simulating spatial partial differential equations with cellular automata by brian strader adviser:...

57
Simulating Spatial Partial Differential Equations with Cellular Automata By Brian Strader Adviser: Dr. Keith Schubert Committee: Dr. George Georgiou Dr. Ernesto Gomez

Upload: stephen-barnett

Post on 29-Dec-2015

215 views

Category:

Documents


0 download

TRANSCRIPT

Simulating Spatial Partial Differential Equations with

Cellular Automata

By Brian Strader

Adviser: Dr. Keith Schubert

Committee: Dr. George Georgiou

Dr. Ernesto Gomez

Introduction & Background

Topics Covered

•Partial Differential Equation, Cellular Automata (CA), & Biology

•Converting Differential Equations to CA

•CA Theoretical Constraints

•Convergence Maps & Guidelines

Introduction & Background

Cellular Automata (CA)

•CA Model uses simple rules about changes with time.

•Rules are localized and involve the values of cell neighbors.

•The set of rules are applied to the cells with the matrix after each time period.

Introduction & Background

Conway’s Game of LifeSurvival Rule: 2-3 Neighbors

Death by Overpopulation: 4+ Neighbors

Introduction & Background

Conway’s Game of LifeDeath by Isolation: 1 or Less Neighbors

Birth: 3 Neighbors

Introduction & Background

Conway’s Game of Life

t = 0

Introduction & Background

Conway’s Game of Life

t = 1

Introduction & Background

Conway’s Game of Life

t = 2

Introduction & Background

Conway’s Game of Life

t = 3

Introduction & Background

Celluar Automata Simulation

Introduction & Background

Celluar Automata Simulation

Introduction & Background

Spatial Partial Diff. Equations

•Changes with respect to time.

•Part of the equation depends on changes in space.

Introduction & Background

Vegetation Patterns

Introduction & Background

CA Advantages

•Simple Rules - easy to understand

•Discretized

•Local Problem View

•Highly Parallelizable

Converting Differential Equations to CA

Diff. Equation Form

Conditions:for n(u) = up where p <= 1for o(u) = up where p <= 1

Converting Differential Equations to CA

Diff. Equation Form

Conditions:for n(u) = up where p <= 1for o(u) = up where p <= 1

Converting Differential Equations to CA

Diff. Equation Form

Conditions:for n(u) = up where p <= 1for o(u) = up where p <= 1

Converting Differential Equations to CA

Discretization Techniques

Converting Differential Equations to CA

Size of hx

Large hx

Small hx

Converting Differential Equations to CA

Euler’s Methods

Forward Euler’s Method:

Converting Differential Equations to CA

Size of ht

Converting Differential Equations to CA

Euler’s Methods

Backward Euler’s Method:

Converting Differential Equations to CA

Euler’s Methods

Forward Euler’s Method:

Backward Euler’s Method:

Converting Differential Equations to CA

Euler’s Methods

Forward Euler’s Method:

1 2 3 4 5i=1jj-1 j+1

3.2 5.7 7.3 9.2 -7.5i=2jj-1 j+1

CA Theoretical Constraints

General Linear Form

CA Theoretical Constraints

Convergence and Divergence

CA Theoretical Constraints

Z-Transform

•Time Domain Frequency Domain

•Discrete Form of Laplace Transform and related to the Fourier Transform

•Transformation makes life easier

•zeros when f(z)=0 poles when g(z)=0

CA Theoretical Constraints

Z-Transform

CA Theoretical Constraints

Z-Transform

1. Perform z-transform

2. Solve for Uj

3. Find poles and zeros for Uj=f(z)/g(z)

4. Set poles and zeros values of z < 1 to converge

CA Theoretical Constraints

Forward Euler’s Constraints

Forward Euler’s Linear Form:

Zeros Constraint:

CA Theoretical Constraints

Forward Euler’s Constraints

Forward Euler’s Linear Form:

Poles Constraint:

CA Theoretical Constraints

Backward Euler’s Constraints

Backward Euler’s Linear Form:

Zeros Constraint:

CA Theoretical Constraints

Backward Euler’s Constraints

Backward Euler’s Linear Form:

Poles Constraint:

Convergence Maps & Guidelines

CA Sim 1 2 3 4 5i=1

jj-1 j+1

1.1 1.9 2.8 2.6 5.4i=2jj-1 j+1

0.1 0.35 0.27 0.4 0.57i=njj-1 j+1

0.11 0.34 0.27 0.4 0.56i=n-1jj-1 j+1

...

< 10-10

Convergence Maps & Guidelines

CA Sim 1 2 3 4 5i=1

jj-1 j+1

1.1 1.9 2.8 2.6 5.4i=2jj-1 j+1

541 -5623 -897 456 878i=njj-1 j+1

1.2 872 927 -722 -256i=n-1jj-1 j+1

...

> 1010

Convergence Maps & Guidelines

CA Sim 1 2 3 4 5i=1

jj-1 j+1

1.1 1.9 2.8 2.6 5.4i=2jj-1 j+1

1.1 2.1 3 4 5.1i=4000jj-1 j+1

1 2.1 3.1 3.9 5i=3999jj-1 j+1

...

Convergence Maps & Guidelines

Forward Convergence Map

Convergence Maps & Guidelines

Backward Convergence Map

Convergence Maps & Guidelines

a Parameters

Convergence Maps & Guidelines

a Parameters

a1

Convergence Maps & Guidelines

a Parameters

a2

Convergence Maps & Guidelines

Forward Constraints Poles Constraint:

Convergence Maps & Guidelines

Backward Constraints

Convergence Maps & Guidelines

Simulation Speed

Convergence Maps & Guidelines

a3 Vertical Constraint

Convergence Maps & Guidelines

a3 Vertical Constraint

Zeros Constraint:

Convergence Maps & Guidelines

Substituting Uj-1 and Uj+1

•Boundary Zero Values

0.11 0.34 0.27 0.4 0.56

jj-1 j+1

0 0

Convergence Maps & Guidelines

Zeros Boundary Constraint

Convergence Maps & Guidelines

Zeros Boundary Constraint

Convergence Maps & Guidelines

Guidelines

If ((upperZero and lowerPole intersects) and (intesection < initial point)) then

htMax = intersection * safetyBuffer;Else

htMax = initial point * safetyBuffer;End

ht = userInput( < htMax);hx=lowerPole(ht);

Convergence Maps & Guidelines

Guidelines Example

Conclusion

Partial Diff -> CA

Conclusion

Theoretical Constraints

Zeros Constraint:

Poles Constraint:

Conclusion

Guidelines

If ((upperZero and lowerPole intersects) and (intesection < initial point)) then

htMax = intersection * safetyBuffer;Else

htMax = initial point * safetyBuffer;End

ht = userInput( < htMax);hx=lowerPole(ht);

Conclusion

Future Work

• Proofs of Observations

•Quadratic General Form:

•Efficient Parallelization

•Simulation Error

Conclusion

References Paul Rochester. Euler's Numerical Method for Solving Differential Equations. November 2009. http://people.bath.ac.uk/prr20/ma10126webpage.html

Region of Convergence. Wikipedia. November 2009. http://en.wikipedia.org/wiki/Z-transform

Keith Schubert. Cellular automaton for bioverms, October 2008.

Jane Curnutt, Ernesto Gomez, and Keith Evan Schubert. Patterned growth in extreme environments. 2007.

Cell Image - http://askabiologist.asu.edu/research/buildingblocks/images/cell.jpg

Martin Gardner. The fantastic combinations of john conway’s new solitaire game”life”. Scientific American, (223):120–123, 1970.

T.A. Burton, editor. Modeling and Differential Equations in Biology. Pure andApplied Mathematics. Marcel Dekker Inc., 1980.

J. von Hardenberg, E. Meron, M. Shachak, and Y. Zarmi1. Diversity of vegetationpatterns and desertification. Physical Review Letters, 87(19), November 2001.

Conclusion

Acknowledgements

and Questions?