG}{}}

38: \newcommand{\volview}{{\tt volview}{}}

39: \newcommand{\evtimg}{{\tt evt2img}{}}

40: \newcommand{\ftf}{{\em file $\rightarrow$ tool $\rightarrow$ file}{}}

41: \newcommand{\vwhere}{{\tt VWhere}{}}

42: \newcommand{\fcalc}{{\tt fcalc}{}}

43: \newcommand{\fv}{{\tt fv}{}}

44: %\newcommand{\bfi}[1]{{\bf \em #1}}

45: \newcommand{\bfi}[1]{{#1}}

46: 

47: %% You can insert a short comment on the title page using the command below.

48: 

49: %\slugcomment{}

50: 

51: %% If you wish, you may supply running head information, although

52: %% this information may be modified by the editorial offices.

53: %% The left head contains a list of authors,

54: %% usually a maximum of three (otherwise use et al.).  The right

55: %% head is a modified title of up to roughly 44 characters.

56: %% Running heads will not print in the manuscript style.

57: 

58: \shortauthors{Noble et al.}

59: 

60: %% This is the end of the preamble.  Indicate the beginning of the

61: %% paper itself with \begin{document}.

62: 

63: \begin{document}

64: 

65: %% LaTeX will automatically break titles if they run longer than

66: %% one line. However, you may use \\ to force a line break if

67: %% you desire.

68: 

69: \title{Beyond \xs: Towards Highly Configurable Astrophysical Analysis}

70: \shorttitle{Beyond \xs: Towards Highly Configurable Astrophysical Analysis}

71: 

72: %% Use \author, \affil, and the \and command to format

73: %% author and affiliation information.

74: %% Note that \email has replaced the old \authoremail command

75: %% from AASTeX v4.0. You can use \email to mark an email address

76: %% anywhere in the paper, not just in the front matter.

77: %% As in the title, use \\ to force line breaks.

78: 

79: \author{M. S. Noble, M. A. Nowak}

80: \affil{Kavli Institute for Astrophysics and Space Research\\

81: 	Massachusetts Institute of Technology\\

82: 	Cambridge, MA 02139}

83: \email{mnoble@space.mit.edu}

84: 

85: %% Mark off your abstract in the ``abstract'' environment. In the manuscript

86: %% style, abstract will output a Received/Accepted line after the

87: %% title and affiliation information. No date will appear since the author

88: %% does not have this information. The dates will be filled in by the

89: %% editorial office after submission.

90: 

91: \begin{abstract}

92: 

93: We present a quantitative comparison between software features of the

94: defacto standard X-ray spectral analysis tool, \xs\, and \isis\, the

95: Interactive Spectral Interpretation System. Our emphasis is on

96: customized analysis, with \isis\ offered as a strong example of

97: configurable software.  While noting that

98: \xs\ has been of immense value to astronomers, and that its

99: scientific core is moderately extensible---most commonly via the

100: inclusion of user contributed ``local models"---we identify a series of

101: limitations with its use beyond conventional spectral modeling.  We

102: argue that from the viewpoint of the astronomical user, the \xs\

103: internal structure presents a \bbp, with many of its important

104: features \bfi{hidden} from the top-level interface, thus \bfi{discouraging}

105: user customization. Drawing from examples in custom modeling, numerical

106: analysis, parallel computation, visualization, data management, and

107: automated code generation, we show how a \bfi{numerically} scriptable,

108: modular, and extensible analysis platform such as \isis\ facilitates

109: many forms of advanced astrophysical inquiry.

110: 

111: \end{abstract}

112: 

113: %% Keywords should appear after the \end{abstract} command. The uncommented

114: %% example has been keyed in ApJ style. See the instructions to authors

115: %% for the journal to which you are submitting your paper to determine

116: %% what keyword punctuation is appropriate.

117: 

118: %\keywords{methods: data analysis --- methods: statistical --- X-rays: general}

119: \keywords{Data Analysis and Techniques}

120: 

121: %% From the front matter, we move on to the body of the paper.

122: %% In the first two sections, notice the use of the natbib \citep

123: %% and \citet commands to identify citations.  The citations are

124: %% tied to the reference list via symbolic KEYs. The KEY corresponds

125: %% to the KEY in the \bibitem in the reference list below. We have

126: %% chosen the first three characters of the first author's name plus

127: %% the last two numeral of the year of publication as our KEY for

128: %% each reference.

129: 

130: 

131: %% Authors who wish to have the most important objects in their paper

132: %% linked in the electronic edition to a data center may do so by tagging

133: %% their objects with \objectname{} or \object{}.  Each macro takes the

134: %% object name as its required argument. The optional, square-bracket 

135: %% argument should be used in cases where the data center identification

136: %% differs from what is to be printed in the paper.  The text appearing 

137: %% in curly braces is what will appear in print in the published paper. 

138: %% If the object name is recognized by the data centers, it will be linked

139: %% in the electronic edition to the object data available at the data centers  

140: %%

141: %% Note that for sources with brackets in their names, e.g. [WEG2004] 14h-090,

142: %% the brackets must be escaped with backslashes when used in the first

143: %% square-bracket argument, for instance, \object[\[WEG2004\] 14h-090]{90}).

144: %%  Otherwise, LaTeX will issue an error. 

145: 

146: \section{Introduction}

147: \label{intro}

148: The pursuit of science pushes not only the boundaries of knowledge, but

149: also the limits of technology used to acquire and analyze data from which

150: new knowledge can be distilled.  In this sense software systems for

151: scientific analysis are never truly complete, so it is crucial that they

152: be extensible.  This enables scientists to incorporate custom, experimental

153: codes into

154: the system and drive it in directions not yet supported, or perhaps even

155: envisioned, by its creators.  A corollary is that the mechanisms for

156: and results of the customization should be made as simple, powerful,

157: and general as possible, freeing the practitioner to concentrate more

158: on algorithmic or scientific concerns and less on the plumbing details

159: of a given computational platform.

160: 

161: This paper compares the extent to which two open-source spectral analysis

162: applications actively used in astronomy,  \xs\ \citep{1996ASPC..101...17A}

163: and \isis\ \citep{2002hrxs.confE..17H}, meet these general criteria.

164: The features of both systems are broadly surveyed, with contrasts

165: drawn from a series of examples representative of \bfi{astronomy in

166: practice}.  A heavy emphasis is placed upon the roles of interpreted arrays,

167: and the capability to quickly generate and numerically manipulate them

168: as powerful enablers of analysis.

169: The emergent theme is that numerical and architectural discontinuities

170: between the internals of \xs\ and its controlling \tcl\ interpreter,

171: referred to in the large as the \bbp, limit its full exploitation for

172: scientific tasks that are \bfi{within the reach of more configurable}

173: interactive analysis systems.  We demonstrate the strong modularity

174: and extensible scriptability of \isis\, and suggest that such open and

175: highly configurable systems offer deeper and wider promise for meeting

176: the needs of exploratory scientific analysis.

177: 

178: The latest major versions of \isis\ and \xs, {\tt 1.4.x} and {\tt 12.x}

179: respectively, were used in our study.  While \xs\ 11 remains in wide

180: use (the first public, non-beta release of \xs\ 12 was

181: circa 2005), it is more difficult to extend with a variety of custom

182: models\footnote{For example, \xs\ 11 supports only local models coded

183: in single precision \fort 77.}

184: and so in terms of extensibility would compare less favorably than does

185: \xs\ 12.

186: 

187: Guiding our endeavor will be the sense of what a typically motivated

188: scientist can achieve in each system with reasonable effort.  

189: On the premise that a strong programmer can eventually replicate

190: virtually any result using nearly any combination of languages or tools,

191: we are not concerned with what might in principle be achieved {\em from}

192: either \xs\ or \isis\, but rather what can be demonstrated {\em within}

193: each.  By avoiding hypotheticals we aim to transform a potentially

194: open-ended debate into a tractable, concrete discussion.

195: Towards this end we define {\em working within an application} to mean

196: the execution of commands, or portions thereof, which only access

197: or modify the code or data within the applications addressable memory.

198: In this scheme, for example, issuing an operating system call from \isis\ to

199: spawn an external visualization tool would not be ``working in \isis,"

200: however enlarging its address space by importing a shared object module,

201: say to smooth an image in memory, would be.

202: 

203: We recognize the utility of integrating complementary but distinct

204: programs into a single logical entity.  Building on trust

205: in known tools, it is a sensible scheme for managing complexity, fostering

206: reuse, and balancing deadlines with goals.  Cooperating applications can,

207: with sufficient effort, be made to appear as seamless as a single

208: program.  \bfi{One} should not, however, equate the act of

209: \begin{itemize}

210: \item dumping internal state to disk files, spawning a series

211: of tools to read this state and create temporary data products

212: along the way, then loading these new data back into the parent

213: application

214: \end{itemize}

215: with the ability to

216: \begin{itemize}

217: \item invoke a function to operate upon a data structure, with both

218: resident in memory.

219: \end{itemize}

220: \bfi{In this paper we will argue that the latter offers greater immediacy

221: of use, as well as significantly more algorithmic, data, and visualization

222: flexibility.  It also addresses

223: real scalability concerns \label{scalability}} as datasets increase in

224: size or tasks increase in time complexity.  A stark example of this

225: is given in \cite{2005ASPC..347..444D}, which discusses the methods

226: employed to analyze the megasecond observation of Cassiopeia A

227: \citep{2004ApJ...615L.117H} by the Chandra X-Ray Observatory in 2004,

228: which yielded the first ever map of electron acceleration in a supernova

229: remnant \citep{2006NatPh...2..614S}.  Over 9 million PHA spectra

230: were extracted from over 300 million events, with hundreds of thousands

231: of spectral fits performed in \isis\, to yield arc-second resolution spatial

232: maps of plasma temperatures, line emission, and Doppler velocities.  Custom

233: modules were developed to manage the multi-gigabyte files involved and

234: distribute the spectral extractions

235: and fits across a network of workstations using the Parallel Virtual

236: Machine \citep{Geist:1994:PPV}.  Based upon the time

237: required to extract a single spectrum from these enormous datasets, with

238: the standard Chandra {\tt dmextract} tool in CIAO (ca. version 3.2), the

239: authors extrapolated that the spectral extractions alone would have

240: required nearly 10 years had a tool-based approach been taken.  In

241: contrast, each run through the entire PVM-based analysis sequence

242: required less than 12 days to complete.

243: 

244: \section{Feature Survey}

245: \label{feature_survey}

246: \xs\ and \isis\ have been primarily concerned with fitting models to 1D

247: spectra:

248: a theoretical model of photon counts is calculated, convolved with an

249: instrument response, and compared to the actual photon counts observed by

250: the detector, using a given fit statistic (typically $\chi^2$).  Each

251: provides mechanisms for indexed loading and management of data

252: (observed counts, instrument responses, et cetera), as well as the

253: capacity to visually inspect, using \pgplot\ to construct a variety

254: of 2D plots, the data, models, residuals, fit statistics, and so

255: forth.  While initially targeted for X-ray analysis, both \xs\ and \isis\

256: have been utilized in multiple wavebands.

257: The primary authors of \xs\ and \isis\ are practicing astronomers with

258: active publication records in the domains served by their applications.

259: Both applications are thus subject to continuous, rapid, and

260: rigorous scientific vetting. \isis\ and \xs\ update cycles are measured

261: in terms of days and weeks, which can be beneficial for users under time

262: constraints before their proprietary data go public.

263: 

264: \xs\ is the tool most widely used in X-ray astronomy for spectral fitting;

265: with a legacy spanning more than two decades and hundreds of citations,

266: its value to the scientific community is beyond question.  \xs\ bundles

267: more than 50 theoretical models, each of which

268: may also be used by external programs simply by linking to

269: its libraries.  The core set of models may be extended with either

270: precomputed file-based table models or compiled codes.  The bulk

271: of compiled \xs\ models are coded in \fort, however as part of a significant

272: redesign effort support for custom C and C++ models was recently added

273: (ca. 2005).

274: \xs\ was also one of the earliest astronomical applications to adopt a

275: widely-used, general-purpose scripting language for interactive use

276: and batch control; the incorporation of \tcl, introduced in version 10

277: and stabilized in version 11, provided programmability and a clear path

278: to graphical interfaces with the Tk widget set.

279: 

280: \isis\ was conceived to support analysis of high-resolution Chandra X-Ray

281: gratings spectra, then quickly grew into a general-purpose analysis system. 

282: \isis\ is essentially a superset of \xs, combining all of its models (though

283: they are not required for \isis\ operation) and more with the \slang\

284: scripting language \bfi{\citep{SLangRef}, whose mathematical performance rivals

285: both commercial packages such as MatLab \citep{Gilat.2004.MATLAB} or

286: IDL \citep{Kling.1999.IDL} as well as the numerical extensions for

287: popular open source languages \citep[\S \ref{numerics}]{Noble.2007.OpenMP}.}

288: 

289: As with \xs, users may define custom models in compiled code, external

290: tables, or as a string specifying simple arithmetic combinations of

291: existing models, but \isis\ takes

292: it further by also allowing models to be interpreted \slang\ functions;

293: this supports rapid prototyping and, because of the high-performance of

294: \slang\ numerics (\S \ref{numerics}), need not sacrifice speed for the

295: convenience of using an interpreter.  \xs\ does not support the use of

296: \tcl\ procedures as models.

297: % NB: mdefine command provides a crude approximation, but is not in XS 12

298: %and given that \tcl\ is poorly suited for numerical work (\S \ref{numerics})

299: %such may not even be advisable.

300: 

301: The chief means of controlling \xs\ and \isis\ interactively is by

302: textual prompt, with both providing VI and EMACS key bindings through

303: GNU readline, as well as filename completion and persistent command

304: history.  \xs\ accepts commands entered in abbreviated form, such as

305: {\tt mo} for {\tt model}, but does not auto-complete \tcl\ command or

306: variable names from partial input.  \isis\ provides strong auto-completion

307: facilities, allowing any \slang\ function or variable name to be

308: % NB: submit an ISIS patch to auto-complete . commands, too!

309: identified from minimal partial input.

310: In addition to the command prompt, optional packages providing

311: graphical interfaces, such as the

312: Gtk-based \vwhere\ (\S \ref{vwhere}) or the SAOds9 \citep{2003ASPC..295..489J}

313: image display module (\S \ref{ds9}), have been publicly available and

314: utilized within \isis\ for several years.

315: Graphical extensions for \xs\ have been discussed in the literature

316: \citep{1994ASPC...61...71J} and online, but, other than wholly external

317: tools like \fv\ \citep{1997ASPC..125..261P}, none seem publicly

318: available for use within \xs\ proper.

319: 

320: The \xs\ prompt presents an appealingly simple,

321: {\em everything-is-a-command} interface, and fosters ease of use by

322: offering brief, germane command names, assuming sensible defaults, and

323: in some cases transparently combining multiple operations into one: e.g.

324: loading both data and background files with a single command.  A typical

325: command is shorter and requires less punctuation in \xs, although the

326: auto-completion capability of \isis\ can compensate for its relative

327: verbosity.  In short, \xs\ makes it easy to perform many common spectral

328: analysis tasks.

329: 

330: \begin{figure}

331:   \begin{centering}

332:   \includegraphics[angle=-90,scale=0.3]{f1.eps}

333:   \caption{Ionization fractions for Fe XVII through XXI.}

334:   \label{ionfrac}

335:   \end{centering}

336: \end{figure}

337: 

338: The \isis\ prompt was designed to be a proxy \slang\ interpreter

339: (Fig. \ref{arch}), making \isis\ programmable and extensible since

340: inception.

341: Its {\em everything-is-a-function} philosophy permeates the implementation

342: and user experience, providing, at the cost of a steeper initial learning

343: curve, considerable flexibility and power.

344: Consider for illustration the Astrophysical Plasma Emission Database

345: (APED) of \cite{2001ApJ...556L..91S}.   Because \isis\ provides

346: programmatic access to virtually all of APED, curves like those of

347: Fig. \ref{ionfrac}, showing

348: the ionization fractions at 25 temperatures for Fe XVII through XXI,

349: can be created with just a few statements

350: { \small

351:   \begin{verbatim}

352:      temperature = 10.0^[6.0:7.2:0.05];

353:      foreach ion ( [17, 18, 19, 20, 21] ) {

354:         fraction = ion_frac(Fe, ion, temperature);

355:         oplot (temperature, fraction);

356:      }

357: \end{verbatim}

358: }

359: \noindent

360: either directly from the \isis\ prompt or within analysis scripts.

361: The importance to X-ray analysis of this sort of facile manipulation of

362: atomic data is heavily underscored in

363: \cite{kallman-2007-79}.

364: 

365: \subsection{The Black Box Problem}

366: \bfi{

367: It is not clear how a similar ad-hoc query of APED could be formulated

368: in \xs, nor how the results of such might be used within it for further

369: analysis.  One problem is that the use of APED in \xs\ is highly constrained,

370: due to it being hidden, e.g., within the internals of the {\tt apec} model

371: and {\tt identify} commands.

372: Another obstacle is that high-level plotting in \xs\ is not generic: the

373: {\tt plot} and {\tt iplot} commands are also opaque, in that they take}

374: their vectors from internal state representing the current data or

375: model under inspection.  Intimately

376: tied to the semantics of 1D spectral analysis, these commands may not

377: be used until after a dataset has been loaded.

378: 

379: While \isis\ contains

380: similar visualization commands specifically tailored to spectral analysis,

381: the {\tt oplot} function shown above is part of a family of routines which

382: \bfi{allows one} to create arbitrary image, scatter, histogram,

383: and contour [over]plots from arbitrary data, independently of the current

384: spectra being analyzed.  This makes exploratory coding and visualization like

385: {\small

386: \begin{verbatim}

387:           isis>  x = [PI/2: 2*PI: PI/100]

388:           isis>  plot(x, cos(x))

389:           isis>  oplot(x, sin(x))

390: \end{verbatim}

391: }

392: \noindent

393: \bfi{as natural in \isis} as it is in other interactive analysis

394: systems such as IDL or PyRAF \citep{2001ASPC..238...59D}.  In \xs\ such

395: dynamic creation of interpreted arrays and direct mathematical manipulation

396: and visualization of them is more difficult, involving a melange of

397: \tcl\ code, \xs\ commands, intermediate files, and low-level QDP/PLT

398: directives.\footnote{Or going outside of \xs\ to use tools like \fv\

399: or XIMAGE.}

400: \begin{figure}[t]

401:   \begin{centering}

402:   \includegraphics[angle=-90,scale=0.30]{f3.eps}

403:   \caption{Performance of IDL 6.1 (binary distribution) and

404:         \slang\ (statically linked), on $\sqrt{b^2 - 4ac}$

405:         for arrays of various sizes.}

406:   \label{idl-plot}

407:   \end{centering}

408: \end{figure}

409: The issue is not that interpreted numeric arrays cannot be utilized in

410: \xs, because the \tcl\ extensions BLT and NAP (see next section) make

411: this possible; nor does the problem lie with \pgplot, since it provides

412: more capabilities than \xs\ currently exploits.  Rather, it is the \xs\

413: architecture which does not connect the two in a \bfi{straightforward manner,

414: leaving the user little choice but to look elsewhere for open-ended

415: numerics and visualization.  This opacity and disconnectedness is evident

416: in other functional areas of \xs, notably the manner in which data are

417: input \& managed (\S \ref{input}) and how custom models are specified

418: (\S \ref{custom-models}).

419: \begin{figure*}[t]

420:   \begin{centering}

421:     \includegraphics[scale=0.50]{f2.eps}

422:     \caption{ The \isis\ { \tt main()} program is a thin ANSI C layer above

423: 	the \slang\ interpreter, used primarily to establish hooks to

424: 	modules and gather user input.  Taking modularity

425: 	to this extreme, where carefully orthogonalized components provide

426: 	the bulk of capability, provides a clean and flexible implementation.

427: 	Most modules may also be

428: 	used outside of \isis.}

429:   \label{arch}

430:   \end{centering}

431: \end{figure*}

432: 

433: Echoing similar criticisms levied upon graphical interfaces when compared

434: to command lines \citep{NextUIBrkThru,OscarCLI}, we collectively refer

435: to these issues as the \bbp: by hiding complexity to make common tasks

436: easy, uncommon or novel tasks can be made difficult or impossible.}

437: 

438: The ease of use that has been a hallmark of \xs\---written by astronomers

439: to do things astronomers need, in a way natural to them---has served the

440: community well.  A challenge, though, is that as instruments and the

441: data collected from them

442: grow in size and complexity, causing us to consider new questions and

443: possibly develop new techniques to answer them, the \bbp\ can engender

444: {\em You May Only!} patterns of analysis,

445: rather than foster rapid, ad hoc explorations of {\em What If?} scenarios.

446: More modular, configurable, and open implementations \citep{BeyondBlackBox}

447: can help resolve this tension, allowing applications to rapidly evolve to

448: suit specific user needs while freeing their primary authors of the

449: burden of coding and maintaining such enhancements themselves.

450: 

451: \section{High Performance Numerics}

452: \label{numerics}

453: Compact multidimensional numerics are a \bfi{major basis

454: for the popularity} of commercial toolsets such as IDL or MatLab.  The

455: innate capability of \slang\ in this regard was a primary motivator for its

456: adoption in \isis,\footnote{Another motivator was wide availability:

457: \slang\ is utilized in numerous open source projects and, in part by

458: virtue of being bundled as a core component of every major Linux 

459: distribution, is available on millions of machines worldwide.}

460: \bfi{and endows it with analytic expressiveness and

461: scripting performance not equaled in \xs}.

462: 

463: As indicated in the earlier code fragments, 

464: complex mathematical abstractions may be stated concisely in \slang,

465: without regard to whether they will operate upon scalars, vectors,

466: multidimensional arrays, or some combination of each, and are computed with

467: performance on par with commercial software (e.g., as in Fig. \ref{idl-plot}).

468: \bfi{In this section we discuss the more involved example of Fig.

469: \ref{orbmodel}, consisting of an orbital model implemented in pure \slang\

470: and taken ``as is" from an MIT research effort

471: The model was fit to a coarsely binned lightcurve (providing an example of

472: how \isis\ can model non-spectroscopic data directly from in-memory \slang\

473: arrays, as discussed in \S \ref{input}), and represents a steady amplitude

474: curve interrupted by an eclipse with a quadratic ingress and egress, with

475: low-level emission during the eclipse also modeled by a quadratic function.}

476: \citep{TrowbridgeNowakWilms2007}.

477: 

478: As a means of assessing

479: the numerics and extensibility of \xs\ we attempted to translate this

480: model into \tcl; for completeness we also made Perl and Python translations,

481: since these languages are actively employed by astronomers in systems like

482: PDL and PyRAF and so are useful contrasts for gauging the numerical

483: capabilities of \slang.  Because neither \tcl\ , Perl, nor Python are

484: intrinsically vectorized we used numerical extension modules for each

485: conversion: BLT \& NAP for \tcl, PDL for Perl, and Numeric, NumArray,

486: \& NumPy for Python.

487: \noindent The Python and Perl conversions were straightforward, with

488: Python being the somewhat easier of the two; we did not complete the

489: conversion to \tcl, however, largely because neither BLT nor NAP provided

490: a clear equivalent to the {\tt where} command or a means of using the

491: results of such for array slicing.  This model would therefore need to be

492: recoded in a compiled language before it could be used in \xs.

493: A plot of the performance of Perl 5.8.4 and Python 2.4 implementations

494: of this model, divided by the corresponding \slang\ 2.0.7 performance,

495: is given in Fig. \ref{orbmodel-plot}.

496: 

497: The testing 

498: methodology behind Figs. \ref{idl-plot} and \ref{orbmodel-plot} is

499: described in \cite{Noble.2007.OpenMP}, along with memory statistics

500: and additional tests showing similar performance trends.

501: Briefly, the datapoints in each curve are the mean times of 1000

502: invocations of the model on a given grid size (31 in all, from 1 to 1e6

503: bins); all codes were compiled using GCC 3.3 with -O3 optimization, and

504: executed on a dual Athlon (1.8Ghz) machine running Debian 3.1.

505: Although not a comprehensive series of benchmarks, these results

506: \bfi{hint that the numerical engine of \isis\ is among the strongest

507: available.  High

508: performance scripting means that rapid development techniques---irrespective

509: of language---can be applied to a broad} scope of analysis problems,

510: \bfi{allowing the writing of compiled code to gradually become a last

511: resort instead of the primary avenue of attack}.

512: \slang\ bears a strong resemblance to C and IDL, arguably the most popular

513: \begin{figure}[h]

514:   \begin{centering}

515:   \includegraphics[scale=0.80]{f4.eps}

516:   \caption{Pure \slang\ orbital model, fit

517:   by \isis\ directly to in-memory data arrays.

518: As described in \cite{TrowbridgeNowakWilms2007}, the model

519: represents a lightcurve of constant amplitude {\tt a}, with an eclipse

520: centered at {\tt c}, with a width {\tt w2} from the start of ingress

521: to the end of egress.  The ingress and egress have a quadratic form

522: ({\tt m} and {\tt b} parameters), and there is a quadratic

523: ``bounceback" ({\tt s} parameter) with width {\tt w1}.}

524:   \label{orbmodel}

525:   \end{centering}

526: \end{figure}

527: scripting language used by astronomers, and in fact we and colleagues

528: have converted numerous IDL and MatLab scripts to \slang\ for use in 

529: \isis\, with \bfi{relative} ease.

530: \begin{figure}[t]

531:   \begin{centering}

532:   \includegraphics[angle=-90,scale=0.3]{f5.eps}

533:   \caption{Performance of orbital models implemented in Perl 5.8.4

534:    and Python 2.4, relative to \slang\ 2.0.7.

535:    \tcl\ is not represented here because neither of its numerical 

536:    extensions, BLT and NAP, offered a clear equivalent to

537:    the {\tt where()} function.}

538:   \label{orbmodel-plot}

539:   \end{centering}

540: \end{figure}

541: 

542: \section{Array-Based Input}

543: \label{input}

544: As we found with the orbital model, converting such scripts to \tcl\ for

545: use in \xs\ would pose a \bfi{considerable challenge}: although

546: the primary interface of

547: \xs\ is a scriptable interpreter in which multidimensional arrays may

548: be created and mathematically manipulated---with the aid of the BLT or

549: NAP---the results of such cannot be straightforwardly utilized for

550: spectroscopy.  The fact that the internal data tables of \xs\ are not exposed

551: for direct population from the \tcl\ interpreter is another instance of

552: the \bbp; data may only be read from FITS files, and only via

553: the {\tt data} command or its INTEGRAL mission-specific variant

554: {\tt SPIdata}.  There is no documented provision by which interpreted

555: arrays may be used, for instance, to specify observed counts or

556: responses.

557: 

558: Reflecting a fundamental difference in \bfi{approach}, \xs\ is

559: more static than \isis, taking the position that extensive data

560: preparation happens outside the application.  An advantage of this

561: approach is self-documentation: well-written tools record the

562: path along which a FITS file travels as history keywords in its header.

563: A drawback is the underlying implication that input data need

564: little massaging during interactive analysis; when the need for such

565: data manipulation arises, the solution typically involves dumping

566: \xs\ state to disk files, and/or running FTOOLS or other programs

567: to generate new file products, then reloading these data back into \xs.

568: 

569: \bfi{This {\em files-only} data management paradigm can lead to slower

570: performance and more cumbersome analytics (\S \ref{scalability},

571: \S \ref{vwhere}).

572: It can also be an evolutionary disadvantage, i.e. by potentially limiting

573: the pool of individuals able or willing to contribute new I/O codes to \xs.}

574: As exemplified by the implementation of the {\tt SPIdata} command,

575: endowing \xs\ with the ability to directly read additional file or

576: mission formats requires detailed mastery of its internals.

577: This means general-purpose code generators like \swig\ \citep{1996SWIG}

578: or \slirp\ \citep{Noble.2007.OpenMP} cannot be leveraged to automate the

579: wrapping of I/O libraries and simplify the use of new formats in \xs. To

580: foster the widest possible use, by design such wrapper generators know

581: nothing about the applications in which the bindings they create will

582: be used.  They emit code targeted to the scope of a scripting language,

583: rather than application-specific C++ methods like

584: {\tt SPI\_Data::read()} which is suitable only for internal \xs\ use.

585: 

586: \subsection{Towards Dynamic Data Management}

587: 

588: \bfi{

589: \isis\ aims for a more dynamic and configurable approach to data management.

590: In addition to supporting file-based input in the manner of \xs, but with

591: ASCII in addition to FITS, \isis\ also permits most facets of the modeling

592: process to be specified directly from interpreted \slang\ arrays in memory,

593: including the theoretical and observed counts, ARF, RMF, and background.

594: 

595: The problem of augmenting spectroscopy with data in foreign formats

596: is thus reduced from having to master \xs\ internals to merely being

597: able to create \slang\ arrays from such files.  This is within the reach

598: of end users, not just application authors, particularly because automatic

599: code generators can then be employed to simplify the creation of scriptable

600: wrappers for the relevant I/O libraries.}

601: 

602: \bfi{As an example of how these considerations can matter in practical use,}

603: consider the HDF5 file format and I/O library \citep{1999IEEE...99...273F}.

604: While FITS is the standard format for archiving and distributing

605: astrophysical observations, HDF5 has become the defacto standard 

606: for storing astrophysical simulations such as those generated by

607: FLASH \citep{2001NuPhA.688..172F}.  Having the ability to easily

608: compare and contrast \bfi{observations with simulations, e.g., using

609: simulation output as source model input, could foster more sophisticated

610: analyses.  We have explored such questions in ISIS as part of the

611: HYDRA\footnote{http://space.mit.edu/hydra} project, using \slirp\ 

612: to generate the SLh5

613: module\footnote{http://space.mit.edu/cxc/software/slang/modules/slh5}

614: for reading and writing HDF5 data directly to and from in-memory \slang\

615: arrays.  The resulting objects may be sliced with \slang\ array manipulators

616: or mathematically transformed in arbitrary ways (\S \ref{numerics}) before

617: further analysis, such as being treated as fit data or model data, or

618: passed to 2D plotting or 3D volumetric routines for visualization.

619: }

620: This module also obviates

621: the need for a conversion tool to migrate data from HDF5 to FITS for

622: spectroscopy\footnote{http://www.hdfgroup.org/RFC/fits2h5/fits2h5.htm

623: describes preliminary efforts to go in the other direction.}.

624: 

625: This spirit of imposing fewer constraints on how its internal tables may

626: be populated or manipulated \bfi{leads to more flexible and mathematically

627: diverse analysis.  For example,}

628: consider the problem of grouping data to ensure an

629: adequate signal to noise ratio is obtained for each detector channel.

630: With \xs\ data are grouped prior to input, usually with the {\tt grppha}

631: tool. The assigned grouping persists for the life of the loaded dataset,

632: and may only be changed by re-running {\tt grppha} outside of \xs,

633: deleting the original dataset in \xs, and reloading it with the new

634: grouping.  With \isis\ spectra may be dynamically rebinned in memory---

635: using the functions {\tt rebin\_data()}, {\tt group\_data()}, or

636: variants---with the associated dataset remaining intact.  \isis\ also

637: provides options not available in the \xs\ {\tt grppha} paradigm, such

638: as grouping by signal to

639: noise ratio, or allowing datasets to be combined directly in memory,

640: bringing to bear the full numerical power discussed in \S \ref{numerics}

641: and \S \ref{beyond1D}, instead of shelling out to run tools such as

642: {\tt mathpha}, {\tt marfrmf}, or {\tt addrmf}---an approach of

643: more limited mathematical generality.

644: \isis\ also supports combining non-linear (e.g. piled-up) data, which

645: in general cannot be done with tools because of the model evaluation

646: involved.

647: 

648: 

649: \section{Custom Models}

650: 

651: \label{custom-models}

652: In \S \ref{numerics} we detailed how the performance of \slang\ couples

653: with the ability of \isis\ to use purely interpreted models to provide

654: an alternative to the traditional method of using compiled

655: codes as custom spectral models.  This significant feature, not

656: available to \xs\ users since it does not permit \tcl\ procedures to be

657: used as models, \bfi{can encourage rapid experimentation with short,

658: highly analytic codes.  These brief, self-contained scripts are

659: easily exchanged among users, as they may be loaded immediately into

660: \isis\, from any directory.   This is in contrast to code that

661: must be compiled into a shareable object then accessed from fixed

662: locations and in conjunction with external description files, such

663: as \$LOCAL\_MODEL\_DIRECTORY and lmodel.dat are respectively used in \xs.}

664: 

665: Moving to the realm of compiled models, while version 12 of \xs\ provides

666: many improvements over version 11---notably support for C and C++,

667: double precision numerics, and the \initp\ automated model wrapping

668: command---a number of limitations persist.

669: To begin, using local models at all means one must first build \xs\ from

670: source.  Although HEASOFT generally builds well, this can still be

671: daunting, and is unnecessary with \isis\ as both interpreted and compiled

672: user models may be used from a binary installation.

673: Next, model components in \xs\ must be designated as having a specific

674: form, e.g. additive, multiplicative, convolution, etc.  This too is 

675: unnecessary in \isis, as it places no restriction upon the computational

676: form of a model---any conceivable mathematical function may be used.

677: 

678: As \xs\ has been the defacto standard for so long, it is also easy

679: for spectral model developers to code expressly for \xs, by relying too

680: heavily upon its internal functions (such as {\tt udmget} for memory

681: allocation or {\tt xwrite} for message output) when crafting models.

682: When \xs\ was the only X-ray spectral modeling tool available this was

683: not a concern, but today it makes reusing models in other environments

684: more difficult.  For example, pulling the {\tt pexriv} model out of

685: \xs\ 11, which remains widely used, required

686: {\small

687: \begin{verbatim}

688:           -lxspec -lxspec_lfn -lxspec -lwcs

689:           -lxanlib -lreadline -lcfitsio -lpgplot

690:           -lcurses -ltermcap

691: \end{verbatim}

692: }

693: \noindent on the link line, plus the introduction of a ``dummy" function

694: to resolve an unused symbol name \citep{2004ApJ...609..972M}.

695: Standing out here is the recursive use of {\tt-lxspec},

696: and the need for auxiliary terminal management and plotting libraries

697: that provide no numerical or physics contribution to the model.

698: Although this situation is improved in \xs\ 12, these show that it is

699: not enough to simply make source code available

700: to the public: {\em orthogonality} is an important aspect of open and

701: flexible software.  To promote external reusability, code must also be

702: {\em separable} into logically distinct units free of unwanted

703: dependencies.  These are core implementation principles of \isis\

704: (Fig. \ref{arch}).

705: 

706: \subsection{Automated Model Wrapping}

707: While \initp\ does help automate the use of custom models in

708: \xs\ 12, it is not a generic, application-neutral wrapper generator in

709: the style of \slirp\ or \swig.  The result is that custom modeling cannot

710: be scripted in \xs\ with the same level of flexibility as in \isis.  For

711: example, \initp\ does not expose the routines it wraps to the top-level

712: \tcl\ interpreter, but rather makes them accessible only through the

713: the \xs\\ {\tt model} command.  Not being able to call

714: models directly from \tcl\ means the convenience of an interpreter

715: cannot be leveraged when testing and using the models.

716: 

717: This can be easy to overlook, because many model writers 

718: devise their own nominal tests in whatever compiled language they

719: used to write the model.  However, it would be valuable for both

720: writers and users of models if they could craft \tcl\ commands or

721: scripts to exercise them (in a more powerful and streamlined manner

722: than the multiplexing {\tt tclout} command, or faking a dataset) or

723: pass their output to other features in the application for further

724: numerical processing.  Consider Fortran common blocks, for instance,

725: and the \xs\ {\tt warmabs} model in particular:  after evaluation

726: the {\tt ewout} block in this model contains a list of the strongest

727: lines, sorted by element and ion.  Wrapping {\tt warmabs} with \slirp\

728: makes the content of this and every other common block in the model

729: accessible to \isis\ as \slang\ variables.\footnote{Common block values

730: may also be modified using regular \slang\ assignment

731: expressions.}  This provides scientists another means of accessing

732: the internals of a code, without needing to master the internals

733: of the application for which that code was initially written.

734: 

735: In addition, \initp\ wraps only the outermost interface routine of a

736: model: if the spectrum returned by a model is actually computed in

737: several steps by calling additional routines, those lower level routines

738: are not exposed by \initp\ for use in \xs\ outside the model evaluation.

739: As an example of how exposing the low level routines could be useful,

740: we consider

741: the {\tt pshock} and {\tt vpshock} models.  Internally, both rely upon

742: the lower-level {\tt ionseqs()} routine to calculate nonequilibrium

743: ionization ionic concentrations.  In response to a collaborator

744: (Ji 2007, priv. comm.) who wished to obtain these ionic

745: concentrations independently, we used \slirp\ to wrap {\tt ionseq.f}

746: in a matter of only minutes, which allowed {\tt ionseqs()} to be called

747: directly from a \slang\ script in \isis.

748: 

749: The jet model of \cite{2005ApJ...635.1203M} provides another good

750: example.  The output spectrum is summed from 4 individual components

751: (disk, comptonization, synchrotron, and synchrotron self-comptonization

752: spectra), each computed by distinct routines within a single Fortran

753: source file.  To analyze or visualize the independent contribution

754: of each component it is necessary to isolate each from the total flux.

755: To achieve this the model can be executed in different modes, one of

756: which will execute {\tt write} statements to output component contributions

757: to disk files as they are computed.  Because our \slirp\ wrapper

758: for this jet model makes every routine within the Fortran file visible

759: to the top-level \slang\ interpreter, not just the outermost model

760: routine, the component fluxes can in principle now be accessed by making

761: the relevant routine calls, or accessing the associated common blocks,

762: directly from \isis.  To our knowledge the only way to achieve a similar

763: result in \xs\ would be to wrap the code twice, once with \initp\ to make

764: the custom model routine visible to the {\tt model} command, and again as

765: a \tcl\ extension module to expose the remaining routines to the

766: \tcl\ interpreter.  Tools like {\tt ftcl} or

767: {\tt critcl}\footnote{http://ftcl.sourceforge.net, http://wiki.tcl.tk/2523}

768: can simplify bindings generation in the latter case, but not automate it

769: to the level that \slirp\ does; it is also unclear whether they support

770: a range of features seen in general scientific codes, such as common blocks,

771: string and multidimensional arrays, or complex datatypes.

772: 

773: Using \initp\ for models which call external libraries can also be

774: problematic.  For example, collaborators are extending the above

775: jet model with codes that call GNU Scientific Library routines, but

776: \initp\ documents no mechanism by which the GSL can be linked in at build time.

777: Following compiler convention, this is achieved with \slirp\ by specifying

778: {\tt -lgsl} on the command line.

779: 

780: Finally, functions wrapped by \slirp\ can be automatically vectorized,

781: allowing them to operate over entire arrays in a single call, and at

782: the speed of compiled C code instead of using slower, interpreted looping

783: constructs.  Vectorized wrappers can also be tuned for

784: parallelization with OpenMP \citep{Noble.2007.OpenMP}, allowing scientists

785: to take better advantage of their multicore machines during \isis\ analysis.

786: To our knowledge autogenerated vector parallel wrappers are currently

787: unique to \slirp, and therefore inaccessible in \xs.

788: 

789: \section{Parallel and Distributed Computation}

790: 

791: Parallel computing is not new, but to this day few astronomers employ

792: parallelism for standard problems in data analysis.  To provide a 

793: quantitative sense of this assertion relative to X-ray spectroscopy

794: \citep{2006ASPC..351..481N},

795: we performed the following full text searches on all published papers

796: indexed in the ADS \citep{1993ASPC...52..132K} by September of 2005.

797: \begin{table}[h]

798:  \footnotesize

799:  \begin{center}

800:   \begin{tabular}{cc}

801:   Keywords & Number of Hits \\

802:   \hline

803:   parallel AND pvm & 38  \\

804:   message AND passing AND mpi & 21 \\

805:   xspec & 832 \\

806:   xspec AND parallel AND pvm & 0 \\

807:   xspec AND message AND passing AND mpi & 0 \\

808:   \end{tabular}

809:  \caption{Published references to parallelism and \xs, as indexed in ADS

810:     through 2005.}

811:  \end{center}

812: \end{table}

813: Extra keywords were included with PVM and MPI (the Message Passing

814: Interface, \citealt{330577}) so as to cull false matches (e.g. with

815: the Max Planck Institute).  Queries in ADS on other modeling tools,

816: or with other search engines such as Google,

817: all yielded similar trends: astronomers and

818: astrophysicists have employed parallel computing, but mainly for highly

819: customized, large-scale problems in simulation, image processing, or

820: data reduction.

821: Even though a majority of papers published in observational astronomy

822: over recent decades

823: result from fitting models within established software systems, virtually

824: no one is employing parallelism to do so, especially in the interactive

825: context.

826: 

827: As discussed in \cite{2006ASPC..351..481N} and earlier

828: in \S \ref{scalability}, for several years PVM has been used in

829: ISIS\footnote{PVM was used due its

830: longstanding tolerance of faults on heterogeneous clusters

831: of workstations, but in principle an MPI module could be used just as

832: easily within \isis.} to apply

833: pools of machines to coarse-grained calculations that do not fit

834: within the compute space of one workstation.

835: Consider relativistic Kerr disk models, for example.  Historically,

836: implementors have opted to use precomputed tables to gain speed at

837: the expense of limiting flexibility in searching parameter space.

838: However, by recognizing that contributions from individual radii may

839: be computed independently the model has been parallelized in \isis\ to

840: avoid this tradeoff \citep[A. Young, priv.  comm.,][]{2006ApJ...652.1028B}.

841: Likewise, the aforementioned jet model can be

842: expensive to compute, particularly when calculating error bars: generating

843: 90\% confidence limits for 12 free parameters, at the

844: very coarse tolerance of 0.5, required 4 days on a single 2.66 GHz

845: Intel Core 2 Duo processor.

846: To increase performance we assigned the confidence limit search to 12

847: Intel Xeon 3.4GHz processors\footnote{Using new versions of the

848: {\tt cl\_master} and {\tt cl\_slave} scripts described on the \isis\

849: website.} within

850: a PVM.  This reduced the runtime at 0.5 tolerance to under 24 hours, a

851: greater than 75\% speedup; it also allowed us to discern finer features

852: in parameter space by increasing the tolerance resolution by a factor

853: of 500, to 0.001, while keeping the overall runtime to 50 hours, or

854: approximately half of the serial runtime at 0.5 resolution.

855: 

856: Temperature mapping is another problem that was straightforward to

857: parallelize with \isis. For instance, \cite{2004cosp...35.3997W}

858: provides a

859: map of heating in the intracluster medium of Perseus, computed from

860: 10,000 spectral extractions and fits on 20+ CPUs in just several hours.

861: Additional efforts have also led to improvements in related areas of

862: research, such as {\tt pvm\_xstar}, a

863: parallelizing wrapper for XSTAR which has made it feasible for us

864: to probe thousands of photoionized gas physical scenarios in the time

865: it has previously taken to compute only a handful of such models

866: (Noble \& Ji, in prep.).

867: 

868: At the other end of the architectural scale, we have also shown

869: that \isis\ can make transparent use of OpenMP to exploit shared

870: memory multiprocessors \citep{Noble.2007.OpenMP}.  This is especially

871: relevant in light of the emergence of multicore architectures: soon

872: most astronomers will have parallel computers on their desktops, but

873: few astronomy applications are poised to take advantage of them.

874: Analysis systems which do not embrace parallelism

875: can process at most the workload of only 1 CPU, resulting in a

876: dramatic $1/n$ underutilization of resources as more CPU cores are

877: added.  However, astronomers are well versed in scripting, particularly

878: with very high-level, array-oriented numerical packages like IDL, PDL,

879: and \slang, to name a few.  They combine easy manipulation of mathematical

880: structures of arbitrary dimension with most of the performance of compiled

881: code, with the latter due largely to moving array traversals from the

882: interpreted layer into lower-level code like this C fragment

883: {\small

884: \begin{verbatim}

885:         case SLANG_TIMES:

886:              for (n = 0; n < na; n++)

887:                  c[n] = a[n] * b[n];

888: \end{verbatim}

889: }

890: \noindent

891: which provides vectorized multiplication in \slang.

892: This suggests that much of the strength and appeal of numerical scripting

893: stems from relatively simple loops over regular structures, making

894: them ripe for automatic parallelization.  This pattern was exploited

895: in \slirp\ and \slang\ to effect the OpenMP parallelizations

896: described in \citep{Noble.2007.OpenMP}.

897: 

898: 

899: We believe that \isis\ is the only general purpose spectroscopy system

900: in which such a range of parallelism---from single processors on

901: multiple machines to multiple processors on single machines---has

902: been demonstrated.  In both cases speedups were obtained without

903: exposing parallelism at the \isis\ application level.  For instance,

904: the same serial implementations of spectroscopic models have been used

905: for both single- and multi-processor execution.  This minimizes two

906: traditional barriers to the use of parallelism by non-specialists:

907: learning how to program for concurrency and recasting sequential

908: algorithms in parallel form.  We believe these considerations are

909: important because the emerging ubiquity of multicore architectures,

910: combined with the ever-growing size of datasets and analysis complexity,

911: makes the regular use of parallelism in astronomy not a question of

912: {\em if} it will occur, but rather one of merely {\em when}.

913: 

914: \section{Beyond 1D Spectroscopy}

915: \label{beyond1D}

916: 

917: Data from modern telescopes are getting larger and more detailed,

918: with the use of multiple datasets from multiple missions becoming

919: commonplace.  Making sense of it all often requires techniques

920: more sophisticated than traditional plots or images.  Canned data

921: reduction threads, while an important piece of the puzzle, can only

922: go so far.  Data openness and architectural modularity

923: (Fig. \ref{arch}) make it easy to push

924: \isis\ beyond its original mission of 1D spectroscopy to address

925: this broader set of analysis problems.  As we illustrate here,

926: it has become unnecessary to use entirely separate applications

927: (as XImage, XSelect, or Xronos are used alongside \xs) or be constrained

928: by a strictly \ftf\ model, in order to

929: supplement spectral modeling with, for example, advanced

930: filtering, imaging, or timing analysis.  For each of the examples

931: shown here it is not clear how a similar result might be

932: obtained so directly within \xs.

933: \subsection{Timing Analysis}

934: The optional SITAR module\footnote{http://space.mit.edu/CXC/analysis/SITAR}

935: endows ISIS with timing analysis capability, obviating the need

936: to use a separate timing application such as XRONOS. The

937: power spectrum in Fig. \ref{timing} was generated entirely in ISIS:

938: the data were read, the FFTs were performed and averaged,

939: the Power Spectrum was logarithmically binned over Fourier frequency, the

940: constant + two Lorentzians model was fit, and the results plotted, 

941: operating directly upon arrays in memory, all without ever leaving

942: the program.

943: \subsection{Modular Imaging with \slang\ Numerics}

944: \label{ds9}

945: Many excellent imaging tools are available to astronomers, but it is

946: \begin{figure}[t]

947:   \begin{centering}

948:   \includegraphics[scale=0.35]{f6.eps}

949:   \caption{Power Spectrum of X1820-303, generated from the PCA in RXTE

950:   observation P10075 (data set 10075-01-01-02).}

951:   \label{timing}

952:   \end{centering}

953: \end{figure}

954: impossible for any one of them to do everything.  Here again, the

955: extensibility and modularity of the \isis\ paradigm helps avoid the

956: fate of ``doing nothing" while waiting for the implementation of new

957: features desired in one's work.  In such cases it can be expedient

958: to craft new modules which add missing features, or make certain

959: functionality more easy to incorporate into modeling and analysis.

960: 

961: For example, consider Fig. \ref{ngc7714}, drawn from Chandra observation

962: 4800 of interacting galaxy pair NGC7714,

963: and the following series of commands:

964: {\small

965: \begin{verbatim}

966:     isis> require("ds9")

967:     isis> require("gsl")

968:     isis> image = ds9_get_array()

969:     isis> hist = sum(image, 0)

970:     isis> range = [440:680]

971:     isis> hplot(range-1, range, hist[range])

972:     isis> xa = [440:680:4]; ya = hist[xa]

973:     isis> smoothed = interp_cspline(range, xa, ya)

974:     isis> oplot(range, smoothed)

975: \end{verbatim}

976: }

977: First the SAOds9 and GNU Scientific Library extension modules

978: \citep{2005ASPC..347...17P} are loaded; we

979: show this explicitly for didactic purposes, but \isis\ can be configured

980: to load either module automatically when the given functions are invoked.

981: The 1024$\times$1024 image is then retrieved directly into a

982: properly byteswapped

983: 2D \slang\ pixel array, using an XPA \citep{1995xpa.v1M} binary transfer 

984: instead of intermediate file I/O.  This image

985: is collapsed along its X axis using the native \slang\ { \tt sum} function,

986: to create an intensity histogram plot. The GSL cubic spline function

987: is then called to smooth the brightest portion

988: of the intensity histogram, interpolating over every fourth point therein,

989: and finally we overplot the result (dotted red line).

990: Although the task here is relatively straightforward, it again shows how

991: open-ended

992: analysis objectives can be achieved by weaving existing tools together 

993: in new ways, using an array-based interpreter as the thread.  DS9 is

994: extended beyond its essential role as a qualitative display tool and

995: into the realm of quantitative analysis, while the \isis\ user is able

996: to exploit the imaging capabilities of DS9 within interactive or

997: scripted analysis.

998: \begin{figure*}

999:   \begin{centering}

1000:   \hspace*{3.5mm}

1001:   \includegraphics[angle=-90,scale=0.35]{f7a.eps}

1002:   \includegraphics[angle=-90,scale=0.35]{f7b.eps}

1003:   \caption{ACIS image from 2004 Chandra observation of NGC7714, with

1004:      corresponding square and smoothed pixel intensity histograms.}

1005:   \label{ngc7714}

1006:   \end{centering}

1007: \end{figure*}

1008: 

1009: \begin{figure*}

1010:   \begin{centering}

1011:   \includegraphics[scale=0.4]{f8a.ps}

1012:   \hspace*{4mm}

1013:   \includegraphics[scale=0.4]{f8b.ps}\\

1014:   \vspace*{2mm}

1015:   \includegraphics[scale=0.37]{f8c.ps}

1016:   \hspace*{10mm}

1017:   \includegraphics[scale=0.37]{f8d.ps}

1018:   \caption{

1019:   Interactive 3-color image creation in \evtimg, using the level-2 event file

1020:   acisf00120N001\_evt2.fits from Chandra observation 120 of supernova

1021:   remnant E0102-72.  One may pan around (by clicking to center the image),

1022:   zoom in and out, and set various scaling properties by clicking on the

1023:   corresponding buttons. The image in the left column shows a fair amount

1024:   of high energy background noise, as evidenced by the blue

1025:   background pixels.  These were removed in realtime by adjusting the

1026:   colored energy band sliders that define the RGB levels (bottom panels),

1027:   to yield the improved image in the right column.}

1028:   \label{evt2img}

1029:   \end{centering}

1030: \end{figure*}

1031: 

1032: Another example is \evtimg,\footnote{http://space.mit.edu/cxc/software/slang/modules/hist/evt2img.html} a simple guilet\footnote{This

1033: term is used in \cite{2005ASPC..347..237N}

1034: to refer to graphical applets embedded within a primarily command-line

1035: application, and typically coded in a scripting language, allowing

1036: the use of graphical interfaces when convenient while avoiding an

1037: exclusively graphical interface.}

1038: which combines the histogram and Gtk modules to provide an interactive

1039: mechanism for creating 3-color images directly from event lists.

1040: Images are created in \evtimg\ by defining 3 energy band filters, 

1041: binning the events selected by each filter with the {\tt hist2d} function,

1042: and overlaying the resulting red, green, and blue monochrome images 

1043: within a Gtk display window.  As shown in Fig. \ref{evt2img}, a useful

1044: feature of \evtimg\ is its ability to plot the energy spectrum as a 1D

1045: histogram in order to tune the color band filters in real time, simply

1046: by adjusting

1047: the individual red, green, and blue sliders.  For interactive exploration

1048: this method is faster and

1049: more intuitive than the traditional \ftf\ approach, and produces

1050: no temporary file litter while experimenting with color assignment filters.

1051: {\tt Evt2img} may be run standalone from the operating system prompt or

1052: as a function in \isis, and in the latter case allows event data to be

1053: input directly from in-memory interpreted arrays in addition to traditional

1054: event files.

1055: The HYDRA project at MIT provides a wealth of more sophisticated

1056: examples\footnote{http://space.mit.edu/hydra/E2D\_demo/e2d\_demo.html}

1057: of how advanced imaging can be tied more directly to numerical modeling

1058: and analysis in \isis.  Under the auspices of the NASA AISRP program, a

1059: broad suite

1060: of 2D and 3D fitting, geometry, and visualization routines have been

1061: developed, e.g. to do forward folding and comparison of models with 2D

1062: event-based data sets.  These are described in detail online,

1063:  but are briefly illustrated

1064: here in the context of an analysis of cavities in Hydra A, using the

1065: Chandra observation 4969.  A model is defined by combining traditional

1066: spectral components (e.g.  mekal and powerlaw) with 3D geometric components

1067: (e.g. an AGN modeled as a sphere).  The fitting is performed directly in

1068: \isis, as is the generation of the residual, data, and model images, and

1069: 2D and 3D model component projections (Fig. \ref{e2d}).

1070: \begin{figure*}[h]

1071:   \begin{centering}

1072:   \includegraphics[scale=0.5]{f9a.ps}\\

1073:   \vspace*{10mm}

1074:   \includegraphics[scale=1.05]{f9b.eps}

1075:   \hspace*{10mm}

1076:   \includegraphics[scale=0.47]{f9c.eps}

1077:   \caption{Images created in \isis\ by HYDRA {\tt Event-2D} and {\tt Source-3D}

1078:   routines, while analyzing cavities within the Chandra observation 4969

1079:   of Hydra A.  Top: residuals, data, and model.  Bottom: 2D and 3D

1080:   projections.}

1081:   \label{e2d}

1082:   \end{centering}

1083: \end{figure*}

1084: \begin{figure*}

1085:   \begin{centering}

1086:   \includegraphics[scale=0.70]{f10a.ps}

1087:   \hspace*{0.2in}

1088:   \includegraphics[scale=0.70]{f10b.ps}

1089:   \caption{Entire datacube from Spitzer observation 3310; the two clumpy

1090: 	   areas are instantly recognizable as regions of strong emission.}

1091:   \label{spitzer-volume}

1092:   \end{centering}

1093: \end{figure*}

1094: \subsection{Volume Visualization}

1095: Astrophysical observation and simulation are in the midst of a transition

1096: beyond 1D spectra and 2D images, and into the realm of three-dimensional

1097: phenomena. A number of astronomy software tools exist which enable

1098: visualization of so-called 3D data cubes, but a problem shared by many of

1099: them is that they are limited to displaying one 2D slice at a time,

1100: optionally in series as an animation.

1101: Here we consider a 99 spectra data cube generated by

1102: CUBISM\footnote{http://ssc.spitzer.caltech.edu/archanaly/contributed/cubism}

1103: from Spitzer observation 3310 of Cassiopeia A, with dimensions RA,

1104: DEC, and $\lambda$.  To identify where emission is strongest in

1105: the spatial and wavelength domains, the entire FITS datacube

1106: can be passed directly to \volview\ (Fig. \ref{spitzer-volume})

1107: {\small

1108: \begin{verbatim}

1109:   isis> require("volview")

1110:   isis> cube = fits_read_image("casa_ll1_s12.fits")

1111:   isis> volview(cube)

1112: \end{verbatim}

1113: }

1114: \noindent

1115: This visualization makes it instantly clear that the clumpy regions

1116: correspond to bright emission at fairly specific wavelengths.

1117: The \volview\ guilet

1118: is our

1119: interface to the Volpack

1120: rendering library \citep{192283}, which, although somewhat dated,

1121: is small, has no external dependencies, requires no special

1122: hardware, and is very fast.  The shear-warp factorization algorithm

1123: in Volpack has been generally recognized as the fastest rendering

1124: technique available \citep{353903}, a crucial factor for interactive

1125: analysis.  In contrast to high-end tools like ParaView and

1126: VISIT\footnote{http://www.paraview.org,\hspace*{2mm}http://www.llnl.gov/visit},

1127: volview provides a simple, low buy-in path to volume rendering and

1128: is actively used in the HYDRA project.\footnote{http://space.mit.edu/hydra/v3d.html}.

1129: 

1130: \subsection{Visual Correlation and Multidimensional Filtering}

1131: Having the ability to cut, visualize, and correlate data, in complex

1132: ways and from multiple missions, is invaluable to analysis.  This

1133: exploratory process ultimately seeks to derive sets of constraints

1134: {\em \bfseries C} = \{$C_{0}, ..., C_{n}$\} upon input data

1135: {\em \bfseries D} by iterating through a series of models, comparisons

1136: to data, and refinements of assumptions and model parameters.

1137: Because its analytic process does not encompass whole-array

1138: numerics (\S \ref{numerics}, \S \ref{input}), an iteration of this cycle

1139: in \xs\ can involve dumping application state to disk, followed

1140: by the execution of multiple tools such as {\tt fselect} or \fcalc\

1141: to transform or cut file data, possibly \fv\ or XIMAGE to visualize

1142: % NB: fv cannot filter table columns by laying regions down upon plots,

1143: % nor can data "shrunk" by the table calculator [e.g. where(x > a) is

1144: % not possible, b/c where() doesn't exist and b/c the fv calculator is

1145: % table-based, and so operates upon whole columns with #rows kept constant

1146: filtered data subsets, and \xs\ to incorporate the results back into analysis.

1147: This process creates temporary file litter products which can quickly

1148: become difficult to manage, and requires slow disk I/O passes over files

1149: when applying each unique set of filter constraints.  Other limiting

1150: factors are that it restricts the scope of exploration to data expressed

1151: in the FITS file format, and that \tcl-based I/O in tools like \fv\ does

1152: not scale well to datasets containing millions of events.

1153: 

1154: \label{vwhere}

1155: In contrast, \vwhere\ \citep{2005ASPC..347..237N}, a graphical extension

1156: of the \slang\ { \tt where} array slicing function, unifies the constraint

1157: cycle by enabling each phase to be performed within a single

1158: application.  The result can be an intuitive, faster, and

1159: more powerful alternative to file-based data mining. 

1160: By supporting the use of interpreted \slang\ arrays \isis\ \& \vwhere\ make

1161: it easy to combine data from multiple sources and formats, not just

1162: FITS files, and examine them in natural ways, as well as foster

1163: scalability.  \vwhere\ has been used to

1164: simultaneously mine hundreds of observations from multiple telescopes,

1165: e.g. for evidence of transitions between the soft and hard states of

1166: Cygnus X-1 \citep{2005ApJ...626.1006N}, correlate model parameters with

1167: fit results---directly in memory from dozens of fits---for scores of

1168: observations \citep{2006A&A...447..245W, MarkoffM81InPrep, NowakCygX1InPrep},

1169: and quickly corroborate colleague results \citep{2007ApJ...663L..93H}.

1170: 

1171: Filtering in \vwhere\ amounts to manipulating regions of interest

1172: on plots generated from \slang\ arrays (Fig. \ref{vwhere1}), with

1173: no syntax required and the effects upon any combination of vectors

1174: automatically visualized for inspection (Fig. \ref{vwhere2}).

1175: This approach reveals data correlations that can be difficult and

1176: time-consuming to ferret out with tool-based techniques.  New data

1177: vectors may be created on the fly and with no additional I/O overhead,

1178: using anything from simple arithmetic combinations of existing vectors

1179: to any mathematical transformation that can be specified in \slang\

1180: or imported from external C, C++, or \fort\ codes.

1181: 

1182: In contrast, file-based filtering tools require the use of syntax which

1183: may conflict in subtle ways from package to package.  Specifying

1184: mathematical operations to such tools, e.g., in the form of

1185: quoted Unix shell syntax, can be tedious and cumbersome, and does not

1186: approach the expressive power (consider recursion, for example) or

1187: performance of \isis\ numerics.

1188: As an illustration, consider the use of calculator tools for

1189: the discriminant computation in \S \ref{numerics}: the 1-million

1190: element \slang\ array datapoint in Fig. \ref{idl-plot} corresponds

1191: to a runtime of ca. 0.15 seconds:

1192: {\small

1193: \begin{verbatim}

1194:    isis>  a = [1:1e6+1];  b = 3*a; c = 2*a

1195:    isis> tic; d = sqrt(b^2 - 4*a*c); toc

1196:    0.151172

1197: \end{verbatim}

1198: }

1199: \noindent

1200: Using tools to perform similar operations upon files

1201: {\small

1202: \begin{verbatim}

1203:    fcalc in out d "sqrt(b^2 - 4*a*c)"

1204:    ftcalc in out d "sqrt(b^2 - 4*a*c)"

1205:    dmtcalc in out expression="d=sqrt(b^2 - 4*a*c)"

1206: \end{verbatim}

1207: }

1208: \noindent (from a locally compiled -O2 distribution of HEASOFT 6.1 and

1209: a CIAO3.3 binary distribution, respectively) was consistently 1 to 2

1210: orders of magnitude slower: roughly 1.8 sec for \fcalc, 2.3 sec for

1211: {\tt ftcalc} and 70 sec for {\tt dmtcalc}.  If we include the time to write

1212: {\small

1213: \begin{verbatim}

1214:    isis>  s = struct{a,b,c}; s.a=a; s.b=b; s.c=c

1215:    isis>  fits_write_binary_table("in","arrays",s)

1216: \end{verbatim}

1217: }

1218: \noindent the arrays to disk and then read them back in

1219: {\small

1220: \begin{verbatim}

1221:    isis>  d = fits_read_table("out").d

1222: \end{verbatim}

1223: }

1224: \noindent the performance penalty of the tool approach is even greater.

1225: Moreover, the output product of each tool is not directly useful:  to

1226: be visually inspected or used in

1227: further analysis, for instance, it first needs to be loaded into another

1228: program (e.g. \fv, or \xs), incurring additional I/O and

1229: application overheads that are not seen in \isis\ because its

1230: result arrays may be immediately manipulated or passed to subsequent

1231: computations.

1232: 

1233: % time ftcalc in shmem://h1 d "sqrt(b^2 - 4*a*c)"

1234: % real    2m45.378s	user    1m40.602s	sys     1m4.659s

1235: % Fri Aug 31 14:49:22 EDT 2007

1236: % time ftcalc in shmem://h1 d "sqrt(b^2 - 4*a*c)"

1237: % real    2m45.787s	user    1m40.132s	sys     1m5.536s

1238: Shared memory can, in principle, mitigate the disk I/O cost,

1239: but in this case it did not help:  instructing {\tt ftcalc}

1240: to write to {\tt shmem://h1} consistently resulted in runtimes of

1241: nearly 3 minutes, while the CIAO dmtcalc tool would not permit the

1242: creation of a shared memory FITS table.  Even if shared memory

1243: were faster than files in this case, the point raised earlier in

1244: \S \ref{input} still remains, namely that \xs\ documents no

1245: clear provision though which such generic data arrays may be utilized

1246: in analysis.

1247: It must also be noted that few, if any, file-based transformation

1248: tools are extensible.  If we wished to use a hypergeometric function in

1249: \vwhere\ it is as simple as loading the GSL module with \verb+require("gsl")+

1250: and calling the relevant function.

1251: If the mathematical operations required for their research are not

1252: supported, tool users would have to either make an enhancement request

1253: and wait or create their own solution.

1254: 

1255: \bfi{

1256: \subsection{Aside On Reproducibility}

1257: 

1258: We have been refining the approach to analysis espoused in this paper for

1259: much of the past decade.  During that time, one of the more persistent

1260: concerns we have privately encountered is

1261: that it can lead to diminished reproducibility, particularly in contrast

1262: to the \ftf\

1263: model.\footnote{This is relevant to the overall paper because it is

1264: the dominant means through which data are prepared for analysis in \xs,

1265: but applies to other analysis packages as well.}

1266: Reproducibility is a cornerstone of science, and remains a

1267: topic of debate in wider circles \citep{HCollins.1992, 2006Natur.442..344G},

1268: but is not treated in depth within the astronomy literature.  We

1269: do not attempt to fill that gap here.  Rather, we aim only to address the

1270: perception that configurable, interpreter-based methods compromise

1271: reproducibility relative to file-based tool methods, and hope that

1272: this may contribute to a more

1273: complete treatment of reproducibility elsewhere.

1274: 

1275: Central to reproducibility is the recording of history.  Many tools

1276: assist that process by annotating modified files with FITS history

1277: keywords.  This is of clear utility, especially when tracing pipelines

1278: and other well-defined data reduction procedures.  For open-ended

1279: analysis, however, we do not believe it is superior to the forms of

1280: history recorded by full-fledged analysis applications.

1281: Consider what is being captured in history records written by tools: each

1282: keystroke of the command used to invoke the tool.  This capability is

1283: not unique to tools.  \xs\ and \isis\ record keystroke history by virtue

1284: of their GNU readline command recall, editing, and logging capabilities,

1285: as do many other systems.  Focusing on

1286: keystrokes alone, though, obscures a larger point:  regardless of whether

1287: one prefers logs or header keywords, indiscriminately long lists of

1288: commands typed are of little value without some sense of their

1289: relevance to publishable results.

1290: 

1291: In the end what matters most is that results may be regenerated so that

1292: conclusions may be plausibly confirmed by others, rather than having

1293: every bump or wrong turn along the way reproduced in high fidelity.

1294: Scripts, the

1295: logical endpoint of our interpreter-based approach to analysis, confer

1296: this conceptual prioritization by making explicit the data and algorithms

1297: of greatest significance.  Such scripts arise in tool-based analysis as 

1298: well, only they tend to be expressed in system-oriented languages like

1299: Bourne shell or Perl, rather than intrinsically whole-array numeric

1300: languages like \slang\ or IDL.  It can be argued, then, that scripts lead

1301: to higher forms of duplicability than FITS history records alone.  They

1302: are also easier to annotate with additional commentary.  We therefore

1303: conclude that that tool- and interpreter-based approaches to analysis

1304: are approximately equivalent in the degree to which they facilitate

1305: reproducibility.

1306: Care must be still taken when ``feature creep" introduces incompatibilities

1307: that make the use of older scripts problematic with newer software

1308: versions, so it is important that version information be recorded and

1309: that older software remains accessible on the internet.

1310: }

1311: \begin{figure*}[p]

1312:   \begin{centering}

1313:   \includegraphics[scale=0.6]{f11a.eps}

1314:   \hspace*{0.51in}

1315:   \includegraphics[scale=0.4]{f11b.ps}

1316:   \caption{\vwhere\ in \isis\, using SITAR and the AGLC package to

1317:      generate a lightcurve for Chandra observation 3814 of Cygnus X-1.

1318:      Configurable plots are generated from an axis expression window

1319:      containing several text fields.  Each X/Y expression may contain

1320:      any valid \slang\ statement, including calls to C, C++ or \fort\ 

1321:      functions within external modules.  The chief constraint upon an

1322:      expression is that, when evaluated, it generate a numeric vector

1323:      equal in length to the existing data vectors.

1324:      Here we show \vwhere\ launched with a struct

1325:      containing 3 input vectors, as well as how easily new vectors (e.g.

1326: 	{\tt lchan + mchan}) can be fabricated from them on the fly.

1327:      Each vector expression is saved in the {\tt Choose} dropdown menu,

1328: 	for easy reuse with no retyping.

1329:      Vectors of arbitrary size may also be overplotted, while the 

1330:      {\tt S-Lang>} prompt may be used as an interactive command line,

1331: 	e.g., to issue \isis\ commands or execute arbitrary \slang\ code.}

1332:   \label{vwhere1}

1333:   \end{centering}

1334: \end{figure*}

1335: 

1336: \begin{figure*}

1337:   \begin{centering}

1338:   \includegraphics[scale=0.4]{f12a.ps}

1339:   \hspace*{0.1in}

1340:   \includegraphics[scale=0.4]{f12b.ps}

1341:   \caption{A polygon filter applied to the color intensity diagram, and

1342:   its effect upon the lightcurve  (Hanke et al, in prep.).}

1343:   \label{vwhere2}

1344:   \end{centering}

1345: \end{figure*}

1346: 

1347: \section{Conclusion}

1348: 

1349: Progress in science can be measured by our ability to pose increasingly

1350: advanced, open-ended questions and address them with flexible techniques

1351: of commensurate sophistication.  Since the age of Newton, one implication

1352: of this is that scientists must also practice mathematics, and since the

1353: age of Turing it has also meant they must practice programming.

1354: 

1355: In this paper we have compared two open-source spectral analysis applications,

1356: \xs\ and \isis, in an attempt to gauge how their scientific and mathematical

1357: reach may be extended with custom programming.  Contrasts between the two

1358: have been drawn from the context of current research efforts utilizing high

1359: performance computation, numerical modeling, atomic physics, visualization,

1360: automated code generation, and data management.  We have

1361: demonstrated how these have led to the incorporation of new analytic

1362: techniques in \isis, in ways that are either unexplored or problematic

1363: for the current \xs\ architecture and its associated \ftf\ model of

1364: analysis.

1365: 

1366: \xs\ has been of tremendous value to the X-ray community, continues

1367: to be actively developed, and has a strong history of community enhancement

1368: via local models (and to some extent, \tcl/TK scripts).  However, we have

1369: argued that the top-level simplicity of

1370: the \xs\ interface, long and rightfully one of the pillars of its appeal,

1371: can shroud much of its internal computational and data handling mechanisms.

1372: This in turn can render \xs\ less adaptable---by the typical user---for

1373: evolving research needs.  Commands which operate as black-boxed common

1374: denominators of analysis, and permitting only file-based data input, may

1375: not be enough to probe some of the more computationally challenging

1376: problems facing astronomers.

1377: 

1378: We conclude that analysis applications such as \isis, endowed with

1379: interpreted whole-array numerical capabilities and an open, modular,

1380: and scriptable architecture designed expressly for high configurability,

1381: are more favorably equipped to support ad-hoc research needs while not

1382: appreciably compromising reproducibility.  Originally envisioned as a

1383: tool for Chandra gratings spectroscopy, \isis\ has heavily influenced

1384: the development of additional Chandra analysis software, while also

1385: receiving NASA AISRP funding to continue its evolution within the

1386: aforementioned HYDRA project.

1387: Although some of the \isis\ capabilities we've emphasized do exist in

1388: other astronomy software, we are unaware of any publications demonstrating

1389: how similar breadths of flexibility and computational power have been

1390: collected under the umbrella of a single open-source analysis application

1391: and brought to bear on published research in spectral analysis and

1392: X-ray astronomy in the manner discussed here.

1393: 

1394: \acknowledgments

1395: We would like to thank Daniel Dewey, J\"{o}rn Wilms, and our MIT

1396: colleagues for reviewing drafts of the paper, as well as Tracey DeLaney

1397: for supplying the Spitzer 3D spectral cubes of CasA.  We are grateful

1398: to our anonymous referee for constructive criticism, and to NASA for

1399: funding this work through the AISRP grant NNG06GE58G.

1400: 

1401: %% To help institutions obtain information on the effectiveness of their

1402: %% telescopes, the AAS Journals has created a group of keywords for telescope

1403: %% facilities. A common set of keywords will make these types of searches

1404: %% significantly easier and more accurate. In addition, they will also be

1405: %% useful in linking papers together which utilize the same telescopes

1406: %% within the framework of the National Virtual Observatory.

1407: %% See the AASTeX Web site at http://www.journals.uchicago.edu/AAS/AASTeX

1408: %% for information on obtaining the facility keywords.

1409: 

1410: %% After the acknowledgments section, use the following syntax and the

1411: %% \facility{} macro to list the keywords of facilities used in the research

1412: %% for the paper.  Each keyword will be checked against the master list during

1413: %% copy editing.  Individual instruments or configurations can be provided 

1414: %% in parentheses, after the keyword, but they will not be verified.

1415: 

1416: %{\it Facilities:} \facility{Nickel}, \facility{HST (STIS)}, \facility{CXO (ASIS)}.

1417: 

1418: %% Appendix material should be preceded with a single \appendix command.

1419: %% There should be a \section command for each appendix. Mark appendix

1420: %% subsections with the same markup you use in the main body of the paper.

1421: 

1422: %% Each Appendix (indicated with \section) will be lettered A, B, C, etc.

1423: %% The equation counter will reset when it encounters the \appendix

1424: %% command and will number appendix equations (A1), (A2), etc.

1425: 

1426: %\appendix

1427: 

1428: %% The reference list follows the main body and any appendices.

1429: %% Use LaTeX's thebibliography environment to mark up your reference list.

1430: %% Note \begin{thebibliography} is followed by an empty set of

1431: %% curly braces.  If you forget this, LaTeX will generate the error

1432: %% "Perhaps a missing \item?".

1433: %%

1434: %% thebibliography produces citations in the text using \bibitem-\cite

1435: %% cross-referencing. Each reference is preceded by a

1436: %% \bibitem command that defines in curly braces the KEY that corresponds

1437: %% to the KEY in the \cite commands (see the first section above).

1438: %% Make sure that you provide a unique KEY for every \bibitem or else the

1439: %% paper will not LaTeX. The square brackets should contain

1440: %% the citation text that LaTeX will insert in

1441: %% place of the \cite commands.

1442: 

1443: %% We have used macros to produce journal name abbreviations.

1444: %% AASTeX provides a number of these for the more frequently-cited journals.

1445: %% See the Author Guide for a list of them.

1446: 

1447: %% Note that the style of the \bibitem labels (in []) is slightly

1448: %% different from previous examples.  The natbib system solves a host

1449: %% of citation expression problems, but it is necessary to clearly

1450: %% delimit the year from the author name used in the citation.

1451: %% See the natbib documentation for more details and options.

1452: 

1453: \bibliographystyle{apalike}

1454: \begin{thebibliography}{}

1455: 

1456: \bibitem[{Arnaud}, 1996]{1996ASPC..101...17A}

1457: {Arnaud}, K.~A. (1996).

1458: \newblock {XSPEC: The First Ten Years}.

1459: \newblock In {Jacoby}, G.~H. and {Barnes}, J., editors, {\em ASP Conf. Ser.

1460:   101: Astronomical Data Analysis Software and Systems V}, pages 17--+.

1461: 

1462: \bibitem[{Ayers}, 1996]{SLangRef}

1463: {Ayers}, L. (1996).

1464: \newblock {S-Lang Applications for Linux}.

1465: \newblock {\em The Linux Gazette}, 12.

1466: \newblock {http://linuxgazette.net/issue12/slang.html}.

1467: 

1468: \bibitem[{Beazley}, 1996]{1996SWIG}

1469: {Beazley}, D. (1996).

1470: \newblock {SWIG : An Easy to Use Tool for Integrating Scripting Languages with

1471:   C and C++.}

1472: \newblock In {\em 4th Tcl/Tk Workshop Proceedings}.

1473: 

1474: \bibitem[Bland et~al., 2007]{OscarCLI}

1475: Bland, W., Naughton, T., Vallee, G., and Scott, S.~L. (May 2007).

1476: \newblock Design and implementation of a menu based oscar command line

1477:   interface.

1478: \newblock {\em High Performance Computing Systems and Applications, 2007. HPCS

1479:   2007. 21st International Symposium on}, pages 25--25.

1480: 

1481: \bibitem[{Brenneman} and {Reynolds}, 2006]{2006ApJ...652.1028B}

1482: {Brenneman}, L.~W. and {Reynolds}, C.~S. (2006).

1483: \newblock {Constraining Black Hole Spin via X-Ray Spectroscopy}.

1484: \newblock {\em \apj}, 652:1028--1043.

1485: 

1486: \bibitem[Collins, 1992]{HCollins.1992}

1487: Collins, H.~M. (1992).

1488: \newblock {\em Changing Order: Replication and Induction in Scientific

1489:   Practice}.

1490: \newblock University of Chicago Press.

1491: 

1492: \bibitem[{Davis} et~al., 2005]{2005ASPC..347..444D}

1493: {Davis}, J.~E., {Houck}, J.~C., {Allen}, G.~E., and {Stage}, M.~D. (2005).

1494: \newblock {Analyzing the Cas A Megasecond in Less than a Megasecond}.

1495: \newblock In {Shopbell}, P., {Britton}, M., and {Ebert}, R., editors, {\em

1496:   Astronomical Society of the Pacific Conference Series}, pages 444--+.

1497: 

1498: \bibitem[{de La Pe{\~n}a} et~al., 2001]{2001ASPC..238...59D}

1499: {de La Pe{\~n}a}, M.~D., {White}, R.~L., and {Greenfield}, P. (2001).

1500: \newblock {The PyRAF Graphics System}.

1501: \newblock In {Harnden}, Jr., F.~R., {Primini}, F.~A., and {Payne}, H.~E.,

1502:   editors, {\em Astronomical Data Analysis Software and Systems X}, volume 238

1503:   of {\em Astronomical Society of the Pacific Conference Series}, pages 59--+.

1504: 

1505: \bibitem[{Folk} et~al., 1999]{1999IEEE...99...273F}

1506: {Folk}, M., {McGrath}, R., and Yeager, N. (1999).

1507: \newblock {HDF: an update and future directions}.

1508: \newblock In {\em IEEE Geoscience and Remote Sensing Symposium, 1999, Vol. 1},

1509:   pages 273--275.

1510: 

1511: \bibitem[{Fryxell} et~al., 2001]{2001NuPhA.688..172F}

1512: {Fryxell}, B., {Zingale}, M., {Timmes}, F.~X., {Lamb}, D.~Q., {Olson}, K.,

1513:   {Calder}, A.~C., {Dursi}, L.~J., {Ricker}, P., {Rosner}, R., {Truran}, J.~W.,

1514:   {MacNeice}, P., and {Tufo}, H. (2001).

1515: \newblock {Numerical simulations of thermonuclear flashes on neutron stars}.

1516: \newblock {\em Nuclear Physics A}, 688:172--176.

1517: 

1518: \bibitem[Geist et~al., 1994]{Geist:1994:PPV}

1519: Geist, A., Beguelin, A., Dongarra, J., Jiang, W., Manchek, R., and Sunderam,

1520:   V.~S. (1994).

1521: \newblock {\em {PVM}: Parallel Virtual Machine: {A} Users' Guide and Tutorial

1522:   for Networked Parallel Computing}.

1523: 

1524: \bibitem[Gilat, 2004]{Gilat.2004.MATLAB}

1525: Gilat, A. (2004).

1526: \newblock {\em {MATLAB: An Introduction with Applications, 2nd Ed.}}

1527: \newblock John Wiley \& Sons.

1528: 

1529: \bibitem[{Giles}, 2006]{2006Natur.442..344G}

1530: {Giles}, J. (2006).

1531: \newblock {The trouble with replication}.

1532: \newblock {\em \nat}, 442:344--+.

1533: 

1534: \bibitem[Gropp et~al., 1999]{330577}

1535: Gropp, W., Lusk, E., and Skjellum, A. (1999).

1536: \newblock {\em Using MPI (2nd ed.): portable parallel programming with the

1537:   message-passing interface}.

1538: \newblock MIT Press, Cambridge, MA, USA.

1539: 

1540: \bibitem[{Heinz} et~al., 2007]{2007ApJ...663L..93H}

1541: {Heinz}, S., {Schulz}, N.~S., {Brandt}, W.~N., and {Galloway}, D.~K. (2007).

1542: \newblock {Evidence of a Parsec-Scale X-Ray Jet from the Accreting Neutron Star

1543:   Circinus X-1}.

1544: \newblock {\em \apjl}, 663:L93--L96.

1545: 

1546: \bibitem[{Houck}, 2002]{2002hrxs.confE..17H}

1547: {Houck}, J.~C. (2002).

1548: \newblock {ISIS: The Interactive Spectral Interpretation System}.

1549: \newblock In {Branduardi-Raymont}, G., editor, {\em High Resolution X-ray

1550:   Spectroscopy with XMM-Newton and Chandra, Mullard Space Science Laboratory of

1551:   University College, London}.

1552: 

1553: \bibitem[{Hwang} et~al., 2004]{2004ApJ...615L.117H}

1554: {Hwang}, U., {Laming}, J.~M., {Badenes}, C., {Berendse}, F., {Blondin}, J.,

1555:   {Cioffi}, D., {DeLaney}, T., {Dewey}, D., {Fesen}, R., {Flanagan}, K.~A.,

1556:   {Fryer}, C.~L., {Ghavamian}, P., {Hughes}, J.~P., {Morse}, J.~A.,

1557:   {Plucinsky}, P.~P., {Petre}, R., {Pohl}, M., {Rudnick}, L., {Sankrit}, R.,

1558:   {Slane}, P.~O., {Smith}, R.~K., {Vink}, J., and {Warren}, J.~S. (2004).

1559: \newblock {A Million Second Chandra View of Cassiopeia A}.

1560: \newblock {\em \apjl}, 615:L117--L120.

1561: 

1562: \bibitem[{Jordan} et~al., 1994]{1994ASPC...61...71J}

1563: {Jordan}, J.~M., {Jennings}, D.~G., {McGlynn}, T.~A., {Bonnell}, J.~T.,

1564:   {Gliba}, G.~W., {Ruggiero}, N.~G., and {Serlemitsos}, T.~A. (1994).

1565: \newblock {An Xwindows/Motif Graphical User Interface for Xspec}.

1566: \newblock In {Crabtree}, D.~R., {Hanisch}, R.~J., and {Barnes}, J., editors,

1567:   {\em ASP Conf. Ser. 61: Astronomical Data Analysis Software and Systems III},

1568:   pages 71--+.

1569: 

1570: \bibitem[{Joye} and {Mandel}, 2003]{2003ASPC..295..489J}

1571: {Joye}, W.~A. and {Mandel}, E. (2003).

1572: \newblock {New Features of SAOImage DS9}.

1573: \newblock In {Payne}, H.~E., {Jedrzejewski}, R.~I., and {Hook}, R.~N., editors,

1574:   {\em ASP Conf. Ser. 295: Astronomical Data Analysis Software and Systems

1575:   XII}, pages 489--+.

1576: 

1577: \bibitem[Kallman and Palmeri, 2007]{kallman-2007-79}

1578: Kallman, T.~R. and Palmeri, P. (2007).

1579: \newblock Atomic data for x-ray astrophysics.

1580: \newblock {\em Reviews of Modern Physics}, 79:79--133.

1581: 

1582: \bibitem[Kiczales, 1996]{BeyondBlackBox}

1583: Kiczales, G. (Jan 1996).

1584: \newblock Beyond the black box: open implementation.

1585: \newblock {\em Software, IEEE}, 13(1):8, 10--11.

1586: 

1587: \bibitem[Kling, 1999]{Kling.1999.IDL}

1588: Kling, R. (1999).

1589: \newblock {\em {Application Development with IDL}}.

1590: \newblock Ronn Kling Consulting.

1591: 

1592: \bibitem[{Kurtz} et~al., 1993]{1993ASPC...52..132K}

1593: {Kurtz}, M.~J., {Karakashian}, T., {Grant}, C.~S., {Eichhorn}, G., {Murray},

1594:   S.~S., {Watson}, J.~M., {Ossorio}, P.~G., and {Stoner}, J.~L. (1993).

1595: \newblock {Intelligent Text Retrieval in the NASA Astrophysics Data System}.

1596: \newblock In {Hanisch}, R.~J., {Brissenden}, R.~J.~V., and {Barnes}, J.,

1597:   editors, {\em Astronomical Data Analysis Software and Systems II}, volume~52

1598:   of {\em Astronomical Society of the Pacific Conference Series}, pages 132--+.

1599: 

1600: \bibitem[Lacroute and Levoy, 1994]{192283}

1601: Lacroute, P. and Levoy, M. (1994).

1602: \newblock Fast volume rendering using a shear-warp factorization of the viewing

1603:   transformation.

1604: \newblock In {\em SIGGRAPH '94: Proceedings of the 21st annual conference on

1605:   Computer graphics and interactive techniques}, pages 451--458, New York, NY,

1606:   USA. ACM Press.

1607: 

1608: \bibitem[{Mandel} et~al., 1995]{1995xpa.v1M}

1609: {Mandel}, E., {Swick}, R., and D., T. (1995).

1610: \newblock {The X Public Access Mechanism: Software Cooperation for Space

1611:   Science and Beyond}.

1612: \newblock {\em Proceedings of the 9th Annual X Technical Conference, Boston,

1613:   MA}.

1614: 

1615: \bibitem[{Markoff} and {Nowak}, 2004]{2004ApJ...609..972M}

1616: {Markoff}, S. and {Nowak}, M.~A. (2004).

1617: \newblock {Constraining X-Ray Binary Jet Models via Reflection}.

1618: \newblock {\em \apj}, 609:972--976.

1619: 

1620: \bibitem[{Markoff} et~al., 2005]{2005ApJ...635.1203M}

1621: {Markoff}, S., {Nowak}, M.~A., and {Wilms}, J. (2005).

1622: \newblock {Going with the Flow: Can the Base of Jets Subsume the Role of

1623:   Compact Accretion Disk Coronae?}

1624: \newblock {\em \apj}, 635:1203--1216.

1625: 

1626: \bibitem[{Markoff} et~al., 2007]{MarkoffM81InPrep}

1627: {Markoff}, S., {Young}, A., and {Nowak}, M.~A. (2007).

1628: \newblock In preparation.

1629: 

1630: \bibitem[Mei{\ss}ner et~al., 2000]{353903}

1631: Mei{\ss}ner, M., Huang, J., Bartz, D., Mueller, K., and Crawfis, R. (2000).

1632: \newblock A practical evaluation of popular volume rendering algorithms.

1633: \newblock In {\em VVS '00: Proceedings of the 2000 IEEE symposium on Volume

1634:   visualization}, pages 81--90, New York, NY, USA. ACM Press.

1635: 

1636: \bibitem[{Noble}, 2005]{2005ASPC..347..237N}

1637: {Noble}, M.~S. (2005).

1638: \newblock {VWhere: A Visual, Extensible `where' Command}.

1639: \newblock In {Shopbell}, P., {Britton}, M., and {Ebert}, R., editors, {\em ASP

1640:   Conf. Ser. 347: Astronomical Data Analysis Software and Systems XIV}, pages

1641:   237--+.

1642: 

1643: \bibitem[Noble, 2007]{Noble.2007.OpenMP}

1644: Noble, M.~S. (2007).

1645: \newblock {Getting More From Your Multicore: Exploiting OpenMP From An Open

1646:   Source Numerical Scripting Language}.

1647: \newblock {\em Computation and Concurrency: Practice and Experience}.

1648: \newblock {In press. arXiv e-print at

1649:   http://adsabs.harvard.edu/abs/2007arXiv0706.4048N}.

1650: 

1651: \bibitem[{Noble} et~al., 2006]{2006ASPC..351..481N}

1652: {Noble}, M.~S., {Houck}, J.~C., {Davis}, J.~E., {Young}, A., and {Nowak}, M.

1653:   (2006).

1654: \newblock {Using the Parallel Virtual Machine for Everyday Analysis}.

1655: \newblock In {Gabriel}, C., {Arviset}, C., {Ponz}, D., and {Enrique}, S.,

1656:   editors, {\em ASP Conf. Ser. 351: Astronomical Data Analysis Software and

1657:   Systems XV}, pages 481--+.

1658: 

1659: \bibitem[Norman, 2007]{NextUIBrkThru}

1660: Norman, D. (2007).

1661: \newblock The next ui breakthrough: command lines.

1662: \newblock {\em interactions}, 14(3):44--45.

1663: 

1664: \bibitem[{Nowak} et~al., 2005]{2005ApJ...626.1006N}

1665: {Nowak}, M.~A., {Wilms}, J., {Heinz}, S., {Pooley}, G., {Pottschmidt}, K., and

1666:   {Corbel}, S. (2005).

1667: \newblock {Is the ``IR Coincidence'' Just That?}

1668: \newblock {\em \apj}, 626:1006--1014.

1669: 

1670: \bibitem[{Nowak} et~al., 2008]{NowakCygX1InPrep}

1671: {Nowak}, M.~A., {Wilms}, J., and {Pottschmidt}, K. (2008).

1672: \newblock In preparation.

1673: 

1674: \bibitem[{Pence} et~al., 1997]{1997ASPC..125..261P}

1675: {Pence}, W., {Xu}, J., and {Brown}, L. (1997).

1676: \newblock {FV: A New FITS File Visualization Tool}.

1677: \newblock In {Hunt}, G. and {Payne}, H., editors, {\em ASP Conf. Ser. 125:

1678:   Astronomical Data Analysis Software and Systems VI}, pages 261--+.

1679: 

1680: \bibitem[{Primini} et~al., 2005]{2005ASPC..347...17P}

1681: {Primini}, F., {Noble}, M., and {The CXC Science Data Systems Group} (2005).

1682: \newblock {Extending the Capabilities of CIAO with S-Lang-based Tools}.

1683: \newblock In {Shopbell}, P., {Britton}, M., and {Ebert}, R., editors, {\em ASP

1684:   Conf. Ser. 347: Astronomical Data Analysis Software and Systems XIV}, pages

1685:   17--+.

1686: 

1687: \bibitem[{Smith} et~al., 2001]{2001ApJ...556L..91S}

1688: {Smith}, R.~K., {Brickhouse}, N.~S., {Liedahl}, D.~A., and {Raymond}, J.~C.

1689:   (2001).

1690: \newblock {Collisional Plasma Models with APEC/APED: Emission-Line Diagnostics

1691:   of Hydrogen-like and Helium-like Ions}.

1692: \newblock {\em \apjl}, 556:L91--L95.

1693: 

1694: \bibitem[{Stage} et~al., 2006]{2006NatPh...2..614S}

1695: {Stage}, M.~D., {Allen}, G.~E., {Houck}, J.~C., and {Davis}, J.~E. (2006).

1696: \newblock {Cosmic-ray diffusion near the Bohm limit in the Cassiopeia A

1697:   supernova remnant}.

1698: \newblock {\em Nature Physics}, 2:614--619.

1699: 

1700: \bibitem[Trowbridge et~al., 2007]{TrowbridgeNowakWilms2007}

1701: Trowbridge, S.~R., Nowak, M.~A., and Wilms, J. (2007).

1702: \newblock {Tracking the Orbital and Super-orbital Periodicities of SMC X-1}.

1703: \newblock {In press}.

1704: 

1705: \bibitem[{Wilms} et~al., 2006]{2006A&A...447..245W}

1706: {Wilms}, J., {Nowak}, M.~A., {Pottschmidt}, K., {Pooley}, G.~G., and {Fritz},

1707:   S. (2006).

1708: \newblock {Long term variability of Cygnus X-1. IV. Spectral evolution

1709:   1999-2004}.

1710: \newblock {\em \aap}, 447:245--261.

1711: 

1712: \bibitem[{Wise} and {Houck}, 2004]{2004cosp...35.3997W}

1713: {Wise}, M. and {Houck}, J. (2004).

1714: \newblock {Mapping Heating in Clusters of Galaxies}.

1715: \newblock In {\em 35th COSPAR Scientific Assembly}, volume~35 of {\em COSPAR,

1716: Plenary Meeting}, pages 3997--+.

1717: 

1718: \end{thebibliography}

1719: 

1720: \clearpage

1721: 

1722: %% Use the figure environment and \plotone or \plottwo to include

1723: 

1724: %\begin{figure}

1725: %  \begin{centering}

1726: %  \includegraphics[angle=-90,scale=0.3]{f1.eps}

1727: %  \caption{Ionization fractions for Fe XVII through XXI.}

1728: %  \label{ionfrac}

1729: %  \end{centering}

1730: %\end{figure}

1731: 

1732: %\begin{figure}

1733: %  \begin{centering}

1734: %    \includegraphics[scale=0.50]{f2.eps}

1735: %    \caption{ The \isis\ { \tt main()} program is a thin ANSI C layer above

1736: %	the \slang\ interpreter, used primarily to establish hooks to

1737: %	modules and gather user input.  Taking modularity

1738: %	to this extreme, where carefully orthogonalized components provide

1739: %	the bulk of capability, provides a clean and flexible implementation.

1740: %	Most modules may also be

1741: %	used outside of \isis.}

1742: %  \label{arch}

1743: %  \end{centering}

1744: %\end{figure}

1745: 

1746: %\begin{figure}

1747: %  \begin{centering}

1748: %  \includegraphics[angle=-90,scale=0.30]{f3.eps}

1749: %  \caption{Performance of IDL 6.1 (binary distribution) and

1750: %        \slang\ (statically linked), on $\sqrt{b^2 - 4ac}$

1751: %        for arrays of various sizes.}

1752: %  \label{idl-plot}

1753: %  \end{centering}

1754: %\end{figure}

1755: %\begin{figure}

1756: %  \begin{centering}

1757: %  \includegraphics[scale=0.92]{f4.eps}

1758: %  \caption{Pure \slang\ orbital model, fit

1759: %  by \isis\ directly to in-memory data arrays.

1760: %As described in \cite{TrowbridgeNowakWilms2007}, the model

1761: %represents a lightcurve of constant amplitude {\tt a}, with an eclipse

1762: %centered at {\tt c}, with a width {\tt w2} from the start of ingress

1763: %to the end of egress.  The ingress and egress have a quadratic form

1764: %({\tt m} and {\tt b} parameters), and there is a quadratic

1765: %``bounceback" ({\tt s} parameter) with width {\tt w1}.}

1766: %  \label{orbmodel}

1767: %  \end{centering}

1768: %\end{figure}

1769: 

1770: %\begin{figure}[t]

1771: %  \begin{centering}

1772: %  \includegraphics[angle=-90,scale=0.3]{f5.eps}

1773: %  \caption{Performance of orbital models implemented in Perl 5.8.4

1774: %   and Python 2.4, relative to \slang\ 2.0.7.

1775: %   \tcl\ is not represented here because neither of its numerical 

1776: %   extensions, BLT and NAP, offered a clear equivalent to

1777: %   the {\tt where()} function.}

1778: %  \label{orbmodel-plot}

1779: %  \end{centering}

1780: %\end{figure}

1781: 

1782: %\begin{figure}

1783: %  \begin{centering}

1784: %  \includegraphics[scale=0.35]{f6.eps}

1785: %  \caption{Power Spectrum of X1820-303, generated from the PCA in RXTE

1786: %  observation P10075 (data set 10075-01-01-02).}

1787: %  \label{timing}

1788: %  \end{centering}

1789: %\end{figure}

1790: 

1791: %\begin{figure}

1792: %  \begin{centering}

1793: %  \hspace*{3.5mm}

1794: %  \includegraphics[angle=-90,scale=0.35]{f7a.eps}

1795: %  \includegraphics[angle=-90,scale=0.35]{f7b.eps}

1796: %  \caption{ACIS image from 2004 Chandra observation of NGC7714, with

1797: %     corresponding square and smoothed pixel intensity histograms.}

1798: %  \label{ngc7714}

1799: %  \end{centering}

1800: %\end{figure}

1801: 

1802: %% figures and captions in your electronic submission.

1803: %% To embed the sample graphics in

1804: %% the file, uncomment the \plotone, \plottwo, and

1805: %% \includegraphics commands

1806: %%

1807: %% If you need a layout that cannot be achieved with \plotone or

1808: %% \plottwo, you can invoke the graphicx package directly with the

1809: %% \includegraphics command or use \plotfiddle. For more information,

1810: %% please see the tutorial on "Using Electronic Art with AASTeX" in the

1811: %% documentation section at the AASTeX Web site,

1812: %% http://www.journals.uchicago.edu/AAS/AASTeX.

1813: %%

1814: %% The examples below also include sample markup for submission of

1815: %% supplemental electronic materials. As always, be sure to check

1816: %% the instructions to authors for the journal you are submitting to

1817: %% for specific submissions guidelines as they vary from

1818: %% journal to journal.

1819: 

1820: %% This example uses \plotone to include an EPS file scaled to

1821: %% 80% of its natural size with \epsscale. Its caption

1822: %% has been written to indicate that additional figure parts will be

1823: %% available in the electronic journal.

1824: 

1825: %\begin{figure}

1826: %\epsscale{.80}

1827: %\plotone{f1.eps}

1828: %\caption{Derived spectra for 3C138 \citep[see][]{heiles03}. Plots for all sources are available

1829: %in the electronic edition of {\it The Astrophysical Journal}.\label{fig1}}

1830: %\end{figure}

1831: 

1832: \clearpage

1833: 

1834: %% Here we use \plottwo to present two versions of the same figure,

1835: %% one in black and white for print the other in RGB color

1836: %% for online presentation. Note that the caption indicates

1837: %% that a color version of the figure will be available online.

1838: %%

1839: 

1840: %\begin{figure}

1841: %\plottwo{f2.eps}{f2_color.eps}

1842: %\caption{A panel taken from Figure 2 of \citet{rudnick03}. 

1843: %See the electronic edition of the Journal for a color version 

1844: %of this figure.\label{fig2}}

1845: %\end{figure}

1846: 

1847: %% This figure uses \includegraphics to scale and rotate the still frame

1848: %% for an mpeg animation.

1849: 

1850: %\begin{figure}

1851: %\includegraphics[angle=90,scale=.50]{f3.eps}

1852: %\caption{Animation still frame taken from \citet{kim03}.

1853: %This figure is also available as an mpeg

1854: %animation in the electronic edition of the

1855: %{\it Astrophysical Journal}.}

1856: %\end{figure}

1857: 

1858: %% If you are not including electronic art with your submission, you may

1859: %% mark up your captions using the \figcaption command. See the

1860: %% User Guide for details.

1861: %%

1862: %% No more than seven \figcaption commands are allowed per page,

1863: %% so if you have more than seven captions, insert a \clearpage

1864: %% after every seventh one.

1865: 

1866: %% Tables should be submitted one per page, so put a \clearpage before

1867: %% each one.

1868: 

1869: %% Two options are available to the author for producing tables:  the

1870: %% deluxetable environment provided by the AASTeX package or the LaTeX

1871: %% table environment.  Use of deluxetable is preferred.

1872: %%

1873: 

1874: %% Three table samples follow, two marked up in the deluxetable environment,

1875: %% one marked up as a LaTeX table.

1876: 

1877: %% In this first example, note that the \tabletypesize{}

1878: %% command has been used to reduce the font size of the table.

1879: %% We also use the \rotate command to rotate the table to

1880: %% landscape orientation since it is very wide even at the

1881: %% reduced font size.

1882: %%

1883: %% Note also that the \label command needs to be placed

1884: %% inside the \tablecaption.

1885: 

1886: %% This table also includes a table comment indicating that the full

1887: %% version will be available in machine-readable format in the electronic

1888: %% edition.

1889: 

1890: \clearpage

1891: 

1892: %% If the table is more than one page long, the width of the table can vary

1893: %% from page to page when the default \tablewidth is used, as below.  The

1894: %% individual table widths for each page will be written to the log file; a

1895: %% maximum tablewidth for the table can be computed from these values.

1896: %% The \tablewidth argument can then be reset and the file reprocessed, so

1897: %% that the table is of uniform width throughout. Try getting the widths

1898: %% from the log file and changing the \tablewidth parameter to see how

1899: %% adjusting this value affects table formatting.

1900: 

1901: %% The \dataset{} macro has also been applied to a few of the objects to

1902: %% show how many observations can be tagged in a table.

1903: 

1904: \end{document}

1905: