Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Physics and Complexity
David Sherrington University of Oxford & Santa Fe Institute
Blaise Pascal Medal in Physics 2010 European Academy of Sciences
Athens 5 Nov 2010
Blaise Pascal (1623-1662)
Mathematician
Physicist
Engineer &
Philosopher
Innovative Genius
Physics
Dictionary definition: Branch of science concerned with the nature and properties of matter and energy
But today I want to use it as a mind-set with valuable methodologies
and to show application to complex systems in many different arenas
Complexity
• Many body systems
• Cooperative behaviour complex – not simply anticipatable from microscopics
• occurs even with simple individual units and simple interaction rules
– with surprising conceptual similarities between superficially different systems
Aim today
Illustrate use of statistical physics methodology to understand complexity and its ubiquity
via simple models, pictures and comparisons
Give flavour of concepts
Point out some relationships to pioneering Pascal
Key ingredients
Frustration Conflicts
Disorder Quenched / self-induced / time-dependent
Methodologies
• Analysis • Experiment
• Conceptualization • Computer simulation
Symbiosis
But only a broad brush picture today
The Dean’s Problem
• Dean to allocate N students to two dorms
• Some students like one another; prefer same dorm • Others dislike one another; prefer different dorms
• Cannot satisfy all →
• Best compromise for whole student body?
- Frustration or or…
J: Inter-student friendship: +/-
The Dean’s Problem as combinatorial optimization
H = + JijSiS j
( ij )∑
Dorm A/B
Maximise a Happiness function
Students, i,j
J: Inter-student friendship: +/-
The Dean’s Problem as combinatorial optimization
H = + JijSiS j
( ij )∑
Dorm A/B
Maximise a Happiness function
Students, i,j
Very difficult for general {J} with both positive & negative Jij ; 2N choices; NP-complete N students: N large
J: Inter-student friendship
The Dean’s Problem as combinatorial optimization
H = + JijSiS j
( ij )∑
Dorm A/B
Maximise a Happiness function
Students, i,j
Very difficult for general {J} with both positive & negative Jij ; 2N choices; NP-complete
Random {J}: characterize by drawing from a probability distribution P(J)
N students: N large
Aside: Pascal and Fermat
Invented the mathematical theory of probability
Spin Glasses
• Magnetic alloys; e.g. CuMn, AuFe
• Theory: All sites occupied, random exchange
• Computer simulation: experiments on model systems similar behaviour
Pascal: Computing
Invented and built the first mechanical
calculating machine
Pascaline
Fore-runner of modern computers
Recognised and honoured by naming of a computer programming language
‘Spin glass’ properties • Non-periodic/amorphous collective ordering
• Non- or very slow equilibration
• Aging, rejuvenation, memory
AuFe
Dean’s model ~ Range-free Spin Glass Model (SK)
.
H = − JijSiS j
( ij )∑
Hamiltonian Exchange interactions Spins ~ magnetic moments
Unhappiness Friendship Dorm allocation
A B
Dean’s problem
Spin glass
Our first equivalence
Minimise
Paradigmic control function
.
H = − JijSiS j
( ij )∑
Statistical physics: Typical properties Expectation values of physical observables
Range-free: soluble but very subtle
with several different identifications of symbols and different types of disorder and frustration
For this talk
Random: Characterize by probability distribution P(J): quenched
Finite-temperature equilibrium: P(S) ~ exp(-H{J}(S)/T)
Paradigmic cartoon
Rugged Landscape
Many metastable states
Hierarchy Valleys within valleys
Hard to minimise/maximize: sticks: glassy
Cost to minimise
Coordinates
Simple algorithms
Smooth local
motion
c.f. Fitness to maximise
Where does this cartoon come from?
Simulations, analytic calculations, anzätze
qSS '
= N −1 σ ii∑ S
σ i
S '
P(q) = WSWS 'δ (q − qSS ' )
SS '∑
qαδ = qβδ ≤ qβδ
Conventional system: single δ fn Complex system: structure
e.g. SK: ultrametric phylogenetic tree
Overlap
Overlap distribution
Via some novel mathematical structures that I believe Pascal would have loved
Parisi’s ansatz
Replica Symmetric
Replica Symmetry Broken Hierarchical sub-division
q0 q1
q2
qαβ ; α ,β = 1,..n, n→ 0 n x n matrix but with non-integral n 0
Spin glass
Pascal’s triangle
.
Pascal_modulo 2
Sierpinski gasket (Hierarchical fractal)
Reminder
Temperature • Natural for real physics
• Often useful for practical optimization to introduce an artificial temperature
and take limit as it is reduced
• Other analogues in other problems
P(S) ~ exp(-H{J}(S)/TA)
Landscape paradigm for hard optimization
Cost
Peaks are obstacles
Steepest descent gets stuck
Simulated annealing
Probabilistic hill-climbing Add artificial annealing ‘temperature’
Variables
Cost TA
Annealing temperature
P(move) ~ exp(−ΔC / TA )
Simulated annealing
Gradually reduce TA
Variables
Cost TA
Annealing temperature
Simulated annealing
Variables
Cost
Hopefully
H = − JijSiS jij∑
+/- ; excitatory/inhibitory
Neural network: Hopfield model
Neurons, rate of firing
T ~ synaptic sigmoidal response rounding
T O
I I O
Synapses
Ii = Jij
j∑ S j
Control function
Attractors
• Associative memory ‘attractors’ memorized patterns
• Retrieval basins
• Many memories many attractors
require frustration
Stored in {J}
Phase space
Neural ‘rugged landscape’
Valleys ~ attractors Sculpture ~ learning {Si } {Jij}
Different timescales fast retrieval slow learning
Proteins
Must fold fairly easily
Minimal frustration
Proteins: Heteropolymrers
Many amino acids Frustrated interactions
Random heteropolymers
In general, very frustrated Fold poorly, glassy
Evolution:
Initial random soup
Fast: attempt to fold
Slower time-scale: Reproduction/mutation Good folders selected
Folding funnel Wolynes et. al.
Analogies
Glassy/slow More minimal frustration/faster
Spin glass Neural network SK Hopfield
Random heteropolymer Protein Wolynes
Price
Time
Different strategies (Disorder)
Common information
(Mean field) (Range-free)
Learn from Experience (Dynamics II)
Not all can win (Frustration)
Buy & sell (Dynamics I)
Stockmarket
Minority game N agents 2 choices
Aim to be in minority
Individual strategies Collective consequence • act on common information (e.g. minority choice for last m steps) • preferences modified by experience (keep point-score)
Simple minimalist model
Phase transitions & ergodicity-breaking
Non-ergodic
Ergodic
Scaled size of information space
Ergodic Non-ergodic
Spin glass susceptibility Minority game: volatility
Suggests similar concepts & methodologies
AuFe
Experimentt
Simulation
.
Recall
Minority game
a
Phase space
H = + Jijsiij∑ sj
Jij= ξ
i
µ
µ
∑ ξj
µ
Many repellors
Analogy with Hopfield neural network but different
c.f. attractors in neural network
Effective control function
Stochastic noise/’temperature’
• Energy Free Energy – Temperatures smoothes free energy
• Reduces ruggedness
– Neural networks • Noise reduces false minima in effective
landscape
– Minority game • Stochastic uncertainty reduces oscillations • & tabala raza volatility
Random K-SAT; K-spin glass
HARD-SAT
N/M
0
UNSAT/ Spin Glass
SAT/ Para
αc-1
α d-1
Simple algorithms stick Theoretically achievable limit
Phase transitions T
Dynamical transition
Thermodynamic trans
Extension
Fascinating Physics Novel maths
Complexity Science
Transfers Opportunities
.
Blaise Pascal (1623-1662)
I think Pascal would have enjoyed the challenges posed by complex many-body systems
I am honoured by the award of this prize bearing his name
and by the EAS recognition
Th
Collaborators
Tomaso Aste Jay Banavar Arnaud Buhot Andrea Cavagna Premla Chandra Tuck Choy Ton Coolen Dinah Cragg Lexie Davison Malcolm Dunlop Alex Duering Sam Edwards David Elderfield
Fernando Nobre Dominic O’Kane Reinhold Oppermann Richard Penney Albrecht Rau Hans-Juergen Sommers Nicolas Sourlas Byron Southern Mike Thorpe Tim Watkin Andreas Wendemuth Werner Wiethege Michael Wong
Julio Fernandez Juan Pedro Garrahan S.K.Ghatak Irene Giardina Peter Gillin Paul Goldbart Lev Ioffe Peter Kahn Scott Kirkpatrick Stephen Laughton Esteban Moro Peter Mottishaw Normand Mousseau Hidetoshi Nishimori
Thanks Teachers, collaborators,
students, postdocs, friends
Tomaso Aste
Jay Banavar
Ludovic Berthier
Stefan Boettcher
Arnaud Buhot
Andrea Cavagna
Premla Chandra
Tuck Choy
Ton Coolen
Dinah Cragg
Lexie Davison
Andrea De Martino
Malcolm Dunlop
Alex Duering
David Elderfield
Julio Fernandez
Tobias Galla
Juan Pedro Garrahan
S.K.Ghatak
Irene Giardina
Peter Gillin
Paul Goldbart
Lev Ioffe Robert Jack Alexandre Lefevre
Turab Lookman
Peter Kahn
Scott Kirkpatrick
Helmut Katzgraber
Stephen Laughton
Francesco Mancini
Marc Mezard
Esteban Moro
Peter Mottishaw
Normand Mousseau
Hidetoshi Nishimori Fernando Nobre Dominic O’Kane
Reinhold Oppermann
Giorgio Parisi
Richard Penney
Albrecht Rau
Avadh Saxena
Manuel Schmidt
Hans-Juergen Sommers
Nicolas Sourlas
Byron Southern
Mike Thorpe
Tim Watkin
Andreas Wendemuth
Werner Wiethege
Stephen Whitelam
Peter Wolynes
Michael Wong
Phil Anderson Sam Edwards Arvid Herzenberg Frank Jackson Harry Kelly Walter Kohn