1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: %% %%
3: %% TITLE: Facing the LISA Data Analysis Challenge %%
4: %% AUTHOR: Louis J. Rubbo %%
5: %% DATE: October 23, 2006 %%
6: %% %%
7: %% NOTES: This document requires the jpconf package to compile. %%
8: %% %%
9: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
10:
11:
12: \documentclass[a4paper]{jpconf}
13:
14: %#### PACKAGES ####################################
15:
16: % Packages for inserting colored eps figures
17: \usepackage[dvips]{color, graphicx}
18:
19:
20:
21: %#### LENGTHS #####################################
22:
23: % This call sets the amount of space on either side of the equals sign
24: % in the equation array environment
25: \setlength{\arraycolsep}{2pt}
26:
27:
28:
29: %#### NEW COMMANDS ################################
30:
31: % Note environment
32: \newcommand{\note}[1]{[$\blacktriangleright$~\textbf{#1}~$\blacktriangleleft$]}
33:
34: % Shortcut for a solar mass
35: \newcommand{\Msun}{\textrm{M}_{\odot}}
36:
37: % Shortcut for a black hole mass
38: \newcommand{\Mbh}{M_{\bullet}}
39:
40:
41:
42: %#### INFORMATION #################################
43:
44: \begin{document}
45:
46: \title{Facing the LISA data analysis challenge}
47:
48: \author{Louis J Rubbo}
49:
50: \address{Center for Gravitational Wave Physics, Pennsylvania State
51: University, University Park, PA 16802 }
52:
53: \ead{rubbo@gravity.psu.edu}
54:
55:
56: \date{\today}
57:
58:
59:
60: %#### MAIN DOCUMENT ###############################
61:
62: %==== Abstract ====================================
63:
64: \begin{abstract}
65: By being the first observatory to survey the source rich low
66: frequency region of the gravitational wave spectrum, the Laser
67: Interferometer Space Antenna (LISA) will revolutionize our
68: understanding of the Cosmos. For the first time we will be able to
69: detect the gravitational radiation from millions of galactic
70: binaries, the coalescence of two massive black holes, and the
71: inspirals of compact objects into massive black holes. The signals
72: from multiple sources in each class, and possibly others as well,
73: will be simultaneously present in the data. To achieve the enormous
74: scientific return possible with LISA, sophisticated data analysis
75: techniques must be developed which can mine the complex data in an
76: effort to isolate and characterize individual signals. This
77: proceedings paper very briefly summarizes the challenges associated
78: with analyzing the LISA data, the current state of affairs, and the
79: necessary next steps to move forward in addressing the imminent
80: challenges.
81: \end{abstract}
82:
83:
84:
85: %==== Text ========================================
86:
87: \section{Introduction}
88:
89: When launched the Laser Interferometer Space Antenna (LISA) will be
90: the first low frequency ($3 \times 10^{-5} - 0.1$~Hz) gravitational
91: wave detector~\cite{Danzmann:2003}. Whenever a new detector is
92: proposed, especially when it is the first of its kind, there are
93: plenty of doubts about its capabilities. LISA is no different.
94: However, even though the LISA technology is ambitious, its the ability
95: to analyze LISA data that many think will be the mission's Achilles
96: heel.
97:
98: The LISA observatory will return a finite set of time series. Encoded
99: within these time series will be the superposition of all
100: gravitational wave signals received during the mission's observational
101: run, co-added to a complicated, time dependent instrumental noise
102: signal. The goal of LISA data analysis is to coax out an individual
103: signal from these correlated time series in order to make scientific
104: inferences about the emitting system or population. The real
105: challenge arises because LISA will observe in excess of $10^{8}$
106: stellar mass galactic binaries, in addition to $0.1 - 10^{5}$ massive
107: black holes binaries (MBHBs) per year, and up to $10^{3}$ extreme mass
108: ratio inspirals (EMRIs) per year. While a daunting task, preliminary
109: investigations suggest that the LISA data analysis challenge can be
110: conquered. This proceedings paper very briefly reviews the
111: difficulties, achievements, and future directions that the LISA
112: science community has and will face.
113:
114: The layout for this paper follows the necessary steps required in
115: building an analysis routine. Section~\ref{sec:expectations} briefly
116: reviews what are our expectations for low frequency sources of
117: gravitational radiation. Section~\ref{sec:response} discusses
118: modeling the detector response and incorporating these models into the
119: analysis routines. Section~\ref{sec:analysis} explores what has been
120: achieved in analyzing simulated data. The last section points to
121: future advancements and necessary steps that must be accomplished.
122:
123:
124:
125: \section{Expectations for astrophysical sources} \label{sec:expectations}
126:
127: Within the LISA band there are three main classes of sources: a large
128: galactic population of compact stellar mass binaries; the inspiral,
129: merger, and ringdown of massive ($10^{4-7}~\Msun$) black hole binaries
130: -- MBHBs; and the capture, inspiral, and eventual merger of compact
131: stellar mass objects into massive black holes -- EMRIs. Each source
132: class presents unique challenges for data analyst, but as equally
133: unique is the scientific content that each signal carries.
134:
135: A number of the greatest challenges associated with LISA data analysis
136: are concerned with the overwhelming number of stellar mass galactic
137: binaries. Initial estimates place the number of individually
138: resolvable LISA binaries in the several thousands, with millions more
139: forming an unresolvable background~\cite{Nelemans:2001, Timpano:2006}.
140: Due to the large orbital periods and low chirp masses associated with
141: galactic binaries, radiation reaction effects will not drive the
142: binaries to coalescence during the mission lifetime. In turn, their
143: signals will be ever present in the detector output.
144:
145: Conversely, MBHBs are semi-continuous sources. They begin as a
146: continuous source during the inspiral phase but eventually fade out
147: during the ringdown. Considering the large masses and small orbital
148: separations, which lead to highly relativistic orbits, the
149: signal-to-noise ratios (SNRs) for MBHB mergers will make them visible
150: from throughout the Universe. The predicted coalescence event rate
151: ranges from 0.1 per year to a confusion background depending on the
152: galaxy evolution and black hole growth models~\cite{Berti:2006}.
153: The wide range in possible event rates hinders the development of data
154: analysis routines. An algorithm that searches for the rare MBHB would
155: be different than one that attempts to isolate an individual signal
156: within a background.
157:
158: The predicted detection event rate for EMRIs is $10^{2-3}$ events per
159: year out to $z \approx 1$ with captures of $10~\Msun$ black holes
160: accounting for the majority of the rate~\cite{Gair:2004}. This
161: estimate was derived using a set of assumptions about analysis
162: capabilities and a particular astrophysical model. Nevertheless, it
163: is evident that a major challenge is detecting EMRIs at large
164: distances where there is a transition from individual detections to a
165: possible EMRI background \cite{Barack:2004}.
166:
167:
168:
169: \section{Forward modeling} \label{sec:response}
170:
171: As with any measurement in astronomy the telescope acts as a filter
172: between the incident radiation and the data analyst. Understanding
173: the filtering process, sometimes referred to as forward modeling, is
174: essential in order to extract the full scientific potential hidden in
175: the data. Forward modeling plays a significant role for spaceborne
176: gravitational wave detectors because it is through the continual
177: orbital motion of the detector that only certain information (e.g. sky
178: location) becomes encoded in the data.
179:
180: The LISA mission consists of three identical spacecraft in separate,
181: slightly eccentric, heliocentric orbits inclined with respect to the
182: ecliptic plane. The orbits are chosen such that the constellation
183: will form an equilateral triangle with a mean spacecraft separation of
184: $5~\times~10^{6}$~km. The constellation center will have an orbital
185: radius of 1~AU and trail the Earth by $20^{\circ}$. The spacecraft
186: motion introduces amplitude, frequency, and phase modulations into the
187: gravitational wave signals. In modeling LISA's response it is
188: critical to incorporate these modulations.
189:
190: An early, complete description for LISA's response was derived by
191: modifying the response function for terrestrial interferometric
192: gravitational wave detectors~\cite{Cutler:1998}. However, since
193: ground-based detectors operate in the small antenna approximation, the
194: extension to LISA is only valid for frequencies below $\sim\!10$~mHz,
195: the point where the gravitational wavelength is on the order of the
196: detector size. A higher fidelity response has also been
197: formulated~\cite{Cornish:2003a}. Based on this, or a similar
198: description, multiple open software packages have been developed that
199: simulate LISA's response to an arbitrary gravitational wave
200: signal~\cite{Cornish:2003d, Vallisneri:2003, Petiteau:2006}.
201:
202:
203:
204: \section{Data analysis} \label{sec:analysis}
205:
206: LISA data analysis is in its early exploratory phase, The typical
207: strategy undertaken is to develop analysis techniques for each source
208: class separately with the intent to combine several independent
209: algorithms to formulate a yet undetermined global analysis procedure.
210: The following subsections briefly review a few of the highlights in
211: data analysis developments.
212:
213:
214: \subsection{Galactic binaries} \label{sub:gb}
215:
216: Strategies for identifying and characterizing individual bright
217: galactic binaries include Doppler demodulation
218: methods~\cite{Hellings:2003, Cornish:2003b}, an iterative subtraction
219: scheme~\cite{Cornish:2003c}, a tomographic search~\cite{Mohanty:2006},
220: Markov Chain Monte Carlo (MCMC) approaches~\cite{Umstatter:2005,
221: Cornish:2005}, and a genetic algorithm~\cite{Crowder:2006}. While
222: each technique shows promise, in this limited space we will highlight
223: the MCMC approach since it appears to be a viable strategy for other
224: source classes as well.
225:
226: The central challenge in detecting bright galactic binaries is that in
227: a small bandwidth ($\Delta f \approx 10^{-6}$~Hz) there may be on the
228: order of 10 bright binaries with the exact number not known \textit{a
229: priori}~\cite{Timpano:2006}. Since each source is described by at
230: least 7 parameters, the associated parameter space for a small snippet
231: of the spectrum can be large. The advantage of a MCMC approach is
232: that it can quickly (in a comparative sense) explore the parameter
233: space and return estimates for the parameter values. In
234: \cite{Cornish:2005} they demonstrated the ability to detect and
235: characterize 10 binary signals when the number of systems was a given.
236: Using a toy model for the signals, \cite{Umstatter:2005} relaxed the
237: \textit{a priori} assumption concerning the number of systems deriving
238: its value from the data along with the source parameters.
239:
240:
241: \subsection{Massive black hole binaries} \label{sub:mbhb}
242:
243: Only recently has work been done on MBHBs. The delay may be
244: attributed to two factors: the large uncertainty in the event rate,
245: which influences the type of algorithm to design; and the large SNRs
246: along with their unique ``chirping'' signals implied it would be a
247: minor exercise to develop a MBHB binary analysis routine.
248:
249: With this sentiment in mind, a few investigations jumped straight to
250: the problem of identifying and characterizing a single MBHB signal in
251: the presence of other signals. Using a Metropolis-Hastings sampling
252: with simulated annealing, \cite{Cornish:2006b} was able to isolate a
253: MBHB signal within a noisy data stream that included a galactic
254: background. Separately \cite{Cornish:2006a} and \cite{Wickham:2006}
255: used a MCMC and a Reversible Jump MCMC (RJMCMC) respectively to
256: characterize a MBHB signal. Moreover, \cite{Wickham:2006}
257: investigated the issue of characterizing a dimmer galactic binary
258: superimposed by a brighter MBHB and found that it should be possible
259: to study weaker signals buried beneath brighter signals.
260:
261: In these isolated examples, the analysis has only focused on the
262: inspiral phase and, therefore, did not use information from the merger
263: or ringdown phases. Also, the MBHB signals where simplified by
264: ignoring component spins and assuming a circular orbit. However,
265: their results are encouraging and suggests future advancements in the
266: algorithms can account for the neglected effects.
267:
268:
269: \subsection{Extreme mass ratio inspirals} \label{sub:emri}
270:
271: Much like with MBHBs, initial analysis algorithms for EMRIs have just
272: recently been formulated. However, in contrast to the MBHB case, the
273: delay for an EMRI analysis algorithm was due to their complicated and
274: intrinsically weaker signals. Unlike bright galactic binaries and
275: MBHBs, the amplitude of a typical EMRI is an order of magnitude below
276: the instrumental noise. Only by tracking multiple wave cycles is
277: enough SNR accumulated to allow a confident detection.
278:
279: Using scaling arguments, it is possible to show that a standard
280: template matching routine would require $\sim\!10^{40}$ templates for
281: an EMRI detection, making it computational prohibitive. However, this
282: assumes a fully coherent search. If instead a hierarchal search is
283: done by piecing together coherent searches over short observational
284: periods ($\sim\!3$ weeks) then it may be possible to use a template
285: based algorithm to search for EMRI signals in the data
286: \cite{Gair:2004}.
287:
288: An alternative tactic is to use a time-frequency method in which short
289: segments of the data time series is Fourier transformed and stacked
290: together to form a spectrogram~\cite{Wen:2005, Gair:2005}. By
291: searching for excess power in the spectrogram it is possible to detect
292: an EMRI out to distance of $\sim\!2.25$~Gpc, which is about half the
293: capabilities of the semi-coherent method described previously.
294:
295: The above analyses are only able to return limited information about
296: the sources themselves. In a recent conference proceedings paper
297: \cite{Stroeer:2006} demonstrated the use of a RJMCMC to characterize a
298: simplified EMRI signal. While this analysis only considered leading
299: order effects to an already approximate EMRI signal, this case study
300: implies that future improvements may lead to a full EMRI
301: characterization algorithm.
302:
303:
304:
305: \section{Concluding remarks} \label{sec:conclusion}
306:
307: While the LISA data analysis challenge seems difficult, the early
308: returns indicate that it will be met. The (RJ)MCMC appears to be a
309: viable approach for addressing many of the analysis challenges.
310: However, other techniques may also play a significant role, especially
311: in the early detection stage where the (RJ)MCMC appears to be limited
312: in its capabilities.
313:
314: The next challenges that need to be addressed include processing full
315: bandwidth (simulated) data without human intervention, cross comparing
316: existing algorithms, and developing robust routines capable of
317: analyzing data that contain multiple classes of gravitational wave
318: sources. These next steps should be attainable using existing
319: methods. However, with LISA data analysis still in its infancy, it is
320: appropriate and necessary to explore other alternatives in an attempt
321: to find methods that will maximize our scientific return on the LISA
322: data.
323:
324:
325:
326: %==== Acknowledgments =============================
327:
328: \ack
329: This work was supported by the Center for Gravitational Wave Physics
330: which is supported by the National Science Foundation under
331: Cooperative Agreement PHY 01-14375.
332:
333:
334:
335: %==== Appendices ==================================
336:
337:
338:
339: %==== Bibliography ================================
340:
341: \section*{References}
342:
343: \bibliographystyle{unsrt}
344:
345: \begin{thebibliography}{10}
346:
347: \bibitem{Danzmann:2003}
348: Danzmann K and R{\"u}diger A
349: \newblock 2003 {\em Class. Quantum Grav.} \textbf{20} S1--9
350:
351: \bibitem{Nelemans:2001}
352: Nelemans G, Yungelson~L R and {Portegies Zwart}~S F
353: \newblock 2001 {\em A\&A} \textbf{375} 890--8
354:
355: \bibitem{Timpano:2006}
356: Timpano~S E, Rubbo~L J and Cornish~N J
357: \newblock 2006 {\em Phys. Rev.}~D \textbf{73} 122001
358:
359: \bibitem{Berti:2006}
360: Berti E
361: \newblock 2006 {\em Class. Quantum Grav.} \textbf{23} S785--97
362:
363: \bibitem{Gair:2004}
364: Gair~J R, Barack L, Creighton T, Cutler C, Larson~S L, Phinney~E S and
365: Vallisneri M
366: \newblock 2004 {\em Class. Quantum Grav.} \textbf{21} S1595--606
367:
368: \bibitem{Barack:2004}
369: Barack L and Cutler C
370: \newblock 2004 {\em Phys. Rev.}~D \textbf{70} 122002
371:
372: \bibitem{Cutler:1998}
373: Cutler C
374: \newblock 1998 {\em Phys. Rev.}~D \textbf{57} 7089--102
375:
376: \bibitem{Cornish:2003a}
377: Cornish~N J and Rubbo~L J
378: \newblock 2003 {\em Phys. Rev.}~D \textbf{67} 022001
379:
380: \bibitem{Cornish:2003d}
381: Cornish~N J and Rubbo~L J
382: \newblock 2003 {\textsf{The LISA Simulator}}
383: \newblock \texttt{http://www.physics.montana.edu/LISA/}
384:
385: \bibitem{Vallisneri:2003}
386: Vallisneri M
387: \newblock 2003 {\textsf{Synthetic LISA}}
388: \newblock \texttt{http://www.vallis.org/syntheticlisa/}
389:
390: \bibitem{Petiteau:2006}
391: Petiteau A, Auger G, Halloin H, Jeannin O, Pireaux S, Plagnol E, Regimbau T
392: and Vinet J-Y
393: \newblock 2006 {\textsf{LISACode}}
394: \newblock \texttt{http://www.apc.univ-paris7.fr/LISA-France/analyse.phtml}
395:
396: \bibitem{Hellings:2003}
397: Hellings~R W
398: \newblock 2003 {\em Class. and Quantum Grav.} \textbf{20} 1019--29
399:
400: \bibitem{Cornish:2003b}
401: Cornish~N J and Larson~S L
402: \newblock 2003 {\em Class. and Quantum Grav.} \textbf{20} S163--70
403:
404: \bibitem{Cornish:2003c}
405: Cornish~N J and Larson~S L
406: \newblock 2003 {\em Phys. Rev.}~D \textbf{67} 103001
407:
408: \bibitem{Mohanty:2006}
409: Mohanty~S D and Nayak~R K
410: \newblock 2006 {\em Phys. Rev.}~D \textbf{73} 083006
411:
412: \bibitem{Umstatter:2005}
413: Umst{\"a}tter R, Christensen N, Hendry M, Meyer R, Simha V, Veitch J, Vigeland
414: S and Woan G
415: \newblock 2005 {\em Phys. Rev.}~D \textbf{72} 022001
416:
417: \bibitem{Cornish:2005}
418: Cornish~N J and Crowder J
419: \newblock 2005 {\em Phys. Rev.}~D \textbf{72} 043005
420:
421: \bibitem{Crowder:2006}
422: Crowder J, Cornish~N J and Reddinger~J L
423: \newblock 2006 {\em Phys. Rev.}~D \textbf{73} 063011
424:
425: \bibitem{Cornish:2006b}
426: Cornish~N J and Porter~E K
427: \newblock 2006 gr-qc/0605135
428:
429: \bibitem{Cornish:2006a}
430: Cornish~N J and Porter~E K
431: \newblock 2006 {\em Class. Quantum Grav.} \textbf{23} S761--7
432:
433: \bibitem{Wickham:2006}
434: Wickham E~D L, Stroeer A and Vecchio A
435: \newblock 2006 {\em Class. Quantum Grav.} \textbf{23} S819--27
436:
437: \bibitem{Wen:2005}
438: Wen L and Gair~J R
439: \newblock 2005 {\em Class. and Quantum Grav.} \textbf{22} S445--51
440:
441: \bibitem{Gair:2005}
442: Gair J and Wen L
443: \newblock 2005 {\em Class. and Quantum Grav.} \textbf{22} S1359--71
444:
445: \bibitem{Stroeer:2006}
446: Stroeer A, Gair J and Vecchio A
447: \newblock 2006 gr-qc/0609010
448:
449: \end{thebibliography}
450:
451: \end{document}