Top Banner
Complex System Science John Finnigan CSIRO Atmospheric Research
37

Complex System Science John Finnigan CSIRO Atmospheric Research.

Mar 26, 2015

Download

Documents

Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Complex System Science John Finnigan CSIRO Atmospheric Research.

Complex System Science

John Finnigan

CSIRO Atmospheric Research

Page 2: Complex System Science John Finnigan CSIRO Atmospheric Research.

Contents

• Complex systems Science• Systems• Complexity-the idea of emergent structure• Farming systems as ‘Complex Adaptive Systems’

• Three Approaches to Understanding• Network Theory• Cellular Automata• Agent Based Models

• Summary• The CSIRO Centre for Complex Systems

Science

Page 3: Complex System Science John Finnigan CSIRO Atmospheric Research.

Complex System Science

Has two elements:

• Systems-collections of interacting things

• Complexity-the essence of which is the property of self-organisation or emergence of structure from the interaction between the constituent parts of the system

Page 4: Complex System Science John Finnigan CSIRO Atmospheric Research.

Emergence or Self-Organisation

•We recognise this phenomenon over a vast range of physical scales and degrees of complexity

•From Galaxies ~ 106 Light Years

Page 5: Complex System Science John Finnigan CSIRO Atmospheric Research.
Page 6: Complex System Science John Finnigan CSIRO Atmospheric Research.

To cyclones

~ 100 km

Page 7: Complex System Science John Finnigan CSIRO Atmospheric Research.

And Chemical reactions ~ 10 cm

Page 8: Complex System Science John Finnigan CSIRO Atmospheric Research.
Page 9: Complex System Science John Finnigan CSIRO Atmospheric Research.
Page 10: Complex System Science John Finnigan CSIRO Atmospheric Research.
Page 11: Complex System Science John Finnigan CSIRO Atmospheric Research.
Page 12: Complex System Science John Finnigan CSIRO Atmospheric Research.
Page 13: Complex System Science John Finnigan CSIRO Atmospheric Research.

Ribosome

E Coli

Root Tip

Amoeba

To Gene expression and cell interaction

Page 14: Complex System Science John Finnigan CSIRO Atmospheric Research.

The processing of information by the brain

Page 15: Complex System Science John Finnigan CSIRO Atmospheric Research.

To animal societies and the

emergence of culture

Page 16: Complex System Science John Finnigan CSIRO Atmospheric Research.

And the social artefacts of human society such as economies

Page 17: Complex System Science John Finnigan CSIRO Atmospheric Research.

The Concept of Self-Organisation has consequences at several levels

• At the whole system level (in the present case, farming systems) it means that ‘no one is in charge’ and optimal or command and control solutions to system problems usually fail (sooner or later)

• At the level of analysis, self organising processes provide us with powerful tools

Page 18: Complex System Science John Finnigan CSIRO Atmospheric Research.

Foot and Mouth Disease in the UK An example of failure caused by focussing on one part of the system and ignoring the links between biophysics and economics

Economic rationalization of abattoirs and bizarre EU subsidies increased the connections between herds to a critical point.

Changes to F&M reporting rules may have delayed the isolation of infectious animals.

The relationship between these actions and the epidemiology of F&M was not appreciated in advance (at least where it mattered) because the livestock industry was not viewed as an integrated system.

Page 19: Complex System Science John Finnigan CSIRO Atmospheric Research.

Farming Systems at the gross level involve Economics, People, their Social Networks as well as Biophysics such as hydrology,

soil science, Agronomy and Biology.

• We can attempt to understand and model the whole system or parts of it

• To model the whole system we need first a mental map and then some techniques to capture the parts and interactions of the mental map

Page 20: Complex System Science John Finnigan CSIRO Atmospheric Research.

The MarketClimate

A regional scale social-ecological system including farming, as a complex adaptive system

In a CAS, there is no Fat Controller. The system behaviour is an emergent property

Page 21: Complex System Science John Finnigan CSIRO Atmospheric Research.

Illustrative Examples of Dynamic Process Interactions

Argonne National Laboratory

NATURALPROCESSES

SOCIETALPROCESSES FORMATION

OF ELITES

LONGDISTANCE

TRADE

IRRIGATION

WEATHER &CLIMATE

DIFFERENTIATIONOF ROLES

HOUSEHOLDMGMT.

SOILCONDITION

TOWNFORMATION

STATEFORMATION

SURFACEEXCHANGE

EVOLUTION OFHOUSEHOLDS

DEMO-GRAPHICS

CONFLICT

KINSHIP-BASED

BEHAVIORS

AGRICULTURE

HYDROLOGY

PLANTGROWTH

Page 22: Complex System Science John Finnigan CSIRO Atmospheric Research.

The simulation framework implementsan "agent-based" simulation in whichdozens to thousands of Households

interact and conduct their activities in anenvironment that responds to their

actions and updates itself daily.

Modeled Households can respond toenvironmental stress (repeated droughts, etc.) by,for example:

violating fallow (plant each field, each year) manuring fields clearing more distant fields shifting emphasis toward pastoralism importing food, emigrating, etc.

MM5Mesoscale

WeatherModel

AtmosphereObject

(one of manyalternativemodels)

EPIC includes: crop growth soil condition soil erosion nutrient

cycling, etc.

SWAT/EPIC

Crop Model

FieldObjects

Decide Crop Strategy

Cultivate Crop

Borrow Grain

Obtain Plow Team

Release Field to Forage

HouseholdObjects

Societal ProcessSimulation Models

Natural ProcessSimulation Models

... many others, basedon historical records...

... and cultural analogs

generally severalFields per Household

Objects and Associated Models

Argonne National Laboratory

Page 23: Complex System Science John Finnigan CSIRO Atmospheric Research.

We can build models of Complex adaptive Systems using techniques like ‘Agent based Modelling’ but to understand and predict their

behaviour, we need a science of systems

The understanding we need is coming from a evolving blend of at least three different approaches:

1. Dynamical Systems Theory

2. Network Theory

3. Evolutionary or adaptive computing

Page 24: Complex System Science John Finnigan CSIRO Atmospheric Research.

1 Studying Ecosystems as dynamical systems

A minimal model of an ecosystem describes the change over time of ecosystem state.

The trajectories indicate stable states of the ecosystem as external conditions change

Ecosystems that display two (or more) alternative stable states include :

•lakes(oligotrophic/eutrophic),

• grasslands/woodlands,

•coral reefs (pristine/algal covered),

•marine ecosystems as measured in fish catch….

(Figs from Scheffer et al, 2001, Nature)

Page 25: Complex System Science John Finnigan CSIRO Atmospheric Research.

Basins of attraction and Ecosystem Resilience

A minimal model of an ecosystem describes the change over time of an unwanted ecosystem property, x such as lake turbidity

a represents an environmental factor that promotes x, b represents the rate at which x decays in the system, r is the rate at which x recovers again as a function f of x

The form of f(x) determines whether multiple stable states or attractors will exist

dxa bx r f x

dt

p p pf x x x h

(Figs from Scheffer et al, 2001, Nature)

Page 26: Complex System Science John Finnigan CSIRO Atmospheric Research.

What do we mean by stable states?

Linear dynamics Non-linear dynamics Boundaries of Strange

Periodic attractor Strange Attractor Attractors are Fractal

Page 27: Complex System Science John Finnigan CSIRO Atmospheric Research.

2 We can represent most systems as networks with interactions across the links-

Network Topologies control System behaviour

Regular Network: each node has the same number of connections

Homogeneous network: Number of connections per node varies but there is a clear average value. Networks like this can result from randomly connecting nodes. Near the phase transition they are vulnerable to random removal of links

Heterogeneous or ‘scale free’ network: There is no average number of connections per node: Living networks that grow by accretion often have this dendritic form. They are resilient to random removal of links but vulnerable to a targeted attack that removes a key node

Page 28: Complex System Science John Finnigan CSIRO Atmospheric Research.

3 Adaptive Systems can be illustrated simply using Cellular Automata. CAs are Systems that evolve

on lattices according to local interaction rules

The simplest rules: the state of a cell at time T+1 is determined by its own state and that of its two neighbours at time T

Page 29: Complex System Science John Finnigan CSIRO Atmospheric Research.

Discretization of PDEs yields Cellular Automata

2 1 1y

y yx

2

21 2 0 1

yy y y

x

2

2

y y y

t x x

Advection-diffusion equation

t=n

t=n+1

-1 0 +1X=

Page 30: Complex System Science John Finnigan CSIRO Atmospheric Research.

We can form new types of Cellular Automata by changing the interaction rules or the wiring or both

T=0

T=1

T=2

Dynamics on networks can evolve either by changes in the interaction rules

Or by changes in the ‘wiring’ of the network

Page 31: Complex System Science John Finnigan CSIRO Atmospheric Research.

The Cellular Automaton as a computer:Evolving the local rules that will perform a

computational task by applying a global selection pressure

T=0

T=1

T=2

The colour that a cell adopts at the next timestep depends only on the colours of itself and its neighbours at the present time step

Rules are recombined (bred) and selected according to Darwinian principles to find the set of local rules that will solve the density problem

Page 32: Complex System Science John Finnigan CSIRO Atmospheric Research.

Moving Away from Classical Mathematics

• With complete freedom to stipulate rules and wiring between elements of our CAs in the virtual world of the computer and then to let them evolve as part of the computation, we can form mathematical objects that are very difficult to capture using the approaches of conventional mathematics but which match very well what we observe in living systems.

• Agent Based Models exploit this freedom• Analysis using network theory and similar techniques is

leading to increasing understanding of these systems-but so far we have few general principles

Page 33: Complex System Science John Finnigan CSIRO Atmospheric Research.

Summary

• Complex Systems Science brings together systems approaches and a rapidly developing science of systems

• Rather than being any particular set of techniques it is primarily the adoption of a different point of view

• That is, to admit the prevalence of self- organisation in complex systems together with the behaviours that flow from that and the techniques necessary to study it.

Page 34: Complex System Science John Finnigan CSIRO Atmospheric Research.

The CSIRO Centre for Complex Systems Science-A Virtual Centre

The Core Group comprises a permanent Science Director, a Communication and Training Manager, Post Docs, PhDs and visitors. It interacts with Division based projects to do basic research in CSS.

Projects are located within CSIRO Divisions (and partner Institutions).

A key function of the Core group is to manage interaction between the projects. The Core and the Division-based Projects are closely networked.

Page 35: Complex System Science John Finnigan CSIRO Atmospheric Research.

The Compass of Complex System Science: Projects in the CSIRO Centre for CSS-1

• Inference of complex systems properties from fragmentary information (Mantle dynamics and mineralization: State Space reconstruction)

• Ensemble Prediction of Atmospheric and Ocean-Atmosphere Regime Transitions (Dynamical Systems theory)

• The stability of the Southern Ocean overturning circulation (Dynamical Systems theory)

• Critical states in bushfires (Dynamical Systems theory, Agent Based Modelling)

• The effects of model structure and dimensionality on the emergent properties of ecosystem models (Estuarine systems: Dynamical Systems theory, ABM)

• Rapid shifts in state and resilience in river systems (ABM)

• Targeting Drug-like properties in Chemical libraries (Genetic Algorithms, Evolutionary computing)

Page 36: Complex System Science John Finnigan CSIRO Atmospheric Research.

The Compass of Complex System Science: Projects in the CSIRO Centre for CSS-2

• Tracking Air Borne Chemical Signals (Fractal turbulence + AI, ABM)

• Adaptation and resilience in regional socio-economic systems (Managed rangelands: ABM)

• Multiscale modelling in Industrial and Natural systems (Lattice-Boltzmann methods, Non-Equil Thermodynamics)

• Interactions, information sharing and simulated reasoning of fishers in an agent-based, Bayesian network model of fishing behaviour (ABM)

• The Future of the Swan River: Governance and Agent Based Modeling (ABM, Evolutionary game theory)

• Our National Electricity Market as a Complex Adaptive System (ABM)

• Links between resilience and information in complex adaptive systems (ABM, Evolutionary game theory)

Page 37: Complex System Science John Finnigan CSIRO Atmospheric Research.

The Purpose of This Workshop

• Is to bring together workers with awareness of the problems and workers with knowledge of CSS Techniques

• And to start a process of developing projects for future joint funding

• We plan a funding round built around the 2nd CSIRO CSS workshop in Sydney 27-29 August, 2003.