Supporting Material is available for this work. For more information, follow the link from
the Table of Contents to "Accessing Supporting Material".
Table of Contents
On Mathematical Structures for Systems Archetypes
Rafael E. Bourguet-Diaz, Gloria Pérez-Salazar
Department of Industrial and Systems Engineering,
ITESM Campus Monterrey
Av. Garza Sada 2501 Sur
64849 Monterrey, N.L. Mexico
Phone (83) 8328-4114; Fax (83) 8358-2000 ext. 5479
bourguet@itesm.mx , gloria.perez@itesm.mx
Abstract
A proposal on mathematical structures for systems archetypes is presented. The
mathematical structures are based on systems of differential equations and the
concept of state variable representation. Vensim is used to construct block
diagrams and execute simulations. Finally, benefits are discussed by using both
representations as conventional System Dynamics models and as differential
equation systems.
Key Words: mathematical models, systems archetypes, differential equation
models.
I. Introduction
The main motivations for this work are the search and learning of generic structures that allow
designing policies to improve problematic situations in human organizations. It is known that
growth, decline, goal seeking, and oscillations are consequences of feedback loop dynamics
(Forrester, 1994, 54.) System Dynamics models and differential equations are two effective
representations to express changes of things through time.
System Dynamics uses symbolic and graphical representations as well as computer simulation
models to represent and understand dynamics of a situation. This latter approach is also found in
disciplines in disciplines such as ecology, electrical engineering, chemical engineering, among
others but these disciplines use differential equations as their representation tool (Zill, 1997),
(Lomen and Lovelock, 1999). Thus it seems to be useful to understand how to pass from one
representation to another.
This work has been organized in five sections. The first section is the introduction. The second
deals with concepts of state-variable description. The third is the introduction of the
mathematical representation for system archetypes using the state-variable description. The
fourth section is conclusions. The fifth section presents the references for this work.
II. State-Variable Representation
Some reasons that make natural the use state-variable representation in System Dynamics are:
1. The exact match between the concepts of state variable x(t) and rate of change dx(t)/dt
with the concepts of Stock and Flow, respectively.
2. Itis a general representation that allows us to handle time varying and nonlinear systems.
Its realization and solution can be obtained using the concept of analog-computer
simulation.
4. First-order differential equations of the state-space description are easily and accurately
evaluated on a digital computer.
The state-variable representation is the structure of a dynamic system accomplished by means of
a set of n first-order differential equations (Rohrs, Melsa, and Schultz, 1993, 28.) The state-
variable description of a system is not unique. The system can be described by many different
sets of state variables (Kailath, 1980, 53.) The general state-variable representation is given by:
x’ =f(x,u,7)
where f is a nx1 state nonlinear vector function, x is the nx1 state vector, n represents the
number of states considered in the system, called the order of the system, u is an mx1 input
vector, m is the number of inputs, and ¢ denotes time dependency.
Linear systems are considered a special class of nonlinear systems. Linear systems take the form:
¥1 (1) = a, X,(1) + ayyXQ(O) +--+ Ay, X, (0) + BU)
3 (2) = yX1 (1) + a9Xp (1) +++ + Ay, X,(D) + yu)
X(t) = yx, () + 4,.%) (1) +++ +.4,,X, (1) +5,u(0)
where the n variables x;(t) are the state-variables and u(t) is the exogenous input. Additionally,
an output expression y(t) as function of the states-variables is used to complete the
representation:
WD = Ey) +e,x, + +6,x, (0)
The general linear state-space equations for an n-states, m-inputs, and k-outputs system has the
form (Wiberg, 1971).
x = A(f)x(¢) + But)
y(t) = C()x(t) + Dut)
where x(t) is an n-vector; u(t) is an m-vector; y(t) is a k-vector; A(t) is an nxn matrix; B(t)
isan mxm matrix; C(t) isan kxn matrix, and D(t) isa kxm matrix.
The state of a physical object is any property of the object, which relates input to output such that
knowledge of the input time function for ¢ 2 ¢, and state at time ¢ = 4, completely determines a
unique output for ¢> 1, (Wiberg, 1971). A state can be seen as the answer to the question: given
{u(t), t > ty} and the mathematical relationships of the abstract object, what additional
information is needed to completely specify {y(0), t2 ty}?
The vector solution x(t) describes the trajectories on time of the internal variables. Elements of
x(t) are the integrator outputs in any realization. It is clear that if the values of the integrator
outputs are known for any given time f=, and the inputs u(t) for ¢2 7, also known then al/
present and future values of the outputs y(¢), and integrator outputs (indeed, of any signal
anywhere in the simulation) can be calculated.
The advantage of this state-space description is that there is no need to know the past of all the
system to establish the present and future behavior. x(¢,) provides a sufficient statistical
information to calculate the future {r>1,} response to a new input {u(),¢>z,} without
worrying about fu(), <1}. In this sense, x(¢)) is a minimal sufficient statistics (Kailath,
1980.) Therefore, it is natural to call the integrator outputs at any time ¢ the state of the system
realization. This interpretation is not restricted to analog-computer realizations but also applies to
any set of state-space equations, no matter how they are obtained -as realization of differential
equations or as description of physical systems (Kailath, 1980, 63.)
Now the state variable concept is applied to provide a mathematical structure for the systems
archetypes described by (Senge, 1990) in Appendix 2: Systems Archetypes.
III. Mathematical structures for Systems Archetypes
A, Eroding Goals
"A shifting the burden type of structure in which the short-term solution involves letting a long-
term, fundamental goal decline" (Senge, 1990, 383.)
XI’
XI Pressures to
Adjust Goal
Goal
ap
Condition
Actions to Improve
x2 Conditions
X2'
Fig. 1 Block diagram for Eroding Goal Archetype
Eroding Goals
100
loo
50
50
x,(0) = xp and x,(0) = Xp
,
;
For this example, 7;= 5 and 7) = 10 are the ° et St Fe inca)
corresponding time constants of the | coa:curen
balanced loops. 7, >7; to represent the
delay included as definition in Bz by the
original archetype in (Senge, 1990, 383.)
x19 = 100 and x29 = 40.
(Conditioe : Cugeent
(01) Actions to Improve Conditions = Gap/10
(02) Condition= INTEG (Actions to Improve Conditions, 40)
(03) FINAL TIME = 12
The final time for the simulation.
(04) Gap = Goal-Condition
(05) Goal= INTEG (-Pressures to Adjust Goal, 100)
(06) INITIAL TIME =0
The initial time for the simulation.
(07) Pressures to Adjust Goal = Gap/5S
(08) SAVEPER = TIME STEP
The frequency with which output is stored.
(09) TIME STEP =0.125
The time step for the simulation.
B. Escalation
"Two people or organizations each see their welfare as depending on a relative advantage over
the other. Whenever one side gets ahead, the other is more threatened, leading it to act more
aggressively to reestablish its advantage, which threatens the first, increasing its aggressiveness,
and so on. Often each side sees its own aggressive behavior as a defensive response to the other's
aggression; but each side acting "in defense" results in a buildup that goes far beyond either
side's desires" (Senge, 1990, 384.)
XI x2
XI’ x2!
CS A's Results Bs Results <?-54—}
Activity
by A
Ds
Rela
Fig. 2 Block diagram for Escalation archetype
Escalation
w=, aif) a
sien) | A Seer
x,(0) = x1) and x,(0) = x5 o 6 WW oF 30 36 42 48 S4 60
Time (Day)
Where R, and R; refer to the desired a
relations x//x2 by A and B, respectively.
Parameters a and b are the x,'s rates of the
description. For the example, R, = 2/1,
and R, =1/l,a = 1, b=1, x19 = 40, and x2
=10.
(01) A's Results = INTEG (Activity by A, 40)
(02) Activity by A = 1*((2/1)-Results of A Relative to B)
(03) Activity by B = 1*(Results of A Relative to B - (1/1))
(04) B's Results = INTEG (Activity by B, 10)
(05) FINAL TIME =60
The final time for the simulation.
(06) INITIAL TIME =0
The initial time for the simulation.
(07) Results of A Relative to B = A's Results/B's Results
(08) SAVEPER = TIME STEP
The frequency with which output is stored.
(09) TIMESTEP =1
The time step for the simulation.
C. Fixes that Fail
"A fix, effective in the short term, has unforeseen long-term consequences which may require
even more use of the same fix" (Senge, 1990, 388.)
XI
Problem
Long-term
Consequences
a An) ay
Unintended
Consequences Rate of Change
X2 X2'
Fig. 3 Block diagram for Fixes that Fails archetype
Fixes that Fails
x{(1)= ax, (1)—bx, (0) i
100
x5 (t) = cx, (t -d) -
x,(0) = x, and x,(0) = x4, a0
0
Where d is the delay in time units, and a, b 0
, ¢ are proportionality parameters. In this a
example, a = 0.5, b = 0.5, ¢ = 0.40, d = 5, | rroviem:curen
X19 = 50, and x29 = 0. "Long-term Consequences"; Caren
(01) Capacity= INTEG (Investment and Capacity, 80)
(02) Demand= INTEG (Net Rate of Change, 1)
(03) FINAL TIME = 12
The final time for the simulation.
(04) Growing Action= 1/100*Demand
(05) Growth and Underinvestment = Performance Standard-Performance
(06) INITIAL TIME =0
The initial time for the simulation.
(07) Investment and Capacity = 0.01*Growth and Underinvestment
(08) Net Rate of Change = 0.75*Growing Action*DELAY3(Performance, | )
(09) Performance = Capacity-Demand
(10) Performance Standard = 100
(11) SAVEPER = TIME STEP
The frequency with which output is stored.
(12) TIME STEP =0.125
The time step for the simulation.
D. Growth and Underinvestment
"Growth approaches a limit which can be eliminated or pushed into the future if the firm, or
individual, invest in additional "capacity." But the investment must be aggressive and
sufficiently rapid to forestall reduced growth, or else it will never get made. Oftentimes, key
goals or performance standards are lowered to justify underinvestment" (Senge, 1990, 389-390.)
/™ x
[Net Rate of Change
Growing Ar)
Action
Au Demand Gp
\ Performance
9 Standard
X2| Capacity 4») ‘.. y
Underinvestment
+
Investment and
Capacity
X2'
Fig. 4 Block diagram for Growth and Underinvestment archetype
Growth and Underinvestment
(= 2 (fs 6-d)-» 4)
xh(t)=c[Ps+x,(1)—x,()] Fy
x,(0) = xp and x;(0) = Xp
Where Ps is the performance standard, and . re
a, b and, c are model parameters. For this | — pemana: current
example, Capacity : Curent
Ps = 100, a=0.75, b = 100, ¢ = 0.01, x19 =
J, and x29 = 80.
(01) Capacity = INTEG (Investment and Capacity, 80)
(02) Demand= INTEG (Net Rate of Change, 1)
(03) FINAL TIME = 12
The final time for the simulation.
(04) Growing Action = 1/100*Demand
(05) Growth and Underinvestment = Performance Standard-Performance
(06) INITIAL TIME =0
The initial time for the simulation.
(07) Investment and Capacity = 0.01*Growth and Underinvestment
(08) Net Rate of Change = 0.75*Growing Action*DELAY3(Performance, | )
(09) Performance = Capacity-Demand
(10) Performance Standard = 100
(11) SAVEPER = TIME STEP
The frequency with which output is stored.
(12) TIME STEP =0.125
The time step for the simulation.
E. Limits to Growth - Model 1
"A process feeds on itself to produce a period of accelerating growth or expansion. Then the
growth begins to slow (often inexplicably to the participants in the system) and eventually comes
to a halt, and may even reverse itself and begin an accelerating collapse.
"The growth phase is caused by a reinforcing feedback process (or by several reinforcing
feedback processes.) The slowing arises due to a balancing process brought into play as a "limit"
is approached. The limit can be a resource constraint, or an external or internal response to
growth. The accelerating collapse (when it occur) arises from the reinforcing process operating
in reverse, to generate more and more contraction" (Senge, 1990, 379.)
Limiting Condition
x 4
Condition
Growing Slowing
Action Ae) G} Action
+
ea
Fig. 5 Block diagram for Limits to Growth archetype
x'(t)= ax(r)- a(t 420)
reordering
thus
x)= {1-D\x
x(0) = xy
Where L is the limit of growth and a is the
maximum fractional growth. For this
example,
L=100, a = 0.1, and xo = 1.
Limits to Growth 1
0 Tt
0 0 2
Condition : Current:
30
40
30
Time (Day)
60
Too
(1)
Condition = INTEG (Growing Action-Slowing Action, 1)
Slowing Action = 0.1*Condition*(1-((Limiting Condition-Condition)/Limiting
(2) FINAL TIME = 100
The final time for the simulation.
(3) | Growing Action = 0.1*Condition
(4) INITIAL TIME =0
The initial time for the simulation.
(5) — Limiting Condition = 100
(6) SAVEPER = TIME STEP
The frequency with which output is stored.
7)
Condition))
(8) TIMESTEP =0.125
The time step for the simulation.
Limits to Growth - Model 2
Limit
XI X2' X2
Slowing Growing
Level Level
Rate of
Ry
Fig. 6 Block diagram for Limits to Growth archetype - model 2
since
x,()=L-x,(0)
then by reordering
; %
x) (1) = di - ae (t)
X,(0) = 9 and x,(0) = xy
Where L is the limit of growth and a is the
maximum fractional growth. For this
example,
L = 100, a = 0.1, X19 = L-1, and x29 = 1.
Limits to Growth 2
o 0 2 3 49 30 60 70 80 90
Time (Day)
Too
Growing Level : Current
Rate of Change : Current
(1) FINAL TIME = 100
The final time for the simulation.
(2) Growing Level=INTEG(__ Rate of Change, 1)
(3) INITIAL TIME =0
The initial time for the simulation.
(4) Limit = 100
(5) Rate of Change = (0.1/Limit)*Growing Level*Slowing Level
(6) SAVEPER = TIME STEP
The frequency with which output is stored.
(7) Slowing Level = INTEG (-Rate of Change, 100-1)
(8) TIMESTEP =0.125
The time step for the simulation.
Limits to Growth - Model 3
ae)
+
Limiting Condition
x’
Net Rate S*
ts)
of Change \
Cx
Gap
Fraction
Condition
xX
Fig. 7 Block diagram for Limits to Growth archetype - model 3
reordering
x(0) = xy
Where L is the limit of growth and a is the
maximum fractional growth. For this
example,
L=100,a = 0.1, and xo = 1.
Limits to Growth 3
o 0 2 3 4 50 60 70 80 00
Time (Day)
Condition : Curr
Net Rate of Chan;
Current
Too
(1) Condition = INTEG (Net Rate of Change, 1)
(2) FINAL TIME = 100
The final time for the simulation.
(3) — Gap Fraction = (Limiting Condition-Condition)/Limiting Condition
(4) INITIAL TIME =0
The initial time for the simulation.
(5) — Limiting Condition = 100
(6) Net Rate of Change = 0.1*Condition*Gap Fraction
(7) SAVEPER = TIME STEP
The frequency with which output is stored.
(8) TIMESTEP =0.125
The time step for the simulation.
F. Shifting the Burden
"A short term "solution" is used to correct a problem, with seemingly positive immediate results.
As this correction is used more and more, more fundamental long-term corrective measures are
used less and less. Over time, the capabilities for the fundamental solution may atrophy or
become disabled, leading to even greater reliance on the symptomatic solution" (Senge, 1990,
381.)
Symptomatic
"Solution"
+ X2'
XI Rate of Change
Problem R
Symptom x
Side Effect
Fundamental ye
Solution
z,
Fig. 8 Block diagram for Shifting The Burden archetype
x,
(1) = —ax, ()- b “OX, Ok, ()
reordering, the system is:
x()=-(a+b)x,()+4, (Ox)
x4(t) = adx,(¢)
%,(0) = x) and x,(0) = X49
For this example, a = 0.10, b = 0.05, c =
1.0, d=0.01, X10 = 20, and x2 = 0.
Shifting the Burden
02 4 6 8 0 1 14 16 18 20 22 24 26 28 30
Time (Day)
Problem Symptom : Cusren————————__________
"Symptomatic \"Solution\" : Curren
Fundamental Solution : Curgene
(01) FINAL TIME =20
The final time for the simulation.
(02) Fundamental Solution = (0.05-1*Side Effect)*Problem Symptom
(03) INITIAL TIME =0
The initial time for the simulation.
(04) Problem Symptom = INTEG (-Fundamental Solution-"Symptomatic
\"Solution\"", 20)
(05) Rate of Change = 0.01*"Symptomatic \"Solution\""
(06) SAVEPER = TIME STEP
The frequency with which output is stored.
(07) Side Effect = INTEG (Rate of Change, 0)
(08) "Symptomatic \"Solution\"" = 0.1*Problem Symptom
(09) TIME STEP =0.125
The time step for the simulation.
G. Success to the Successful
"Two activities compete for limited support or resources. The more successful one becomes, the
more support it gains, thereby starving the other" (Senge, 1990, 385.)
XI XI'
Resources to A
C,
Allocation to
A instead of B
a>
Success of B =
Success of A
op
Resouces to B
x2 x2"
Fig. 9 Block diagram for Success To The Successful archetype
Success to the Successfull
xi(1)= ax,())—ax,() 10
x5 (t) = bx; (0) + dx, (1)
¥,(0) =X) and x,(0) = xy :
For this example, a = 0.1, b = 0.1, x10 =| 0
5.5, and o1 23 4s 6 7 8 9 10 2
to = 45 Time (Month)
‘Sueeess of A: Curgent
‘Sueeess of B Cuprent
(01) Allocation to A instead of B = Success of A-Success of B
(02) FINAL TIME = 12
The final time for the simulation.
(03) INITIAL TIME =0
The initial time for the simulation.
(04) Resouces to B = -0.1*Allocation to A instead of B
(05) Resources to A = 0.1*Allocation to A instead of B
(06) SAVEPER = TIME STEP
The frequency with which output is stored.
(07) Success of A= INTEG (Resources to A, 5.5)
(08) Success of B = INTEG (Resouces to B, 4.5)
(09) TIME STEP =0.125
The time step for the simulation.
H. Tragedy of the Commons
"Individuals use a commonly available but limited resource solely on the basis of individual
need. At first they are rewarded for using it; eventually, they get diminishing returns, which
causes them to intensify their efforts. Eventually, the resource is either significantly depleted,
eroded, or entirely used up" (Senge, 1990, 387.)
Gain Rate for A
x1
.
Net Gains for
Individual A's A.
Activity
RI
of Ari) G) Resource
Limit
+ x3" Gain per
sous HES # tee Individual | X3
Activity
Total Activity
te)
Net Gains for
B
‘
x2
a forB
Fig. 10 Block diagram for Tragedy of the Commons archetype
‘Tragedy of the Commons
H=(22-5},0 :
30
x4 ()= [2-« @)
c
xi(t)=x,(t-e)+x,(t-e) 3
¥4(0) =x ps ¥4(0) = aq, and x(0) =X
s 6 7 8 9 1 MW 2
Time (Month)
‘Net Gains for A : Current
Net Gains for B : Current
For this example, a = 1000*12, b = 0.4, ¢
= 1000*12, d = 0.4, e = 5, x10 = 20, x =
15, and x39 = 100.
(01) FINAL TIME = 12
The final time for the simulation.
(02) Gain per Individual Activity = INTEG (-Total Activity, Resource Limit)
(03) Gain Rate for A = Gain per Individual Activity/1000*12-0.4
(04) Gain Rate for B = Gain per Individual Activity/1000*12-0.4
(05) Individual A's Activity = Gain Rate for A*Net Gains for A
(06) Individual B's Activity = Net Gains for B*Gain Rate for B
(07) INITIAL TIME =0
The initial time for the simulation.
(08) Net Gains for A = INTEG (Individual A's Activity, 20)
(09) Net Gains for B= INTEG(_ Individual B's Activity, 15)
(10) Resource Limit = 100
(11) SAVEPER = TIME STEP
The frequency with which output is stored.
(12) TIME STEP =0.0078125
The time step for the simulation.
(13) Total Activity = IF THEN ELSE( Gain per Individual Activity>0,
DELAY 3 (Individual A's Activity +Individual B's Activity, 5), 0)
IV. Conclusions
Proposed mathematical structures for systemic archetypes has been presented. The mathematical
structures are based on systems of differential equations and the concept of state variable
representation.
Formal models of dynamic systems represented by differential equations have proved to be
effective elements to transmit knowledge, ideas, and experiences among different disciplines of
science.
System Dynamics as a modeling method that enhances learning of complex systems by using
feedback principles and computer simulation models have the great potential to become a
conventional provider of formal models for the social sciences. The following and natural step
seems to be to represent mathematically System Dynamics models.
Then a wider channel of communication may be established among other disciplines by sharing
System Dynamics models in the form of differential equations, and taking advantages in the
same way of those models developed and considered as deep knowledge by other areas.
So the goals of System Dynamics would not change at all. As it is known, learning and policy
designs is what SD is about. However, the scope of the implementation stage would widen with
the availability of a second representation of the results of the modeling process. The benefits of
having this second standard communication mechanism might be substantial.
V. References
Forrester, J. W. 1994. Policies, Decisions, and Information Sources for Modeling. In John D. W.
Morecroft and John D. Sterman (editors). Modeling for Learning Organizations. Portland,
OR: Productivity Press, System Dynamics Series, 51-84.
Kailath, T. 1980. Linear Systems. Prentice-Hall.
Lomen, D. and Lovelock, D. 1999. Differential Equations: Graphs, Models, Data. John Wiley
and Sons, Inc.
Rohrs, Ch. E., Melsa, J. L., and Schultz, D. G. 1993. Linear Control Systems. McGraw-Hill, Inc.
Senge, P. M. 1990. The Fifth Discipline: The Art and Practice of the Learning Organizations.
NY: Doubleday/Currency.
Wiberg, D. M. 1971. Theory and Problems of State Space and Linear System. Schaum’s Outline
Series. McGraw-Hill, Inc.
Zill, D. G. 1997. A First Course in Differential Equations. 5" edition. Brooks/Cole Publishing.
CC BY-NC-SA 4.0