Www.great08challenge.info 1/19 PASCAL Challenge.

www.great08challenge.info www.cosmocoffee.info 1/19

PASCAL Challenge

www.great08challenge.info www.cosmocoffee.info 2/19

www.great08challenge.info www.cosmocoffee.info 3/19

Contents of our Universe

70% Dark Energy

25% Dark Matter

5% Baryonic Matter

www.great08challenge.info www.cosmocoffee.info 4/19

Gravitational Lensing

Light from a distant galaxy is bent around a mass

www.great08challenge.info www.cosmocoffee.info 5/19

Gravitational Lensing

Galaxies seen through dark matter distribution analogous to

Streetlamps seen through your bathroom window

www.great08challenge.info www.cosmocoffee.info 6/19

www.great08challenge.info www.cosmocoffee.info 7/19

Cosmic Lensing

Real data:gi~0.03

gi~0.2

www.great08challenge.info www.cosmocoffee.info 8/19

Atmosphere and Telescope

Convolution with kernel

Real data: Kernel size ~ Galaxy size

www.great08challenge.info www.cosmocoffee.info 9/19

Pixelisation

Sum light in each square

Real data: Pixel size ~ Kernel size /2

www.great08challenge.info www.cosmocoffee.info 10/19

Noise

Mostly Poisson. Some Gaussian and bad pixels.Uncertainty on total light ~ 5 per cent

www.great08challenge.info www.cosmocoffee.info 11/19

www.great08challenge.info www.cosmocoffee.info 12/19

Typical starUsed for finding Convolution kernel

Typical galaxyused for cosmicshear analysis

www.great08challenge.info www.cosmocoffee.info 13/19

www.great08challenge.info www.cosmocoffee.info 14/19

GREAT08 Data

One galaxy per imageKernel is givenOne shear per setNoise is Poisson

150 000 galaxies

27 000 000 galaxies

3 000 000 galaxies

150 000 galaxies

All divided into sets containing 10 000 galaxies each

www.great08challenge.info www.cosmocoffee.info 15/19

GREAT08 Data

www.great08challenge.info www.cosmocoffee.info 16/19

GREAT08 Results

You submit g1, g2 for each set of images

www.great08challenge.info www.cosmocoffee.info 17/19

GREAT09 and Beyond• Kernel is not given (just kernel images)• Galaxies with more complicated shapes• Kernel is more realistic• Many objects in each image• Correlated noise• Poisson + Gaussian noise + bad pixels• Shear is different for each galaxy• Kernel is not the same for every galaxy• Cosmic rays, satellite tracks, saturated stars• Multiple exposures• Pixels not exactly square or on a grid

www.great08challenge.info www.cosmocoffee.info 18/19

GREAT08 Timeline• Feb 2008 GREAT08 Handbook public• Jun 2008 Internal release of simulations• Aug 2008 First simulations public• 27 Oct 2008 Launch of public challenge

– Leaderboard starts containing internal results

• 5 Jan 2009 – mid-term workshop at UCL• 30 Apr 2009 Competition deadline• ~June 2008 Workshop; Release final report

– Input shears public

www.great08challenge.info

www.great08challenge.info www.cosmocoffee.info 19/19

GREAT08 Summary

• 30 million images• 1 galaxy per image• De-noise, de-convolve, average → shear

• gi ~ 0.03 to accuracy 0.0003 → Q~1000 → Win!