=== Andrea Lommen - pulsar timing overview ===

- NANO-Grav: North American Nano-Hertz Observatory of Gravitational Waves
	- Arecibo, Green Bank... later ATA
- Falsification of 3C 66B 10^10 Msun BH binary
- Monte Carlo simulation to yield upper limit for stochastic background
	- 10000 GWs with amplitude randomly generated as member of a model spectrum
	- Spectrum by interpolation + double differencing + FFT
- Ask not: what is the minimum amplitude required to detect background 95% of time
- Ask: what is the value of A that leads to the conclusion that background is absent 95% of the time
- Limit places an interesting constraint on SMBH population
	- Equal to Wyithe and Loeb (2003), larger than Jaffe Backer (2005) and Enoki (2004)
- Pulsar response to standard and longitudinal polarization modes (best way to measure polarization)
- Directional burst detection also possible with pulsar correlations

=== Vuk Mandic - stochastic background sources ===

=== Tania Regimbau - astrophysical sources of stochastic gravitational-wave backgrounds ===

- astrophysical vs. cosmological backgrounds
- spectral properties; duty cycle = event duration / time interval
- D << 1: burst data-analysis, optimal filtering
- D ~  1: maximum likelihood statistic, probability event horizon
- D >> 1 (Gaussian): cross-correlation statistic

--> Inspiral session <--

=== Chris Messenger - stochastic template placement ===

- matched filtering requires prior waveform knowledge
- CPU limited parameter-space exploration
- need efficient template placement

- metric: construct a measure of distance in parameter space equivalent to signal-template overlap
	- use eigensystem to define local unit and directions
	- in globally flat spaces, get mathematical covering problem
	- simplest n-dimensional lattice is cubic; "best known" is A_n for n < 24 (hexagonal basic cell in 2D)
	- lattice normalized thickness theta = R^n = sqrt(|g_ij|); theta is proportional to number of templates

- random template placement:
	- place single template; what is probability that it's outside a distance circle? 1 - V_n R^n / V_s
	- derive probability of achieving mismatch mu: P' = 1 - (1 - V_n R^n)^N / V_s^N
	- the covering is never total for any number of templates
	- random covering has better theta than lattices at 10 dimensions if a little covering can be sacrificed: lattices go to a lot of effort to cover last few % of space
	- can we use "lazy" lattices? Yes, but random covering still beats cubic

- for nonflat spaces, need only to control the determinant of the metric:
	- generate template density function, use it to fill parameter space
	- but placing random bank is far simpler than lattice
	- all searches are statistical anyway

- Stas: explore ambiguity function to decide random bank placement

=== Drew Keppel - improvements to LSC binary inspiral search ===

- so far:
	- investigation of SPA templates up to 35 Msun
	- using nonspinning for somewhat spinning
	- using circular for eccentric

- for S5:
	- 2PN SPA for low-mass region, component masses 1-34 Msun, maximum M = 35 Msun, cutoff at ISCO
	- 2PN EOB time-domain for high-mass, component masses 1-99 Msun, total mass 25-100 Msun, cutoff at light ring
	- background triggers are very different in BNS and BBH mass regions: higher effective SNR for the latter
	- estimate background using nonphysical time-slide coincidences: excess of H1H2L1 background by sliding H1 and H2 separately
	- new clustering and coincidence based on template-placement metric

=== Lucia Santamaria - Including information from numrel into inspiral searches ===

- numrel: breakthrough in 2005; stable, accurate codes producing waveforms; NR can simulate most promising sources
- GW DA: LIGO has completed five science runs, and analyzed lots of data

- extraction of h(t) from numerical evolution: psi_4 or Zerilli function
- decompose h+ - ihx into spin-weighted spherical harmonics of weight -2; these must be provided by numerical groups as functions of time and incorporated into LAL searches

- in practice: use metadata file to find files, rescale them for total mass, compute h(t) = F+ h+(t) + Fx hx(t), run injection
- check sanity by injecting T1 3.5PN PN into white noise
- then try injecting and recovering hybrid Ajith waveforms

=== Ruslan Vaulin - Estimating statistical significance of the candidate events in LSC compact binary coalescence search ===

=== John Veitch - An Evidence Based Approach to Inspiral Followups ===

- compute probability that a candidate is really a gravitational wave
- probability ratio (odds ratio) for signal vs. null hypothesis; does not need probability of the data
- use Gaussian model of noise, integrate over signal parameters
- code runs in 2hrs for 100s data
- try with Gaussian noise only, model favors noise only
- inject signal, Bayes factor increases with SNR
- tried polluting Gaussian noise with time-domain Poisson components, noise model still significantly favored for noise only, but sensitivity to signal is reduced
- how does technique behave with unmodeled glitches in noise? Try with nonphysical ringdowns... show robust behavior
- ideal follow-up---conceptually straightforward, can be used for other signals than inspirals, too

posters:	Tagoshi on coherent-search strategies
			McKechan on higher harmonics -- check out
			van der Sluys on Bayesian inference for inspirals -- check out his errors
			Bose on coherent searches