Presentation is loading. Please wait.

Presentation is loading. Please wait.

Optimizing Foam, A Monte Carlo Event Generator

Similar presentations


Presentation on theme: "Optimizing Foam, A Monte Carlo Event Generator"— Presentation transcript:

1 Optimizing Foam, A Monte Carlo Event Generator
Thomas Sandell February 1, 2016 University of Michigan

2 Overreaching Project Goal
Find ttH collisions easily from the ATLAS experiment and quickly filter out “noisy” collisions, specifically ttbar collisions

3 The Higgs Boson

4 Problems with Finding Higgs
Background decays occur much more often than Higgs decays These processes look very similar, and it is difficult to filter out background decays, making finding the Higgs very difficult Need to find an effective way to filter out the noise and find where a Higgs is actually formed

5 A Solution: The Foam Method
Foam is a probability-density estimation method Essentially, split an n-dimensional space into many many hyperrectangles and measure the probability density in each space Treat this as a look-up table to compare events and check if they satisfy the traits of a ttbar event Each “dimension” is actually an input variable

6 Foam Modeling a 5-D Distribution

7 Problems with Foam Most physics problems are not 5-D, and require many more bins Especially if there are strongly peaked functions, many bins are required to estimate the functions accurately We can use coordinate transformations to change most strongly peaked functions into more reasonable functions, but running foam on a 16-D function still takes multiple days

8 Why this isn’t a big deal
Luckily, Foam does not have to run each time we compare an event to a sample distribution It is only ran once on sample events or a sample function, and then the output is compared with the distribution output However, for development purposes it would still be great if foam ran much faster than it currently does

9 A solution: Running Multiple Processes
A huge inefficiency with foam is that because it takes advantage of self-adapting binning, you cannot split up multiple cells at the same time However, if we spend a lot of computing power (2-3 minutes of processing time) to find the absolute optimal first few cuts, we can split each “subfoam” into separate processes, making the code run in an eighth of the time

10 My Initial Project! First, spend a lot of computing time finding the optimal split point(s) of the foam Next, create a subfoam for each of the split hyperrectangles of the foam Finally, combine each of these subfoams into one “master foam” which will be returned by the program

11 Without Multiple Processes

12 First Half

13 Second Half

14 Combined

15 My Projects at CERN In the long run, I will be working on optimizing the Foam code in general Once I implement multithreading, I will continue to work on other Foam issues For instance, I will be combining the functionality of a foam that works with inputting sample events and this foam, which works on inputted functions

16 My Projects in Europe

17 Citations Foam: A General-Purpose Cellular Monte Carlo Event Generator. S. Jadach, 2002 PDE-Foam – a probability-density estimation method using self-adapting phase-space binning. Dannheim et al., 2008 Search for the Standard Model Higgs boson produced in association with top quarks and decaying into a bb pair in pp collisions at sqrt(s) = 13 TeV with the ATLAS detector. The Atlas Collaboration, 2016


Download ppt "Optimizing Foam, A Monte Carlo Event Generator"

Similar presentations


Ads by Google