Simulation of discrete spatio-temporal systemsliacs.leidenuniv.nl/~csnaco/CSA/slides/CSA8.pdfThe...
Transcript of Simulation of discrete spatio-temporal systemsliacs.leidenuniv.nl/~csnaco/CSA/slides/CSA8.pdfThe...
Systems with many variables
� Iterative function systems describe systems with a single variable
� A iterative system with two variables was given by the Julia set
� Systems in economy, meteorology, ecology, sociology, etc., consists often of a large number of variables, interacting with each other
� Besides chaos many new phenomena occur in such many variable systems
� These are for example evolution, self-organization, emergence, self-reproduction, phase-transitions
Complexity, Chaos, and Anti-Chaos
� The study of spatio-temporal systems will reveal that complexity and chaos are not the same
� Chaotic processes can produce simple patterns called anti-chaos
� The emergence of order out of chaos can be observed (Cohen, Stewart 1994)
S.A. Kauffman (1991) Antichaos and adaptation, Scientific American,265 (2), 64-70J. Cohen and I. Stewart 1994: The collapse of chaos, Penguin, NY
Simple models of complex systems
� A realistic treatment of complex systems is computationally expensive and often intractable
� Scientists are looking for simple models of complex systems
� Often the predictions made with these simple models are surprisingly realistic or provide deep insights in the dynamics of real world systems (e.g. explaining phase-transitions).
� However, the science of complexity is in its very beginning and a new frontier in nonlinear systems dynamics; definitions are evolving and scientists have not yet discovered unifying theories;
T. Bohr et al. (1998): Dynamical systems approach to turbulence, CambridgeUniversity Press, New York
Locally interacting cell arrays
� One of the simplest models involve a spatial array of cells
� The cells interact with nearby cells by simple rules
� These systems often exhibit spatio-temporal chaos– Spatial patterns in time are aperiodic and difficult
to predict
– Complex, often self-similar,patterns evolve in time
Cellular automata
� Von Neumann introduced cellular automata 1966, Wolfram studied them extensively and classified them (“A new kind of science”)
� CA are perhaps the most simple models of spatio-temporal systems, but their behavioral spectrum is wide and interesting to study
Wolfram, S. (1986) Theory and application of cellular automata, WorldScientific, Singapore
Von Neumann (1966) Theory of self-reproducing cellular automata, Univ.Of Illinois Press,Urbana, Il.
Wolfram S. (2002) A new kind of science, Wolfram Press
A motivating example – the XOR 1D Automaton
� Consider a ring of people
� Each one is wearing a cap with the bill forwards, except one who is wearing the bill to the back
� Now, each one is looking at his/her left and right neighbor, and adapts using these rules:
– Left and right neighbor have bill forwards � wear bill forwards
– Left and right neighbor have bill backwards � wear bill forwards
– If only one neighbor has bill forwards � wear bill backwards
� Evolve system over a number of generations for a large ring of people
Sierpinski cones and maximal-speed of information
� Long term behavior shows self-similar cone-like structures
� They resemble Sierpinskitriangles
� The maximal speed-of-information gives rise to the boundary of the cones, similar to the speed-of-light giving rise to Minkowski’s space-time cones
� Indeed, in CA literature the set of possible states that influenced the system in some past are called ‘light cones’
� Complex organization, but no chaos is evident
Formal expression
� The given game is an example of a 1-D cellular automaton
� Several ways to express a cellular automata rule
– X(t+1,i)=(X(t,i-1)+X(t,i+1)) modulo 2
– X(t+1,i)=X(t,i-1) XOR X(t,i+1)
� The XOR statement is a logical function
� 24 = 16 logical functions could be tried instead of XOR
x411
x301
x210
x100
t+1,it,i+1t,i-1
Rule of a cellular automaton
Initial state
� The evolution of a cellular automaton is defined also by its initial state
� The left figure shows the evolution of a cellular automaton with random initial condition using the XOR function
� The behavior is not chaotic, but propagate the initial conditions in an ordered way forward in time
� Other rules may give rise to chaotic behavior from ordered starting conditions
The size of the rule space
� The discussed XOR automaton is an example of an 1-D Cellular automaton
� The size of the neighborhood and the number of possible states determine the number of possible rules for a cellular automaton
� If we consider N nearest neighbors to each side, the number of possible rules would grow to:
Why?S(N) = 222N
Four dynamical classes of Cellular automata
� Cellular automata were classified by Wolfram (2002), into four classes based on their dynamics
1. Class 1 reach a homogeneous state with all cells the same for all initial conditions
2. Class 2 reach a non-uniform state that is either constant or periodic in time, with a pattern depending on initial conditions
3. Class 3 have somewhat random patterns, are sensitive to initial conditions, and small scale local structure
4. Class 4 have relatively simple localized structures that propagate and interact in very complicated ways
� The four classes correspond roughly to fixed points, periodicity, and chaos in dynamical systems � examples will follow
Langton’s λ λ λ λ quantity
� Langton’s quantity λ is the number of state configurations that map to 1 divided by the total number of state configurations
� For instance in the left figure λ=3/8
� As the numbers of 0 equals 1- λ, only the range from 0 to 0.5 is of particular interest
0111
0011
0101
1001
0110
0010
1100
1000
t+1,it,i+1t,it,i-1
Langton, C. (1986) Studying artificial life with cellular automata, Physica D 22, 120-49
Langton’s λ λ λ λ quantity and dynamic behaviour
Solid at zero temperature
Melting FluidSolid at finite temperature
Turbulent Fluid
Melting fluid
Solid at finite temperature
λ
By increasing λ from 0 to 0.5 (1 downto 0.5) roughly the system goes through the same states than the logistic map for different values of the constant a� Assignment
Higher dimensional Cellular automata
� Cellular automata can be defined not only for 1-D arrays but also on higher dimensional arrays
� Some mathematical notation:
– 1-D arrays are called chains
– 2-D arrays are called grids
– Arrays of any dimension d are called d-dimensional lattices
Cellular automata in 2-D
� A classical cellular automaton was defined by Conway –Conway’s game of life
� Consider a ‘game’ played on a rectangular grid, each grid cell can have two states – dead or alive
� The neighbors of a center cell are the nearest neighbors to the north, south, east, west, north-west, south-west, south-east, north-east
� This is termed the Moore Neighborhood
SESSW
ECW
NENNW
Cellular automata in 2-D
� Rule– A cell that is alive, stays
alive, if it has two or three living neighbors
– A dead cell becomes alive, when it has exactly three living neighbors
– For all other cases a cell dies or remains dead
� Example of outer totalistic rule, i.e. a rule that involves only the sum of neighbor states
SESSW
ECW
NENNW
Evolution of the game of life
Starting froman initially randomConfiguration Colonies of cellsemerge, someof them periodicsome of them fixedor moving throughspace, shooting pixels (glider guns*)etc.
*Berlekamp et al. (1982) Winning ways for your mathematical plays, AcademicPress, New York
The glider gun
� Conway offered 50$ for everyone, who could find an endlessly growing configuration or prove that none exists
� William Gosper and 5 other MIT students discovered the glider gun and won the price
– The glider gun shoots a copy of itself
– On an infinite grid it would grow and evolve without limit
Other possible configuration spaces
� Regular tilings of the 2-D plane (there are three possibilities)
� More than 3-dimensional configuration spaces
� Most generally:– Configuration spaces
represented by Caleygraphs of some group
All possible regular tilings ofthe 2-D plane,i.e. tilings consistingonly of the same objects
hexagonalgrid
General CA definition via Caley graph
� Groups describe symmetric structures
� (M,+) is a group, iff ∀a,b,c∈M:– a+b∈ M and a+(b+c)=(a+b)+c
– There exists e∈M with e+a =a
– each a∈M has an inverse called (-a), such that a+(-a)=e.
� We define a group M via a set of generators X ⊆ M, such that for every element a∈M and generator x, both a+x∈M and a+(-x)∈M; Moreover, all elements belong to the group that can be obtained by concatenated application of generators.
� Given a generator {x1,… ,xm} we can define a Caley graph C=(V,E) of the group:
– vertex set: V=M
– edges E are given by (v1,v2)∈E, iff v1 = v2+a, or v1=v2+(-a) for some a in X.
� The fundamental neighborhood N(C)of the Caley graph is defined by the union of the set of generators and the set of its inverse elements.
� For each element in the graph we can get its neighbors by using the generator elements in N(C).
� A cellular automaton (C, N(C),A, T) is defined as a tuple of a Caley graph with labeled vertices, its fundamental neighborhood, an finite alphabet, and a transition function T
� Node labels are chosen from a finite local state space A
� Transition rule T: A|N(C)+1|�A assigns each element of a cell a new value based on the neighbors in the Caleygraph, obtained by applying the generator.
Fd = 〈x1, . . . , xm|〉 Free group
Zd = 〈x1, . . . , xd|∀m,n : xm + xn = xn + xm〉
Zm × Zn = 〈a, . . . , b|ab = ba,ma = nb〉 Torus
Neighborhood types and sizes
� Von Neumann neighborhood and Moore neighborhood are most commonly used in 2-D grids
� The radius of these neighborhoods can be increased, e.g. by applying group generators twice
Examples of groups and their Caleygraphs
Fd = 〈x1, . . . , xm|〉 Free group
Zd = 〈x1, . . . , xd|∀m,n : xm + xn = xn + xm〉
ab
x1
x2
The group Z2 and its fundamental neighborhood
Freegroup
Zm × Zn = 〈a, . . . , b|ab = ba,ma = nb = e〉 Torus
Questions
� How many transition rules can we define on a cellular automaton (C, N(C), A, T)?
� What could be the set of generators for the -2D integer lattice with Moore neigborhood?
Extensions of Cellular automata
� A multidimensional state space
– In Lattice gas models each cell is assigned a vector (velocity of the fluid flow)
� Memory of states in t-1, t-2, etc.
� Dynamic rule sets, dynamic neighborhoods, etc.
Finite cellular automata and chaos
� Finite CA cannot be truly chaotic because the number of states is finite, and thus the system will eventually return to some previous state and be trapped in a circle from then on
� To obtain maximal periods, prime numbers are chosen as cell array sizes
Self-organization
� Simple rules such as the game of life can cause an initially chaotic state to evolve into a highly ordered one � Self-organization
� This somehow contradicts the third law of thermodynamics (3LT), that the entropy is always increasing
� Haken attributed the self-organizing behavior to cooperative effects of the systems components (synergetics)
� The 3LT is motivated by deterministic systems, but in fact also stochastic systems can self-organize
Forest simulation model by Sprott
� Consider a forest with trees placed on grid cells, 0=fur, 1=oak
� We choose a random tree that dies
� We replace this tree with a new tree
� Five trees in the neighborhood are chosen randomly
� If the vast majority (s=4,5) is oak, the new tree gets an oak
� If the vast majority is fur (s=0,1), the new tree gets a fur
� Otherwise (s=2,3), the same tree than before will grow
� Connected patterns emerge from a random starting set
Broken symmetry
� It is surprising, that despite the highly symmetrical starting conditon the emerging system does not converge to a symmetrical object
� This phenomenon is called spontaneous symmetry breaking and can be observed in highly ordered systems, deterministic systems (Wolfram 2002)
Self-organized critically
� So-called dissipative structures will emerge
� Connected regions with a strange but not necessarily fractal boundary (fat fractals)
� The size distribution of the clusters follows a power laws
� Dissipative patterns are observed in many spatio-temporal processes
– Animal migration– Spread of diseases– Vegetation patterns– Clouds and mud
Prigogine, I. (1997) The end of certainty: time, chaos, and the new laws ofnature, the free press, new york
Self-organized critically
� Systems like the forest converge to a pattern for which there is no characteristic scale size
� Size distributions of objects often obey power laws (this they share with the fractals), i.e. the distribution can be fitted to a function
� Recently, power laws are applied in all kind of applications
– Gene regulatory networks– DNA pattern– Stock prices– City distributions– Letter frequency in human/ape
generated random strings (Zipf)
� Not always SOC is the explanation for the Power law
� In case of city size distribution it related to a least effort principle (Zipf).
d ∼ 1/fα
Diffusion
� Diffusion can be modelled via:
� Note, that there is a conservation rule fulfilled
� Task: Implement diffusion system in 2-D in MATLAB and visualize its behaviour over time
aj(t+1)−aj(t)=aj+1(t)−2aj(t)+aj−1(t)
Sand Pile – the prototype of a SOC system
� Consider a pile of sand to which we add sand continuously
� The sand-pile steepens until it reaches an angle of repose, whereupon avalanches keep the sandpile close to this angle
� The avalanches obey a power law scaling in their size distribution and in their duration
Bak’s CA simulation of a pile of sand
� Bak simulated a pile of sand using the following CA model
� The pile is represented by a N ×N matrix of integers
� Initially all cells are chose between 1 and 3
� At each time step choose a random cell i,j and set Z(t+1,i,j)=Z(t,i,j)
� Cells outside the boundary are kept as 0� All other cells which exceed Z=3, and
their von-Neumann neighbors are updated with:Z(t+1,i,j)=Z(t,i,j)-4Z(t+1,i±1,j)=Z(t,i±1,j)+1Z(t+1,I,j±1)=Z(t,I,j±1)+1
� Strictly speaking, this is not a cellular automaton, as it evolves not autonomously
d=2
Power spectrum of sum(Z(i,j)) over t
Emergence vs. Reductionism
� Reductionism assumes simple laws that govern natural processes and that these simple laws help to understand/explain global behaviour
� Emergence holds that high level structure is generally unpredictable from low level processes, and does not even depend very much on its properties
� Due to Sprott (2006), systems are complex, if they exhibit emergent behaviour
How to measure degree of (self-)organization
� The term of self-organization is used since 1947, but up to know there is no standard definition except “I know it when I see it”
� Thermodynamic entropy measures the degree of a system’s “mixedupedness”(to use Gibbs’s word), or how far it departs from being in a pure state
� Organisms are essentially never in pure states, and are highly mixed up at the molecular level, but are the paradigmatic examples of organization.
� Furthermore, there are many different kinds of organization, and entropy ignores all the distinctions and gradations between them
W. R. Ashby, “Principles of the self-organizing dynamic system,” Journal of General Psychology 37, pp. 125–128, 1947.
How to measure a degree of self-organization and complexity?
� Another school of thought has been put forward by Kolmogorov and Solomonoff
“A complex phenomena is one which does not admit of descriptions which are both short and accurate”
� Problem exactness: Coin tossing, produces sequences of maximal Kolmogorovcomplexity, though dynamics are simple to describe.
� Grassberger gave a more general definition: ‘The complexity of a process as the minimal amount of information about its state needed for maximally accurate prediction’
� Crutchfield and Young gave operational definitions of “maximally accurate prediction”and “state”
� The Crutchfield-Young “statistical complexity”, C
µ, of a
dynamical process is the Shannon entropy (information content) of the minimal sufficient statistic for predicting the process’s future.
� Shalizi and Shalizi used this measure recently to quantify self-organization in CA practically
� They used cyclic CA to assess their method
Cyclic CA
� Cyclic cellular automata (CCA) are simple models of chemical oscillators.
� Started from random initial conditions, they produce several kinds of spatial structure, depending on their control parameters.
� They were introduced by David Griffeath, and extensively studied by Fisch
� Transition rule
– Each site in a square two-dimensional lattice is in one of κ colors.
– A cell of color k will change its color to k + 1 mod κ if there are already at least T cells of that color in its neighborhood
– Otherwise, the cell retains its current color
Fisch, R. (1990a). "The one-dimensional cyclic cellular automaton: A system with deterministic dynamics that emulates an interacting particle system with stochastic dynamics". Journal of Theoretical Probability 3 (2): 311–338.
Cyclic CA
� The CCA has three generic forms of long-term behavior, depending on the size of the threshold relative to the range.
� At high thresholds, the CCA forms homogeneous blocks of solid colors, which are completely static — so-called fixation behavior.
� At very low thresholds, the entire lattice eventually oscillates periodically;
– sometimes the oscillation takes the form of large spiral waves which grow to engulf the entire lattice.
� There is an intermediate range of thresholds where incoherent traveling waves form, propagate for a while, and then disperse;
– this is called “turbulence”, but whether it has any connection to actual fluid turbulence is unknown.
Spirallingwaves
Spirals engulfing the space for Moore neighborhood
� Cyclic CA for a Moore neighborhood and T=2
� For Moore neighborhood the following transitions can be found:
– T=1: local oscillations
– T=2: spiraling waves
– T=3: turbulence, often metastable in very long run (then spirals can take over)
Cellular Automata and beyond
� Statistical complexity of cyclic CA over time
Cosma Rohilla Shalizi and Kristina Lisa Shalizi: Quantifying Self-Organization in Cyclic Cellular Automata, http://arxiv.org/abs/nlin/0507067v1
Cellular automata and beyond
Partial Differential equations
ContinousContinuousContinuous
DiscreteContinuousContinuous
ContinuousDiscreteContinuous
DiscreteDiscreteContinuous
Coupled Flow Lattices
ContinuousContinuousDiscrete
DiscreteContinuousDiscrete
Coupled Map Lattices
ContinuousDiscreteDiscrete
Cellular AutomatonDiscreteDiscreteDiscrete
ModelStateTimeSpace
Summary (1)
� Cellular automata are defined on a Caley graph (with state labels) with a neighbourhood and transition rule mapping the state of a center cell to a new state based on its neighbor states.
� The number of possible transition rules grows exponentially with the size of the local state space and neighborhood
� Common neighborhood types are von Neumann and Moore neighborhood, and the k-neighbors in 1-D arrays (with periodic boundary conditions)
Summary (2)
� CA are simple models of natural systems� Despite their simplicity the behavior of CA can be
extremely complex and difficult to predict� CA serve as models for studying emergent
phenomena and self-organization� Self organized systems are often at the boundary of
chaotic and ordered states; many open questions remain, and definitions are not yet clarified
� An interesting question if the type of global behavior can be predicted from properties of the rules (e.g. Langtons lambda)
Summary (3)
� As simulators CA models are easy to implemented (also in parallel) and can be used to model phenomena such as diffusion, cell systems, flow, pattern formation, etc.
� CA can be seen as discrete counterparts of partial differential equations
� As such they belong to the class of spatio-temporal models