Profile Fitting with Genetic Algorithms

libyantawdryΤεχνίτη Νοημοσύνη και Ρομποτική

23 Οκτ 2013 (πριν από 4 χρόνια και 20 μέρες)

103 εμφανίσεις

Profile Fitting with Genetic
Algorithms

Barry Fitzgerald

Plan!


Genetic Algorithms
-

an explanation



Applications to Astrophysics


P cygni profiles



Results of fitting


Backround


In the early 1960’s computer scientists
tried to find new ways of tackling
optimization problems.


Many looked to nature for ways of
solving problems


Obvious example was Darwin’s
“survival of the fittest”

Genetic Algorithms
-

Theory


Create a “population” of possible solutions.


Allow the solutions breed (mix) and mutate.


Apply a selection criteria which gives better
solutions greater chance of surviving.


Repeat until adequate solution is reached

GA’s a general plan

GA’s
-

Advantages


Completely general problem solving
method


Gives multiple solutions


Needs little information about the
problem

GA’s Disadvantages


Hard to tailor to specific problems


Needs to maintain a population of
solutions


can be slow


Non
-
deterministic


run it twice and get
two different answers

Application to Astrophysics


Common problem for fitting spectra is
trying to fit theory to reality.


This can be transformed into an
optmization problem by plotting the
theoretical spectrum against the
observed one and then calculating a
numerical value of “goodness of fit”.

Least square fitting

P
-
cygni profiles


Many stars emit not just radiation but
also matter. Often in the form of stellar
winds.


A sensitive indicator of stellar winds are
P Cygni profiles.


These are resonance lines usually seen
in the UV range with a distinctive
shape.

Typical P Cygni profile


P Cygni emission
line profile of triply
ionized carbon at
1548.2 Å in the
central star of the
cat’s eye planetary
nebula, NGC 6543.


Theory

Expanding envelope
surrounding stellar surface.
Observer views system from
bottom of figure. Color ranges
from blue for largest blueshift
to white for zero Doppler shift
to red for largest redshift.
(constantly accelerating radial
wind case)


Explanation

Fitting P cygni profiles


To fit these profiles a program that
reads in 16 parameters about a stellar
wind and outputs a theoretical profile
was used.


Using a GA this process was run in
reverse


a profile was read in to the
computer then the set of parameters
that best fit that profile was searched
for.


Sample run theoretical data

Real Observed Data

Conclusions


Genetic algorithms provide a effective
means of fitting profiles

In theory!


However as they are a blind method
unphysical solutions are often found.


Further Work


Apply program to fit doublets


may
remove some amibugities.


Try to solve the program in stages first
fitting one part of spectrum then next.


Add restrictions to stop non physical
solutions.