Goal: Code validation (known parameters) and single source analysis
Release date: June 2006
Due date: December 2006 [known source analysis reported in September 2006]
Note: I would propose to distribute two data sets (with just slightly different source parameters) for each sub-challenge: (i) one with known parameters mainly for code validation (this is a “known source analysis, and we should have report around Septermber) and (ii) one with unknown parameters

• a) 1 year of Gaussian stationary instrument noise + 1 exactly monochromatic WD with SNR ~10 → 20 over 1 year (for single IFO) at frequency 1 mHz
  
• b) 1 year of Gaussian stationary instrument noise + 1 exactly monochromatic WD with SNR ~10 → 20 over 1 year (for single IFO) at frequency a 3 mHz
  
• c) 1 year of Gaussian stationary instrument noise + 1 exactly monochromatic WD with SNR ~10 → 20 over 1 year (for single IFO) at frequency a 10 mHz
  

• 1 year of Gaussian stationary instrument noise with 5 true verification binaries (parameters chosen according to our current best guess) + 15 mock verification binaries in the frequency range 0.5 mHz → 12 mHz. All the signals will be monochromatic (no fdot). Since these are verification binaries, the sky locations and orbital frequencies will be given. The orientation parameters and distances will be considered unknown.

• 1 year of Gaussian stationary instrument noise with 20 resolvable binaries (parameters chosen from the BLT population synthesis code) with frequencies spread over the LISA band (0.1 mHz - 10 mHz). All the signals will be monochromatic (no fdot).

• 1 year of Gaussian stationary instrument noise with ~50 WD binaries chosen around 3 mHz in a band \pm 1.5 micro Hz with SNR > 5. Sources are drawn from the BLT population synthesis code. All the signals will be monochromatic (no fdot). The number of sources will not be given.

• 1 year of Gaussian stationary instrument noise + 1 Schwarzschild SMBH binary with time to coalescence U[178-20, 178+20] days and masses m1 U[1, 5]x10^6 solar masses and m2 = m1/x, where x is U[1,4] and U stands for uniform. The signals have SNR ~ 500 in one IFO. The waveform model for the inspiral is restricted to 2PN approximation with no spin-orbit nor spin-spin modultations.

• 1 year of Gaussian stationary instrument noise + 1 Schwarzschild SMBH binary with time to coalescence U[400-20, 400+20] days and masses m1 U[1, 5]x10^6 solar masses and m2 = m1/x, where x is U[1,4] and U stands for uniform. The signals have SNR ~ 20 → 100 in one IFO. The waveform model for the inspiral is restricted to 2PN approximation with no spin-orbit nor spin-spin modultations.

Since this challenge has different completion date, I’d suggest to have more data streams. We have agreed 5 data streams with SNR more or less equally spaced between ~30 and 100. Data set duration is 2 years. [Results are actually expected in June 2007, as part of challenge 2]

• 2 years of Gaussian stationary noise (instrumental only) + 1 EMRI and combined SNR between 30 and 100 (D~1Gpc), depending on the data set. Parameters are distributed as follows: mu U[5,15] solar mass, M U[0.5,2]x10^6 solar mass, spin U[0.2, 0.8] M^2, duration lies in [1,2] years and [2,4] years with slight bias toward right end, eccentricity at the plunge is on [0.1, 0.25] with a small pick at 0.15. Use cadence 15 sec (2^22 an 2^23 points)

Comment on "2 and 4" years, some participants might not have PC with large RAM, but probably would want to read the whole data at 
once, that is the main reason behind 2 years of data. (SB)

Comment on physical params: I think the early challenges should concentrate on the stronger sources--BHs of about 10 Msun--
which we also expect to detect more of. LIkewise, might as well make the MBH about 10^6 (CC).

Comment on e=0.0: the inspirals get more eccentric at the end, not less, so I don't think e=0 is very realistic (CC).

Comment on challenge set-up: I'd favor breaking the EMRI challenge into 2: very bright sources (like the SNR =100 above) and 
the presumably more typical ones with SNR = 30 (in single IFO).  I think we should issue 2 different data sets: one with
a bright SNR=100 source, and one with SNR =30, since the methods for the 2 problems might be quite different.  The weaker
source should come will a restricted parameter range (e.g., we tell people a small patch on the sky to look in, and perhaps
a narrow mass range, to make up for our poor computational resources relative to 2015 (CC).

Comment on number of sources: As we discussed in the telecon, for each SNR (100 and 30) I think we should produce 2 data
sets: one w/ a single source and one with ~10 overlapping sources, since I can imagine some methods that might succeed for
the former but have real troubles w/ the latter (CC).

Comment on e=0.20: of course we have to say "when" it's 0.2; I'd suggest e=0.2 at very end of inspiral, or maybe
e=0.3  when the m=2 harmonic sweeps thru 3mHz (which amounts to pretty much the same thing, for a 10/10^6 pair) (CC).

Comment on e=0.20: I (AV) propose that we use: m1 = 10^6 and m2 = 10; e=0.3  when the m=2 harmonic sweeps thru 3mHz;
random parameters for the geometry and distance is adjusted in order to have the desired SNR

/* —————————– */

Goal: Entirely dedicated to “global data analysis”
Release date: December 2006
Due date: June 2007

A ~2-yr long (to be precise: 2^22 data points with cadence 15 seconds) data set containing:

• Radiation from galactic binaries generated using a population synthesis code (~ 30 Million sources) - radiation from binaries is (as in Challenge 1) exactly monochromatic (in the source reference frame)
• Verification binaries (same sky locations and frequencies as Challenge 1)
• Gaussian and stationary instrumental noise
• Assume (as done in challenge 1) perfect TDI cancellation of laser frequency noise

Note: This data set provides also the first challange for a non-stationary and anisotropic stochastic background

A ~2-yr long (to be precise: 2^22 data points with cadence 15 seconds) long data set containing:

• Radiation from galactic binaries generated using a population synthesis code (~ 30 Million sources) - radiation from binaries is (as in challenge 1) exactly monochromatic (in the source reference frame)
• Verification binaries – exactly the same frequency and sky location as in challenge 2.1
• 4-6 signals from MBHBs.
  • Waveform: exactly as in Challenge 1
    • For the future (NOT FOR CHALLENGE 2.2): Replace tapering with a mock-up of the signal coming from the plunge (that resembles in the key features the recent results obtained by the numerical relativity community) followed by the ring-down (only the leading order l=m=2”: damped sinusoid)
  • Signal parameters Same mass ranges as used in Challenge 1 [note that the SNR quoted here is only for the inspiral and assume coherent matched-filtering using 1 TDI observable]:
    • The BBH signals will include
      1. A loud (SNR ~ 2000) signal (with in-spiral “in band”, that is from f = 0.1 mHz, lasting about 1 week) and coalescing 3 months +/- 1 month after the beginning of the observations
      2. A moderate-to-low SNR (~ 20) signal with coalescence between 25 and 26 months from the beginning of the observations
      • The BBH signals may also include
      1. A loud (SNR ~ 1000) signal coalescing between 6 and 24 months from beginning of observations
      2. A loud (SNR ~ 200) signal coalescing between 6 and 24 months from beginning of observations
      3. A loud (SNR ~ 100) signal with coalescence taking place between 18 and 21 months from the start of the observations
      4. A low SNR (~ 10) signal with coalescence taking place between 27 and 28 months from the beginning of the observations
• 5 EMRIs of the type considered in challenge 1, with SNR uniformly distributed between 30 and 100:
  1. mu – U[9.5,10.5] solar mass
  2. M – 1 with U[0.95,1.05]x10^7 solar masses, 2 with U[4.95,5.05]x10^6 solar masses, 2 with U[0.95,1.05]x10^6 solar masses
  3. spin – U[0.5, 0.7] M^2
  4. duration lies in [1.5,2] years
  5. eccentricity at the plunge is on [0.1, 0.25] with a small peak at 0.15
• Gaussian and stationary instrumental noise
• Assume (as done in challenge 1) perfect TDI cancellation of laser frequency noise

Goal: Test analysis adding more complex waveforms, new classes of signals and asymmetric instrument noise
Release date: December 2007
Due date: December 2008

The plan is to have a mixture of easy and difficult data sets that introduce new facets to the analysis. One new ingredient in some of the data sets will be “asymmetric” instrument noise, in which the strain spectral density of the shot and acceleration noise differs in each component.

• Isolated chirping galactic binaries with circular orbits. May have increasing or decreasing orbital frequency. Injected signals will have astrophysically relevant frequency evolution modeled up to the second frequency derivative. A description of the astrophysical model will be provided to aid in the construction of priors. The instrument noise model will be given.

• a) Two year data stream with Gaussian instrument noise and a single chirping binary within 30 micro Hz of 5 mHz.
• b) Two year data stream with Gaussian instrument noise and a single chirping binary within 30 micro Hz of 10 mHz.
• c) Two year data stream with Gaussian instrument noise and a single chirping binary within 30 micro Hz of 15 mHz.

• A two year data stream with symmetric instrument noise and the signals from ~30 million galactic binaries. Their frequency evolution will be described by the same astrophysical model used in 3.1.1. The instrument noise model will be given.

• A two year data stream with asymmetric instrument noise and the signals from ~30 million galactic binaries. Their frequency evolution will be described by the same astrophysical model used in 3.1.1. The instrument noise will be stationary and gaussian, but the noise levels in each component (photo diode, disturbance reduction system) will differ by as much as a factor of two, and will not be disclosed.

• A one year data stream containing symmetric instrument noise and the signal from a single spinning black hole binary that colesces during the observation time. The orbits will be circular so the signal is described by 15 parameters. The signals will be modeled to 3PN order and will have multiple harmonics. The merger and ringdown portion of the waveform will not be included.

• A one year data stream containing symmetric instrument noise and the signal from a single spinning black hole binary that colesces roughly a month after the observation period. The orbits will be circular so the signal is described by 15 parameters. The signals will be modeled to 3PN order and will have multiple harmonics.

• A two year data stream with symmetric instrument noise and an unknown number of EMRIs (similar to those used in past challenges). Very little information will be given about the choice of parameters.

• One year of symmetric, stationary and gaussian instrument noise and a number of bursts from cosmic cusps, say one every couple of weeks - of different SNR (between 5 and 1000).

• One year of symmetric, stationary and gaussian instrument noise and a number of “unkown” bursts (with wierd, randomly generated waveform morphologies), say one every couple of weeks - of different SNR (between 10 and 1000).

• One year of data with non-stationary, glitchy instrument noise and a number of “unkown” bursts (with wierd, randomly generated waveform morphologies), say one every couple of weeks - of different SNR (between 10 and 1000).

• One year of Gaussian and stationary instrumental noise + an isotropic stochastic background of spectrum S_h(f) \propto f^{-7/3} and amplitude equal to the instrumental noise at 2 mHz. No other sources present in the data stream.

Goal: Test analysis with realistic noise behaviour (non-stationarities, non-Gaussianity, data dropouts..)
Release date: December 2007
Due date: June 2008

/* —————————– */

/* —————————– */

As I have mentioned in the previous teleconference, I think we are missing a very important branch of challenges: we need to include more sophisticated (mean realistic) models in the latest challenges. It would be good to have another branch of challenges which aims at the detection of more complicated signals. For instance for galactic binaries it would mean including eccentricity, interacting binaries, etc,; for SMBH it would mean going to higher PPN orders, including spins, eccentricity and higher harmonics; for EMRIs it is including numerical kludge or even waveform based on the Teukolsky formalism.

* It was mentioned several times that we should not oversimplify the problem. Here I suggest to include study of more complicated signals starting with a “single source” challenge and introducing the confusion problem much later. Some numbers given for simplified models might not work that well for more complicated signals. One good example is parameter estimation for SMBH. We know that including the spins and higher harmonics can improve parameter estimation, and, since we want to study problem of issuing “early warning “, I think it is important to cover this in our challenges as well.

* My second point is that we can learn a lot from the study when the injected signal and templates do not match exactly. We use this strategy successfully for ground-based ifos. I am sure we can use the simplified models for detection in many cases, we need to learn whether we can estimate parameters as well (or rather “how well”). One of the problem we will face is removal of strong signals and estimation of residuals. Another reason is that we use some assumption when we model the GW signal, and the “best-of-our-knowledge” model might not match the true signal perfectly.

* My third point was already mentioned by Curt, the use of simple model and nice (hopefully) result of simplified analysis might make people too comfortable. This false comfort can hide the amount of work which has to be done in reality! Use of more complicated models might require NEW data analysis methods. This challenges should be introduced early enough, so people can start working on it. I do not suggest here to start running before we learned to crawl, I rather suggest to give people more time to think and start working in the directions outlined above.

Stas Babak, 19th April 2006