Mathematical Foundation of System Dynamics PUGACHEVA ELENA, Associate Professor Visiting Academic Scholar Radboud University Nijmegen (The Netherlands)
MathematicalFoundationofSystemDynamics
PUGACHEVAELENA,AssociateProfessor
Visiting Academic ScholarRadboud University Nijmegen
(The Netherlands)
• Softwareofsystemdynamics(Stella,VensimandPowerSim)makesitpossibletobuildacomplicateddynamicmodelwithoutknowingsystemdynamicsanddifferentialequations
FromDynamicstoSystemDynamicsTheaimistointroducesystemdynamics
fromadynamicsviewpointWhy?
• Dynamicanalysisisafoundationofsystemdynamics
• Dynamicanalysishelpstounderstandthemechanismsthroughwhichunpredictable,unknownandemergentchangehappens
Twodifferentconceptsoftime
• 1.Timeasamomentoftimeorapointintime,denotedhereasτ (τ =1,2,3,....);
• 2.Timeasaperiodoftimeoranintervaloftime,denotedhereast,suchthatt=1st,2nd,3rd,...,ormorelooselyt=1,2,3,…
• Unitsoftheperiodcouldbeasecond,aminute,anhour,aweek,amonth,aquarter,ayear,adecade,acentury,amillennium,etc.,dependingonthenatureofthedynamicsinquestion.
Stocks
• Stocksareaccumulations• Stocksholdthecurrentstateofthesystem• Stocksfullydescribetheconditionofthesystematanypointintime
• So,stockistheamountthatexistsataspecificpointintime.
• Letxbesuchanamountofstockataspecificpointintimeτ .Thenstockcanbedefinedasx(τ)whereτ canbeanyrealnumber.
Flows
• Flowsdothechanging• Flowsincreaseanddecreasestocksnotjustonce,buteveryunitoftime
• Flowisdefinedaschangeofstockduringaunitinterval,anddenotedhereby𝑓(𝑡).
Stock-flowrelation
𝑥(𝜏 + 1) = 𝑥(𝜏) + 𝑓(𝑡)𝜏𝑎𝑛𝑑𝑡 = 0; 1; 2; 3;…
τ+1τ
𝑥(𝑡 + 1)
𝑥(𝑡) 𝑓(𝑡)
x(t+1)=x(t)+f(t)t=0;1;2;3;…
𝑥 𝑡 = 𝑥 0 +3𝑓(𝑖)567
89:
StockFlow
Definedataperiodoftime
Definedatamomentoftime
ContinuousFlow
• Theinfinitesimalamountofflowthatisaddedtostockataninstantaneouslysmallperiodintimecanbewrittenas
dx = f(t)dtContinuous flow and stock are transformed todifferential equation, and the amount of stock at𝑡is obtained by solving the differential equation.
;<;5
=f(t)
𝑥 𝑡 = 𝑥 0 + = 𝑓 𝑢 𝑑𝑢5
:
ConstantFlow𝑓(𝑡) = 𝑎
Discreteinterpretation:𝑥(𝑡 + 1) = 𝑥(𝑡) + 𝑎
𝑥(𝑡) = 𝑥(0) + 𝑎𝑡
Continuousinterpretation𝑑𝑥𝑑𝑡
= 𝑎
𝑥 𝑡 = 𝑥 𝑜 + ∫ 𝑎𝑑𝑢5: = 𝑥(0) + 𝑎𝑡
Stock2000
1500
1000
500
00 12 24 36 48 60 72 84 96 108 120
Time (Month)
custo
mer
s
Stock : Current
Changestakeplaceata“constant”rate
CustomersCustomer
Acquisition Rate
Constant growthper month
Feedback
• Flowbecomesafunctionofstock
BankBalanceInterest Payments
Interest Rate
𝑥 𝑡 + 1 = 𝑥 𝑡 + 𝑓 𝑥 𝑡 , 𝑡 = 0; 1; 2; …𝑑𝑥/𝑑𝑡 = 𝑓(𝑥(𝑡))
PositiveFeedback
Let a>0beinflowfraction𝑓(𝑥) = 𝑎𝑥(𝑡)
Bank Balance2000
1500
1000
500
00 5 10 15 20 25 30 35 40 45 50
Time (Year)
$
Bank Balance : Current
𝑥 𝑡 + 1 = 𝑥 𝑡 + 𝑎𝑥 𝑡 ; 𝑡 = 0; 1; 2; …
𝑑𝑥/𝑑𝑡=𝑎𝑥,𝑎 > 0𝑋(𝑡) = 𝑥(0)𝑒S5,e=2.71…
Note:theinitialvalueofthestock𝑥(0)cannotbezero,sincenon-zeroamountofstockisalwaysneededasaninitialcapitaltolaunchagrowthof flow
anincreaseinflow↑→anincreaseinstock↑→anincreaseinflow↑
self-increasingrelation
Stock2000
1500
1000
500
00 5 10 15 20 25 30 35 40 45 50
Time (Month)
$
Stock : Current
StockFlow
Decay Factor
Self-regulatingorbalancingrelation
NegativeFeedback
𝑑𝑥/𝑑𝑡=−𝑏𝑥,𝑏 > 0
anincreaseinflow↑→adecreaseinstock↓→adecreaseinflow↓
Let𝑏 > 0 beoutflowfraction
AddingConstantFlowsInterestRate=10%Payments=$1
Bank Balance80
40
0
-40
-800 5 10 15 20 25 30 35 40 45 50
Time (Year)
$
Bank Balance : Starting capital = 8Bank Balance : Starting capital = 12Bank Balance : Starting capital = 10
Criticalvalue:𝑥(𝑡 + 1) = 1,1𝑥(𝑡)– 1𝑥∗ = 𝑥(𝑡 + 1) = 𝑥(𝑡)
𝑥 ∗= 1,1𝑥∗ − 1𝑥∗ = 1/0,1 = 10
;<;5=0,1𝑥 − 1𝑑𝑥𝑑𝑡
= 0
0,1𝑥 ∗ −1 = 0,𝑥 ∗= 10
BankBalance
Interest Payments
Interest Rate
Payments
CombiningFeedbackStock
Inflow Outflow
Inflow Fraction Outflow Fraction
Stock400
300
200
100
00 1 2 3 4 5 6 7 8 9 10 11 12
Time (Month)
$
Stock : Outflow Fraction DominateStock : Inflow Fraction DominateStock : Inflow Fraction = Outflow fraction
𝑑𝑥𝑑𝑡
= 𝑎 − 𝑏 𝑥
DynamicalSystems
• Dynamicalsystemisasystemthatevolvesintimeaccordingtoawell-definedunchangingrule.
• Oneofthegoalsofthestudyofdynamicalsystemsistoclassifyandcharacterizethesortsofbehaviorsseeninclassesofdynamicalsystems
TwoTypesofDynamicalSystems
DifferenceEquations• Theoutputofonestepis
usedastheinputforthenext.
• 𝑥5X7 = 𝑓(𝑥5)•
• Y𝑥5X7 = 𝑓(𝑥5)𝑥𝑜
• Timeisdiscrete.
DifferentialEquations
• Z;<;5= 𝑓(𝑥)
𝑥 0 = 𝑥:• Ifthefunctionf(x)is
continuousanddifferentiable,thanthesolutionexists andisunique.
• Thederivative;<;5
istheinstantaneousrateofchange𝑥.
• Timeiscontinues.
fX f(X)
FixedPoints
• Apoint𝑥 isfixedpointifitdoesnotchange(equilibriumpoint)
• Iteratedfunction:𝑓(𝑥 ∗) = 𝑥*• Differentialequations:𝑑𝑥/𝑑𝑡 = 0• Afixedpointisstable ifnearbypointsmoveclosertothefixedpointwhentheyareiterated(attractor)
• Afixedpointisunstableifnearbypointsmovefurtherawayfromthefixedpointwhentheyareiterated(repellor)
LogisticEquation
• PopulationModel• Population(nextyear)=
FunctionofPopulation(thisyear)• 𝑃5X7 = 𝑟𝑃5,• wherer– growthrate
(parameter)
Population2
1.5
1
.5
00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1
Time (Month)
rabb
its
Population : r greater than 1Population : r = 1Population : r less than 1
𝑟 > 1 – populationtendstowardinfinity𝑟 = 1 – populationstaysthesame0 < 𝑟 < 1 – populationapproacheszero
LogisticEquation• 𝑓(𝑝) = 𝑟𝑝(1 − 𝑝/𝐴)• 𝑝`X7 = 𝑟𝑝𝑛(1 − 𝑝𝑛/𝐴)• 𝑟 – growthrate• 𝐴- annihilationpopulation
• If𝑝 = 𝐴,𝑓(𝑝) = 0,• If𝑝<< 𝐴,𝑓(𝑝) ≈ 𝑟𝑝
LogisticEquationinaStandardForm
A
𝑥 – populationexpressedasfractionofannihilationpopulation𝑥`X7 = 𝑟𝑥𝑛(1 − 𝑥𝑛)𝑓(𝑥) = 𝑟𝑥(1 − 𝑥)
Experiment
Fixedpoint=0,333
Experiment
Cycle:0,560,76
Experiment
Cycle:0,8750,3830,8270,501
Experiment
SensitiveDependenceonInitialConditions
• Adynamicalsystemhassensitivedependenceoninitialconditions(SDIC)ifarbitrarilysmalldifferencesininitialconditionseventuallyleadtoarbitrarilylargedifferencesintheorbits.
ButterflyEffect
• Averysmallerrorintheinitialconditiongrowsextremelyrapidly
• Long-termpredictionisimpossible• Adeterministic(rule-based)systemcanbehaveunpredictably
DefinitionofChaos
• Adynamicalsystemischaoticif:• 1.Thedynamicalsystemisdeterministic.• 2.Thesystem’sorbitsarebounded.• 3.Thesystem’sorbitsareaperiodic;i.e.,theyneverrepeat.
• 4.Thesystemhassensitivedependenceoninitialconditions.
DifferentialEqs vs.IteratedFunctions
DifferentialEquations• 𝑑𝑃/𝑑𝑡 =𝑟𝑃(1−𝑃/𝐾)• Timeiscontinuous• Piscontinuous
• Cyclesandchaosarenotpossible
Iteratedfunctions• 𝑝5X7 = 𝑟𝑝5(1 − 𝑝5/𝐴)• Timemovesinjumps• Pmovesinjumps
• Cyclesandchaosarepossible
TheLogisticDifferentialEquation
• 𝑑𝑃/𝑑𝑡 = 𝑟𝑃(1 − 𝑃/𝐾)• risagrowthparameter• Kisthecarryingcapacity
• Equilibrium:• 𝑑𝑃/𝑑𝑡 = 0• 𝑃 = 0; 𝑃 = 𝐾
Population100
75
50
25
00 2 4 6 8 10 12 14 16 18 20
Time (Month)
rabb
its
Population : r greater than 1
0 K
LogisticEquationwithHarvest• 𝑑𝑃/𝑑𝑡 = 𝑟𝑃(1 − 𝑃/𝐾)– ℎ• r- growthparameter• K- carryingcapacity• h- harvestrate
r=3K=100
𝑃7 =dX de6fdg/h�
jd/h
𝑃j =d6 de6fdg/h�
jd/h
Bifurcation:ℎ = 𝑟𝐾/4
ImportantLessonsforManagers
• Sometimespropertiesofcontinuousmodelsarediscontinuous;
• Atbifurcationpointthebehaviorofdynamicalsystemchangessuddenlyandqualitatively;
• Profitoptimizationleadstoacatastrophicvalue;• Thedecisionconcerningtheamountofexploitation(harvesting,taxation,obligatorypayments)shouldbedonenotonthebaseofprescriptiveguidelines,butonthebasisoffeedback.
LogisticMap
Iteration• 𝑥(𝑛+1)=𝑟𝑥𝑛(1−𝑥𝑛)• 𝑓(𝑥)=𝑟𝑥(1−𝑥)• 0 ≤ 𝑥 ≤ 1• 0 ≤ 𝑟 ≤ 4
• Fixedpoints:𝑥 ∗= 𝑓(𝑥 ∗)• 𝑥 ∗= 0• 𝑥 ∗= 1 − 1/𝑟
CobwebPlot
r=0,5
r=2,9
r=3,2
r=3,5
r=3,9
BifurcationDiagram
r
...6692016,4
,
12
1lim
=
=
+-
+
-+
¥®
d
d
nrnr
nrnr
n
Feldman,DavidP.ChaosandFractals:AnElementaryIntroduction.OxfordUniversityPress,2012.
Chaos
• islong-termbehaviorofnonlineardynamicalsystem;
• lookslikearandomfluctuation,butstilloccursincompletelydeterministic,simpledynamicalsystems;
• exhibitssensitivitytoinitialconditions;• occurswhennoperiodictrajectoriesarestable;• isaprevalentphenomenonthatcanbefoundeverywhereinnature,aswellasinsocialreality.
ImportantResultsofModelling
• 1.Smallchangesinparametercanshiftthedynamicalbehaviorofthesystemfromstabletochaotic.
• 2.Achaoticsystembehavesasifitisrandom,notgovernedbyadeterministicrule.Sodeterministicsystemscanproducerandom,unpredictablebehavior.
• 3.Theperiod-doublingroutetochaosisuniversalscenario.Universalitygivesussomereasontobelievethatwecanunderstandcomplexsystemswithsimplemodels.
• Not only in research, but in the world of politics and economics, we would all be better off if more people realized that simple non-linear systems do not necessarily possess simple dynamical properties.
• R. M. May et al., “Simple mathematical models with very complicated dynamics,”
Nature, vol. 261, no. 5560, pp. 459–467, 1976.
Readinglist
• R.M.May,“Simplemathematicalmodelswithverycomplicateddynamics,”Nature,vol.261,no.5560,pp.459–467,1976.(beforeclass)
• KaoruYamaguchi“Stock-FlowFundamentals,DeltaTime(DT)andFeedbackLoop-FromDynamicstoSystemDynamics”,JournalofBusinessAdministration,OsakaSangyoUniversity,Vol.1No.2,March2000(afterclass)
AdditionalReading
• Gleick,James.Chaos:Makinganewscience.RandomHouse,1997.
• Stewart,Ian.DoesGodplaydice?:Thenewmathematicsofchaos.PenguinUK,1997.
• Feldman,DavidP.ChaosandFractals:AnElementaryIntroduction.OxfordUniversityPress,2012.
• JosLeys,EtienneGhys,andAurelien Alvarez,Chaos:AMathematicalAdventurehttp://www.chaos-math.org/en
Thankyouforattention